-------------------------Indexing Terms------------------------- 
REPORTNUM:   GAO-04-96		

TITLE:     Energy Markets: Effects of Mergers and Market Concentration 
in the U.S. Petroleum Industry

DATE:   05/27/2004 
				                                                                         
----------------------------------------------------------------- 

This is the accessible text file for GAO report number GAO-04-96 
entitled 'Energy Markets: Effects of Mergers and Market Concentration 
in the U.S. Petroleum Industry' which was released on May 27, 2004.

This text file was formatted by the U.S. General Accounting Office 
(GAO) to be accessible to users with visual impairments, as part of a 
longer term project to improve GAO products' accessibility. Every 
attempt has been made to maintain the structural and data integrity of 
the original printed product. Accessibility features, such as text 
descriptions of tables, consecutively numbered footnotes placed at the 
end of the file, and the text of agency comment letters, are provided 
but may not exactly duplicate the presentation or format of the printed 
version. The portable document format (PDF) file is an exact electronic 
replica of the printed version. We welcome your feedback. Please E-mail 
your comments regarding the contents or accessibility features of this 
document to [email protected].

This is a work of the U.S. government and is not subject to copyright 
protection in the United States. It may be reproduced and distributed 
in its entirety without further permission from GAO. Because this work 
may contain copyrighted images or other material, permission from the 
copyright holder may be necessary if you wish to reproduce this 
material separately.

Report to the Ranking Minority Member, Permanent Subcommittee on 
Investigations, Committee on Governmental Affairs, U.S. Senate: 

May 2004: 

ENERGY MARKETS: 

Effects of Mergers and Market Concentration in the U.S. Petroleum 
Industry: 

[Hyperlink, http://www.gao.gov/cgi-bin/getrpt?GAO-04-96]: 

GAO Highlights: 

Highlights of GAO-04-96, a report to the Ranking Minority Member, 
Permanent Subcommittee on Investigations, Committee on Governmental 
Affairs, U.S. Senate

Why GAO Did This Study: 

Starting in the mid-1990s, the U.S. petroleum industry experienced a 
wave of mergers, acquisitions, and joint ventures, several of them 
between large oil companies that had previously competed with each 
other. For example, as shown in the figure, Exxon, the largest U.S. oil 
company, acquired Mobil, the second largest, thus forming ExxonMobil. 
 
GAO was asked to examine the effects of the mergers on the U.S. 
petroleum industry since the 1990s. For this period, GAO examined (1) 
mergers in the U.S. petroleum industry and why they occurred, (2) the 
extent to which market concentration (the distribution of market shares 
among competing firms) and other aspects of market structure in the 
U.S. petroleum industry have changed as a result of mergers, (3) major 
changes that have occurred in U.S. gasoline marketing, and (4) how 
mergers and market concentration in the U.S. petroleum industry have 
affected U.S. gasoline prices at the wholesale level. Commenting on a 
draft of GAOï¿½s report, FTC asserted that the models were flawed and the 
analyses unreliable. GAO used state-of-the-art econometric models to 
examine the effects of mergers and market concentration on wholesale 
gasoline prices. The models used in GAOï¿½s analyses were peer reviewed 
by independent experts. Thus, GAO believes its analyses are sound.

What GAO Found: 

Over 2,600 mergers have occurred in the U.S. petroleum industry since 
the 1990s. The majority occurred later in the period, most frequently 
among firms involved in exploration and production. Industry officials 
cited various reasons for the mergers, particularly the need for 
increased efficiency and cost savings. Economic literature also 
suggests that firms sometimes merge to enhance their ability to control 
prices. 

Market concentration has increased substantially in the industry, 
partly because of these mergers. Concentrated markets can enable firms 
to raise prices above competitive levels but can also lead to cost 
savings and lower prices. Evidence suggests mergers also have changed 
other factors that affect competition, such as the ability of new firms 
to enter the market.

According to industry officials, two major changes have occurred in 
U.S. gasoline marketing related to these mergers. First, the 
availability of generic gasoline, which is generally priced lower than 
branded gasoline, has decreased substantially. Second, refiners now 
prefer to deal with large distributors and retailers, which has 
motivated further consolidation in distributor and retail markets.

GAOï¿½s econometric analyses indicate that mergers and increased market 
concentration generally led to higher wholesale gasoline prices in the 
United States from the mid-1990s through 2000. Six of the eight mergers 
GAO modeled led to price increases, averaging about 1 cent to 2 cents 
per gallon. GAO found that increased market concentration, which 
reflects the cumulative effects of mergers and other competitive 
factors, also led to increased prices. For conventional gasoline, the 
predominant type used in the country, the change in wholesale price due 
to increased market concentration ranged from a decrease of about 1 
cent per gallon to an increase of about 5 cents per gallon. For 
boutique fuels sold in the East Coast and Gulf Coast regions, wholesale 
prices increased by about 1 cent per gallon, while prices for boutique 
fuels sold in California increased by over 7 cents per gallon. 

Selected Recent Major Petroleum Mergers: 

[See PDF for image]

[End of figure]

www.gao.gov/cgi-bin/getrpt?GAO-04-96

To view the full product, including the scope and methodology, click on 
the link above. For more information, contact Jim Wells at (202) 
512-3841 or [email protected].

[End of section]

Contents: 

Letter: 

Executive Summary: 

Purpose: 

Background: 

Results in Brief: 

Principal Findings: 

Recommendations for Executive Action: 

Agency Comments and GAO's Evaluation: 

Chapter 1: Introduction: 

The Petroleum Industry Consists of Three Main Segments: 

Different Entities Have Historically Exerted Influence over the World 
Petroleum Market: 

FTC and DOJ Review Proposed Mergers to Preserve Market Competition: 

Objectives, Scope, and Methodology: 

Chapter 2: All Segments of the Petroleum Industry Experienced Mergers 
for Several Reasons: 

Mergers Occurred in All Three Segments, but Most Frequently in the 
Upstream: 

Several Reasons Were Cited for Mergers in the Petroleum Industry: 

Chapter 3: Mergers Contributed to Increases in Market Concentration and 
Other Changes in Market Structure: 

Market Concentration Increased Mostly in the Downstream Segment of the 
Petroleum Industry During the 1990s: 

Mergers Have Caused Changes in Other Aspects of Market Structure, but 
the Extent of These Changes Is Not Easily Quantifiable: 

Chapter 4: Gasoline Marketing Has Changed in Two Major Ways: 

The Availability of Unbranded Gasoline Decreased: 

Refiners Prefer Dealing with Large Distributors and Retailers: 

Chapter 5: Mergers and Increased Market Concentration Generally Led to 
Higher Wholesale Gasoline Prices in the United States: 

Econometric Models Developed to Estimate the Effects of Mergers and 
Market Concentration on Wholesale Gasoline Prices: 

Mergers in the Second Half of the 1990s Mostly Led to Increases in 
Wholesale Gasoline Prices: 

Increased Market Concentration Generally Led to Higher Wholesale 
Gasoline Prices: 

Other Factors Also Resulted in Higher Wholesale Gasoline Prices: 

Our Findings Are Generally Consistent with Previous Studies' Empirical 
Results: 

Agency Comments and Our Evaluation: 

Appendixes: 

Appendix I: Companies, Agencies, and Organizations Contacted by GAO: 

Appendix II: Experts Who Reviewed GAO's Econometric Models: 

Appendix III: Correlation Analysis of Mergers and Market Concentration 
in the U.S. Petroleum Industry: 

Wholesale Gasoline Market Concentration by State: 

Correlation Analysis of Mergers and Market Concentration: 

Appendix IV: Econometric Analyses of the Effects of Specific Mergers 
and Market Concentration on U.S. Wholesale Gasoline Prices: 

GAO's Econometric Models of Wholesale Gasoline Prices Built on Previous 
Studies and Market Analysis: 

Data Sources and Sample Selection: 

Specification of Econometric Models and Estimation Methodology: 

Econometric Results: 

Our Econometric Methodology Had Some Limitations: 

Appendix V: Comments from the Federal Trade Commission's Commissioners: 

GAO's Comments: 

Appendix VI: Comments from the Federal Trade Commission's Bureau of 
Economics Staff: 

GAO's Comments: 

Appendix VII: GAO Contacts and Staff Acknowledgments: 

GAO Contacts: 

Acknowledgments: 

Bibliography: 

Tables: 

Table 1: FTC/DOJ Horizontal Merger Guidelines on the General Standards 
for Evaluating Postmerger Market Concentration: 

Table 2: Selected Vertical Mergers in the Petroleum Industry Since the 
1990s: 

Table 3: Types of Wholesale Prices Paid for Gasoline: 

Table 4: Selected Oil Industry Mergers Affecting Wholesale Gasoline 
Markets, 1994-2000: 

Table 5: Estimated Changes in Conventional Wholesale Gasoline Prices 
Associated with Individual Mergers (1994-2000): 

Table 6: Estimated Changes in Reformulated Wholesale Gasoline Prices 
Associated with Individual Mergers (1995-2000): 

Table 7: Estimated Changes in CARB Reformulated Wholesale Gasoline 
Prices Associated with Individual Mergers (1996-2000): 

Table 8: Estimated Changes in Conventional Wholesale Gasoline Prices 
Associated with Increased Market Concentration (1994-2000): 

Table 9: Estimated Changes in Boutique Fuels Wholesale Prices Associated 
with Increased Market Concentration (1995-2000): 

Table 10: State-level HHI for Wholesale Gasoline (1994-2002): 

Table 11: Correlation between the Average Transaction Value of Mergers 
and Market Concentration (HHI) for Petroleum Refining by PADD (1991-
2000): 

Table 12: Correlation between the Average Transaction Value of Mergers 
and Market Concentration (HHI) for Wholesale Gasoline (1994-2001): 

Table 13: Expected Effects of Key Explanatory Variables on Wholesale 
Gasoline Prices: 

Table 14: Variables in Our Econometric Analysis of Wholesale Gasoline 
Prices: 

Table 15: Effects of Mergers on Conventional Wholesale Gasoline Prices 
(1994-2000): 

Table 16: Effects of Mergers on Reformulated Wholesale Gasoline Prices 
(1995-2000): 

Table 17: Effects of Mergers on CARB Wholesale Gasoline Prices (1996-
2000): 

Table 18: Effects of Market Concentration on Conventional Wholesale 
Gasoline Prices (1994-2000): 

Table 19: Effects of Market Concentration on Wholesale Prices of 
Boutique Fuels (1995-2000): 

Table 20: Selected Summary Statistics for Conventional Wholesale 
Gasoline Markets: 

Table 21: Econometric Estimates of Mergers' Effects on Conventional 
Wholesale Gasoline Prices: 

Table 22: Econometric Estimates of Mergers' Effects on Reformulated 
Wholesale Gasoline Prices: 

Table 23: Econometric Estimates of Mergers' Effects on CARB Wholesale 
Gasoline Prices: 

Table 24: Econometric Estimates of Market Concentration on Conventional 
Wholesale Gasoline Prices: 

Table 25: Econometric Estimates of Market Concentration on Conventional 
Wholesale Gasoline Prices: Eastern Region (PADDs I-III): 

Table 26: Econometric Estimates of Market Concentration on Conventional 
Wholesale Gasoline Prices: Western Region (PADDs IV-V): 

Table 27: Econometric Estimates of Market Concentration on Reformulated 
Wholesale Gasoline Prices: 

Table 28: Econometric Estimates of Market Concentration on CARB 
Wholesale Gasoline Prices: 

Figures: 

Figure 1: U.S. Petroleum Industry Chain: 

Figure 2: Product Yield from a Barrel of Crude Oil, 2000: 

Figure 3: Major Events in the World Petroleum Market: 

Figure 4: Shares of the World's Conventional Crude Oil Reserves 
(February 2003): 

Figure 5: World's Estimated Excess Production Capacity (February 2003): 

Figure 6: Percentage of Mergers That Occurred in Each Segment of the 
Petroleum Industry (1991-2000): 

Figure 7: Petroleum Industry Merger Trends (1991-2000): 

Figure 8: Selected Major Petroleum Mergers (1996-2002): 

Figure 9: Percentage of Merger Transactions within the Downstream 
Segment by Type of Key Assets Acquired: 

Figure 10: Range of Reported Merger Transaction Values (1991-2000): 

Figure 11: Petroleum Administration for Defense Districts: 

Figure 12: Market Concentration for the Upstream Segment, as Measured 
by the HHI (1990-2000): 

Figure 13: Refining Market Concentration for PADD I Based on Crude Oil 
Distillation Capacity (1990-2000): 

Figure 14: Refining Market Concentration for PADD II Based on Crude Oil 
Distillation Capacity (1990-2000): 

Figure 15: Refining Market Concentration for PADD III Based on Crude 
Oil Distillation Capacity (1990-2000): 

Figure 16: Refining Market Concentration for PADD IV Based on Crude Oil 
Distillation Capacity (1990-2000): 

Figure 17: Refining Market Concentration for PADD V Based on Crude Oil 
Distillation Capacity (1990-2000): 

Figure 18: Percentage of U.S. States with Unconcentrated, Moderately 
Concentrated, and Highly Concentrated Wholesale Gasoline Markets (1994, 
2000, and 2002): 

Figure 19: Wholesale Gasoline Market Concentration by State in Each 
PADD (1994 and 2002): 

Figure 20: The Flow of Gasoline Marketing: 

Figure 21: Percentage Volume of Gasoline Sold through Different 
Marketing Channels: 

Figure 22: Normalized Inventories and Expected Demand for Wholesale 
Gasoline (1994-2000): 

Figure 23: Ratio of Inventories to Expected Demand for Wholesale 
Gasoline (1994-2000): 

Abbreviations: 

API: American Petroleum Institute: 

BP: British Petroleum: 

CARB: California Air Resources Board: 

cpg: cents per gallon: 

DOE: Department of Energy: 

DOJ: Department of Justice: 

DTW: Dealer-tankwagon: 

EAI: Energy Analysts International, Inc.

EIA: Energy Information Administration: 

FERC: Federal Energy Regulatory Commission: 

FRS: Financial Reporting System: 

FTC: Federal Trade Commission: 

HHI: Herfindahl-Hirschman Index: 

MAP: Marathon Ashland Petroleum: 

mmb/d: million barrels per day: 

MTBE: methyl tertiary butyl ether: 

OGJ: Oil and Gas Journal: 

OPEC: Organization of Petroleum Exporting Countries: 

OPIS: Oil Price Information Service: 

PADD: Petroleum Administration for Defense Districts: 

PMAA: Petroleum Marketers Association of America: 

RFG: reformulated gasoline: 

UDS: Ultramar Diamond Shamrock: 

WTI: West Texas Intermediate: 

Letter May 17, 2004: 

The Honorable Carl Levin: 
Ranking Minority Member:
Permanent Subcommittee on Investigations: 
Committee on Governmental Affairs:
United States Senate: 

Dear Senator Levin: 

This report responds to your request that we examine the effect of the 
wave of mergers that occurred in the U.S. petroleum industry in the 
1990s. The report examines the segments of the petroleum industry that 
were involved in mergers, the extent to which market concentration and 
other aspects of market structure that affect competition have changed 
in the U.S. petroleum industry because of mergers, and major changes 
that have occurred in U.S. gasoline marketing because of mergers. 
Finally, the report estimates the effects of mergers and market 
concentration on U.S. gasoline prices at the wholesale level.

As agreed with your office, unless you publicly announce its contents 
earlier, we plan no further distribution of this report until 30 days 
after the date of this letter. At that time, we will send copies to 
appropriate congressional committees, the Chairman of the Federal Trade 
Commission, the Secretary of Energy, the Attorney General, and other 
interested parties.

Please contact me at (202) 512-3841 if you or your staff have any 
questions. Major contributors to this report are listed in appendix 
VII.

Sincerely yours,

Signed by: 

Jim Wells: 
Director, Natural Resources and Environment: 

[End of section]

Executive Summary: 

Purpose: 

Since the 1990s, the U.S. petroleum industry has experienced a wave of 
mergers, acquisitions, and joint ventures (hereafter referred to as 
mergers), several of them between large oil companies that had 
previously competed with each other for the sale of petroleum products. 
For example, in 1998 British Petroleum (BP) and Amoco merged to form 
BP-Amoco, which later acquired ARCO in 2000. In 1999, Exxon, the 
largest U.S. oil company, acquired Mobil, the second largest, thus 
forming ExxonMobil. Increasing concerns about potential 
anticompetitive effects have caused some policy makers and consumer 
groups to suggest that these mergers may have reduced competition in 
the United States and ultimately led to higher gasoline prices.

In this context, the Ranking Minority Member, Permanent Subcommittee on 
Investigations, Senate Committee on Governmental Affairs, asked GAO to 
examine the effect of the mergers that have occurred in the U.S. 
petroleum industry since the 1990s. GAO examined (1) mergers in the 
U.S. petroleum industry from the 1990s through 2000 and why they 
occurred, (2) the extent to which market concentration (the 
distribution of market shares among competing firms within a market) 
and other aspects of market structure in the U.S. petroleum industry 
have changed as a result of mergers, (3) major changes that have 
occurred in U.S. gasoline marketing since the 1990s, and (4) how 
mergers and market concentration in the U.S. petroleum industry have 
affected U.S. gasoline prices at the wholesale level.

To address these issues, GAO purchased and analyzed a large body of 
data on mergers and wholesale gasoline prices, as well as data on other 
relevant economic factors. GAO also developed econometric models for 
examining the effects of eight specific mergers and increased market 
concentration on U.S. gasoline wholesale prices. In doing so, GAO 
isolated the effects of mergers and market concentration from several 
other factors that could influence wholesale gasoline prices, such as 
crude oil costs, gasoline inventories relative to demand, refinery 
capacity utilization rates, and gasoline supply disruptions. GAO also 
differentiated among fuel formulations in its analyses. Other factors-
-including taxes--can affect the retail gasoline prices that consumers 
ultimately pay, but GAO did not examine the effects of such factors 
because this study focuses on wholesale gasoline prices.

In the course of its work, GAO consulted with Dr. Severin 
Borenstein,[Footnote 1] a recognized expert in the modeling of gasoline 
markets; interviewed officials across the industry spectrum; and 
reviewed relevant economic literature and numerous related studies. 
Furthermore, GAO used an extensive peer review process to obtain 
comments from experts in academia and relevant government agencies.

Background: 

The U.S. petroleum industry consists of many firms of varying sizes 
that operate in one or more of three broad segments--the upstream, 
which consists of exploration for and production of crude oil and 
natural gas; the midstream, which consists of pipelines and other 
infrastructure used to transport these products; and the downstream, 
which consists of refining crude oil and marketing petroleum products 
such as gasoline and heating oil. While some firms engage in only one 
or two of these activities, fully vertically integrated oil companies 
participate in all of them. Before the 1970s, major oil companies that 
were fully vertically integrated controlled the global network for 
supplying, pricing, and marketing crude oil. However, the structure of 
the world crude oil market has dramatically changed as a result of such 
factors as the nationalization of oil fields by oil-producing 
countries, the emergence of independent oil companies, and the 
evolution of futures and spot markets in the 1970s and 1980s. Moreover, 
U.S. oil prices, controlled by the government since 1971, were 
deregulated in 1981. Consequently, the price of crude oil is now 
largely determined in the world oil market, which is mostly influenced 
by global factors, especially Organization of Petroleum Exporting 
Countries' (OPEC) supply decisions and world economic and political 
conditions.

The United States currently imports over 60 percent of its crude oil 
supply. In contrast, the bulk of the gasoline used in the United States 
is produced domestically. In 2001, for example, gasoline refined in the 
United States accounted for over 90 percent of the total domestic 
gasoline consumption. Companies that supply gasoline to U.S. markets 
also post the domestic gasoline prices. Historically, the domestic 
petroleum market has been divided into five regions, known as Petroleum 
Administration for Defense Districts (PADD)--PADD I is the East Coast 
region, PADD II is the Midwest region, PADD III is the Gulf Coast 
region, PADD IV is the Rocky Mountain region, and PADD V is the West 
Coast region.

Proposed mergers in all industries, including the petroleum industry, 
are generally reviewed by federal antitrust authorities--including the 
Federal Trade Commission (FTC) and the Department of Justice (DOJ)--to 
assess the potential impact on market competition. According to FTC 
officials, FTC generally reviews proposed mergers involving the 
petroleum industry because of the agency's expertise in that industry. 
FTC analyzes these mergers to determine if they would likely diminish 
competition in the relevant markets and result in harm, such as 
increased prices. To determine the potential effect of a merger on 
market competition, FTC evaluates, among other things, how the merger 
would change the level of market concentration. Conceptually, the 
higher the concentration, the less competitive the market is and the 
more likely that firms can exert control over prices. The ability to 
maintain prices above competitive levels for a significant period of 
time is known as market power.

Market concentration is commonly measured by the Herfindahl-Hirschman 
Index (HHI), calculated by summing the squares of the market shares of 
all the firms within a given market. According to the merger guidelines 
jointly issued by DOJ and FTC, market concentration is ranked into 
three separate categories based on the HHI: a market with an HHI under 
1,000 is considered to be unconcentrated; if the HHI is between 1,000 
and 1,800 the market is considered moderately concentrated; and if the 
HHI is above 1,800, the market is considered highly concentrated.

While concentration is an important aspect of market structure--the 
underlying economic and technical characteristics of an industry--other 
aspects of market structure that may be affected by mergers also play 
an important role in determining the level of competition in a market. 
These aspects include barriers to entry, which are market conditions 
that provide established sellers an advantage over potential new 
entrants in an industry, and vertical integration, which is the 
participation of firms in more than one successive stage of production 
or distribution in a market.

Results in Brief: 

GAO's analysis indicates that from 1991 through 2000 all three segments 
of the U.S. petroleum industry experienced mergers--over 2,600 
transactions in all. The majority of the mergers occurred during the 
second half of the decade, most frequently in the upstream (exploration 
and production) segment. Petroleum industry officials cited various 
reasons for this wave of mergers, particularly the need for increased 
efficiency and cost savings. Economic literature suggests that firms 
also sometimes use mergers to enhance their market power. However, the 
reasons cited by both sources generally relate to the merging 
companies' desire to ultimately maximize profit or shareholder wealth.

Market concentration, as measured by HHI, has increased substantially 
in the downstream segment of the U.S. petroleum industry since the 
1990s, partly as a result of merger activities, while changing very 
little in the upstream segment. Increased market concentration can 
result in greater market power, potentially increasing prices above 
competitive levels. On the other hand, it can lead to efficiency gains 
through cost savings, which may be passed on to consumers in the form 
of lower prices. The impact--either positive or negative--of increased 
market concentration on prices ultimately depends on whether market 
power or efficiency dominates. In the downstream (refining and 
marketing) segment, market concentration in refining increased from 
moderately to highly concentrated in the East Coast and from 
unconcentrated to moderately concentrated in the West Coast; it 
increased but remained moderately concentrated in the Rocky Mountain 
region. Concentration in the wholesale gasoline market increased 
substantially from the mid-1990s so that by 2002, most states had 
either moderately or highly concentrated wholesale gasoline markets. On 
the other hand, market concentration decreased somewhat in the upstream 
(exploration and production) segment and remained unconcentrated by the 
end of the 1990s. While mergers occurred in the midstream 
(transportation) segment, GAO could not determine the extent to which 
concentration changed in this segment because of a lack of relevant 
data and difficulties in defining markets. Anecdotal evidence and 
economic analysis by some industry experts suggest that mergers not 
only affected market concentration but also enhanced vertical 
integration and barriers to entry. GAO could not, however, determine 
the extent to which these other aspects of market structure changed in 
the petroleum industry because adequate data do not exist. Like 
increased market concentration, increased vertical integration can 
result in higher or lower consumer prices. Barriers to entry are 
important in a market because firms that operate in concentrated 
industries with high barriers to entry are more likely to have market 
power.

According to industry officials, two major changes have occurred in 
U.S. gasoline marketing since the 1990s, partly related to mergers. 
First, the availability of unbranded (generic) gasoline has decreased 
substantially. Unbranded gasoline is generally priced lower than 
branded gasoline, which is marketed under the refiner's trademark. 
Industry officials generally attributed the decreased availability of 
unbranded gasoline to, among other factors, a reduction in the number 
of independent refiners that typically supply unbranded gasoline. GAO 
could not, however, statistically quantify the extent to which the 
supply of unbranded gasoline has decreased because relevant data are 
not available. The second change identified by industry officials is 
that refiners now prefer dealing with large distributors and retailers. 
This preference, according to the officials, has motivated further 
consolidation in both the distributor and retail markets, including the 
rise of hypermarkets--a relatively new breed of gasoline market 
participants that includes such large retail warehouses as Wal-Mart and 
Costco.

GAO's econometric analyses show that oil industry mergers and increased 
market concentration generally led to higher wholesale gasoline prices 
(measured in this report as wholesale prices less crude oil prices) for 
different gasoline types in the United States in the second half of the 
1990s, although prices sometimes decreased. Six of the eight specific 
mergers GAO modeled--which mostly involved large, fully vertically 
integrated companies--generally resulted in increases in wholesale 
prices for branded and/or unbranded gasoline of about 2 cents per 
gallon, on average. Two of the mergers generally led to price 
decreases, of about 1 cent per gallon, on average. For conventional 
gasoline--the predominant type used in the United States except in 
areas that require special gasoline formulations to meet clean air 
standards--the change in wholesale prices ranged from a decrease of 
about 1 cent per gallon to an increase of about 5 cents per gallon. The 
preponderance of price increases over decreases indicates that the 
market power effects, which tend to increase prices, for the most part 
outweighed the efficiency effects, which tend to decrease prices. 
Increased market concentration, which captures the cumulative effects 
of mergers as well as other market structure factors, also generally 
led to higher prices for conventional gasoline, which is sold 
nationwide, and for boutique fuels--gasoline that has been reformulated 
for certain areas in the East Coast and Gulf Coast regions and in 
California, to lower pollution. The price increases were particularly 
large in California, where they averaged about 7 cents per gallon. 
Higher wholesale gasoline prices were also a result of other factors: 
low gasoline inventories, which typically occur in the summer driving 
months; high refinery capacity utilization rates; and supply 
disruptions, which occurred in the Midwest and the West Coast.

GAO's findings are generally consistent with previous studies of the 
effects of specific oil mergers and of market concentration on 
wholesale and retail gasoline prices. GAO used extensive peer review to 
obtain comments from outside experts, which were incorporated as 
appropriate. GAO believes that this is the first study to model the 
impact of the petroleum industry's 1990s merger wave on wholesale 
gasoline prices for the primary gasoline specifications for the entire 
United States, an effort that required GAO to acquire large datasets 
and perform complex analyses.

Principal Findings: 

Mergers Occurred in All Segments of the U.S. Petroleum Industry in the 
1990s for Several Reasons: 

Over 2,600 merger transactions occurred from 1991 through 2000 
involving all three segments of the U.S. petroleum industry. Almost 85 
percent of the mergers occurred in the upstream segment (exploration 
and production), while the downstream segment (refining and marketing 
of petroleum) accounted for about 13 percent, and the midstream segment 
(transportation) accounted for over 2 percent. The vast majority of the 
mergers--about 80 percent--involved one company's purchase of a segment 
or asset of another company, while about 20 percent involved the 
acquisition of one company's total assets by another so that the two 
became one company. Most of the mergers occurred in the second half of 
the decade, including those involving large partially or fully 
vertically integrated companies.

Petroleum industry officials and experts GAO contacted cited several 
reasons for the industry's wave of mergers in the 1990s, including 
achieving synergies, increasing growth and diversifying assets, and 
reducing costs. Economic literature indicates that enhancing market 
power is also sometimes a motive for mergers. These reasons mostly 
relate to companies' ultimate desire to maximize profit or stock 
values.

Mergers Contributed to Increases in Market Concentration and to Changes 
in Other Aspects of Market Structure That Affect Competition: 

Mergers in the 1990s have contributed to increases in market 
concentration in the downstream segment of the U.S. petroleum industry, 
while the upstream segment experienced little change overall. Increased 
market concentration can result in greater market power, potentially 
allowing firms to increase prices above competitive levels. On the 
other hand, increased market concentration may also lead to efficiency 
gains that can be passed on to consumers as lower prices. Whether 
increased market concentration results in higher or lower prices 
depends on which effect predominates. GAO found that market 
concentration, as measured by the HHI, decreased slightly in the 
upstream segment, based on crude oil production activities at the 
national level, from 290 in 1990 to 217 in 2000. Moreover, based on 
benchmarks established jointly by DOJ and FTC, the upstream segment of 
the U.S. petroleum industry remained unconcentrated at the end of the 
1990s. The increases in market concentration in the downstream segment 
varied by activity and region. For example, the HHI of the refining 
market in the East Coast region increased from a moderately 
concentrated level of 1136 in 1990 to a highly concentrated level of 
1819 in 2000. In the Rocky Mountain and the West Coast regions it 
increased from 1029 to 1124 and from 937 to 1267, respectively, in that 
same period. Thus, while each of these refining markets increased, the 
Rocky Mountain region remained within the moderately concentrated range 
but the West Coast region changed from unconcentrated in 1990 to 
moderately concentrated in 2000. The HHI of refining markets also 
increased from 699 to 980 in the Midwest region and from 534 to 704 in 
the Gulf Coast region during the same period, although these markets 
remained unconcentrated. In wholesale gasoline markets, GAO found that 
market concentration increased broadly throughout the United States 
between 1994 and 2002. Specifically, GAO found that 46 states and the 
District of Columbia had moderately or highly concentrated markets by 
2002, compared to 27 in 1994. For both the refining and wholesale 
markets of the downstream segment, GAO found that merger activity and 
market concentration were highly correlated for most regions of the 
country.

Evidence from various sources indicates that in addition to increasing 
market concentration, mergers also contributed to changes in other 
aspects of market structure in the U.S. petroleum industry that affect 
competition--specifically, vertical integration and barriers to entry. 
However, GAO could not quantify the extent of these changes because of 
a lack of relevant data. Vertical integration can conceptually have 
both pro-and anticompetitive effects. Based on anecdotal evidence and 
economic analyses by some industry experts, GAO determined that a 
number of mergers that have occurred since the 1990s have led to 
greater vertical integration in the U.S. petroleum industry, especially 
in the refining and marketing segment. For example, GAO identified 
eight mergers that occurred between 1995 and 2001 that might have 
enhanced the degree of vertical integration, particularly in the 
downstream segment. Concerning barriers to entry, GAO's interviews with 
petroleum industry officials and experts provide evidence that mergers 
had some impact on the U.S. petroleum industry. Barriers to entry could 
have implications for market competition because companies that operate 
in concentrated industries with high barriers to entry are more likely 
to possess market power. Industry officials pointed out that large 
capital requirements and environmental regulations constitute barriers 
for potential new entrants into the U.S. refining business. For 
example, the officials indicated that a typical refinery could cost 
billions of dollars to build and that it may be difficult to obtain the 
necessary permits from the relevant state or local authorities. At the 
wholesale and retail marketing levels, industry officials pointed out 
that mergers may have exacerbated barriers to entry in some markets. 
For example, the officials noted that mergers have contributed to a 
situation where pipelines and terminals are owned by fewer, mostly 
integrated companies that sometimes deny access to third-party users, 
especially when supply is tight--which creates a disincentive for 
potential new entrants into such wholesale markets.

U.S. Gasoline Marketing Has Changed in Two Major Ways: 

According to some petroleum industry officials that GAO interviewed, 
gasoline marketing in the United States has changed in two major ways 
since the 1990s. First, the availability of unbranded gasoline has 
decreased, partly due to mergers. Officials noted that unbranded 
gasoline is generally priced lower than branded. They generally 
attributed the decreased availability of unbranded gasoline to one or 
more of the following factors: 

* There are now fewer independent refiners, who typically supply mostly 
unbranded gasoline. These refiners have been acquired by branded 
companies, have grown large enough to be considered a brand, or have 
simply closed down.

* Partially or fully vertically integrated oil companies have sold or 
mothballed some refineries. As a result, some of these companies now 
have only enough refinery capacity to supply their own branded needs, 
with little or no excess to sell as unbranded.

* Major branded refiners are managing their inventory more efficiently, 
ensuring that they produce only enough gasoline to meet their current 
branded needs.

GAO could not quantify the extent of the decrease in the unbranded 
gasoline supply because the data required for such analyses do not 
exist.

The second change identified by these officials is that refiners now 
prefer dealing with large distributors and retailers because they 
present a lower credit risk and because it is more efficient to sell a 
larger volume through fewer entities. Refiners manifest this preference 
by setting minimum volume requirements for gasoline purchases. These 
requirements have motivated further consolidation in the distributor 
and retail sectors, including the rise of hypermarkets.

Mergers and Increased Market Concentration Generally Led to Higher U.S. 
Wholesale Gasoline Prices: 

GAO's econometric modeling shows that the mergers GAO examined mostly 
led to higher wholesale gasoline prices in the second half of the 
1990s. GAO's analysis shows that the majority of the eight specific 
mergers examined--Ultramar Diamond Shamrock (UDS)-Total, Tosco-Unocal, 
Marathon-Ashland, Shell-Texaco I (Equilon), Shell-Texaco II (Motiva), 
BP-Amoco, Exxon-Mobil, and Marathon Ashland Petroleum (MAP)-UDS--
resulted in higher prices of wholesale gasoline in the cities where the 
merging companies supplied gasoline before they merged. For the seven 
mergers that GAO modeled for conventional gasoline, five led to 
increased prices, especially the MAP-UDS and Exxon-Mobil mergers, where 
the increases generally exceeded 2 cents per gallon. For the four 
mergers that GAO modeled for reformulated gasoline, two--Exxon-Mobil 
and Marathon-Ashland--led to increased prices of about 1 cent per 
gallon, on average. In contrast, the Shell-Texaco II (Motiva) merger 
led to price decreases of less than one-half cent per gallon for 
branded gasoline only. For the two mergers--Tosco-Unocal and Shell-
Texaco I (Equilon)--that GAO modeled for the reformulated gasoline used 
in California, known as California Air Resources Board (CARB) gasoline, 
only the Tosco-Unocal merger led to price increases. The increases were 
for branded gasoline only and exceeded 6 cents per gallon. The effects 
of some of the mergers were inconclusive, especially for boutique fuels 
sold in the East Coast and Gulf Coast regions and in California.

For market concentration, which captures the cumulative effects of 
mergers as well as other competitive factors, GAO's econometric 
analysis shows that increased market concentration resulted in higher 
wholesale gasoline prices. Prices increased for conventional (non-
boutique) gasoline, the dominant type of gasoline sold nationwide from 
1994 through 2000, by less than one-half cent per gallon for branded 
and unbranded gasoline. The increases were larger in the West than in 
the East--the increases were between one-half cent and 1 cent per 
gallon in the West, and about one-quarter cent in the East (for branded 
gasoline only). Price increases for boutique fuels sold in some parts 
of the East Coast and Gulf Coast regions and in California were larger 
compared to the increases for conventional gasoline. The wholesale 
prices increased by about 1 cent per gallon for boutique fuel sold in 
the East Coast and Gulf Coast regions between 1995 and 2000, and by 
over 7 cents per gallon in California between 1996 and 2000.

GAO's analysis shows that wholesale gasoline prices were also affected 
by other factors included in the econometric models--particularly, 
gasoline inventories relative to demand, refinery capacity utilization 
rates, and the supply disruptions that occurred in some parts of the 
Midwest and the West Coast. In particular, wholesale gasoline prices 
were about 1 cent per gallon higher when gasoline inventories were low 
relative to demand, typically in the summer driving months. Also, 
prices were higher by about one-tenth to two-tenths of 1 cent per 
gallon when refinery capacity utilization rates increased by 1 percent. 
The prices of conventional gasoline were about 4 to 5 cents per gallon 
higher on average during the Midwest and West Coast supply disruptions. 
The increase in prices for CARB gasoline was about 4 to 7 cents per 
gallon, on average, during the West Coast supply disruptions.

Recommendations for Executive Action: 

GAO is not making recommendations in this report.

Agency Comments and GAO's Evaluation: 

GAO provided a draft of this report to FTC for its review and comment. 
FTC stated that the draft report was flawed and did not provide a basis 
for reliable judgments about the competitive effects of mergers in the 
petroleum industry. However, GAO believes that its analyses are sound 
and consistent with the views of independent economists and experts 
that peer reviewed GAO's overall modeling approach. In particular, Dr. 
Severin Borenstein, a recognized expert in the modeling of gasoline 
markets, reviewed and commented on GAO's econometric analysis and 
results at several stages. In response, GAO made revisions in the 
course of developing and estimating its models and in its final report, 
as appropriate. In addition, partly in response to FTC's comments, GAO 
re-estimated its models to account for the effects of gasoline supply 
disruptions that occurred in some parts of the West Coast and Midwest 
regions.

FTC focused a substantial portion of its comments on GAO's econometric 
models, outlining five concerns. First, FTC asserted that the models 
did not control for the many factors that could cause gasoline price 
increases, citing the following factors: seasonality, temperature, 
income, changes in gasoline formulations, and supply disruptions in the 
Midwest and West Coast regions. This assertion is not correct. GAO's 
models incorporated key factors that affect wholesale gasoline prices, 
including crude oil prices, refinery capacity utilization rates, and 
gasoline inventory-to-demand ratio--a ratio that captures the effects 
of seasonality and temperature. GAO considered the available data for 
income by city but found that income data did not vary over time and 
therefore would not be appropriate for the estimation technique (fixed-
effects) that GAO used. GAO controlled for changes in gasoline 
formulations between seasons through the inventory-to-demand ratio; 
other changes in formulations either occurred outside the time period 
that GAO examined or were unlikely to significantly affect the results. 
During GAO's December 2002 meeting with FTC staff, the staff agreed 
that the effects of other formulations could be minimal because these 
other formulations are typically a small percentage of the total volume 
of gasoline in the areas that GAO modeled. Regarding the potential 
effects of the Midwest and West coast supply disruptions, GAO believes 
that the models indirectly captured these effects through the 
inventory-to-demand ratio. Nonetheless, in response to FTC's comments, 
GAO included a proxy for these disruptions in its models.

Second, FTC stated that GAO's modeling of the effect of market 
concentration on wholesale gasoline prices was problematic, primarily 
because the agency claimed that the methodology GAO used did not 
meaningfully distinguish correlation from causation. GAO disagrees. 
Modeling using appropriate economic structure is a common basis for 
inferring causation, and GAO's market concentration model is consistent 
with previous studies on prices and market concentration.

Third, FTC said that GAO used geographic markets that were empirically 
unjustified to conduct its analysis. GAO recognizes the importance and 
difficulty of defining relevant geographic markets for gasoline, 
especially at the wholesale level, and discussed this issue with FTC 
and other industry experts. FTC indicated that it could not provide 
specific evidence on what the actual geographic markets for wholesale 
gasoline were across the United States because, when analyzing 
potential mergers, FTC focuses on a limited geographic area and relies 
substantially on proprietary company data. Like other industry experts 
that GAO contacted, FTC staff agreed in a December 2002 meeting that it 
was appropriate to use terminal cities and even states, in some cases, 
as geographic markets for wholesale gasoline. GAO therefore used 
terminal (rack) cities as the geographic unit. In measuring market 
concentration at the wholesale level, the draft report that GAO 
provided to FTC used HHI data from DOE's Energy Information 
Administration (EIA) that were based on sales of prime suppliers of 
wholesale gasoline and available only at the state level. In the final 
report, GAO measured market concentration using HHI data that GAO 
constructed based on refinery capacity at the PADD level, after 
consultation with GAO's expert consultant/reviewer, because GAO 
believes that market concentration at the refining level more 
effectively captures the potential market power of the refiners.

Fourth, FTC said that GAO's modeling results are, in many cases, not 
robust. By robustness, FTC meant that model results yielded by 
alternative modeling approaches should be consistent. GAO believes that 
the results for its models' key variables--mergers and market 
concentration--are robust because these models yielded consistent 
results using alternative model specifications. In particular, when GAO 
estimated its models without including the effects of supply 
disruptions in the affected markets, the effects of the key policy 
variables--mergers and market concentration--were consistent with the 
results obtained when GAO incorporated the effects of supply 
disruptions. Furthermore, because market concentration reflects the 
cumulative effects of the mergers and other competitive factors, one 
would expect the results from both approaches--market concentration 
models and mergers models--to be similar if mergers are the predominant 
contributing factor to market concentration. In GAO's study, the 
overall results for both approaches were consistent. GAO believes these 
are valid demonstrations of the robustness of its model results.

Fifth, FTC said that GAO did not provide complete technical 
documentation for its econometric models. This is not correct. GAO 
provided a detailed and complete description of the basis of its 
econometric models, including data sources, sample selection processes 
(including tables detailing the list of variables, definitions, 
sources, data frequency, and level), specifications of the econometric 
models, and estimation techniques.

In addition to criticizing GAO's models, FTC also criticized GAO's 
findings about the effects of mergers on the structure of the petroleum 
industry and U.S. gasoline marketing. Specifically, the agency 
commented that GAO's findings--that mergers have contributed to 
barriers to entry and vertical integration and that the availability of 
unbranded gasoline has decreased--lacked quantitative foundations and 
were therefore flawed. GAO disagrees with this opinion. Economic 
findings can be qualitative or quantitative. GAO stated in its report 
that it could not quantify the extent to which mergers have affected 
barriers to entry and vertical integration because of a lack of 
comprehensive data to fully measure these factors and because there is 
no consensus on how to appropriately measure them. GAO's finding that 
mergers have contributed to barriers to entry was based on information 
from industry officials who provided examples, which GAO included in 
its report, to validate this finding. While GAO discussed the overall 
importance of barriers to entry in a market, which FTC recognizes in 
its merger guidelines, GAO did not conclude, contrary to FTC's 
assertions, that barriers to entry have harmed or eliminated 
competition in the industry. To validate GAO's finding that mergers 
have contributed to vertical integration, GAO presented examples of 
mergers--particularly in the downstream segment between refiners and 
marketers--that were vertical in nature (that is, the mergers involved 
different functional levels of the merging companies) and would 
contribute to increased vertical integration. GAO also added language 
to its report, as suggested by EIA, acknowledging the shift during the 
1990s toward fully integrated companies' divestiture of certain 
downstream assets, such as refineries, to nonintegrated companies. For 
its finding that unbranded gasoline has become less available, GAO 
relied on extensive interviews with industry participants in different 
regions of the country. While it would be desirable to ascertain this 
finding quantitatively, GAO noted in its report that EIA--the federal 
agency mandated by Congress to collect energy data, including data on 
gasoline supply--told GAO that the agency does not require petroleum 
companies to report gasoline data in the form that would permit the 
identification of branded and unbranded sales.

The full text of FTC's comments and GAO's responses are included in 
appendixes V and VI. Appendix V contains the comments from FTC 
Commissioners and appendix VI contains the comments from FTC's Bureau 
of Economics staff.

[End of section]

Chapter 1: Introduction: 

Since the 1990s, the U.S. petroleum industry has experienced a wave of 
mergers, acquisitions, and joint ventures (hereafter referred to as 
mergers). Some of these mergers involved well known major petroleum 
companies that had previously competed with each other for the sale of 
gasoline and other petroleum products. There were also numerous mergers 
between smaller companies. Some policy makers and consumer groups 
believe that these mergers may have reduced competition in the U.S. 
petroleum industry and ultimately led to higher gasoline prices. During 
the second half of the 1990s, U.S. gasoline prices exhibited periods of 
high price volatility, with fairly frequent price spikes. The price of 
crude oil, the primary input for producing gasoline, was similarly 
volatile.

The Petroleum Industry Consists of Three Main Segments: 

As depicted in figure 1, the U.S. petroleum industry consists of the 
exploration and production segment (upstream); the refining and 
marketing segment (downstream); and a third segment typically referred 
to as the midstream, which consists of the infrastructure used to 
transport crude oil and petroleum products. Some of the petroleum 
companies in the United States, like Exxon-Mobil and Chevron-Texaco, 
operate in all segments of the industry--that is, they are fully 
vertically integrated. Others, like Anadarko and Valero, that operate 
in one or more but not all segments are generally called partially 
vertically integrated or independents.[Footnote 2]

Figure 1: U.S. Petroleum Industry Chain: 

[See PDF for image]

[End of figure]

The Upstream Segment: 

The activities of the upstream segment consist essentially of 
exploration for and production of crude oil and natural gas. Hence, the 
upstream is also referred to as the exploration and production segment. 
Participants in the U.S. upstream include fully vertically integrated 
companies and independent producers. The U.S. upstream segment is 
characterized by a large number of independent producers and a smaller 
number of fully vertically integrated oil companies.

The Energy Information Administration (EIA)--the independent 
statistical and analytical agency within the U.S. Department of Energy 
(DOE)--has classified U.S. upstream operators into three main 
categories according to the size of their production in 2001, not 
according to whether they are integrated or independent: 

* large operators--who produced a total of 1.5 million barrels or more 
of crude, 15 billion cubic feet of natural gas, or both;

* intermediate operators--who produced a total of at least 400,000 
barrels of crude oil, 2 billion cubic feet of natural gas, or both, but 
less than the large operators; and: 

* small operators--who produced less than the intermediate operators.

Based on this classification, EIA estimated that as of 2001, there were 
179 large operators, which accounted for 84.2 percent of crude oil 
production; 430 intermediate operators, which accounted for 5.8 percent 
of crude oil production; and 22,519 small operators, which accounted 
for 10 percent of crude oil production.

Fully vertically integrated companies are generally large operators, 
while independent producers are generally small operators, with a few 
medium and large operators. While the fully vertically integrated 
companies are generally multibillion dollar companies that are publicly 
traded, the independent producers include many extremely small, 
privately owned operations as well as a few multibillion dollar and 
publicly traded companies. In general, the fully vertically integrated 
companies have upstream operations both in the United States and 
overseas and accounted for about 60 percent of U.S. crude oil 
production in 2002. On the other hand, the exploration and production 
activities of the independents occur: 

mostly in the United States and accounted for about 40 percent of the 
crude oil produced in the United States in 2002.[Footnote 3]

The price of crude oil produced in the United States is determined in 
the world oil market because the decontrol of domestic oil prices in 
1981 has effectively linked the U.S. oil market to the world oil 
market. In 2000, the United States contained only about 2 percent of 
world's estimated oil reserves but accounted for about 26 percent of 
the world's oil demand. From 1990 to 2000, U.S. production decreased 
significantly, from about 7.4 million barrels per day (mmb/d), or about 
55.5 percent of total U.S. crude oil supply, to about 5.8 mmb/d, or 39 
percent of total crude oil supply. Nevertheless, the United States was 
still the world's third largest producer of crude oil. U.S. reliance on 
oil imports has increased over the last decade as domestic production 
has dwindled.

The Midstream Segment: 

The midstream segment transports crude oil and petroleum products. 
Petroleum transportation facilities include pipelines, marine tankers 
and barges, railways, and trucks. Pipelines and, to a lesser extent, 
the other carriers transport domestically produced crude oil from the 
production points to the refineries, while marine carriers generally 
transport imported oil. Refined products, such as gasoline, are also 
carried via these modes from refineries to storage terminals, from 
which they are generally transported by trucks to retail stations.

In general, pipelines are the dominant and most efficient mode of 
transporting crude oil and petroleum products in the United States. 
According to data from the Association of Oil Pipelines, pipelines 
transported 66.1 percent of all the crude and petroleum products in the 
United States in 2000. Marine tankers and barges transported 28 
percent, while trucks and railways hauled 3.6 percent and 2.3 percent, 
respectively. According to DOE's Office of Transportation Technology, 
there are more than 200,000 miles of oil pipelines in the United States 
in all 50 states. The Federal Energy Regulatory Commission (FERC) 
regulates the rates on common carrier pipelines. The Association of Oil 
Pipelines told us that FERC currently regulates about 202 pipeline 
companies. According to the pipeline association, 84 percent of the 
pipelines are federally regulated while 16 percent are not.

The Downstream Segment: 

Refining and marketing are the main activities of the downstream 
segment. Refining is the process of transforming crude oil into 
petroleum products ranging from gasoline and distillate fuel oil 
(heating oil) to heavier products such as asphalt. As figure 2 shows, 
gasoline accounted for nearly half of U.S. refinery output in 
2000.[Footnote 4]

Figure 2: Product Yield from a Barrel of Crude Oil, 2000: 

[See PDF for image]

[End of figure]

According to data from EIA, as of January 1, 2002, there were 149 
operable refineries in the United States, with a total crude oil 
distillation capacity of about 16.8 mmb/d. Overall, 60 refining firms, 
including large fully vertically integrated companies and independent 
refiners, owned these refineries.[Footnote 5] The refining companies 
ranged in size from the smallest, with only 880 barrels per day of 
crude oil distillation capacity, to the biggest, with a combined 
refinery capacity of 1.8 mmb/d of crude distillation. Not all of these 
refineries produce gasoline; some, especially those with small 
distillation capacity, produce only asphalt.

Marketing in the downstream involves selling petroleum products to 
customers, who are generally wholesale and retail purchasers. For 
gasoline, as shown in figure 1, refiners arrange to move products from 
the refineries to storage terminals, from which they sell the product 
to wholesale purchasers. As discussed in detail in chapter 4, there are 
different classes of wholesale gasoline purchasers in the United 
States, and the prices they pay depend, in part, on the type of 
relationship they have with the refiners. From the terminals, gasoline 
is distributed to retail stations for sale to final consumers.

Different Entities Have Historically Exerted Influence over the World 
Petroleum Market: 

The world petroleum market, of which the U.S. market is a part, has 
been characterized by eras when a relatively small number of entities 
exerted considerable influence on the market. Three entities in 
particular have significantly influenced the world petroleum market 
during their eras: (1) Standard Oil, (2) the "Seven Sisters," and (3) 
the Organization of Petroleum Exporting Countries (OPEC). Figure 3 
shows a timeline of the major events that shaped the eras dominated by 
these entities.

Figure 3: Major Events in the World Petroleum Market: 

[See PDF for image]

[End of figure]

The Standard Oil Era: 

The Standard Oil Company was established in 1870, about a decade after 
the discovery of crude oil in commercial quantity in the United States, 
and the company quickly became the dominating force in the emerging 
U.S. petroleum industry. During the decade prior to the establishment 
of Standard Oil, the new industry experienced periods of overcapacity 
in both crude oil production and refining. The industry consisted of 
numerous independent producers and refiners. Railroad companies 
provided transportation services for crude oil and refined products. 
Thus, the industry tended to be intensely competitive and, by the end 
of the 1860s, the industry had excess crude oil supply and refinery 
capacity, resulting in frequent price fluctuations and price collapses.

In response to these conditions, Standard Oil adopted a process of 
consolidation that would ultimately lead to the virtual monopolization 
of the industry. Specifically, it employed a combination of tactics 
that included acquisitions and buyouts of competitors, vertical 
integration, control of transportation, and below-cost pricing to force 
competitors out of business. By the time Standard Oil was broken into 
separate companies in 1911 under the Sherman Antitrust Act, the company 
was able to effectively determine the purchase price for American crude 
oil. The breakup of Standard Oil ended its dominance as a single 
company over the U.S. petroleum market. However, the resulting separate 
companies began seeking ways to cooperate among themselves and with 
other foreign oil companies to control the global supply and price of 
oil.

The "Seven Sisters" Era: 

During the decades following the breakup of Standard Oil until about 
1970, seven oil companies--Exxon, Mobil, Chevron, Gulf, Texaco, Royal 
Dutch/Shell, and British Petroleum (BP)--dominated and controlled the 
global network for supplying, pricing, and marketing crude oil. Because 
of their close association and multiple joint ventures, these companies 
ultimately became known as the "Seven Sisters." The strategies the 
companies employed to control the world petroleum market sometimes 
included cooperation and collusion among themselves. For example, as a 
surge of oil supply from the United States and other countries flooded 
the world market in the 1920s, the ensuing competition between some of 
the companies for market share precipitated collapsing oil prices and 
threatened the security of their markets. In response, Exxon, Royal 
Dutch Shell, and BP met to draw up a series of agreements in the late 
1920s and 1930s to curb what they viewed as "ruinous competition" in 
the market. The overall thrust of the agreements was to allocate market 
shares or quotas; fix prices; and eliminate, through acquisitions and 
other means, the potential competitive impact of other oil companies 
outside their group, namely the independent producers and refiners.

Although by the 1960s the Seven Sisters had lost some ground in the 
world petroleum market--especially in the United States where the role 
of the independents continued to increase--as late as 1972, the seven 
companies were still producing 91 percent of the Middle East's crude 
oil and 77 percent of the supply outside the United States and the 
former Soviet Union. By the 1960s and 1970s, the United States had 
become a substantial net importer of oil.

The OPEC Era: 

OPEC was formed in 1960 after members of the Seven Sisters unilaterally 
cut the posted price of Middle Eastern crude oil--upon which they paid 
taxes and royalties to the producing nations--without consulting the 
producing nations. The founding members of OPEC were Saudi Arabia, 
Iraq, Iran, Kuwait, and Venezuela. Over time, the organization's 
membership grew to 13, with the addition of the United Arab Emirates, 
Nigeria, Libya, Qatar, Algeria, Indonesia, Ecuador, and Gabon.[Footnote 
6] The aim of the organization was to create an entity through which 
member countries could jointly confront the Seven Sisters over the 
control of their oil. The group had little or no influence on the world 
oil market during its first 10 years, partly because the international 
oil companies, not OPEC member countries, owned and controlled oil 
reserves in those countries in the 1960s. OPEC also lacked sufficient 
cohesion among its members to effectively challenge the influence of 
the Seven Sisters. Since the 1970s, however, OPEC has been a dominant 
force in the world oil market. OPEC became a major influence in 1973 
when it orchestrated a nearly fourfold price increase in a matter of 
months through an oil embargo by its Arab members against the United 
States and other countries friendly to Israel. Two other major oil 
price episodes resulting from events in OPEC member countries also 
occurred. In 1979 the Iranian revolution caused the doubling of crude 
oil prices from about $14 a barrel to $34 a barrel, and in 1990 the 
Iraqi invasion of Kuwait caused an immediate increase in the crude oil 
price from about $16 a barrel to about $28 barrel.

As a group, OPEC holds the world's largest and lowest-cost reserves of 
crude oil. As figure 4 shows, OPEC countries accounted for over two-
thirds of the world's estimated conventional reserves of about 1 
trillion barrels in 2001 (the latest available data). Persian Gulf OPEC 
countries had by far the largest reserves, with Saudi Arabia alone 
accounting for over one-fourth of world reserves. In contrast, the 
United States contained an estimated 2 percent of world reserves.

Figure 4: Shares of the World's Conventional Crude Oil Reserves 
(February 2003): 

[See PDF for image]

[End of figure]

Moreover, as shown in figure 5, OPEC countries, especially Saudi 
Arabia, also hold most of the world's excess production capacity, which 
means they are the only countries in a position to increase production 
relatively quickly if there is a supply shortage in the world oil 
market. These conditions give OPEC countries considerable flexibility 
to influence world oil prices.

Figure 5: World's Estimated Excess Production Capacity (February 2003): 

[See PDF for image]

[End of figure]

During the 1980s, OPEC nations abandoned their strategy of setting 
"official" prices for their crude oil, but the individual and/or 
collective actions of the organization's member countries can still 
have a significant impact on world oil prices. OPEC now establishes a 
"target" price during its biannual meetings. To achieve this price, 
OPEC sets an aggregate production level, or quota, based on the 
organization's determination of the demand for its oil. OPEC then 
allocates voluntary production quotas among its members, primarily 
based on the size of each member's oil reserves and other negotiated 
factors. Whether or not the target price is achieved depends on the 
discipline exercised in producing oil, as well as the actual demand for 
oil and non-OPEC countries' production levels. If, by adjusting its 
production, OPEC keeps the world's oil supply relatively tight with 
respect to demand, the average world price will likely be close to the 
target price range.

FTC and DOJ Review Proposed Mergers to Preserve Market Competition: 

While crude oil prices are determined by global market forces and 
particularly by OPEC countries' actions, the prices of gasoline and 
other petroleum products are generally influenced by, among other 
things, the extent of domestic market competition. Thus, U.S. antitrust 
laws, which are enforced by the Federal Trade Commission (FTC) and the 
Department of Justice (DOJ), prohibit mergers and other activities that 
may be anticompetitive. As part of their responsibility for enforcing 
the antitrust laws, FTC and DOJ review proposed mergers to ensure they 
would not be anticompetitive. Under the Hart-Scott-Rodino Act of 
1976,[Footnote 7] as amended, companies contemplating a merger valued 
at $15 million or more ($50 million or more from February 1, 2001) and 
meeting certain other conditions must formally notify these agencies. 
There is then a 30-day waiting period to allow FTC or DOJ to review the 
proposed merger to determine its potential effect on 
competition.[Footnote 8] If the review does not indicate a need for 
further investigation, the merger can be consummated at the end of the 
waiting period or earlier if the parties request early termination of 
the waiting period and the request is granted.[Footnote 9] According to 
an FTC official, FTC generally handles mergers in the petroleum 
industry because of its expertise in the area.

The agencies will challenge a merger if it may substantially lessen 
competition or tend to create or enhance market power or to facilitate 
its exercise. Guidelines issued jointly by DOJ and FTC in 1992 outline 
how the agencies generally analyze proposed horizontal mergers and 
indicate when the government is likely to challenge a merger.[Footnote 
10] For a recent GAO report, FTC staff told us that the majority of 
mergers that raise antitrust concerns are horizontal mergers (mergers 
between firms operating in the same market).[Footnote 11] The 
guidelines indicate that horizontal mergers should not be permitted to 
create, enhance or facilitate the exercise of market power, which is 
the ability of one or more firms to profitably maintain prices above 
competitive levels for a significant period of time.

In reviewing proposed horizontal mergers, FTC first examines market 
concentration--a function of the number of firms in a market and their 
respective market shares. Other things being equal, market 
concentration affects the likelihood that one company, or a small group 
of firms, could successfully exercise market power. The merger 
guidelines identify the Herfindahl-Hirshman Index (HHI) as the measure 
used in evaluating market concentration. The HHI reflects the 
composition of a market while giving proportionately greater weight to 
the market shares of the larger firms.[Footnote 12] The higher the HHI, 
the greater the market concentration. According to the guidelines, a 
merger will generally not be challenged in a market where HHI after the 
proposed merger would be: 

* less than 1,000 points (an unconcentrated market);

* 1,000 to 1,800 points (a moderately concentrated market), and the HHI 
would be increased by less than 100 points by the merger; or: 

* over 1,800 points (a highly concentrated market), and the merger 
would increase it by less than 50 points.

Mergers that would increase the concentration above these levels will 
be examined further by the agency. Other factors that affect market 
competitiveness, such as barriers to entry into a market, are also 
considered in deciding whether to challenge a proposed merger. (See 
chapter three of this report for further discussion of these factors 
and HHI).

If FTC determines that a merger has potential anticompetitive effects, 
it can litigate to block the merger; negotiate a settlement to resolve 
anticompetitive aspects of the merger while allowing the transaction to 
go forward; or develop a consent arrangement that allows the merger to 
proceed but requires divestiture of assets to remedy the decrease in 
competition that would otherwise result. FTC has required divestiture 
of assets in some of the mergers in the petroleum industry since the 
1990s. For example, FTC required that Exxon divest all its retail 
stations from New York to New England and that Mobil divest all its 
retail stations from New Jersey to Virginia as a condition for the 
merger between the two companies. Tosco acquired these stations.

Objectives, Scope, and Methodology: 

As requested by the Ranking Minority Member, Permanent Subcommittee on 
Investigations of the Senate Committee on Governmental Affairs, this 
report examines the impact of mergers on the U.S. petroleum industry. 
It includes an econometric modeling of the effects of mergers and 
market concentration on U.S. wholesale gasoline markets. Specifically, 
the report examines: 

* mergers in the U.S. petroleum industry from the 1990s through 2000 
and why they occurred,

* the extent to which market concentration and other aspects of market 
structure in the petroleum industry have changed since the 1990s as a 
result of mergers,

* the major changes that have occurred in U.S. gasoline marketing since 
the 1990s, and: 

* the effect of mergers and market concentration in the U.S. petroleum 
industry on U.S. gasoline prices at the wholesale level.

To examine mergers in the U.S. petroleum industry in the 1990s and why 
they occurred, we analyzed a large body of data on petroleum industry 
merger transactions that occurred in the United States from the 1990s 
through 2000. We purchased data on mergers that occurred in all 
segments of the U.S. petroleum industry from 1990 through 2000 from 
John S. Herold, Inc., and Thompson Financial. We also obtained 
information from EIA on some of the industry's mergers since the 1990s. 
In addition, we interviewed officials from these entities. We also 
interviewed petroleum industry officials, including those whose firms 
were involved in mergers, and experts to obtain their views on the 
reasons for the mergers and reviewed relevant economic literature and 
FTC documents.

To assess the extent to which market concentration and other aspects of 
market structure have changed since the 1990s as a result of mergers, 
we obtained data on petroleum industry market shares from the Oil and 
Gas Journal (OGJ) and EIA. We also used the merger data from John S. 
Herold, Inc., and Thomson Financial. Using these data, we calculated 
and analyzed changes in the HHI--a measure of market concentration--for 
the various segments of the industry and, as necessary, for the 
relevant geographic markets, from the 1990s through 2000 or 2001, where 
data availability allowed.[Footnote 13] We also calculated correlation 
coefficients, where data availability permitted, to determine the 
extent to which changes in market concentration were statistically 
correlated with mergers. Because empirical data on other aspects of 
market structure--essentially vertical integration and barriers to 
entry--are usually not available, particularly at the broad levels that 
our study examined, we relied instead on an extensive body of relevant 
economic literature. Economic research on market structure is abundant 
and well developed, although it has rarely been applied specifically to 
the petroleum industry. We also interviewed oil industry officials and 
experts to obtain their views.

To determine what major changes have occurred in U.S. gasoline 
marketing since the 1990s, we analyzed EIA's data on gasoline 
marketing, reviewed relevant studies and documents from EIA and 
industry sources, and interviewed petroleum industry officials and 
experts and EIA officials.

To examine how mergers and market concentration have affected U.S. 
gasoline prices at the wholesale level, we developed econometric models 
that examined the effect of mergers and of market concentration on U.S. 
wholesale gasoline markets from 1994 through 2000. We chose 1994 as the 
initial year of our analysis because the market concentration (HHI) 
data on wholesale gasoline provided by EIA were available from 1994. 
Also, the Oil Price Information Service (OPIS), the company from whom 
we purchased the wholesale gasoline price data, informed us that it had 
more comprehensive data on U.S. wholesale gasoline prices starting in 
the second half of the 1990s than earlier. We developed two groups of 
econometric models: 

* one to estimate the impact of selected individual mergers on the 
wholesale gasoline price (measured in this report as wholesale gasoline 
price minus crude oil cost) in affected terminal markets and: 

* another to estimate the impact of market concentration, which 
essentially captures the cumulative effects of all the mergers in the 
U.S. wholesale petroleum industry during the 1990s as well as the 
effects of other changes in the structure of U.S. wholesale gasoline 
markets on wholesale gasoline prices in different U.S. geographic 
regions.

In doing so, we isolated the effects of mergers and market 
concentration from several other factors that could influence wholesale 
gasoline prices, such as crude oil costs, gasoline inventories relative 
to demand, refinery capacity utilization rates, and gasoline supply 
disruptions. We also differentiated among fuel formulations in our 
analyses. Retail gasoline prices that consumers ultimately pay may be 
affected by many other factors that vary from location to location, 
including, among other things, taxes, land values, zoning regulations, 
and competition at the retail level. We did not examine the effects of 
such factors because this study focuses on wholesale gasoline prices.

We provided a detailed draft outline of our econometric methodology, 
including a description of the types and sources of data we used, to a 
cross section of experts in academia, industry, and government for peer 
review and comment. We discussed extensively our econometric 
methodology, including data requirements, with the staff of FTC's 
Bureau of Economics. We requested comments from the American Petroleum 
Institute (API) on our econometric methodology, but they did not 
provide any comments. We also provided the same draft outline and our 
estimated results and interpretations to our consultant/peer reviewer, 
Dr. Severin Borenstein, E.T. Grether Professor of Business 
Administration and Public Policy and Director of the California Energy 
Institute at the University of California, Berkeley, for review and 
comment. See appendix II for a list of expert peer reviewers. Based on 
comments from and discussions with these experts and this consultant, 
we revised our models and interpretations as appropriate. Also, we 
interviewed industry officials and representatives involved in all 
aspects of the petroleum industry and in all major U.S. regions, oil 
industry experts, and officials from relevant federal and state 
agencies. In addition, we reviewed numerous economic studies on 
gasoline markets and pricing, including the few studies that have 
modeled the impact of mergers on gasoline markets; several textbooks on 
econometrics and industrial organization; and econometric studies of 
the impact of mergers and market concentration on other industries. 
Appendix IV contains details on our models' methodology, types and 
sources of data we used, and our econometric analysis.

Although in building our models we drew substantial insight from 
existing models, our models differ from most previous ones in three 
principal ways. First, to our knowledge, our study is the first to 
model the impact of the petroleum industry's merger wave in the 1990s 
on the wholesale gasoline prices for the entire United States, while 
isolating the effects of major boutique fuels and unique geographic 
markets as well as the effects of specific (individual) mergers, 
including some of the largest in the industry's history. Doing so 
required us to acquire large and expensive data and make complex 
computations. Second, we studied the behavior of wholesale prices 
because this allows us to capture the net effect of any potential 
market power and efficiency gains from mergers and market 
concentration. Third, we included the effects of refinery capacity 
utilization rates and of gasoline inventories, whereas other studies 
have either omitted these variables entirely or included only one.

Most of the data used for our econometric analysis of the impact of 
mergers and market concentration on wholesale prices were purchased 
from OPIS, a company that collects and sells oil industry information 
to oil companies and other entities. We also obtained data from EIA and 
the Department of Commerce's Bureau of the Census.

This report did not assess the appropriateness of FTC's review or the 
actions they took regarding mergers in the petroleum industry. However, 
we obtained detailed comments from FTC staff and commissioners and EIA 
staff on our modeling approach and revised our models and report where 
appropriate.

We conducted our review between June 2001 and April 2004 in accordance 
with generally accepted government auditing standards.

[End of section]

Chapter 2: All Segments of the Petroleum Industry Experienced Mergers 
for Several Reasons: 

During the 1990s, mergers occurred in all segments of the U.S. 
petroleum industry, but the upstream segment had the most mergers. The 
majority of the mergers--especially mergers among large firms--occurred 
in the second half of the decade. According to petroleum industry 
officials, mergers occurred in the petroleum industry for several 
reasons mostly related to the firms' desire to maximize profits through 
efficiency gains and cost savings. In addition to these reasons, 
economic literature also indicates that firms' desire to enhance their 
market power is a motive for mergers.

Mergers Occurred in All Three Segments, but Most Frequently in the 
Upstream: 

A total of over 2,600 merger transactions occurred in the U.S. 
petroleum industry from 1991 through 2000.[Footnote 14] As shown in 
figure 6, the upstream segment accounted for almost 85 percent of these 
mergers. About 13 percent of the mergers occurred in the downstream 
segment. The midstream segment, specifically pipelines--a key 
infrastructure for moving crude oil and petroleum products--accounted 
for about 2 percent of the mergers.[Footnote 15]

Figure 6: Percentage of Mergers That Occurred in Each Segment of the 
Petroleum Industry (1991-2000): 

[See PDF for image]

Note: When a merger involved the acquisition of assets simultaneously 
in each segment, it was counted in the segment with the largest 
monetary value.

[End of figure]

As shown in figure 7, mergers in all segments occurred more frequently 
in the mid-to late 1990s than in the early 1990s. Some of the mergers 
involving vertically integrated oil companies and large independent 
refiners occurred during this time (see figure 8).[Footnote 16]

Figure 7: Petroleum Industry Merger Trends (1991-2000): 

[See PDF for image]

[End of figure]

Figure 8: Selected Major Petroleum Mergers (1996-2002): 

[See PDF for image]

[End of figure]

Mergers that occurred in the petroleum industry in the 1990s were 
categorized into two broad transaction types: corporate mergers and 
asset mergers.[Footnote 17] About 20 percent of the mergers that 
occurred in the U.S. petroleum industry were corporate mergers,which 
generally involve the acquisition of a company's total assets by 
another so that the two become one company.[Footnote 18] Most of the 
mergers depicted in figure 8, such as Exxon-Mobil, BP-Amoco, Chevron-
Texaco, and Valero-UDS, were corporate mergers.

The majority of the mergers (about 80 percent) were asset mergers, 
which involved one company's purchase of only a segment or asset of 
another company, such as Williams' purchase of three storage and 
distribution terminals from Amerada Hess in 1999 and Tosco's 
acquisition of Unocal's refining and marketing assets on the West 
Coast. Similarly, the majority of the mergers that occurred in each of 
the segments were asset mergers. In the upstream, about 85 percent of 
the merger transactions were asset mergers, involving the acquisition 
of oil and/or gas reserves.[Footnote 19] In the downstream segment, 
about 54 percent were asset mergers, where one or more downstream 
assets--such as refining or gasoline service station assets--were 
purchased. As figure 9 shows, 71 percent of all mergers in the 
downstream segment involved the acquisition of wholesale distribution 
assets.

Figure 9: Percentage of Merger Transactions within the Downstream 
Segment by Type of Key Assets Acquired: 

[See PDF for image]

Note: According to the data from John S. Herold, Inc., the wholesale 
marketing category presented here includes both those establishments 
engaged in wholesale gasoline marketing and those engaged in the 
storage and/or wholesale distribution of crude petroleum, other 
petroleum products, and natural gas (including liquid petroleum gas).

[End of figure]

As shown earlier, mergers involving transportation assets in the 
midstream segment accounted for a small percentage of the total merger 
transactions in the industry. About 65 percent of the midstream merger 
transactions were asset mergers.

The mergers varied widely in terms of transaction values,but the 
highest value mergers were corporate mergers. Our merger data included 
transaction values for about 57 percent of the mergers, and those 
values ranged from less than $1 million to over $10 billion. As 
indicated in chapter 1, under the Hart-Scott-Rodino Act, firms 
contemplating mergers with a transaction valued at $50 million or more 
are required to provide information to FTC and the Department of 
Justice and to observe a waiting period before completing the 
transaction while it is reviewed for potential anticompetitive effects. 
FTC reviews required some petroleum companies that merged to divest 
assets to remedy potential anticompetitive effects.

As figure 10 shows, the majority of the reported transaction values 
were below $50 million, and over 89 percent of these mergers were asset 
transactions. Of the mergers with reported transactions values, about 
32 percent of them exceeded $50 million, and about 3 percent were over 
$1 billion. The latter accounted for over 83 percent of the total 
dollar value reported for all petroleum mergers during the past decade.

Figure 10: Range of Reported Merger Transaction Values (1991-2000): 

[See PDF for image]

[End of figure]

Several Reasons Were Cited for Mergers in the Petroleum Industry: 

Petroleum industry officials and experts that we spoke with cited a 
number of reasons for the wave of mergers in the industry in the 1990s. 
These reasons generally related to the need for increased efficiency 
and cost savings to ultimately maximize profits. Specifically, as 
discussed below, the officials and experts said that mergers were 
motivated by the firms' desire to achieve synergies, diversify their 
assets, reduce costs, enhance stock values, and respond to price 
volatility.[Footnote 20] However, economic literature also indicates 
that the desire to enhance and use market power--as a means to help 
maximize profits--may also have been a motive.

Achieving Synergies: 

Many oil industry officials indicated that achieving synergies--
benefits from the combined strengths of different companies--was an 
important motivation for some of the mergers in the industry in the 
1990s. Firms that engage in different but complementary activities may 
achieve synergies from mergers because it is more efficient and less 
costly for one company to perform two related activities than for two 
specialized firms to perform them separately. Furthermore, economic 
literature states that mergers can create synergies that improve firms' 
growth potential by yielding scale economies in production, marketing, 
research and development, and management, among other things. We found 
several instances of mergers where company officials cited synergies or 
complementary activities as a factor for the transactions. These 
mergers include Marathon Ashland's acquisition of Ultramar Diamond 
Shamrock's Michigan terminals, jobber networks, convenience stores, and 
pipelines in 1999; Sunoco's acquisition of crude oil transportation and 
marketing business assets from Pride Refining in 1999; and Tesoro's 
acquisition of BP Amoco's West Coast marine fuels operations in 1999.

Diversifying Assets: 

According to industry officials, the need for firms to diversify their 
portfolios in order to maintain stable profits played a role in 
petroleum industry mergers. Officials cited the acquisition of natural 
gas assets as a reason for mergers. Within the upstream segment, most 
independent exploration and production firms have both oil and gas in 
their portfolios because crude oil and natural gas are generally 
produced jointly. However, in the 1990s, some companies sought to 
increase their natural gas reserves through acquisition. For example, 
in 1999, Dominion acquired Remington Energy, Ltd., a natural gas 
production and exploration company, and increased its natural gas 
reserves to one trillion cubic feet. For a producer, natural gas could 
become a cushion during periods of low oil prices, all things being 
equal, allowing the producer to develop and produce more gas when oil 
prices are low, and vice versa. Moreover, EIA has reported that natural 
gas demand is likely to increase in coming years due to its relatively 
clean-burning qualities in comparison with other fossil fuels.

Within the downstream segment, some independent refiners acquired 
marketing and retail assets to expand their presence in U.S. retail 
markets. For example, Tosco's acquisition of Unocal's West Coast 
refining, marketing, and transportation assets allowed Tosco to 
diversify into retail operations on the West Coast.

Reducing Costs: 

Industry officials said that some mergers occurred as part of efforts 
to cut costs. Petroleum companies generally view each activity--such as 
exploration and production, refining, wholesaling, and retailing--as an 
individual "profit center." As a prudent business practice, petroleum 
firms assess the performance of each profit center relative to their 
overall business to determine where they could reduce costs or improve 
efficiency by acquiring or divesting assets. For example, one industry 
official said that mergers occurred frequently in the upstream segment 
partly because it is more cost effective and less risky to buy existing 
reserve assets than to discover new ones. Industry officials also told 
us that some firms divested refineries partly because of high operating 
costs and low returns. For companies acquiring these refineries, it was 
more cost effective to acquire an existing refinery than to build one, 
especially given the high cost and stringent environmental requirements 
for refinery construction in the United States.

Enhancing Stock Values: 

Some industry officials said that mergers, especially those involving 
publicly traded companies, were also partly motivated by the need to 
enhance stock values. The value of a company's common stock depends on 
investor expectations regarding its future profits. According to one 
industry official, the technology-fueled stock market boom of the 1990s 
heightened investor expectations for firms to consistently generate 
high stock appreciation. Thus, like other so-called old economy 
sectors, the petroleum industry was under pressure to meet Wall 
Street's expectations for rapid growth. Mergers were seen as a quick 
strategy for achieving this growth. Industry officials also believe 
that companies used mergers as a growth strategy to facilitate access 
to the capital markets, which seemingly favored bigger companies.

Responding to Price Volatility: 

Some industry officials believe that the large number of mergers that 
occurred in the second half of the 1990s may have been related, in 
part, to increased oil price volatility. According to one industry 
official, the collapse in crude oil prices, which dropped from $18.46 
per barrel in 1996 to $10.87 in 1998,dried up access to capital and 
made long-term investment difficult, especially for small firms. As a 
result, some high-cost producers became financially distressed, making 
them valuable yet inexpensive takeover targets.

Enhancing Market Power: 

While many of the reasons that industry officials cited for mergers are 
broadly consistent with achieving efficiency, economic literature also 
cites companies' desire to enhance market power as a motive for some 
mergers.[Footnote 21] As described in the literature, mergers increase 
market concentration and could reduce competition, allowing companies 
to exert greater control over prices. However, while mergers raise 
concern about potential anticompetitive effects,as stated in the 
previous chapter, U.S. antitrust laws are intended to mitigate such 
effects. Chapter 3 of this report examines in more detail the 
relationship between mergers and market concentration and other aspects 
of market structure that can affect competition in the U.S. petroleum 
industry.

[End of section]

Chapter 3: Mergers Contributed to Increases in Market Concentration and 
Other Changes in Market Structure: 

Mergers contributed to substantial increases in market concentration--
the extent to which a small number of firms controls most of an 
industry's sales--in the downstream segment of the U.S. petroleum 
industry, while concentration in the upstream segment changed very 
little by the end of the 1990s. Within the downstream segment, the 
increases were most significant in refining and wholesale gasoline 
markets. The overall impact of mergers is less clear for other aspects 
of petroleum market structure that also affect competition--in 
particular, vertical integration (the extent to which the same firms 
own the various stages of production and marketing of a product) and 
entry barriers (market conditions that provide established sellers in 
an industry an advantage over potential entrants). However, anecdotal 
evidence and economic studies indicate that mergers have affected these 
aspects as well.

Market Concentration Increased Mostly in the Downstream Segment of the 
Petroleum Industry During the 1990s: 

While market concentration in the upstream segment changed very little, 
the downstream segment of the petroleum industry experienced increases 
in concentration by the end of the 1990s that were largely associated 
with mergers during that period.[Footnote 22] Although mergers also 
occurred in the midstream segment, we could not determine the extent of 
midstream concentration during this period.[Footnote 23]

Analyzing Market Concentration in Relation to Mergers: 

Increased market concentration can result in greater market power, 
potentially increasing prices above competitive levels.[Footnote 24] 
Economists have posited that the extent of market power in a given 
market is directly and positively related to the degree of market 
concentration as measured by the Herfindahl-Hirschman Index (HHI) of 
that market, other things held constant. This index, as discussed 
earlier, reflects the composition of a market while giving 
proportionately greater weight to the market shares of the larger 
firms.[Footnote 25] On the other hand, increased concentration may also 
lead to cost savings and efficiency gains, which may be passed on to 
consumers in lower prices. Ultimately, the impact of higher 
concentration on prices depends on whether market power or efficiency 
dominates. (The effects of mergers and market concentration on 
wholesale gasoline prices are analyzed in chapter 5.): 

Economists and federal antitrust agencies have identified mergers as a 
major factor leading to higher market concentration. For example, in a 
1989 study on mergers in the petroleum industry, FTC reported that 
mergers and acquisitions, as well as other factors, affected changes in 
concentration in the petroleum industry. DOJ and FTC pay close 
attention to market concentration when reviewing proposed mergers. In 
the DOJ/FTC 1992 Horizontal Merger Guidelines[Footnote 26] for 
determining the potential anticompetitive effect of a proposed merger 
and whether to challenge such a merger, a central analysis is to assess 
whether the merger would significantly increase market concentration, 
as measured by the HHI, after defining the relevant geographic and 
product market. We based our analysis of market concentration in the 
various segments of the U.S. petroleum industry on the HHI criteria 
established by the guidelines. These guidelines, previously outlined in 
chapter 1, are summarized in table 1.

Table 1: FTC/DOJ Horizontal Merger Guidelines on the General Standards 
for Evaluating Postmerger Market Concentration: 

Postmerger HHI: HHI less than 1,000; 
Degree of market concentration: Unconcentrated; 
Change in HHI that would result from the proposed merger: 
Not applicable; 
Potential competitive consequences and likely need for 
further DOJ/FTC analysis: Mergers in this category require no further 
analysis.

Postmerger HHI: HHI between 1,000 and 1,800; 
Degree of market concentration: Moderately concentrated; 
Change in HHI that would result from the proposed merger: 
HHI increase <100; 
Potential competitive consequences and likely need for 
further DOJ/FTC analysis: No further analysis.

Postmerger HHI: HHI between 1,000 and 1,800; 
Degree of market concentration: Moderately concentrated;
Change in HHI that would result from the proposed merger: 
HHI increase > 100; 
Potential competitive consequences and likely need for 
further DOJ/FTC analysis: Could raise significant competitive concerns, 
depending on other factors.

Postmerger HHI: HHI greater than 1,800; 
Degree of market concentration: Highly concentrated; 
Change in HHI that would result from the proposed merger: 
HHI increase < 50; 
Potential competitive consequences and likely need for 
further DOJ/FTC analysis: No further analysis.

Postmerger HHI: HHI greater than 1,800; 
Degree of market concentration: Highly concentrated; 
Change in HHI that would result from the proposed merger: 
HHI increase > 50; 
Potential competitive consequences and likely need for 
further DOJ/FTC analysis: Could raise significant competitive concerns, 
depending on other factors.

Postmerger HHI: HHI greater than 1,800; 
Degree of market concentration: Highly concentrated; 
Change in HHI that would result from the proposed merger: 
HHI increase > 100; 
Potential competitive consequences and likely need for 
further DOJ/FTC analysis: Likely to create or enhance market power or 
facilitate its exercise. 

Sources: FTC and DOJ.

[End of table]

As table 1 shows, the guidelines establish market concentration into 
three broad categories of market concentration as measured by the HHI: 
an unconcentrated market has an HHI less than 1,000, a moderately 
concentrated market has an HHI between 1,000 and 1,800, and a highly 
concentrated market has an HHI over 1,800. Along with the level of HHI, 
the agencies also consider changes in HHI that would result from a 
proposed merger. In order to examine market concentration and any 
changes in the proper market context, the guidelines stipulate that the 
relevant geographic and product markets be defined on a case-by-case 
basis. For example, firms selling a given product may compete at the 
national level--in which case the relevant geographic market is 
national--while firms selling another product may compete at less than 
the national level--in which case the relevant geographic market could 
be regional, statewide, or smaller.

In analyzing market concentration, we based our choice of relevant 
geographic markets on various criteria, including what FTC officials 
and industry experts told us. We also based our choice of relevant 
market on the availability of data.

For the U.S. petroleum upstream segment, we analyzed market 
concentration, as measured by HHI,[Footnote 27] at the national level. 
For the downstream segment, we examined market concentration separately 
for refining and wholesale gasoline marketing, focusing our HHI 
analyses for refining at the regional (or the Petroleum Administration 
for Defense Districts or PADD) level and, for wholesale gasoline 
marketing, at the state level.[Footnote 28]'[Footnote 29] Figure 11 
depicts the U.S. PADDs and the states within each PADD.

Figure 11: Petroleum Administration for Defense Districts: 

[See PDF for image]

[End of figure]

To determine the extent to which mergers were associated with increased 
market concentration in the U.S. petroleum industry in the 1990s, we 
performed statistical correlation analyses. Correlation numbers (or 
coefficients), which range from -1 to +1, measure the strength and 
direction of the relationship between two variables.[Footnote 30] A 
positive number denotes a positive and direct relationship, while a 
negative number denotes a negative or inverse relationship. Overall, 
the higher the number, the stronger the relationship between the two 
variables being analyzed. (See appendix III for a more detailed 
discussion of our correlation analysis). For our analysis, we used as a 
surrogate for the level of merger activity the average transaction 
value of all the mergers for which such values were reported.[Footnote 
31] We correlated this value with the HHI for the upstream segment, 
refining (at the PADD level), and the wholesale gasoline market (at the 
state level).[Footnote 32] Other factors besides mergers that can 
affect market concentration include firms entering and exiting the 
industry. For example, if a company withdraws from a market and is not 
replaced by a new company, both the market shares of the remaining 
firms and concentration would increase.[Footnote 33]

The Upstream Segment Experienced Little Change in Market Concentration 
and Remained Unconcentrated Over the 1990s: 

Based on crude oil production activities, concentration in the upstream 
segment of the U.S. petroleum industry experienced little change over 
the decade. Specifically, the HHI for the upstream market decreased 
somewhat from 290 in 1990 to 217 in 2000 (see figure 12). Hence, the 
upstream segment of the U.S. petroleum industry remained unconcentrated 
as of the year 2000. Moreover, notwithstanding the level of domestic 
upstream concentration, industry officials and experts believe that 
because crude oil prices are generally determined in the world market, 
individual U.S. companies are not likely to have much influence on the 
global market.

Figure 12: Market Concentration for the Upstream Segment, as Measured 
by the HHI (1990-2000): 

[See PDF for image]

[End of figure]

For the upstream market, we did not find a statistically significant 
correlation between mergers in the 1990s and market concentration, as 
measured by the HHI, for U.S. crude oil production.

Overall, the Downstream Segment of the Market Became More Concentrated: 

In general, the downstream segment--consisting of the refining, 
wholesale, and retail marketing levels--became more concentrated in the 
1990s. However, the extent to which concentration increased varied 
among operating levels and geographic regions.

Refining: 

Overall, the U.S. refining market experienced increasing levels of 
market concentration (based on refinery capacity) during the 1990s, 
especially during the latter part of the decade, but the levels as well 
as the changes of concentration varied geographically.

In PADD I--the East Coast--the HHI for the refining market increased 
from 1136 in 1990 to 1819 in 2000, an increase of 683 (see figure 13). 
Consequently, this market went from moderately concentrated to highly 
concentrated. Compared to other U.S. PADDs, a greater share of the 
gasoline consumed in PADD I comes from other supply sources--mostly 
from PADD III and imports--than within the PADD. Consequently, some 
industry officials and experts believe that the competitive impact of 
increased refiner concentration within the PADD could be 
mitigated.[Footnote 34]

Figure 13: Refining Market Concentration for PADD I Based on Crude Oil 
Distillation Capacity (1990-2000): 

[See PDF for image]

Note: Data for 1996 and 1998 were unavailable.

[End of figure]

For PADD II (the Midwest), the refinery market concentration increased 
from 699 to 980 --an increase of 281--between 1990 and 2000. However, 
as figure 14 shows, this PADD's refining market remained unconcentrated 
at the end of the decade. According to EIA's data, as of 2001, the 
quantity of gasoline refined in PADD II was slightly less than the 
quantity consumed within the PADD.

Figure 14: Refining Market Concentration for PADD II Based on Crude Oil 
Distillation Capacity (1990-2000): 

[See PDF for image]

Note: Data for 1996 and 1998 were unavailable.

[End of figure]

The refining market in PADD III (the Gulf Coast), like PADD II, was 
unconcentrated as of the end of 2000, although its HHI increased by 
170--from 534 in 1990 to 704 in 2000 (see figure 15). According to 
EIA's data, much more gasoline is refined in PADD III than is consumed 
within the PADD, making PADD III the largest net exporter of gasoline 
to other parts of the United States.

Figure 15: Refining Market Concentration for PADD III Based on Crude 
Oil Distillation Capacity (1990-2000): 

[See PDF for image]

Note: Data for 1996 and 1998 were unavailable.

[End of figure]

The HHI for the refining market in PADD IV--the Rocky Mountain region-
-where gasoline production and consumption are almost balanced--
increased by 95 between 1990 and 2000. This increase changed the PADD's 
refining market from 1029 in 1990 to 1124 in 2000, within the moderate 
level of market concentration (see figure 16).

Figure 16: Refining Market Concentration for PADD IV Based on Crude Oil 
Distillation Capacity (1990-2000): 

[See PDF for image]

Note: Data for 1996 and 1998 were unavailable.

[End of figure]

The refining market's HHI for PADD V--the West Coast--increased from 
937 to 1267, an increase of 330, between 1990 and 2000 and changed the 
West Coast refining market, which produces most of the gasoline it 
consumes, from unconcentrated to moderately concentrated by the end of 
the decade (see figure 17).[Footnote 35]

Figure 17: Refining Market Concentration for PADD V Based on Crude Oil 
Distillation Capacity (1990-2000): 

[See PDF for image]

Note: Data for 1996 and 1998 were unavailable.

[End of figure]

We estimated a high and statistically significant degree of correlation 
between merger activity and the HHIs for refining in PADDs I, II, and V 
for 1991 through 2000. Specifically, the corresponding correlation 
numbers are 91 percent for PADD V (West Coast), 93 percent for PADD II 
(Midwest), and 80 percent for PADD I (East Coast). While mergers were 
positively correlated with refining HHIs in PADDs III and IV--the Gulf 
Coast and the Rocky Mountains--the estimated correlations were not 
statistically significant. (See table 11 in appendix III for 
correlation coefficients and associated statistics for each of the 
PADDs.): 

Wholesale Gasoline: 

The overall U.S. wholesale gasoline market--measured at the state 
level[Footnote 36]--also experienced significant increases in and 
higher levels of concentration, based on HHI data for wholesale 
gasoline from 1994 to 2002 that we obtained from the Department of 
Energy's Energy Information Administration (EIA).[Footnote 37] We found 
that all but four states and the District of Columbia experienced 
increases in wholesale gasoline market concentration between 1994 and 
2002. (See table 10, app. III.) Forty-six states and the District of 
Columbia had moderately or highly concentrated wholesale gasoline 
markets in 2002, compared to 27 in 1994. For the years 1994, 2000, and 
2002, figure 18 illustrates how the percentage of states categorized as 
unconcentrated has fallen while the percentage of states categorized as 
moderately to highly concentrated has risen. Specifically, the 
proportion of states categorized as unconcentrated has decreased from 
47 percent to 8 percent, while the percentage of states in the moderate 
category has risen from 43 percent to 75 percent. The percentage of 
states in the highly concentrated category has risen from 10 percent to 
18 percent.

Figure 18: Percentage of U.S. States with Unconcentrated, Moderately 
Concentrated, and Highly Concentrated Wholesale Gasoline Markets (1994, 
2000, and 2002): 

[See PDF for image]

[End of figure]

To determine the degree to which mergers and market concentration in 
wholesale gasoline were related and how closely they moved together 
during this period, we performed a correlation analysis for this 
operating level. We found that mergers, as measured by their 
transaction values, were significantly and highly positively correlated 
with market concentration, as measured by the state HHI, for wholesale 
gasoline. (See table 12, appendix III for the correlation coefficients 
and associated statistics for individual states.)[Footnote 38] This was 
especially the case for states that exhibited high levels of 
concentration or experienced large changes in concentration between 
1994 and 2001.

Figure 19 shows a comparison of concentration levels in individual 
states and the District of Columbia--grouped within PADDs--between 1994 
and 2002.

Figure 19: Wholesale Gasoline Market Concentration by State in Each 
PADD (1994 and 2002): 

[See PDF for image]

[End of figure]

* As can be observed, the wholesale gasoline market in 16 states in 
PADD I (the East Coast) were moderately concentrated in 2002, compared 
to 7 states in 1994. Also, in PADD I, the number of states that had 
unconcentrated wholesale gasoline markets decreased from 10 in 1994 to 
just 1 in 2002. Some key mergers that affected PADD I during this 
period include Exxon-Mobil, BP-Amoco, and Shell-Texaco (Motiva).

* In PADD II (the Midwest) the wholesale gasoline markets in 5 states 
were highly concentrated, 8 were moderately concentrated, and 2 were 
unconcentrated as of 2002. By comparison, in 1994, there were no highly 
concentrated markets, 7 states were moderately concentrated, and 8 
states were unconcentrated in this PADD. Some key mergers that affected 
PADD II during the period included Marathon-Ashland, Marathon-Ultramar 
Diamond Shamrock (UDS), BP-Amoco, Shell-Texaco (Equilon), and UDS-
Total.

* The wholesale gasoline market in all the states in PADD III (the Gulf 
Coast region) except one had become moderately concentrated in 2002, 
compared to 1994 when all were unconcentrated. Key mergers that 
affected PADD III during the period include Exxon-Mobil, Shell Texaco 
(Motiva), Marathon-Ashland, and Valero-UDS.

* For the states included in PADDs IV and V (the Rocky Mountains and 
the West Coast, respectively), wholesale gasoline markets remained in 
the moderately or highly concentrated range in 2002 as in 1994. Within 
this range, concentration levels increased in all but one state in PADD 
IV and in all but one state in PADD V between 1994 and 2002. Key 
mergers that affected PADD IV during this period include Shell-Texaco 
(Equilon), Phillips-Tosco, Conoco-Phillips, and UDS-Total. Key mergers 
that affected PADD V during the period included Tosco-Unocal, Shell-
Texaco (Equilon), Chevron-Texaco, Phillips-Tosco, and Valero-UDS.

Mergers Have Caused Changes in Other Aspects of Market Structure, but 
the Extent of These Changes Is Not Easily Quantifiable: 

Evidence from various sources suggests that in addition to market 
concentration, mergers affected other aspects of market structure--in 
particular, vertical integration and barriers to entry. The extent to 
which they did so, however, could not be easily quantified because, in 
addition to lack of consensus on how to appropriately measure these 
aspects, there are no comprehensive data on them.

Vertical Integration: 

Like increased concentration, increased vertical integration, as 
measured by the extent to which the various stages of production and 
marketing of a product are owned by the same firms, could conceptually 
have both procompetitive and anticompetitive effects, with the net 
effect depending on which effects dominate. One procompetitive view of 
vertical integration is that it promotes efficiencies and leads to 
lower prices by allowing a company to lower costs by making 
transactions that are internal rather than external to the company. On 
the other hand, a high degree of vertical integration in an industry 
could be anticompetitive by creating disincentives for new firms to 
enter a market because of the need to enter at several levels of the 
market in order to compete effectively. Vertical integration could also 
allow firms to use a strategy of "market foreclosure" against their 
non-vertically-integrated rivals by reducing input supply for rivals, 
raising prices paid by rival retailers, or totally refusing to sell 
product to rival retailers. Some studies have recently found that 
increased vertical integration in the U.S. petroleum industry has been 
associated with higher wholesale gasoline prices.[Footnote 39]

While our review was not comprehensive, we found that a number of the 
mergers since the 1990s led to greater vertical integration in the U.S. 
petroleum industry, especially in the downstream market, as shown in 
table 2. EIA has also reported that a substantial number of vertical 
mergers have occurred between independent refiners and marketers in the 
United States since the 1990s.[Footnote 40] Table 2 presents some 
examples of petroleum industry mergers since the mid-1990s that created 
or enhanced vertical integration.

Table 2: Selected Vertical Mergers in the Petroleum Industry Since the 
1990s: 

Year: 1995; 
Acquiring company: Diamond Shamrock; 
Stage of operation: Refining; 
Company acquired: Stop-N-Go; 
Stage of operation for assets purchased: Gasoline retailing.

Year: 1996; 
Acquiring company: Tosco; 
Stage of operation: Refining; 
Company acquired: Circle K; 
Stage of operation for assets purchased: Gasoline retailing.

Year: 1997; 
Acquiring company: Tosco; 
Stage of operation: Refining; 
Company acquired: Unocal Corporation; 
Stage of operation for assets purchased: Refining/marketing/ retail.

Year: 1997; 
Acquiring company: ARCO; 
Stage of operation: Integrated; 
Company acquired: Thrifty; 
Stage of operation for assets purchased: Gasoline retailing in 
California.

Year: 1998; 
Acquiring company: Shell; (Joint Venture); 
Stage of operation: Integrated; (with small downstream market share); 
Company acquired: Texaco; 
Stage of operation for assets purchased: Integrated; (with large 
downstream market share).

Year: 2000; 
Acquiring company: Tosco; 
Stage of operation: Refining; 
Company acquired: Some of Exxon's and Mobil's East Coast retail 
gasoline stations; 
Stage of operation for assets purchased: Retail gasoline stations.

Year: 2001; 
Acquiring company: Phillips; 
Stage of operation: Integrated; 
Company acquired: Tosco; 
Stage of operation for assets purchased: Refining/marketing/retail.

Year: 2001; 
Acquiring company: Valero; 
Stage of operation: Refining; 
Company acquired: Ultramar Diamond Shamrock; 
Stage of operation for assets purchased: Refining/marketing/retail. 

Source: GAO.

[End of table]

Typically, firms in the petroleum industry are either fully vertically 
integrated--operating across the entire industry spectrum from crude 
production to retail gasoline sales--or partially vertically 
integrated--operating in more than one but not all stages of the 
petroleum industry's operation. We included in our analysis mergers 
that have led to either type of vertical integration. Also, we have 
included in our analysis mergers that have enhanced the degree of 
vertical integration in the market--even if the mergers were 
essentially horizontal--such as the acquisition of an independent 
refiner by an already partially or fully vertically integrated company. 
Our analysis of mergers encompassed all these types of vertical 
integration because they all can affect competition in the market.

As shown in table 2, many mergers that contributed to increased 
vertical integration occurred between independents as well as between 
fully vertically integrated companies and independents.[Footnote 41] 
For example, Tosco, a previously independent refiner that had no retail 
operation, acquired several retail assets on the West Coast, such as 
Circle K (a retail chain) and Unocal's retail stations and other 
downstream assets.[Footnote 42] These acquisitions essentially 
transformed Tosco into a partially vertically integrated downstream 
company before Phillips Petroleum, a fully vertically integrated 
company, acquired it. This acquisition most likely boosted Phillips' 
downstream position in both refining and wholesale and retail 
marketing. Also, the acquisition of Thrifty, an independent chain 
retailer on the West Coast, by ARCO, an integrated company, enhanced 
the latter's retail position in the West Coast retail market. Likewise, 
the acquisition of UDS' wholesale gasoline terminals and retail outlets 
in Michigan by Marathon Ashland Petroleum--a joint venture between 
Marathon and Ashland, which are both fully vertically integrated oil 
companies--enhanced Marathon Ashland Petroleum's position in 
Michigan's wholesale and retail market.

Barriers to Entry: 

Our interviews with petroleum industry officials and experts provided 
anecdotal evidence that mergers have had some impact on barriers to 
entry in the U.S. petroleum industry, but there are generally no 
empirical data to quantify the extent of the impact. Barriers to entry 
can be defined as market conditions that provide established sellers in 
an industry an advantage (typically cost advantage) over potential 
entrants. Entry barriers are important in a market because of their 
effect on competitive conditions; theoretically, industries that are 
highly concentrated and have high entry barriers are more likely to 
possess market power. Industry officials that we interviewed indicated 
that large investment capital requirements, regulatory impediments/
environmental concerns, and public opposition to siting facilities 
constitute significant entry barriers that may have been exacerbated by 
mergers.

For example, industry officials told us that in the upstream segment, 
crude oil exploration and production activities moved increasingly 
offshore to areas such as the deep waters of the Gulf of Mexico during 
the 1990s because of the greater likelihood of finding oil. Offshore 
operations are generally riskier and require much higher capital 
investments than onshore operations.[Footnote 43] One official 
estimated that it could cost a company about $40 million to $100 
million just to drill several wells in deep waters and purchase 
equipment, and some operations could cost as much as $1 billion. As a 
result, some firms, mostly large producers that already had the 
wherewithal to engage in offshore activities, merged to further share 
the risks and costs. These mergers tended to help consolidate their 
dominance in offshore activities and made it more difficult for smaller 
firms to enter the market.

For the transportation infrastructure segment--pipelines--the 
potential barriers to entry include high investment costs and large 
economies of scale.[Footnote 44] Moreover, as noted by one source, 
procedural requirements and associated legal costs for entry into the 
pipeline business have limited the number of companies in the segment.
[Footnote 45] Thus, as mergers, and possibly concentration, increased, 
entry barriers also increased because firms must make large and high-
cost investments in order to enter the market and be competitive at 
large scales of operation.

Like the upstream and midstream segments, the downstream segment of the 
U.S. petroleum industry is characterized by pervasive barriers to 
entry, including large capital investment requirements at the refining 
level, and regulatory and permitting impediments at the refining and 
wholesale/retail levels. For example, regarding refining, industry 
officials told us that building a typical refinery or even upgrading an 
existing one is a multibillion dollar investment. Also, they said that 
it is extremely difficult to obtain a permit from the relevant state or 
local authorities to build a new refinery in many parts of the country 
because of regulatory hurdles and public opposition. In addition, they 
noted that federal and state environmental regulations to meet clean 
air requirements have contributed to the high cost of owning and 
operating a refinery. Furthermore, they pointed out that return on 
investment in refining has been relatively low compared to investment 
in other industries. They attributed the failure to build any new 
refineries in the United States in over 20 years to these factors.

We could not quantify the extent to which mergers may have increased or 
decreased these barriers because of the lack of empirical data to 
properly measure entry barriers. Industry officials said that mergers 
have not caused these barriers. Instead, they opined that some of the 
mergers and acquisitions in refining have been partly a result of these 
barriers because merging with or acquiring existing refineries is less 
expensive than building a new one. During the 1990s, many refiners 
expanded through mergers and acquisitions as well as through upgrading 
existing facilities. For example, refiners such as Tosco and Tesoro 
entered the industry through acquisitions in the early 1990s.

Entry barriers also exist at the wholesale gasoline marketing level of 
the downstream segment in the form of high investment capital 
requirements, regulatory/permitting impediments, and infrastructure 
barriers. For example, a potential entrant into the wholesale gasoline 
supply market may enter by operating his own refinery and producing 
gasoline and/or buying from existing domestic refiners or importing 
gasoline for distribution. As a potential refiner, he faces the entry 
barriers in refining discussed above. On the other hand, industry 
officials told us that while it is possible to enter this market as an 
independent purchaser from domestic refiners and/or importers, there 
are potential infrastructure impediments to doing so, such as lack of 
access to pipelines and terminals. They pointed out that although 
shipping gasoline through a third-party, common carrier pipeline 
operator such as Kinder Morgan offers an option in some markets, this 
option may not be available in the particular market that the shipper 
wants to bring gasoline into. Moreover, to ship gasoline through such a 
common-carrier pipeline, the shipper must have access to a terminal on 
that route to receive the product, or the pipeline operator cannot 
accept such shipment. According to some industry officials, oil 
companies who own most of the gasoline terminals around the nation 
sometimes deny access to third-party users, especially when supply is 
tight. Some industry officials indicated that mergers have exacerbated 
entry barriers at the wholesale level in some markets because mergers 
have created a situation in which pipelines and terminals in some 
markets are owned by fewer, mostly integrated companies who use these 
facilities mostly proprietarily. In addition, industry officials 
pointed out that there has been a preference for larger distributors 
over smaller distributors in the market. For example, wholesale 
marketers or distributors need to be large enough to secure credit 
lines to make large volume purchases or minimum volume requirements set 
by refiners. Also, in some markets, such as California, boutique fuel 
specifications to meet clean air requirements limit the ability of 
potential independent wholesalers to enter the market because the 
unique gasoline blends are not widely produced in other refining 
centers.

At the retail level, industry officials pointed out that mergers have 
exacerbated the barriers for potential retail entrants because there 
are fewer companies to supply gasoline to retailers and, as discussed 
in more detail in chapter 4, retailers must operate at a large scale in 
order to meet minimum volume requirements preferred by refiners. They 
also indicated that restrictive land-use laws and permitting processes 
in some areas of the country, such as California and Washington, D.C., 
constitute a barrier for potential retailers seeking to build new 
stations.[Footnote 46]

[End of section]

Chapter 4: Gasoline Marketing Has Changed in Two Major Ways: 

According to industry officials, two major changes have occurred in 
gasoline marketing since the 1990s, partly related to mergers. First, 
the availability of generic (unbranded) gasoline has decreased for 
various reasons; more gasoline is now marketed as branded, under the 
refiner's trademark. Branded gasoline is generally higher priced than 
unbranded. We could not statistically quantify the extent of this 
change because no data on the supply of unbranded gasoline exist. 
Second, refiners now prefer dealing with large distributors and 
retailers. This preference, officials told us, has motivated further 
consolidation in both the distributor and retail sectors, including the 
rise of "hypermarkets"--a relatively new breed of gasoline market 
participants that include such large retail warehouses as Wal-Mart and 
Costco.

The Availability of Unbranded Gasoline Decreased: 

Refiners market either branded or unbranded gasoline through several 
wholesale channels, but since the 1990s the availability of unbranded 
gasoline from refiners has decreased substantially, according to 
industry officials. Officials generally attributed this decrease to a 
reduction in the number of independent refiners, the sale and/or 
mothballing of refineries by mostly fully vertically integrated oil 
companies, and better inventory management by major branded refiners. 
The decrease cannot be precisely quantified because the data are not 
adequate to do so.

Refiners Market Either Unbranded or Branded Gasoline through Several 
Channels: 

The gasoline market consists of various supply arrangements that 
ultimately influence gasoline prices throughout the supply chain. 
Gasoline flows through several marketing channels, as shown in figure 
20. The refiner can market gasoline to the consumer through a direct 
distribution system and an indirect distribution system.

Figure 20: The Flow of Gasoline Marketing: 

[See PDF for image]

[End of figure]

The direct system typically involves the sale and/or supply of branded 
gasoline by a refiner to its company-operated stations or other retail 
outlets operated by lessee dealers who lease the service station and 
basic equipment from the refiner or distributor but operate their own 
retail outlets. Branded gasoline is marketed under the refiner's 
trademark. Refiners can also sell unbranded gasoline directly to 
hypermarkets--including such large retail warehouses as Wal-Mart and 
Costco, as well as grocery store chains such as Safeway--that have over 
the last decade added gasoline retailing to their locations. (The role 
of hypermarkets is discussed later in this chapter.) Retailers of 
unbranded gasoline can sell it as a generic/private brand and tend to 
compete mostly through lower prices than their branded competitors. In 
the direct distribution system, these hypermarkets have, over the last 
decade, taken the place of open dealers who either own their own 
stations or lease them from distributors or third parties in the supply 
structure.

In the indirect distribution system, refiners sell branded or unbranded 
gasoline to independent middlemen--generally called distributors, 
marketers, or jobbers--who resell the gasoline to other retailers or 
sell to consumers through their own retail operations. Branded gasoline 
that flows through the indirect system must also be marketed by 
distributors or retailers under the refiner's trademark, while 
unbranded could be sold under the distributor's or retailer's private 
name. Many market participants told us that much of the gasoline sold 
through both the direct and indirect channels is now branded.

Depending on the type of supply arrangement with the supplier, gasoline 
distributors and retailers may pay one or more of the distinct 
wholesale prices summarized in table 3 below. Under normal market 
conditions, the spot price is the lowest wholesale price, followed by 
the unbranded rack price, branded rack price, and dealer-tankwagon 
price. Because, as discussed below, transfer prices are generally 
considered proprietary, it is not clear how high or low they are 
relative to the other prices.

Table 3: Types of Wholesale Prices Paid for Gasoline: 

Wholesale purchaser of gasoline: Distributor; 
Spot: Yes; 
Unbranded rack: Yes; 
Branded rack: Yes; 
Dealer-tankwagon: No; 
Transfer[A] price: No. 

Wholesale purchaser of gasoline: Company-operated outlet; 
Spot: No; 
Unbranded rack: No; 
Branded rack: No; 
Dealer-tankwagon: No; 
Transfer[A] price: Yes.

Wholesale purchaser of gasoline: Lessee dealer; 
Spot: No; 
Unbranded rack: No; 
Branded rack: No; 
Dealer-tankwagon: Yes; 
Transfer[A] price: No.

Wholesale purchaser of gasoline: Open dealer; 
Spot: No; 
Unbranded rack: Yes; 
Branded rack: Yes; 
Dealer-tankwagon: Yes; 
Transfer[A] price: No.

Wholesale purchaser of gasoline: Hypermarket; 
Spot: Yes; 
Unbranded rack: Yes; 
Branded rack: No; 
Dealer-tankwagon: No; 
Transfer[A] price: No.

Source: GAO.

[A] Transfer prices are internal prices at which refiners and 
distributors supply gasoline to their company-owned and -operated 
stations.

[End of table]

Spot Prices are generally the lowest wholesale price under normal 
market conditions because there is no binding contract between the 
seller and the buyer, and gasoline sold in the spot market is typically 
unbranded. Market participants typically use the spot market when faced 
with surpluses or shortages that may arise from their contractual 
transactions. The spot market accounts for only a small portion of 
domestic gasoline sales, even smaller than it was a decade ago, partly 
because just-in-time inventory management leaves less gasoline for spot 
sales. Nonetheless, spot prices, as well as futures prices, strongly 
influence the other wholesale prices.

Rack Prices are the prices that distributors and retailers pay for 
gasoline supplied at a refiner's wholesale terminal or rack. Typically, 
rack prices are set daily by refiners and are generally influenced by 
prices in the spot and futures markets, as well as by the extent of 
competition among refiners within a particular market. Average rack 
prices are generally higher than spot prices under normal market 
conditions. There are two types of rack prices--branded and unbranded.

* Branded rack prices are paid by distributors who buy gasoline 
supplies from major refiners selling under their trademarks. Branded 
rack prices include a premium reflecting the recognized brand name, the 
costs of issuing company credit cards, and other costs such as 
advertising. In addition, when refiners sell branded gasoline to 
distributors and retailers, the contracts tend to be less flexible than 
contracts for unbranded gasoline but guarantee a more secure supply. 
Thus, branded rack prices may also include a premium for this 
additional security.

* Unbranded rack prices are paid by distributors, hypermarkets, and 
open dealers for unbranded gasoline supplied primarily by independent 
refiners and, to a small extent, by fully vertically integrated 
refiners. Under normal market conditions, unbranded rack prices tend to 
be lower than branded rack prices. Buyers of unbranded gasoline may or 
may not have a binding contractual arrangement with a refiner.[Footnote 
47] Therefore, a buyer of unbranded gasoline may not be guaranteed a 
secure supply or lower prices, particularly during a market shock 
involving a reduction in overall gasoline supply. Thus, when there is a 
disruption in the supply system, such as those caused by pipeline or 
refinery breakdowns, unbranded rack prices can be higher than branded 
rack.[Footnote 48]

Dealer-tankwagon (DTW) prices are contract prices paid by lessee 
dealers and some open dealers to refiners or distributors for branded 
gasoline delivered at the dealers' stations. DTW prices, which are set 
by suppliers, include the cost of transporting the gasoline to the 
stations and a premium associated with the suppliers' brand name. 
Suppliers set their DTW prices using the futures and/or spot prices as 
a reference, as well as the DTW prices of other suppliers in the market 
area. In general, DTW prices are less volatile and higher than spot and 
rack prices.

Transfer Prices are internal prices at which refiners or distributors 
supply gasoline to their company-owned and -operated stations at the 
retail level. Oil companies generally regard their transfer prices as 
proprietary information and do not publicly disclose them. Several oil 
companies told us that transfer prices are based on market prices such 
the DTW, but we were unable to confirm this.

Based on data on gasoline sales reported to EIA by major U.S. energy 
companies under the EIA's Financial Reporting System (FRS),[Footnote 
49] about 46 percent of U.S. refiners' gasoline was marketed through 
distributors in 1990; this share increased to over 50 percent in 2000 
(see figure 21). Despite distributors' significant role in gasoline 
marketing, the distributors we interviewed stated that their marketing 
activities are generally confined to rural or less urban areas. 
Distributors said that refiners who supply them with branded gasoline 
preclude them from operating stations within certain proximities of 
major metropolitan markets where the refiners generally prefer to 
locate their company-owned and -operated and lessee dealer stations--a 
phenomenon the distributors described as "redlining." We did not 
explore the impact, if any, of this practice on the gasoline market 
because it is outside the scope of the present study.

Figure 21: Percentage Volume of Gasoline Sold through Different 
Marketing Channels: 

[See PDF for image]

[End of figure]

In 1990, refiners marketed about 31 percent of their gasoline through 
lessee dealers who pay DTW prices and open dealers who pay rack and/or 
DTW prices; this percentage declined to 26 percent in 2000.[Footnote 
50] Lessee dealers that we spoke with attributed their declining role 
as a marketing channel to high DTW and rent costs charged by their 
suppliers. The dealers also alleged that continued decline in their 
market participation could ultimately lead to reduced competition and 
higher gasoline price to consumers. Again, we did not attempt to 
further analyze these claims because they are not within the scope of 
our study.

The percentage of gasoline sold by refiners to consumers through 
company-operated stations remained virtually unchanged between 1990 and 
2000--16 percent and 13 percent, respectively.

Industry Officials Cited Several Reasons for the Decrease in the 
Availability of Unbranded Gasoline: 

Oil industry officials whom we interviewed indicated that the 
availability of unbranded gasoline has decreased since the 1990s. 
According to these officials, more branded gasoline is now sold at a 
price that is generally higher than that of unbranded gasoline, as 
discussed above. This premium is presumably justified because of 
certain additives in branded gasoline and consumer brand loyalty.

In general, industry officials cited one or more of the following 
reasons for the decrease in the availability of unbranded gasoline.

* Fewer independent refiners are supplying gasoline. Independent 
refiners generally supply unbranded gasoline, but since the 1990s their 
numbers have decreased as they merged with branded companies, grew 
large enough to be considered a brand, or closed down. For example, 
Tosco was one of the largest independent refiners selling unbranded 
gasoline in the United States. However, the company made several 
acquisitions--some involving purchases of retail stations from branded 
companies like British Petroleum--which allowed it to market some of 
its gasoline through branded outlets. Tosco also acquired downstream 
assets on the East Coast that were divested from Exxon and Mobil as a 
condition for their merger. These acquisitions allowed Tosco to market 
gasoline under the Exxon and Mobil brands under a consent agreement 
worked out with FTC and ExxonMobil. Moreover, in 2001 Tosco was 
acquired by Phillips Petroleum, a large branded refiner. According to 
some gasoline distributors who used to purchase unbranded gasoline from 
Tosco, their ability to purchase unbranded gasoline has decreased 
substantially because of Tosco's acquisition by Phillips. They said 
that Phillips now sells a greater share of its gasoline as branded so 
that they no longer have access to as much unbranded gasoline.

* Fully vertically integrated oil companies have decided to sell some 
refineries to independents or to mothball inefficient refineries. Fully 
vertically integrated oil companies have, in recent years, sold off or 
mothballed refineries they deemed to be unprofitable.[Footnote 51] As a 
result, some of them now have only enough refinery capacity to produce 
gasoline to meet their branded supply needs, while others said that 
they have even become net buyers of gasoline. Moreover, independent 
refiners, some of whom bought refineries from the fully vertically 
integrated oil companies, also sell a portion of their gasoline to 
these companies, further reducing the amount of gasoline that the 
independent refiners can sell to unbranded distributors and retailers.

* The major branded refiners have increased the efficiency of their 
inventory management systems. Some unbranded supply came from excess 
gasoline production. Synergies developed through mergers have increased 
the industry's ability to use just-in-time inventory management system, 
which ensures that refiners produce an amount of gasoline sufficient to 
meet their current branded needs without producing any excess that can 
be sold as unbranded. For example, officials from one large fully 
vertically integrated oil company told us that its refineries produce 
just enough gasoline to cover its company-operated stations and lessee 
dealer sales.

Data Are Not Adequate to Precisely Quantify the Decreasing Availability 
of Unbranded Gasoline: 

Although oil industry officials we interviewed overwhelmingly said that 
the supply of unbranded gasoline in the U.S. has decreased 
significantly in the 1990s, we could not statistically quantify this 
change because the data required for such an analysis do not currently 
exist. DOE's EIA is the federal agency mandated by Congress to collect 
energy data. EIA collects data on gasoline supply and prices, but EIA 
officials told us that the agency does not require petroleum companies 
to report gasoline data in the form that would permit the 
identification of branded and unbranded sales for two reasons: (1) the 
agency lacks the resources to properly track these data and (2) the 
industry has sued the agency on several occasions on the grounds that 
tracking this type of information was too burdensome. EIA, however, 
acknowledged that unbranded gasoline provides a low-cost competitive 
option for consumers. EIA also acknowledged that data on unbranded 
gasoline supply would facilitate better monitoring of the overall 
competitive trends in the gasoline market.

Refiners Prefer Dealing with Large Distributors and Retailers: 

Market participants that we spoke with told us that refiners now prefer 
dealing with large distributors and retailers for two reasons: (1) 
large distributors and retailers are a much lower credit risk than 
their smaller counterparts and (2) it is more efficient to sell a 
larger volume through fewer entities than to sell a smaller volume 
through many entities because minimizing the number of transactions 
reduces administrative and distribution costs. As mergers have occurred 
among refiners, fewer supply options exist for distributors. This 
consolidation at the refining level has allowed large refiners to 
dictate the terms of supply contracts, including minimum volume 
requirements. Partly in response, distributors are becoming larger 
through mergers and consolidation. In addition, hypermarkets, which 
often buy gasoline in large quantities from the refiners and so receive 
volume discounts on the unbranded rack price, are becoming major 
unbranded retailers.

Distributors Are Becoming Larger and Fewer in Several Markets: 

Distributors, through whom about half of all gasoline is sold, are 
themselves merging or entering into joint ventures with branded 
refiners to enlarge their scale of operation, which has ultimately led 
to a reduction in their number. For example, at the end of 2002, there 
were about 7,000 distributors and dealers who were members of the 
Petroleum Marketers Association of America (PMAA), compared to about 
10,000 in 1991. PMAA officials attributed this decline mostly to 
mergers and consolidations among their members. According to a PMAA 
official, the trend since the 1990s has been not only for large 
distributors to absorb smaller ones but for large ones to merge among 
themselves to enhance their competitive position. This pattern of 
consolidation has been particularly noted in some areas, such as parts 
of Colorado and Michigan. For example, one industry official told us 
that the number of distributors in some rural Colorado communities has 
decreased from about seven or eight distributors a decade ago to 
generally only one today. According to market participants, 
distributors have several incentives for consolidation. First, it gives 
the distributors the ability to meet minimum volume requirements. 
Second, it increases distributors' ability to negotiate volume 
discounts in supply contracts. Finally, by increasing their scales of 
operation, distributors can enhance their access to capital, allowing 
them greater flexibility in purchasing gasoline on credit.

Hypermarkets Are Becoming a Significant Player in U.S. Gasoline 
Marketing: 

Our interviews with oil industry officials and available data suggest 
that hypermarkets are playing a significant and growing role today in 
U.S. gasoline marketing to consumers. As noted above, hypermarkets 
generally buy directly from the refiner and typically deal in heavy 
volumes. For example, two hypermarkets that we spoke with reported that 
they sold about 420 million and 470 million gallons, respectively, of 
gasoline per year. This is comparable to the volume sold by some of the 
largest distributors we interviewed, whose sales volume ranged between 
200 million and 700 million gallons per year. Hence, the typical 
hypermarket fits the profile of the large wholesale purchasers that 
refiners now prefer to deal with. Hypermarkets purchase and sell almost 
entirely unbranded gasoline and are becoming a channel for the sale of 
the dwindling unbranded gasoline supply.[Footnote 52] Hence, they are 
rapidly displacing the "mom and pop" open dealers who used to dominate 
the unbranded retail market. These dealers are now either "branding 
up"[Footnote 53] or going out of business.

Although the overall market share of hypermarkets in U.S. gasoline 
marketing is currently relatively small, it is projected to grow very 
rapidly, at least in the short term. According to a study by Energy 
Analysts International, Inc. (EAI), a consulting firm that has analyzed 
hypermarkets, there were between 1,230 and 1,250 hypermarket locations 
selling gasoline in the U.S. in 2000, and these locations collectively 
sold over 4 billion gallons of gasoline, or 3.3 percent of total 
gasoline sales to consumers.[Footnote 54] Furthermore, EAI projects 
that by 2005, hypermarkets' gasoline sales will increase more than 
five-fold to 22.7 billion gallons, or 16 percent of gasoline sales to 
consumers.

In general, it appears that hypermarkets are gaining market share in 
gasoline retailing through an aggressive pricing strategy--on average, 
their pump prices are lower than those of their competitors--a 
situation that has raised concern among some of the traditional 
competitors, especially distributors. Many distributors contend that 
hypermarkets use their gasoline as a "loss leader" and subsidize 
gasoline sales with profits from store sales. For their part, 
hypermarkets told us that because they often buy in large volumes, they 
are able to negotiate and receive discounts on their unbranded rack 
price. Lower purchase prices allow them to set lower pump prices. 
However, if the supply of unbranded gasoline continues to dwindle 
because of the attrition, acquisition, and/or vertical integration of 
unbranded refiners, it is not clear how the hypermarkets will respond. 
The hypermarkets that we spoke with said that if they could not obtain 
an adequate supply of unbranded gasoline in the future, they would have 
to switch to branded gasoline. Some of them have considered purchasing 
refineries to produce their own gasoline and/or importing gasoline.

[End of section]

Chapter 5: Mergers and Increased Market Concentration Generally Led to 
Higher Wholesale Gasoline Prices in the United States: 

The results of our econometric analyses suggest that six of eight 
specific oil industry mergers--which mostly involved large, fully 
vertically integrated, companies--generally led to increases in 
wholesale gasoline prices (which in this report are measured by 
wholesale prices less crude oil prices) for branded and/or unbranded 
gasoline of about 2 cents per gallon, on average. Two of the mergers 
generally led to price decreases, of about 1 cent per gallon, on 
average. These findings imply that the combined effects of market power 
(which tends to increase prices) and efficiency gains (which tend to 
decrease prices) from the mergers led to increased prices. These 
findings applied to both conventional gasoline, the dominant type of 
gasoline sold nationwide, and to "boutique fuels"--gasoline that has 
been reformulated for certain geographical areas to mitigate 
environmental pollution.

In a complementary analysis, we found that increased market 
concentration, which captures the cumulative effects of mergers as well 
as other market structure factors, generally resulted in increased 
wholesale prices for conventional and boutique fuels. For conventional 
gasoline, the increases in prices were larger in the western half of 
the United States than in the eastern half, in part because the West 
has limited access to gasoline supplies from abroad and from the Gulf 
Coast region, which has high refinery capacity. For the boutique fuels-
-which are sold only in certain cities in the East Coast and Gulf Coast 
regions, or in California--increased market concentration led to higher 
wholesale prices than for conventional gasoline. This difference likely 
stems from the limited availability of the boutique fuels, which can 
only be produced by a few refiners. The changes in the wholesale prices 
can be attributed partly to the wave of mergers that reduced the number 
of suppliers in the affected geographic regions. Our results also 
suggest that lower gasoline inventories relative to expected demand, 
higher refinery capacity utilization rates, and supply disruptions in 
the Midwest and West Coast led to higher wholesale gasoline prices.

As part of our methodology to model the effects of mergers and market 
concentration, we used extensive peer review to obtain comments from 
outside experts and made changes as appropriate. However, due to the 
complexities of analyzing the effects of mergers and market 
concentration on wholesale gasoline prices, there are some limitations 
to our econometric methodology, including the time periods over which 
we could model the effects of the mergers and the market concentration 
data that we used.

Econometric Models Developed to Estimate the Effects of Mergers and 
Market Concentration on Wholesale Gasoline Prices: 

In developing our econometric models, we relied on information from 
previous studies, industry experts, and our own analysis of the oil 
industry, specifically wholesale gasoline markets. We developed two 
groups of econometric models to estimate the effects of individual 
mergers and increased market concentration on wholesale prices of 
different gasoline types--conventional, reformulated, and 
CARB[Footnote 55]--in the second half of the 1990s. To estimate effects 
on wholesale gasoline prices, we used wholesale prices minus crude oil 
prices,[Footnote 56] with crude oil prices serving as our proxy for 
marginal input costs (crude costs constitute about two-thirds of total 
refining costs).[Footnote 57] We focused our study on wholesale 
gasoline markets because trends in gasoline prices usually are observed 
in wholesale markets before the retail markets and because more 
comprehensive data on volumes were available at the wholesale level 
than the retail level.

For both models, we used panel data--data pooled across all cities 
where wholesale gasoline terminals or racks are located and over time. 
This enabled us to account for variations in prices across rack cities 
(city-specific effects) and over time (time effects). Also, for 
mergers, the panel data allowed us to estimate the effect on prices in 
the rack cities where the merging companies operated, relative to 
prices in rack cities where they did not, taking into account other 
variables. In addition to mergers, which are measured by indicator (or 
dummy) variables for consolidations between the merging companies, and 
market concentration, which is measured by the Herfindahl-Hirschman 
Index (HHI) of refinery capacity, we included in our models other 
relevant variables that could affect wholesale gasoline prices, such as 
gasoline inventories relative to demand and refinery capacity 
utilization rates, and supply disruptions.

There were supply disruptions that caused price spikes in the Midwest 
in 2000 and on the West Coast in 1999 and 2000. The immediate causes of 
the disruptions included refinery outages and pipeline ruptures and, in 
the case of the Midwest, changes in gasoline formulations. It is 
difficult to determine the timing, duration, and the extent of the 
geographical impact of the disruptions, all of which makes it difficult 
to construct reliable and accurate measures of the supply 
disruptions.[Footnote 58] Nonetheless, we constructed crude measures of 
these supply disruptions that we included in our models for the markets 
that were affected.

There are two common approaches for estimating panel data--the "random-
effects" model and the "fixed-effects" model. The random effects model 
is preferred when observations (rack cities) are drawn randomly from a 
common population and any difference in individual effects can only be 
attributed to chance. Otherwise, the fixed effects model is preferred. 
The selection of the rack cities used in our study was based on data 
availability and not random choice. Furthermore, in wholesale gasoline 
markets, unobserved city-specific differences might include unmeasured 
supply or demand effects such as different pricing strategies of the 
refiners at different rack cities and the level of development of the 
transportation system in the different areas. These differences are not 
random. We, therefore, prefer the fixed-effects model, which--unlike 
the random effect model--remains valid even when the unobserved city-
specific effects are not independent of the included explanatory 
variables.

Although we preferred to use wholesale gasoline prices minus crude oil 
prices as the dependent variable for economic and statistical reasons, 
as part of our sensitivity analysis, we reestimated the models with the 
crude oil prices as an explanatory variable. We found that these 
results were similar, but the explanatory power, as expected, increased 
significantly. Also, because some of the explanatory variables are 
likely to be determined simultaneously with gasoline prices 
(particularly gasoline inventories and refinery capacity utilization 
rates), we estimated our models taking this into account. Furthermore, 
it is likely that prices of wholesale gasoline would be correlated 
across nearby racks, partly due to spatial competition. Our estimation 
technique accounts for possible contemporaneous correlations across the 
racks. A complete discussion of our econometric approach, including 
model specifications, variables used, data sources, and estimation 
techniques, is provided in appendix IV.

Mergers in the Second Half of the 1990s Mostly Led to Increases in 
Wholesale Gasoline Prices: 

The results of our econometric modeling indicate that most of the 
individual mergers we examined led to increases in the prices of 
wholesale gasoline for the time periods we analyzed, while a smaller 
number of mergers led to price decreases. Overall, we examined eight 
mergers, shown in table 4, including the two largest in the petroleum 
industry in history--the BP-Amoco and Exxon-Mobil mergers.[Footnote 59] 
We selected these mergers because of their transaction size, FTC's 
review of them,[Footnote 60] or concerns expressed by some industry 
participants and state officials we interviewed about their potential 
anticompetitive effects.[Footnote 61]

Table 4: Selected Oil Industry Mergers Affecting Wholesale Gasoline 
Markets, 1994-2000: 

Merger[A]: Tosco-Unocal; 
Effective date of merger[B]: April 1, 1997[E]; 
Acquirer: Tosco; 
Target: Unocal; 
Relevant geographic region[C, D]: PADD V; 
Markets in which FTC identified competitive concerns: Not applicable.

Merger[A]: UDS-Total; 
Effective date of merger[B]: October 1, 1997[F]; 
Acquirer: UDS; 
Target: Total; 
Relevant geographic region[C, D]: PADD II, III, IV; 
Markets in which FTC identified competitive concerns: Not applicable.

Merger[A]: Marathon-Ashland; 
Effective date of merger[B]: January 5, 1998[E]; 
Acquirer: Joint venture; 
Target: Joint venture; 
Relevant geographic region[C, D]: PADD I, II, III; 
Markets in which FTC identified competitive concerns: Not applicable.

Merger[A]: Shell-Texaco I[G] (Equilon); 
Effective date of merger[B]: February 1, 1998[F]; 
Acquirer: Joint venture; 
Target: Joint venture; 
Relevant geographic region[C, D]: PADD II, III, IV, V; 
Markets in which FTC identified competitive concerns: Refining; 
Wholesale; Retail.

Merger[A]: Shell-Texaco II[G] (Motiva); 
Effective date of merger[B]: July 1, 1998[E]; 
Acquirer: Joint venture; 
Target: Joint venture; 
Relevant geographic region[C, D]: PADD I, II, III; 
Markets in which FTC identified competitive concerns: Pipelines.

Merger[A]: BP-Amoco[H]; 
Effective date of merger[B]: December 31, 1998[E]; 
Acquirer: BP; 
Target: Amoco; 
Relevant geographic region[C, D]: PADD I, II, III; 
Markets in which FTC identified competitive concerns: Wholesale.

Merger[A]: MAP-UDS; 
Effective date of merger[B]: December 13, 1999[E]; 
Acquirer: MAP; 
Target: UDS (Michigan); 
Relevant geographic region[C, D]: PADD II; 
Markets in which FTC identified competitive concerns: NA.

Merger[A]: Exxon-Mobil[H]; 
Effective date of merger[B]: March 1, 2000[F]; 
Acquirer: Exxon; 
Target: Mobil; 
Relevant geographic region[C, D]: PADD I, III; 
Markets in which FTC identified competitive concerns: Refining; 
Pipelines; Retail. 

Legend: 

UDS = Ultramar Diamond Shamrock; 
BP= British Petroleum; 
MAP = Marathon Ashland Petroleum: 

Sources: GAO's analysis of EIA, FTC, OPIS, and Thomson Financial data.

[A] The first company is the acquirer and the second company is the 
target.

[B] The effective dates are either the merger completion date or the 
date when FTC's merger remedies became effective.

[C] Both merging companies operated in rack cities located in these 
geographic regions.

[D] Traditionally, the United States has been divided into five 
Petroleum Administration for Defense Districts (PADD): PADD I, the East 
Coast region; PADD II, the Midwest region; PADD III, the Gulf Coast 
region; PADD IV, the Rocky Mountain region; and PADD V, the West Coast 
region. (See figure 11.): 

[E] The merger completion date.

[F] The date when FTC's merger remedies became effective. The merger 
completion date for the UDS-Total merger was September 25, 1997; for 
the Shell-Texaco I (Equilon) merger, January 23, 1998; and for the 
Exxon-Mobil merger, November 30, 1999.

[G] The Shell-Texaco II joint venture involved Shell, Texaco, and Star 
(jointly controlled by Texaco and Saudi Refining Company). The Shell-
Texaco mergers ended in 2000 when Chevron and Texaco merged to form 
Chevron-Texaco. Shell then acquired the Shell-Texaco assets as a 
condition by FTC to allow the Chevron-Texaco merger.

[H] This merger also involved upstream assets of these companies, but 
our modeling focused on the effects at the wholesale gasoline level.

[End of table]

In tables 5-7 we present the effects of the mergers we modeled on 
wholesale prices of conventional gasoline, reformulated gasoline, and 
CARB gasoline, respectively. The tables show (1) the geographic regions 
that the mergers affected and (2) the estimated changes in wholesale 
gasoline prices associated with each merger in the relevant geographic 
areas.

Our estimates in table 5 show that of the seven mergers we analyzed for 
conventional gasoline sold nationwide, five--the Marathon-Ashland, the 
Shell-Texaco I (Equilon), BP-Amoco, MAP-UDS, and the Exxon-Mobil 
mergers--led to price increases for both branded and unbranded gasoline 
ranging from about 0.39 to 5.00 cents per gallon.[Footnote 62] The two 
other mergers, UDS-Total and the Shell-Texaco II (Motiva), led to price 
decreases for both branded and unbranded wholesale conventional 
gasoline ranging from about 0.89 to 1.77 cents per gallon. Similarly, 
for reformulated gasoline, which is sold mainly in cities in the East 
Coast and Gulf Coast regions, table 6 shows that the mergers of 
Marathon-Ashland and Exxon-Mobil increased wholesale gasoline prices 
from about 0.71 to 1.61 cents per gallon. The Shell-Texaco II merger 
led to decreased prices of about 0.39 cents per gallon for branded 
gasoline and the BP-Amoco merger was not associated with price changes. 
For CARB reformulated gasoline, as shown in table 7, the Tosco-Unocal 
merger led to price increases for branded gasoline of about 6.87 cents 
per gallon, while the Shell-Texaco I merger led to price decreases of 
about 0.69 cents per gallon. Neither the Tosco-Unocal merger nor the 
Shell-Texaco merger affected the prices of unbranded CARB gasoline.

Table 5: Estimated Changes in Conventional Wholesale Gasoline Prices 
Associated with Individual Mergers (1994-2000): 

Merger: UDS-Total[B]: Branded; 
Estimated change in prices (cents per gallon)[A]: - 0.89.

Merger: UDS-Total[B]: Unbranded; 
Estimated change in prices (cents per gallon)[A]: - 1.25.

Merger: Marathon-Ashland[C]: Branded; 
Estimated change in prices (cents per gallon)[A]: 0.70.

Merger: Marathon-Ashland[C]: Unbranded; 
Estimated change in prices (cents per gallon)[A]: 0.39.

Merger: Shell-Texaco I[D]: Branded; 
Estimated change in prices (cents per gallon)[A]: 0.99.

Merger: Shell-Texaco I[D]: Unbranded; 
Estimated change in prices (cents per gallon)[A]: 1.13.

Merger: Shell-Texaco II[E]: Branded; 
Estimated change in prices (cents per gallon)[A]: - 1.77.

Merger: Shell-Texaco II[E]: Unbranded; 
Estimated change in prices (cents per gallon)[A]: - 1.24.

Merger: BP-Amoco[F]: Branded; 
Estimated change in prices (cents per gallon)[A]: 0.40.

Merger: BP-Amoco[F]: Unbranded; 
Estimated change in prices (cents per gallon)[A]: 0.97.

Merger: MAP-UDS[G]: Branded; 
Estimated change in prices (cents per gallon)[A]: 1.38.

Merger: MAP-UDS[G]: Unbranded; 
Estimated change in prices (cents per gallon)[A]: 2.63.

Merger: Exxon-Mobil[H]: Branded; 
Estimated change in prices (cents per gallon)[A]: 3.71.

Merger: Exxon-Mobil[H]: Unbranded; 
Estimated change in prices (cents per gallon)[A]: 5.00. 

Sources: GAO econometric analysis of EIA, FTC, QPIS, and Thomson 
Financial data.

Notes: 

See table 15 in appendix IV for additional information.

The average estimated prices (measured as wholesale gasoline prices 
less crude oil prices) were 19 cents and 17 cents per gallon for 
branded and unbranded gasoline, respectively. (See table 20 in appendix 
IV.): 

[A] The effective date, which is the first date in the postmerger 
period, is based on either the merger completion date or the date when 
FTC's remedial actions became effective. The time periods over which 
the estimates were obtained are provided in table 15 in appendix IV. 
The estimated changes associated with the mergers are statistically 
significant at the 1 percent level or lower.

[B] The UDS-Total merger affected rack cities in the Midwest, Gulf 
Coast, and Rocky Mountain regions.

[C] The Marathon-Ashland merger affected rack cities in the East Coast, 
Midwest, and Gulf Coast regions.

[D] The Shell-Texaco I merger affected rack cities in the Midwest, Gulf 
Coast, Rocky Mountain, and West Coast regions.

[E] The Shell-Texaco II merger affected rack cities in the East Coast, 
Midwest, and Gulf Coast regions.

[F] The BP-Amoco merger affected rack cities in the East Coast, 
Midwest, and Gulf Coast regions.

[G] The MAP-UDS merger affected rack cities in the Midwest region.

[H] The Exxon-Mobil merger affected rack cities in the East Coast and 
Gulf Coast regions.

[End of table]

As shown in table 5, for conventional wholesale gasoline, we found the 
following effects of individual mergers on prices.

UDS-Total: This merger led to price reductions for both branded and 
unbranded gasoline of about 1 cent per gallon. FTC did not identify 
potential anticompetitive concerns for this merger.

Marathon-Ashland: We found statistically significant increases in 
prices of branded gasoline of about 1 cent per gallon and in unbranded 
gasoline of about one-third cent per gallon due to this merger. FTC did 
not identify potential anticompetitive concerns.

Shell-Texaco I (Equilon): This merger led to price increases of about 1 
cent per gallon for both branded and unbranded gasoline. FTC identified 
this merger as raising potential anticompetitive concerns at the 
refining, wholesale, and retail levels in certain markets. Thus, the 
agency sought to preserve competition by taking remedial actions.

Shell-Texaco II (Motiva): This merger led to decreases in prices of 
about 1 cent to 2 cents per gallon for both branded and unbranded 
gasoline. This finding is consistent with FTC's determination that the 
merger was not likely to reduce competition in the affected wholesale 
gasoline markets.

BP-Amoco: We found that this merger led to increases in prices of about 
one-half to 1 cent per gallon for both branded and unbranded gasoline. 
FTC identified many cities or metropolitan areas in the eastern half of 
the United States (East Coast, Midwest, and Gulf Coast) where this 
merger could reduce competition in wholesale markets. The agency, 
therefore, took remedial actions to preserve competition in wholesale 
gasoline markets affected by this merger.

MAP-UDS: This merger led to price increases of about 1 cent to 3 cents 
per gallon for both branded and unbranded gasoline. FTC did not 
identify this merger as raising potential anticompetitive concerns in 
the wholesale gasoline markets and so did not take remedial action.

Exxon-Mobil: This merger led to increases in prices of about 4 to 5 
cents per gallon for both branded and unbranded gasoline. The merger 
was identified by FTC as raising potential anticompetitive concerns in 
some retail markets, but not in wholesale markets. Thus, FTC required 
divestitures of retail assets in the affected wholesale markets.

Table 6: Estimated Changes in Reformulated Wholesale Gasoline Prices 
Associated with Individual Mergers (1995-2000): 

Merger[A]: Marathon-Ashland[C]: Branded; 
Estimated change in prices (cents per gallon)[B]: 0.71[D].

Merger[A]: Marathon-Ashland[C]: Unbranded; 
Estimated change in prices (cents per gallon)[B]: 0.86[D].

Merger[A]: Shell-Texaco II[E]: Branded; 
Estimated change in prices (cents per gallon)[B]: - 0.39[F].

Merger[A]: Shell-Texaco II[E]: Unbranded; 
Estimated change in prices (cents per gallon)[B]: 0.09.

Merger[A]: BP-Amoco[G]: Branded; 
Estimated change in prices (cents per gallon)[B]: 0.55.

Merger[A]: BP-Amoco[G]: Unbranded; 
Estimated change in prices (cents per gallon)[B]: 0.40.

Merger[A]: Exxon-Mobil[H]: Branded; 
Estimated change in prices (cents per gallon)[B]: 1.61[D].

Merger[A]: Exxon-Mobil[H]: Unbranded; 
Estimated change in prices (cents per gallon)[B]: 1.01[F].

Sources: GAO econometric analysis of OPIS, EIA, FTC, and Bureau of 
Labor statistics data.

Notes: 

See table 16 in appendix IV for additional information.

The average estimated prices (measured as wholesale gasoline prices 
less crude oil prices) were 20 cents and 18 cents per gallon for 
branded and unbranded gasoline, respectively. (See table 20 in appendix 
IV).

[A] No estimates are reported for the UDS-Total merger because data are 
available for only one rack city.

[B] The effective date, which is the first date in the postmerger 
period, is based on either the merger completion date or the date when 
FTC's remedial actions became effective. The time periods over which 
the estimates were obtained are provided in table 16 in appendix IV.

[C] The Marathon-Ashland merger affected rack cities in the East Coast, 
Midwest, and Gulf Coast regions.

[D] The estimated changes associated with the mergers are statistically 
significant at the 1 percent level or lower.

[E] The Shell-Texaco II merger affected rack cities in the East Coast 
and Gulf Coast regions.

[F] The estimated changes associated with the mergers are statistically 
significant at the 5 percent level or lower.

[G] The BP-Amoco merger affected rack cities in the East Coast, 
Midwest, and Gulf Coast regions.

[H] The Exxon-Mobil merger affected rack cities in the East Coast and 
Gulf Coast regions.

[End of table]

The results presented in table 6 for reformulated wholesale gasoline 
sold in cities in the East Coast and Gulf Coast indicate the following 
effects of the individual mergers on prices.

Marathon-Ashland: This merger led to increases in prices of about 1 
cent per gallon for both branded and unbranded gasoline. As already 
indicated, FTC did not identify potential anticompetitive concerns.

Shell-Texaco II (Motiva): This merger led to price reductions of about 
0.39 cents per gallon for branded gasoline. As already indicated, this 
finding is consistent with FTC's determination that the merger was not 
likely to reduce competition in the affected wholesale gasoline 
markets.

BP-Amoco: The effects of this merger were inconclusive. As already 
indicated, FTC took remedial actions to preserve competition in 
wholesale gasoline markets affected by this merger.

Exxon-Mobil: This merger led to increases in prices of about 1 cent to 
2 cents per gallon for both branded and unbranded gasoline. As already 
indicated, FTC required divestitures of retail assets in the affected 
wholesale markets.

Table 7: Estimated Changes in CARB Reformulated Wholesale Gasoline 
Prices Associated with Individual Mergers (1996-2000): 

Merger: Tosco-Unocalb[B]: Branded; 
Estimated change in prices (cents per gallon)[A]: 6.87[C].

Merger: Tosco-Unocalb[B]: Unbranded; 
Estimated change in prices (cents per gallon)[A]: -1.58.

Merger: Shell-Texaco Id[D]: Branded; 
Estimated change in prices (cents per gallon)[A]: - 0.69[C].

Merger: Shell-Texaco Id[D]: Unbranded; 
Estimated change in prices (cents per gallon)[A]: -0.24.

Sources: GAO econometric analysis of OPIS, EIA, FTC, and Bureau of 
Labor Statistics data.

Notes: 

See table 17 in appendix IV for additional information.

The average estimated prices (measured as wholesale gasoline prices 
less crude oil prices) were 36 cents and 31 cents per gallon for 
branded and unbranded gasoline, respectively. (See table 20 in appendix 
IV).

[A] The effective date, which is the first date in the postmerger 
period, is based on either the merger completion date or the date when 
FTC's remedial actions became effective. The time periods over which 
the estimates were obtained are provided in table 17 in appendix IV.

[B] The Tosco-Unocal merger affected rack cities in the West Coast 
region (California only).

[C] The estimated changes associated with the mergers are statistically 
significant at the 5 percent level or lower.

[D] The Shell-Texaco I merger affected rack cities in the West Coast 
region (California only).

[End of table]

As shown in table 7, for CARB wholesale gasoline sold in California, we 
found the following price changes associated with individual mergers.

Tosco-Unocal: This merger, which affected both refining and marketing 
(wholesale and retail), led to higher prices of branded gasoline--
increases of about 7 cents per gallon. FTC did not take remedial 
actions in the merger.

Shell-Texaco I (Equilon): This merger led to decreases in prices of 
about 1 cent per gallon. As already indicated, FTC sought to preserve 
competition by taking remedial actions in this merger.

Increased Market Concentration Generally Led to Higher Wholesale 
Gasoline Prices: 

Based on our econometric analyses, we found that increased market 
concentration led to higher wholesale prices for all gasoline types. 
This finding is consistent with the fact that the wave of oil industry 
mergers in the second half of the 1990s reduced the number of 
competitors in the wholesale markets. As shown in table 8, the 
estimated increases in wholesale prices of branded and unbranded 
conventional gasoline from 1994 to 2000 were less than one-half cent 
per gallon for all regions. The increases in prices of wholesale 
gasoline were larger, especially for unbranded gasoline, in the western 
half of the United States, which generally has limited access to 
gasoline supplies from other regions or from abroad, potentially 
exacerbating the effects of market concentration.

Table 8: Estimated Changes in Conventional Wholesale Gasoline Prices 
Associated with Increased Market Concentration (1994-2000): 

All regions[A]; 
Market concentration (HHI): 1994: 803; 
Market concentration (HHI): 2000: 1101; 
Market concentration (HHI): Increase in HHI: 298.

Branded; 
Estimated change in prices due to increase in HHI (cents per gallon): 
0.15[B].

Unbranded; 
Estimated change in prices due to increase in HHI (cents per gallon): 
0.33[B].

Geographic area; 
Market concentration (HHI): Increase in HHI: 298.

Eastern United States[C]; 
Market concentration (HHI): 1994: 773; 
Market concentration (HHI): 2000: 1090; 
Market concentration (HHI): Increase in HHI: 317.

Branded; 
Estimated change in prices due to increase in HHI (cents per gallon): 
0.25[B].

Unbranded; 
Estimated change in prices due to increase in HHI (cents per gallon): 
0.10.

Western United States[D]; 
Market concentration (HHI): 1994: 1032; 
Market concentration (HHI): 2000: 1180; 
Market concentration (HHI): Increase in HHI: 148.

Branded; 
Estimated change in prices due to increase in HHI (cents per gallon): 
0.56[B].

Unbranded; 
Estimated change in prices due to increase in HHI (cents per gallon): 
1.29[E].

[End of table]

Sources: GAO econometric analysis of OPIS, EIA, FTC, and Bureau of 
Labor statistics data.

Note: See table 18 in appendix IV for additional information.

[A] All states except Alaska, California, Connecticut, Delaware, 
District of Columbia, Hawaii, Massachusetts, New Hampshire, New Jersey, 
and Rhode Island.

[B] The estimated changes in prices are statistically significant at 
the 1 percent level or lower.

[C] The Eastern U.S. consists of the East Coast, Midwest, and Gulf 
Coast regions.

[D] The Western U.S. consists of the Rocky Mountains and West Coast 
(excluding California) regions.

[E] The estimated changes in prices are statistically significant at 
the 5 percent level or lower.

As shown in table 9, we also found that increased market concentration 
led to higher wholesale prices of about 1 cent per gallon for 
reformulated gasoline sold in certain cities in the East Coast and Gulf 
Coast regions from 1995 through 2000. As also shown in table 9, for 
CARB gasoline (sold only in California), we estimated that prices of 
both branded and unbranded gasoline increased by about 7 and 8 cents 
per gallon, respectively, from 1996 to 2000. The California market is 
isolated from refinery centers in rest of the United States both 
geographically and in terms of its gasoline type.

Table 9: Estimated Changes in Boutique Fuels Wholesale Prices 
Associated with Increased Market Concentration (1995-2000): 

Reformulated wholesale gasoline 1995-2000[A]: Branded; 
Market concentration(HHI): 1995: 1,237; 
Market concentration(HHI): 2000: 1,477; 
Market concentration(HHI): Increase in HHI: 240; 
Estimated change in price margin due to increase in HHI (cents per 
gallon): 0.98[B].

Reformulated wholesale gasoline 1995-2000[A]: Unbranded; 
Market concentration(HHI): 1995: 1,237; 
Market concentration(HHI): 2000: 1,477; 
Market concentration(HHI): Increase in HHI: 240; 
Estimated change in price margin due to increase in HHI (cents per 
gallon): 0.89[B].

CARB reformulated wholesale gasoline: 1996-2000[C]: Branded; 
Market concentration(HHI): 1995: 965; 
Market concentration(HHI): 2000: 1,267; 
Market concentration(HHI): Increase in HHI: 302; 
Estimated change in price margin due to increase in HHI (cents per 
gallon): 7.19[B].

CARB reformulated wholesale gasoline: 1996-2000[C]: Unbranded; 
Market concentration(HHI): 1995: 965; 
Market concentration(HHI): 2000: 1,267; 
Market concentration(HHI): Increase in HHI: 302; 
Estimated change in price margin due to increase in HHI (cents per 
gallon): 7.94[D]. 

Source: GAO econometric analysis of OPIS, EIA, FTC, and Bureau of Labor 
statistics data.

Note: See table 19 in appendix IV for additional information.

[A] The area consists of Connecticut, Delaware, Maryland, 
Massachusetts, New Jersey, New York, Pennsylvania, Rhode Island, 
Virginia in the East region; Kentucky in the Midwest region; and Texas 
in the Gulf Coast region.

[B] The estimated changes in prices are statistically significant at 
the 1 percent level or lower.

[C] The area consists of California.

[D] The estimated changes in prices are statistically significant at 
the 10 percent level or lower.

[End of table]

Other Factors Also Resulted in Higher Wholesale Gasoline Prices: 

In addition to the price increases resulting from the mergers and 
market concentration, we found that low gasoline inventories relative 
to demand, high refinery capacity utilization rates, and supply 
disruptions in the Midwest and West Coast resulted in higher wholesale 
gasoline prices[Footnote 63]--a finding generally consistent with the 
expected effects.

Our econometric models indicate that when gasoline inventories are low 
relative to demand, there is less protection against unexpected or not 
fully anticipated supply problems, thereby increasing prices.[Footnote 
64] Based on our analysis of EIA's data, low inventories relative to 
demand occurred mostly between May/June and October, the summer driving 
months. We found that wholesale prices were about 1 cent per gallon 
higher between May/June and October compared to the other months of the 
year. As shown in figure 22, both the inventories of gasoline and 
expected demand for wholesale gasoline follow seasonal patterns, but 
they move in opposite directions.

Figure 22: Normalized Inventories and Expected Demand for Wholesale 
Gasoline (1994-2000): 

[See PDF for image]

[End of figure]

The ratio of gasoline inventories to expected demand, shown in figure 
23, demonstrates a seasonal pattern, and prices are expected to 
increase when the ratio is less than one, which is from about May/June 
to October, and to decrease when the ratio exceeds one, which is from 
about November to April.

Figure 23: Ratio of Inventories to Expected Demand for Wholesale 
Gasoline (1994-2000): 

[See PDF for image]

[End of figure]

Our econometric models also indicate that when refinery capacity 
utilization rates were high--averaging about 93 percent over the period 
of our study--wholesale gasoline prices increased. In general, 
refineries are utilized at high rates when gasoline demand increases 
relative to gasoline inventories, all other things being constant. High 
utilization rates increase operating costs, hence higher prices. We 
estimated that a 1 percent increase in refinery capacity utilization 
rates was associated with about one-tenth to two-tenths of 1 cent per 
gallon increase in wholesale prices. The effect of high refinery 
capacity utilization rates on prices has not been analyzed in previous 
studies.

We found that both the Midwest and West Coast supply disruptions led to 
higher wholesale gasoline prices, as expected, in the areas that were 
affected by these disruptions. Specifically, prices of conventional 
gasoline were about 4 to 5 cents per gallon higher on average during 
both the Midwest and West Coast supply disruptions. The increase in 
prices for: 

CARB gasoline was about 4 to 7 cents per gallon, on average, during the 
West Coast supply disruptions.[Footnote 65]

Our Findings Are Generally Consistent with Previous Studies' Empirical 
Results: 

Although our econometric models differed from the few previous studies 
in the 1990s in several aspects, our results are generally consistent 
with previous studies' findings that specific oil industry mergers or 
increased market concentration have generally led to increases in 
wholesale gasoline prices. For example, one study (Hastings and 
Gilbert) examined the effects of changes in vertical and horizontal 
market structures on the wholesale prices of unbranded 
gasoline.[Footnote 66] Two kinds of analyses were performed--one for 26 
rack cities on the West Coast from 1993 to 1997 and the other for the 
effect of the 1997 merger between Tosco and Unocal in 13 West Coast 
cities. The authors found that an increase in vertical integration was 
associated with higher wholesale prices for unbranded gasoline. In 
particular, consistent with the strategic incentive to raise 
competitors' input costs, they found that wholesale gasoline prices 
were higher in cities where there was greater competition between 
integrated refiners and independent retailers. As discussed earlier, 
our model of the effects of the: 

Tosco-Unocal merger, using data from 1996 to 2000 in six rack cities, 
found increases in the prices of branded gasoline.[Footnote 67]

Another study (Hendricks and McAfee) analyzed the effects of the then 
proposed merger between Exxon and Mobil--two fully vertically 
integrated oil companies--on CARB wholesale gasoline prices.[Footnote 
68] The authors found the gasoline industry in California to have 20 
percent price-cost margins (or markup), and that the merger would 
increase the margins by about one or two percentage points for prices. 
In addition, most of the postmerger changes would result from changes 
at the refining rather than retail level, emphasizing the vertical 
links in these markets.[Footnote 69]

In another study (Chouinard and Perloff) examining the determinants of 
wholesale and retail gasoline prices in the United States, using data 
for 48 states covering 1989 to 1997, the authors analyzed the effects 
of 8 mergers affecting refining and wholesale markets in 5 states and 
of 27 mergers affecting retail markets in 19 states.[Footnote 70] 
Unlike our study, this study found that, overall, there were more 
decreases in prices than increases for these mergers. However, none of 
the mergers was large and none affected many regional markets. 
Moreover, the mergers did not include any of the eight specific mergers 
we studied.

We reviewed a study by FTC staff on the effects of the Marathon-Ashland 
merger on reformulated wholesale gasoline prices and retail prices in 
only one city, Louisville, Kentucky, using data from 1997 to 1999. The 
FTC study found that wholesale prices increased, consistent with our 
findings, while retail prices did not increase.[Footnote 71]

Although we developed our models drawing on insights from some of these 
and other studies, and there are some similarities with them, the 
models that we estimated differ from most existing ones in several 
ways. First, ours is a comprehensive study of wholesale gasoline 
markets that examines the effects of major individual mergers that were 
part of the petroleum industry's merger wave in the 1990s. Second, we 
examined the cumulative effects of these mergers as well as the effects 
of other market structure factors, using the market concentration 
index. Third, we performed our study for different types of gasoline--
conventional gasoline sold nationwide and boutique fuels sold in 
California and in certain cities in the East Coast and Gulf Coast 
regions. Fourth, we focused on the changes in wholesale price-crude 
cost margins (wholesale prices less crude oil prices) instead of 
wholesale prices because this allowed us to capture the net effects of 
any potential market power and efficiency gains from mergers and market 
concentration.[Footnote 72] Fifth, unlike most previous studies, we 
included the effects of gasoline inventories and refinery capacity 
utilization rates on wholesale prices, whereas previous studies have 
typically included either none of the factors or only gasoline 
inventories.

As we have already indicated, there are limitations to our methods for 
estimating the effects of individual mergers and market concentration 
on wholesale gasoline prices.[Footnote 73] First, we based the timing 
of the mergers on the effective dates as provided by FTC. These are 
either the merger completion dates or the dates when FTC's merger 
remedies became effective for mergers that were subject to remedies. In 
reality, the effective dates of some of the mergers could be some time 
after the dates we used. However, because the mergers typically 
occurred very close to one another and there were overlaps, we could 
not perform sensitivity analyses on the timing of the mergers since 
changing the timing of one merger could cause it to coincide with the 
timing of another merger. Furthermore, the effective date is what most 
experts use to date mergers, and it is expected that using these dates 
would generally underestimate the effect of the mergers. More 
importantly, we used the dates that FTC indicated as the merger 
effective dates. Second, to estimate the effects of mergers on prices, 
we would have preferred to use market shares of the merged companies. 
These data are not usually available because they are proprietary. We 
therefore determined the effects of the mergers by estimating the 
difference in average prices before and after the effective dates of 
the mergers. Also, because of the closeness in the timing of the oil 
industry mergers in the second half of the 1990s as well as the 
overlapping nature of the mergers, estimates from our econometric 
models captured the mergers' effects on prices over shorter time 
periods. However, because our estimate of a merger's effect starts from 
the date that FTC indicated to be the effective date of the merger, we 
believe that our results are sound and reasonable. Third, the market 
concentration variable, measured by the Herfindahl-Hirschmann Index 
(HHI) of refinery capacity at the refining (or PADD) level, includes 
the production of other products in addition to gasoline. Also, the 
data were not available for 2 years (1996 and 1998), and we constructed 
the missing data using the average of the values of the adjacent years. 
Nonetheless, we believe that in a vertically integrated gasoline 
market, market power is better captured by production of gasoline at 
the refinery level since it captures the ability of refiners to control 
gasoline sales. Also, previous studies have identified some conceptual 
limitations of price-concentration relationships, in particular the 
problem of obtaining meaningful estimates from these relationships due 
to possible endogeneity of market concentration. This issue is less 
relevant to our models because it is not likely that prices at the rack 
cities would affect decisions on refinery capacity, which is made for 
the broader regional (PADD) market. Also, we chose to use the mergers 
and market concentration models and found that the effects from both 
models are generally consistent.

We utilized an expert in econometric modeling of the petroleum 
industry, Dr. Severin Borenstein, as a consultant/peer reviewer, and he 
provided us with comments on our econometric methodology and results, 
which we incorporated in our report. Other experts that reviewed a 
detailed outline of our econometric methodology also provided comments, 
which we incorporated as appropriate.

Agency Comments and Our Evaluation: 

We provided a draft of this report to FTC for its review and comment. 
FTC strongly disagreed with our econometric approach and findings, 
stating that the draft report was flawed and did not provide a basis 
for reliable judgments about the competitive effects of mergers in the 
petroleum industry. However, we believe that its analyses are sound and 
consistent with the views of independent economists and experts that 
peer reviewed our overall modeling approach and with previous studies. 
We also believe that our model specifications captured key variables 
that could affect wholesale gasoline prices. Partly in response to 
FTC's comments, we reestimated its models to account for the effects of 
gasoline supply disruptions that occurred in some parts of the West 
Coast and Midwest regions.

The full text of FTC's comments and our responses are included in 
appendixes V and VI. Appendix V contains the comments from FTC 
Commissioners and appendix VI contains the comments from FTC's Bureau 
of Economics staff.

[End of section]

Appendixes: 

Appendix I: Companies, Agencies, and Organizations Contacted by GAO: 

Integrated Oil Companies: 
British Petroleum: 
ChevronTexaco: 
ExxonMobil: 
Shell Oil Company: 

Exploration and Production Companies: 
Apache Corporation: 
AROC, Inc.: 
Devon Energy Corporation: 
Dominion Oil and Gas: 
Kerr-McGee: 

Independent Refiners: 
Kern Oil: 
Flint Hills Resources, LP (wholly owned by Koch Industries): 
Sunoco: 
Valero: 

Pipelines and/or Terminal Operators: 
Kinder Morgan Energy Partners, LP: 
Holland Terminal Company: 

Independent Distributors: 
Barger: 
Cato, Inc., MD: 
CMS Oil Company: 
Congress Gas & Oil: 
Cross Petroleum: 
Downeast Energy: 
Free Enterprise, Inc: 
Global Petroleum, LLC: 
Holland Oil: 
Johnson and Dicks: 
Karbowski Oil Company: 
Lykins Companies, Inc.: 
Ocean Petroleum: 
Primar Petroleum, Inc: 
Quality Oil: 
Rice Oil Company: 
Rusche Distributors: 
Silco Oil: 
Southern Counties Oil Company: 
Speigel & Sons Oil Company: 
Van Manen Petroleum Group: 
Westco, Inc: 
World Oil: 
X-Vest, Inc.: 

Federal Agencies: 
Federal Trade Commission: 
Department of Energy/Energy Information Administration: 
Federal Energy Regulatory Commission: 

State Agencies: 
Michigan Assistant Attorney General's Office: 
Michigan Public Service Commission: 

Associations Association of Oil Pipelines: 
American Petroleum Institute (API): 
California Independent Oil Marketers Association (CIOMA): 
California Service Station Dealers' Association: 
Colorado Petroleum Marketers Association: 
Independent Petroleum Association of America (IPAA): 
Michigan Petroleum Association/Michigan Association of Convenience 
Stores: 
Michigan Service Station Dealers' Association: 
National Petrochemical & Refiners Association: 
New York Service Station Dealers' Association: 
Ohio Service Station Dealers' Association: 
Pennsylvania Service Station Dealers' Association: 
Petroleum Marketers Association of America (PMAA): 
Society of Independent Gasoline Marketers Association: 
Texas Marketers Association: 

Hypermarkets/Unbranded Retailers: 
Rotten Robbie: 
Fuel Mart: 
Wawa: 
Sheetz: 
Safeway: 
Meijer: 
Costco: 
Kroger: 

Consultants: 
PIRA Energy Group: 

Industry Data Sources: 
John S. Herold, Inc.: 
Oil Price Information Service (OPIS): 
Thomson Financial.

[End of section]

Appendix II: Experts Who Reviewed GAO's Econometric Models: 

Peter Ashton: 
President, Innovation and Information Consultants: 

Hank Banta, Partner: 
Law Firm of Label, Novins, Lamont (Antitrust Issues): 

Severin Borenstein, Ph.D.: 
E. T. Grether Professor of Business Administration and Public Policy, 
Haas School of Business, University of California at Berkeley: 
Director, University of California Energy Institute: 
Research Associate, National Bureau of Economic Research: 

John S. Cook, Ph.D. 
Director, Petroleum Division, and his staff: 
Energy Information Administration: 

Lawrence Goldstein, Ph.D. 
President, Petroleum Industry Research Foundation, Inc.

Justine Hastings, Ph.D.
Assistant Professor of Economics, Yale University: 

Kenneth Hendricks, Ph.D. 
Professor of Economics, University of Texas: 

Louis Silva, Ph.D. 
Assistant Director, Antitrust 1, and other economists 
Federal Trade Commission: 

[End of section]

Appendix III: Correlation Analysis of Mergers and Market Concentration 
in the U.S. Petroleum Industry: 

In this appendix, we present levels of wholesale gasoline market 
concentration as well as the results of our correlation analysis 
between mergers and market concentration for the petroleum refining and 
wholesale gasoline markets. We found that levels of wholesale gasoline 
market concentration increased--in some cases dramatically--in all but 
five states from 1994 to 2002. Our correlation analysis for petroleum 
refining showed a positive and statistically significant correlation 
between the average transaction value of mergers (henceforth mergers) 
and market concentration in three of the five geographic regions of the 
U.S. including--the East Coast, the Midwest, and the West Coast. For 
wholesale gasoline markets, we found a positive and statistically 
significant correlation between mergers and market concentration in 
nearly all states from 1994 through 2001. This correlation was 
generally highest in states that experienced large changes in market 
concentration over this period.

Wholesale Gasoline Market Concentration by State: 

As seen in table 10, most states experienced moderate to high levels of 
market concentration in wholesale gasoline by the year 2002, using 
thresholds defined by the 1992 Horizontal Merger Guidelines jointly 
issued by the Department of Justice and the Federal Trade 
Commission.[Footnote 74] Also, 43 states had increases in market 
concentration of well over 100 points, as measured by the Herfindahl-
Hirschman index (HHI) between the years 1994 and 2002. Only 4 states--
New York, Idaho, Montana, Oregon--and the District of Columbia 
experienced decreases in market concentration during this period, and 
in those cases the decrease did not change the category of market 
concentration under the Guidelines. For PADD I, market concentration 
levels in 2002 ranged from a low of 986 in New Hampshire to a high of 
2,967 in the District of Columbia. In addition, market concentration 
increases ranged from 28 in Maryland to 652 in Rhode Island from 1994 
through 2002. Decreases in concentration during this period were found 
in both New York and the District of Columbia. In PADD II, 
concentration levels in 2002 ranged from a low of 951 in Iowa to a high 
of 2,162 in North Dakota. Increases in concentration in this PADD 
ranged from 152 in Illinois to 911 in Kentucky from 1994 through 2002. 
In PADD III, market concentration levels in 2002 ranged from a low of 
857 in Arkansas to a high of 1,326 in New Mexico, with increases in 
concentration ranging from 228 to 432 in Arkansas and Alabama, 
respectively. PADD IV experienced concentration levels ranging from 
1,222 in Idaho to 2,316 in Montana, with changes in concentration from 
1994 through 2002 ranging from a decrease of 40 in Idaho to an increase 
of 477 in Colorado. Finally, in PADD V, concentration levels in 2002 
ranged from a low of 1,171 in Arizona to a high of 3,123 in Hawaii, 
with changes in market concentration ranging from a decrease of 143 in 
Oregon to an increase of 596 in Hawaii from 1994 through 2002.

Table 10: State-level HHI for Wholesale Gasoline (1994-2002): 

: PADD[A]: I; 
State: CT; 
Year: 1994: 1028; 
Year: 1995: 1110; 
Year: 1996: 1223; 
Year: 1997: 1248; 
Year: 1998: 1292; 
Year: 1999: 1374; 
Year: 2000: 1302; 
Year: 2001: 1418; 
Year: 2002: 1501; 
Change: 473.

PADD[A]: I; 
State: MA; 
Year: 1994: 966; 
Year: 1995: 1023; 
Year: 1996: 1130; 
Year: 1997: 1107; 
Year: 1998: 1218; 
Year: 1999: 1249; 
Year: 2000: 1079; 
Year: 2001: 1185; 
Year: 2002: 1280; 
Change: 314.

PADD[A]: I; 
State: ME; 
Year: 1994: 1171; 
Year: 1995: 1193; 
Year: 1996: 1305; 
Year: 1997: 1385; 
Year: 1998: 1435; 
Year: 1999: 1423; 
Year: 2000: 1349; 
Year: 2001: 1340; 
Year: 2002: 1453; 
Change: 282.

PADD[A]: I; 
State: NH; 
Year: 1994: 749; 
Year: 1995: 810; 
Year: 1996: 917; 
Year: 1997: 844; 
Year: 1998: 855; 
Year: 1999: 997; 
Year: 2000: 884; 
Year: 2001: 920; 
Year: 2002: 986; 
Change: 237.

PADD[A]: I; 
State: RI; 
Year: 1994: 1037; 
Year: 1995: 1073; 
Year: 1996: 1154; 
Year: 1997: 1167; 
Year: 1998: 1180; 
Year: 1999: 1513; 
Year: 2000: 1470; 
Year: 2001: 1647; 
Year: 2002: 1689; 
Change: 652.

PADD[A]: I; 
State: VT; 
Year: 1994: 1061; 
Year: 1995: 1164; 
Year: 1996: 1134; 
Year: 1997: 1111; 
Year: 1998: 1114; 
Year: 1999: 1148; 
Year: 2000: 1015; 
Year: 2001: 1198; 
Year: 2002: 1164; 
Change: 103.

PADD[A]: I; 
State: DC; 
Year: 1994: 3474; 
Year: 1995: 3465; 
Year: 1996: 3249; 
Year: 1997: 3117; 
Year: 1998: 3245; 
Year: 1999: 2997; 
Year: 2000: 3033; 
Year: 2001: 2784; 
Year: 2002: 2967; 
Change: -507.

PADD[A]: I; 
State: DE; 
Year: 1994: 865; 
Year: 1995: 839; 
Year: 1996: 884; 
Year: 1997: 975; 
Year: 1998: 1002; 
Year: 1999: 1076; 
Year: 2000: 1044; 
Year: 2001: 1182; 
Year: 2002: 1283; 
Change: 418.

PADD[A]: I; 
State: MD; 
Year: 1994: 1120; 
Year: 1995: 1095; 
Year: 1996: 1060; 
Year: 1997: 1117; 
Year: 1998: 1041; 
Year: 1999: 1179; 
Year: 2000: 1148; 
Year: 2001: 1105; 
Year: 2002: 1148; 
Change: 28.

PADD[A]: I; 
State: NJ; 
Year: 1994: 828; 
Year: 1995: 834; 
Year: 1996: 882; 
Year: 1997: 907; 
Year: 1998: 889; 
Year: 1999: 1018; 
Year: 2000: 1026; 
Year: 2001: 955; 
Year: 2002: 1130; 
Change: 302.

PADD[A]: I; 
State: NY; 
Year: 1994: 1087; 
Year: 1995: 1094; 
Year: 1996: 1146; 
Year: 1997: 1113; 
Year: 1998: 1130; 
Year: 1999: 1075; 
Year: 2000: 927; 
Year: 2001: 977; 
Year: 2002: 1048; 
Change: - 39.

PADD[A]: I; 
State: PA; 
Year: 1994: 946; 
Year: 1995: 945; 
Year: 1996: 956; 
Year: 1997: 967; 
Year: 1998: 927; 
Year: 1999: 1076; 
Year: 2000: 1167; 
Year: 2001: 1203; 
Year: 2002: 1341; 
Change: 395.

PADD[A]: I; 
State: FL; 
Year: 1994: 839; 
Year: 1995: 860; 
Year: 1996: 844; 
Year: 1997: 828; 
Year: 1998: 743; 
Year: 1999: 997; 
Year: 2000: 1093; 
Year: 2001: 1070; 
Year: 2002: 1043; 
Change: 204.

PADD[A]: I; 
State: GA; 
Year: 1994: 715; 
Year: 1995: 722; 
Year: 1996: 723; 
Year: 1997: 715; 
Year: 1998: 694; 
Year: 1999: 1152; 
Year: 2000: 1151; 
Year: 2001: 1088; 
Year: 2002: 1089; 
Change: 374.

PADD[A]: I; 
State: NC; 
Year: 1994: 831; 
Year: 1995: 888; 
Year: 1996: 886; 
Year: 1997: 893; 
Year: 1998: 846; 
Year: 1999: 1160; 
Year: 2000: 1222; 
Year: 2001: 1138; 
Year: 2002: 1117; 
Change: 286.

PADD[A]: I; 
State: SC; 
Year: 1994: 814; 
Year: 1995: 820; 
Year: 1996: 817; 
Year: 1997: 833; 
Year: 1998: 823; 
Year: 1999: 1007; 
Year: 2000: 1023; 
Year: 2001: 1029; 
Year: 2002: 1023; 
Change: 209.

PADD[A]: I; 
State: VA; 
Year: 1994: 895; 
Year: 1995: 957; 
Year: 1996: 948; 
Year: 1997: 963; 
Year: 1998: 921; 
Year: 1999: 1124; 
Year: 2000: 1083; 
Year: 2001: 1162; 
Year: 2002: 1116; 
Change: 221.

PADD[A]: I; 
State: WV; 
Year: 1994: 1654; 
Year: 1995: 1374; 
Year: 1996: 1446; 
Year: 1997: 1602; 
Year: 1998: 2356; 
Year: 1999: 2487; 
Year: 2000: 2020; 
Year: 2001: 1785; 
Year: 2002: 1744; 
Change: 90.

PADD[A]: II; 
State: IA; 
Year: 1994: 765; 
Year: 1995: 778; 
Year: 1996: 806; 
Year: 1997: 931; 
Year: 1998: 828; 
Year: 1999: 849; 
Year: 2000: 866; 
Year: 2001: 834; 
Year: 2002: 951; 
Change: 186.

PADD[A]: II; 
State: IL; 
Year: 1994: 1147; 
Year: 1995: 1140; 
Year: 1996: 1173; 
Year: 1997: 1176; 
Year: 1998: 1188; 
Year: 1999: 1260; 
Year: 2000: 1253; 
Year: 2001: 1281; 
Year: 2002: 1299; 
Change: 152.

PADD[A]: II; 
State: IN; 
Year: 1994: 1599; 
Year: 1995: 1636; 
Year: 1996: 1644; 
Year: 1997: 1676; 
Year: 1998: 1966; 
Year: 1999: 1983; 
Year: 2000: 1917; 
Year: 2001: 2069; 
Year: 2002: 2117; 
Change: 518.

PADD[A]: II; 
State: KS; 
Year: 1994: 873; 
Year: 1995: 888; 
Year: 1996: 912; 
Year: 1997: 875; 
Year: 1998: 923; 
Year: 1999: 944; 
Year: 2000: 992; 
Year: 2001: 1023; 
Year: 2002: 1214; 
Change: 341.

PADD[A]: II; 
State: KY; 
Year: 1994: 1216; 
Year: 1995: 1285; 
Year: 1996: 1392; 
Year: 1997: 1437; 
Year: 1998: 2170; 
Year: 1999: 2116; 
Year: 2000: 2033; 
Year: 2001: 2161; 
Year: 2002: 2127; 
Change: 911.

PADD[A]: II; 
State: MI; 
Year: 1994: 1150; 
Year: 1995: 1156; 
Year: 1996: 1158; 
Year: 1997: 1091; 
Year: 1998: 1241; 
Year: 1999: 1337; 
Year: 2000: 1838; 
Year: 2001: 1861; 
Year: 2002: 1884; 
Change: 734.

PADD[A]: II; 
State: MN; 
Year: 1994: 1162; 
Year: 1995: 1183; 
Year: 1996: 1204; 
Year: 1997: 1211; 
Year: 1998: 1268; 
Year: 1999: 1298; 
Year: 2000: 1340; 
Year: 2001: 1362; 
Year: 2002: 1383; 
Change: 221.

PADD[A]: II; 
State: MO; 
Year: 1994: 739; 
Year: 1995: 770; 
Year: 1996: 818; 
Year: 1997: 850; 
Year: 1998: 899; 
Year: 1999: 884; 
Year: 2000: 908; 
Year: 2001: 895; 
Year: 2002: 983; 
Change: 244.

PADD[A]: II; 
State: ND; 
Year: 1994: 1761; 
Year: 1995: 1815; 
Year: 1996: 1845; 
Year: 1997: 1916; 
Year: 1998: 1516; 
Year: 1999: 2292; 
Year: 2000: 2278; 
Year: 2001: 1892; 
Year: 2002: 2162; 
Change: 401.

PADD[A]: II; 
State: NE; 
Year: 1994: 898; 
Year: 1995: 878; 
Year: 1996: 856; 
Year: 1997: 844; 
Year: 1998: 874; 
Year: 1999: 943; 
Year: 2000: 898; 
Year: 2001: 987; 
Year: 2002: 1307; 
Change: 409.

PADD[A]: II; 
State: OH; 
Year: 1994: 1540; 
Year: 1995: 1508; 
Year: 1996: 1495; 
Year: 1997: 1536; 
Year: 1998: 2058; 
Year: 1999: 2147; 
Year: 2000: 2132; 
Year: 2001: 2040; 
Year: 2002: 1943; 
Change: 403.

PADD[A]: II; 
State: OK; 
Year: 1994: 895; 
Year: 1995: 927; 
Year: 1996: 991; 
Year: 1997: 944; 
Year: 1998: 944; 
Year: 1999: 957; 
Year: 2000: 1001; 
Year: 2001: 933; 
Year: 2002: 1048; 
Change: 153.

PADD[A]: II; 
State: SD; 
Year: 1994: 845; 
Year: 1995: 934; 
Year: 1996: 927; 
Year: 1997: 887; 
Year: 1998: 760; 
Year: 1999: 898; 
Year: 2000: 957; 
Year: 2001: 943; 
Year: 2002: 1153; 
Change: 308.

PADD[A]: II; 
State: TN; 
Year: 1994: 742; 
Year: 1995: 806; 
Year: 1996: 835; 
Year: 1997: 853; 
Year: 1998: 871; 
Year: 1999: 1215; 
Year: 2000: 1251; 
Year: 2001: 1224; 
Year: 2002: 1231; 
Change: 489.

PADD[A]: II; 
State: WI; 
Year: 1994: 944; 
Year: 1995: 999; 
Year: 1996: 1003; 
Year: 1997: 1123; 
Year: 1998: 1187; 
Year: 1999: 1120; 
Year: 2000: 1201; 
Year: 2001: 1235; 
Year: 2002: 1275; 
Change: 331.

PADD[A]: III; 
State: AL; 
Year: 1994: 713; 
Year: 1995: 718; 
Year: 1996: 762; 
Year: 1997: 809; 
Year: 1998: 779; 
Year: 1999: 1216; 
Year: 2000: 1170; 
Year: 2001: 1150; 
Year: 2002: 1145; 
Change: 432.

PADD[A]: III; 
State: AR; 
Year: 1994: 629; 
Year: 1995: 637; 
Year: 1996: 625; 
Year: 1997: 633; 
Year: 1998: 772; 
Year: 1999: 817; 
Year: 2000: 840; 
Year: 2001: 784; 
Year: 2002: 857; 
Change: 228.

PADD[A]: III; 
State: LA; 
Year: 1994: 897; 
Year: 1995: 912; 
Year: 1996: 941; 
Year: 1997: 936; 
Year: 1998: 845; 
Year: 1999: 1055; 
Year: 2000: 1105; 
Year: 2001: 1098; 
Year: 2002: 1157; 
Change: 260.

PADD[A]: III; 
State: MS; 
Year: 1994: 711; 
Year: 1995: 731; 
Year: 1996: 738; 
Year: 1997: 771; 
Year: 1998: 727; 
Year: 1999: 1025; 
Year: 2000: 1043; 
Year: 2001: 1019; 
Year: 2002: 1063; 
Change: 352.

PADD[A]: III; 
State: NM; 
Year: 1994: 938; 
Year: 1995: 966; 
Year: 1996: 1030; 
Year: 1997: 1111; 
Year: 1998: 1225; 
Year: 1999: 1305; 
Year: 2000: 1192; 
Year: 2001: 1236; 
Year: 2002: 1326; 
Change: 388.

PADD[A]: III; 
State: TX; 
Year: 1994: 794; 
Year: 1995: 825; 
Year: 1996: 852; 
Year: 1997: 850; 
Year: 1998: 837; 
Year: 1999: 941; 
Year: 2000: 977; 
Year: 2001: 970; 
Year: 2002: 1117; 
Change: 323.

PADD[A]: IV; 
State: CO; 
Year: 1994: 1002; 
Year: 1995: 1029; 
Year: 1996: 1039; 
Year: 1997: 1039; 
Year: 1998: 1240; 
Year: 1999: 1282; 
Year: 2000: 1278; 
Year: 2001: 1274; 
Year: 2002: 1479; 
Change: 477.

PADD[A]: IV; 
State: ID; 
Year: 1994: 1262; 
Year: 1995: 1272; 
Year: 1996: 1151; 
Year: 1997: 1120; 
Year: 1998: 1035; 
Year: 1999: 1098; 
Year: 2000: 1130; 
Year: 2001: 1089; 
Year: 2002: 1222; 
Change: -40.

PADD[A]: IV; 
State: MT; 
Year: 1994: 2339; 
Year: 1995: 2290; 
Year: 1996: 2282; 
Year: 1997: 2079; 
Year: 1998: 2024; 
Year: 1999: 2064; 
Year: 2000: 2303; 
Year: 2001: 2380; 
Year: 2002: 2316; 
Change: -23.

PADD[A]: IV; 
State: UT; 
Year: 1994: 1142; 
Year: 1995: 1153; 
Year: 1996: 1161; 
Year: 1997: 1146; 
Year: 1998: 1270; 
Year: 1999: 1320; 
Year: 2000: 1305; 
Year: 2001: 1200; 
Year: 2002: 1391; 
Change: 249.

PADD[A]: IV; 
State: WY; 
Year: 1994: 1115; 
Year: 1995: 1129; 
Year: 1996: 1070; 
Year: 1997: 992; 
Year: 1998: 1107; 
Year: 1999: 1291; 
Year: 2000: 1402; 
Year: 2001: 1325; 
Year: 2002: 1350; 
Change: 235.

PADD[A]: V; 
State: AK; 
Year: 1994: 2505; 
Year: 1995: 2577; 
Year: 1996: 2679; 
Year: 1997: 2975; 
Year: 1998: 2828; 
Year: 1999: 2719; 
Year: 2000: 2599; 
Year: 2001: 2721; 
Year: 2002: 2746; 
Change: 241.

PADD[A]: V; 
State: AZ; 
Year: 1994: 1142; 
Year: 1995: 1069; 
Year: 1996: 1151; 
Year: 1997: 1215; 
Year: 1998: 1427; 
Year: 1999: 1331; 
Year: 2000: 1175; 
Year: 2001: 1045; 
Year: 2002: 1171; 
Change: 29.

PADD[A]: V; 
State: CA; 
Year: 1994: 1122; 
Year: 1995: 1144; 
Year: 1996: 1200; 
Year: 1997: 1310; 
Year: 1998: 1488; 
Year: 1999: 1511; 
Year: 2000: 1356; 
Year: 2001: 1395; 
Year: 2002: 1597; 
Change: 475.

PADD[A]: V; 
State: HI; 
Year: 1994: 2527; 
Year: 1995: 2575; 
Year: 1996: 2339; 
Year: 1997: 2271; 
Year: 1998: 2222; 
Year: 1999: 2813; 
Year: 2000: 2890; 
Year: 2001: 2942; 
Year: 2002: 3123; 
Change: 596.

PADD[A]: V; 
State: NV; 
Year: 1994: 1417; 
Year: 1995: 1463; 
Year: 1996: 1425; 
Year: 1997: 1339; 
Year: 1998: 1361; 
Year: 1999: 1354; 
Year: 2000: 1282; 
Year: 2001: 1359; 
Year: 2002: 1555; 
Change: 138.

PADD[A]: V; 
State: OR; 
Year: 1994: 1867; 
Year: 1995: 1495; 
Year: 1996: 1406; 
Year: 1997: 1445; 
Year: 1998: 1699; 
Year: 1999: 1734; 
Year: 2000: 1594; 
Year: 2001: 1556; 
Year: 2002: 1724; 
Change: -143.

PADD[A]: V; 
State: WA; 
Year: 1994: 1421; 
Year: 1995: 1381; 
Year: 1996: 1398; 
Year: 1997: 1427; 
Year: 1998: 1572; 
Year: 1999: 1557; 
Year: 2000: 1423; 
Year: 2001: 1376; 
Year: 2002: 1579; 
Change: 158. 

Source: GAO analysis of EIA data.

[A] Petroleum Administration Defense Districts (PADD) are regional 
districts defined by the Department of Energy.

[End of table]

Correlation Analysis of Mergers and Market Concentration: 

To determine the degree of association or connection between changes in 
merger activity and market concentration we analyzed correlations for 
both petroleum refining and wholesale gasoline supply.[Footnote 75] For 
both correlations, we used average merger transaction values from John 
S. Herold, Inc., as a proxy for merger activity and market 
concentration data from the Energy Information Administration (EIA) of 
the Department of Energy.[Footnote 76] Transaction values were reported 
for nearly 60 percent of the mergers in the John S. Herold merger 
database, including all mergers during the period valued at over $1 
billion. We performed the correlations using the Pearson correlation 
coefficient from the SAS (Statistical Analysis System) statistical 
package. This coefficient measures the strength of the linear 
relationship between two variables and ranges from -1 to +1, with a 
positive number corresponding to a positive or direct association and a 
negative number corresponding to a negative or inverse association. In 
addition, we used the SAS package to test the statistical significance 
for each pair of variables in the correlation.

Correlation Analysis for Petroleum Refining: 

To perform the correlation analysis for petroleum refining, we used 
market concentration data at the regional or PADD level because we were 
able to obtain data at this level and were told by experts and 
academicians that this was a relevant geographic market for refining. 
We used yearly average merger transaction values and yearly market 
concentration (HHI) data from 1991 to 2000, omitting the years 1996 and 
1998 because market concentration (HHI) data were unavailable for these 
years. Table 11 presents the results of our correlation analysis 
between the average transaction value of mergers and market 
concentration for the petroleum refining market.

Table 11: Correlation between the Average Transaction Value of Mergers 
and Market Concentration (HHI) for Petroleum Refining by PADD (1991-
2000): 

Petroleum Administration for Defense Districts (PADD): Market 
concentration (HHI); 
range of lowest to highest; 
PADD I: East Coast: Region: 1,150-827; 
PADD II: Midwest: Region: 674-004; 
PADD III: Gulf Coast: Region: 520-704; 
PADD IV: Rocky Mountain: Region: 1,029-1,128; 
PADD V West Coast: Region: 877-1,267.

Petroleum Administration for Defense Districts (PADD): Correlation 
coefficient between the average of transaction value of mergers and 
HHI[A]; 
PADD I: East Coast: Region: 0.80[B]; (0.0177); 
PADD II: Midwest: Region: 0.93[B]; (0.0091); 
PADD III: Gulf Coast: Region: 0.53; (0.1773); 
PADD IV: Rocky Mountain: Region: 0.37; (0.3652); 
PADD V West Coast: Region: 0.91[B]; (0.0018). 

Source: GAO analysis of data from John S. Herold, Inc., and the EIA.

Notes: (1) The correlation between mergers and concentration by 
Petroleum Administration for Defense Districts (PADDs) does not include 
years 1996 and 1998 because HHI data are unavailable for these years. 
(2) We calculated the average transaction values of mergers using the 
transaction values of mergers divided by the total number of mergers, 
as reported in the John S. Herold, Inc., dataset. John S. Herold 
defines transaction value as the value of the merger at the time of the 
offer, based on either the value of the seller's assets or the offer 
from the buyer.

[A] Numbers in parenthesis indicate the statistical significance of the 
estimate of correlation.

[B] Eestimates are statistically at the 0.05 level or below.

[End of table]

As table 11 shows, the average transaction values of mergers and 
petroleum refining market concentration (HHIs) are positively 
correlated and highly statistically significant for the regions of PADD 
I (the East Coast), PADD II (the Midwest), and PADD V (the West Coast).

Correlation Analysis for Wholesale Gasoline: 

To perform the correlation analysis for wholesale gasoline supply, we 
used market concentration data at the state level because we were able 
to obtain data at this level and were told by experts and academicians 
that this was the relevant geographic market. We estimated correlations 
between the transaction values of mergers and market concentration 
(HHI) for wholesale gasoline supply from 1994 through 2001. Although we 
were able to obtain monthly HHI data from 1994 through 2002 for each 
state, we only had yearly merger transaction data. Therefore, for this 
correlation, we matched lagged values of the yearly average merger 
transaction from 1993 through 2000 with monthly observations of the HHI 
for each state from 1994 through 2001. Table 12 presents the results of 
our correlation analysis between the yearly average transaction values 
of mergers and market concentration for wholesale gasoline supply. The 
correlation was positive and statistically significant in almost all 
states.

Table 12: Correlation between the Average Transaction Value of Mergers 
and Market Concentration (HHI) for Wholesale Gasoline (1994-2001): 

PADD: I; 
State: GA; 
Correlation coefficient: 0.92; 
Statistical significance: <.0001.

PADD: I; 
State: NC; 
Correlation coefficient: 0.91; 
Statistical significance: <.0001.

PADD: I; 
State: SC; 
Correlation coefficient: 0.86; 
Statistical significance: <.0001.

PADD: I; 
State: VA; 
Correlation coefficient: 0.80; 
Statistical significance: <.0001.

PADD: I; 
State: FL; 
Correlation coefficient: 0.78; 
Statistical significance: <.0001.

PADD: I; 
State: RI; 
Correlation coefficient: 0.77; 
Statistical significance: <.0001.

PADD: I; 
State: NH; 
Correlation coefficient: 0.75; 
Statistical significance: <.0001.

PADD: I; 
State: NJ; 
Correlation coefficient: 0.74; 
Statistical significance: <.0001.

PADD: I; 
State: WV; 
Correlation coefficient: 0.74; 
Statistical significance: <.0001.

PADD: I; 
State: MD; 
Correlation coefficient: 0.68; 
Statistical significance: <.0001.

PADD: I; 
State: CT; 
Correlation coefficient: 0.67; 
Statistical significance: <.0001.

PADD: I; 
State: DE; 
Correlation coefficient: 0.65; 
Statistical significance: <.0001.

PADD: I; 
State: ME; 
Correlation coefficient: 0.61; 
Statistical significance: <.0001.

PADD: I; 
State: PA; 
Correlation coefficient: 0.58; 
Statistical significance: <.0001.

PADD: I; 
State: MA; 
Correlation coefficient: 0.58; 
Statistical significance: <.0001.

PADD: I; 
State: VT; 
Correlation coefficient: 0.36; 
Statistical significance: 0.0003.

PADD: I; 
State: NY; 
Correlation coefficient: -0.36; 
Statistical significance: 0.0004.

PADD: I; 
State: DC; 
Correlation coefficient: -0.37; 
Statistical significance: 0.0002.

PADD: II; 
State: TN; 
Correlation coefficient: 0.91; 
Statistical significance: <.0001.

PADD: II; 
State: OH; 
Correlation coefficient: 0.85; 
Statistical significance: <.0001.

PADD: II; 
State: KY; 
Correlation coefficient: 0.76; 
Statistical significance: <.0001.

PADD: II; 
State: ND; 
Correlation coefficient: 0.76; 
Statistical significance: <.0001.

PADD: II; 
State: IL; 
Correlation coefficient: 0.69; 
Statistical significance: <.0001.

PADD: II; 
State: IN; 
Correlation coefficient: 0.68; 
Statistical significance: <.0001.

PADD: II; 
State: MN; 
Correlation coefficient: 0.63; 
Statistical significance: <.0001.

PADD: II; 
State: MI; 
Correlation coefficient: 0.62; 
Statistical significance: <.0001.

PADD: II; 
State: MO; 
Correlation coefficient: 0.61; 
Statistical significance: <.0001.

PADD: II; 
State: WI; 
Correlation coefficient: 0.56; 
Statistical significance: <.0001.

PADD: II; 
State: KS; 
Correlation coefficient: 0.55; 
Statistical significance: <.0001.

PADD: II; 
State: NE; 
Correlation coefficient: 0.39; 
Statistical significance: <.0001.

PADD: II; 
State: IA; 
Correlation coefficient: 0.31; 
Statistical significance: 0.002.

PADD: II; 
State: OK; 
Correlation coefficient: 0.25; 
Statistical significance: 0.0128.

PADD: II; 
State: SD; 
Correlation coefficient: 0.21; 
Statistical significance: 0.0395.

PADD: III; 
State: AL; 
Correlation coefficient: 0.91; 
Statistical significance: <.0001.

PADD: III; 
State: MS; 
Correlation coefficient: 0.89; 
Statistical significance: <.0001.

PADD: III; 
State: AR; 
Correlation coefficient: 0.82; 
Statistical significance: <.0001.

PADD: III; 
State: NM; 
Correlation coefficient: 0.77; 
Statistical significance: <.0001.

PADD: III; 
State: LA; 
Correlation coefficient: 0.74; 
Statistical significance: <.0001.

PADD: III; 
State: TX; 
Correlation coefficient: 0.73; 
Statistical significance: <.0001.

PADD: IV; 
State: UT; 
Correlation coefficient: 0.83; 
Statistical significance: <.0001.

PADD: IV; 
State: CO; 
Correlation coefficient: 0.78; 
Statistical significance: <.0001.

PADD: IV; 
State: WY; 
Correlation coefficient: 0.74; 
Statistical significance: <.0001.

PADD: IV; 
State: MT; 
Correlation coefficient: -0.03; 
Statistical significance: 0.7746.

PADD: IV; 
State: ID; 
Correlation coefficient: -0.43; 
Statistical significance: <.0001.

PADD: V; 
State: CA; 
Correlation coefficient: 0.76; 
Statistical significance: <.0001.

PADD: V; 
State: HI; 
Correlation coefficient: 0.72; 
Statistical significance: <.0001.

PADD: V; 
State: WA; 
Correlation coefficient: 0.64; 
Statistical significance: <.0001.

PADD: V; 
State: AZ; 
Correlation coefficient: 0.37; 
Statistical significance: 0.0003.

PADD: V; 
State: OR; 
Correlation coefficient: 0.33; 
Statistical significance: 0.0009.

PADD: V; 
State: AK; 
Correlation coefficient: -0.01; 
Statistical significance: 0.9060.

PADD: V; 
State: NV; 
Correlation coefficient: -0.21; 
Statistical significance: 0.0430. 

Source: GAO analysis of data from John S. Herold, Inc., and EIA.

Note: We calculated the average transaction values of mergers using the 
transaction values of mergers divided by the total number of mergers, 
as reported in the John S. Herold, Inc., dataset. John S. Herold 
defines transaction value as the value of the merger at the time of the 
offer, based on either the value of the seller's assets or the offer 
from the buyer.

[End of table]

A comparison of tables 10 and 12 illustrates that, with few exceptions, 
states with large increases in market concentration from 1994 through 
2002 also displayed a high level of correlation between the average 
transaction value of mergers and market concentration for wholesale 
gasoline over this period. For instance, 13 of 18 states in PADD I had 
correlations between mergers and market concentration greater than 
0.60. Of those 13 states, 11 experienced increases in their HHI from 
200 to 400 between 1994 and 2002. In PADD II, states with correlations 
above 0.70 (Tennessee, Ohio, Kentucky, and North Dakota) all had HHIs 
that increased by 400 to over 900 index points over the period. 
Similarly, in PADD III, all states displayed correlations over 0.70 and 
all had increases in their HHI of over 200 to over 400 points. In PADD 
IV, the states of Utah, Colorado, and Wyoming, all had increases in HHI 
of 200 to over 400 points and displayed correlations over 0.70. Lastly, 
in PADD V, California and Hawaii experienced increases in their HHI of 
400 to over 500 points and also had correlations over 0.70.

[End of section]

Appendix IV: Econometric Analyses of the Effects of Specific Mergers 
and Market Concentration on U.S. Wholesale Gasoline Prices: 

This appendix discusses our analysis of the effects of specific 
mergers, market concentration, and other factors on wholesale gasoline 
prices in the United States in the second half of the 1990s. In 
particular, we discuss: 

* the development of two groups of econometric models we used to 
estimate the effects of eight specific oil industry mergers, market 
concentration, and other factors on wholesale gasoline prices of 
different gasoline specifications,

* the data sources and selection of the geographic markets that we 
analyzed,

* specifications of econometric models and estimation methodology,

* our econometric results, and: 

* limitations of our econometric methodology.

GAO's Econometric Models of Wholesale Gasoline Prices Built on Previous 
Studies and Market Analysis: 

We developed two groups of econometric models to determine the effects 
of mergers and market concentration on U.S. wholesale prices of 
different gasoline specifications--conventional, reformulated, and 
CARB--in the second half of the 1990s. The first group of models 
(mergers models) determined the effects of eight individual oil 
industry mergers on wholesale gasoline prices using a broad panel data 
that included racks where the merging companies operated before they 
merged.[Footnote 77] The second group of models (market concentration 
models) determined the effects of market concentration on wholesale 
gasoline prices nationwide. The market concentration models capture the 
cumulative effects of mergers as well as other structural factors such 
as barriers to entry. We relied on information from previous studies, 
industry experts, and from our own analysis of the oil industry, 
specifically the wholesale gasoline market, to develop our econometric 
models.

Oil Industry Mergers in the Second Half of the 1990s Affected Wholesale 
Gasoline Markets: 

Several oil industry mergers involving large and partially or fully 
vertically integrated companies in the second half of the 1990s 
affected wholesale gasoline markets. These mergers generally reduced 
the number of suppliers at the relevant wholesale racks, except in the 
cases where the Federal Trade Commission (FTC) required the merging 
parties to divest assets to a third party.[Footnote 78]

The second half of the1990s witnessed a wave of oil industry mergers, 
and we examined the eight transactions listed below--which we refer to 
generally as mergers since they led to consolidation of assets, 
although some of the transactions were identified as joint ventures.

* Tosco-Unocal: On April 1, 1997, Tosco bought Unocal's West Coast 
refining and marketing (wholesale and retail) assets in Petroleum 
Administration for Defense District (PADD) V.[Footnote 79]

* UDS-Total: On September 25, 1997, Ultramar Diamond Shamrock (UDS) 
merged with Total, affecting wholesale markets mainly in PADDs II, III, 
and IV.

* Marathon-Ashland: On January 5, 1998, Marathon formed a joint venture 
with Ashland, creating Marathon Ashland Petroleum LLC (MAP), a refining 
and marketing (wholesale and retail) company, affecting PADDs I, II, 
and III.

* Shell-Texaco I (Equilon): On January 23, 1998, Shell formed a joint 
venture with Texaco to combine their refining and marketing (wholesale 
and retail) businesses mainly in PADDs II, III, IV, and V, creating 
Equilon.[Footnote 80]

* Shell-Texaco II (Motiva): On July 1, 1998, Shell formed a joint 
venture with Texaco and Star (jointly controlled by Texaco and Saudi 
Refining Company) to combine their refining and marketing assets mainly 
in PADDs I, II, and III, creating Motiva.

* BP-Amoco: On December 31, 1998, British Petroleum (BP) merged with 
Amoco, affecting wholesale markets in PADDs I, II, and III.

* Exxon-Mobil: On November 30, 1999, Exxon merged with Mobil, affecting 
wholesale markets in PADDs I and III.

* MAP-UDS: On December 13, 1999, MAP bought the assets of UDS located 
in Michigan (PADD II), including its distribution network and rack 
facilities in the wholesale market.

Some of the mergers had the potential to directly reduce competition in 
wholesale gasoline markets because the merging companies supplied 
wholesale gasoline in overlapping markets before they merged. For 
instance, FTC identified the BP-Amoco merger as having potential 
anticompetitive effects on wholesale gasoline markets in 30 cities or 
metropolitan areas in the eastern part of the U.S. Mergers that reduce 
competition in other levels of the downstream segment of the petroleum 
industry--such as refining or retail--could also indirectly reduce 
competition in wholesale markets if one of the merging companies is 
partially or fully vertically integrated. For instance, the Exxon-Mobil 
merger, which had the potential to reduce competition in refining in 
the West Coast and in retail markets on the East Coast, could have had 
competitive implications in the relevant wholesale markets.

Several Factors, Including Mergers and Market Concentration, Are 
Expected to Affect Wholesale Gasoline Prices: 

We used the same general econometric specification to estimate the 
effects of individual mergers, market concentration, and other factors 
on prices in wholesale gasoline markets. In U.S. wholesale gasoline 
markets, wholesalers (consisting of affiliated and independent 
distributors)[Footnote 81] buy gasoline from refiners (consisting of 
integrated refiners and independents)[Footnote 82] at racks and truck 
the gasoline to retail gasoline stations. In models of vertically 
integrated markets such as gasoline marketing, market power can be 
assigned to either the sellers or to buyers.[Footnote 83] It has been 
previously assumed that the refinery (upstream) market is imperfectly 
competitive while the wholesale (downstream) market, in contrast, is 
generally competitive.[Footnote 84] We focused on prices the refiners 
post at the racks ("rack prices") because, on average, that is the most 
dominant form of wholesale market transaction nationwide and there are 
no publicly available data on transfer prices and no reliable 
systematic data on dealer-tankwagon sales.[Footnote 85]

Dependent Variable--Wholesale Prices: 

Our dependent variable is wholesale prices--measured by wholesale 
gasoline prices less crude oil prices--because this approach enables us 
to assess the combined market power and efficiency effects of mergers 
and market concentration on wholesale prices.[Footnote 86] We used the 
average rack prices at the rack cities for both the mergers' models and 
the market concentration models--for the mergers' models, we used the 
average prices instead of prices of only the merging companies because 
the average rack prices better capture competition at the racks before 
and after the mergers.

Explanatory Variables: 

Several of the explanatory variables we used in our models have been 
used in previous studies of wholesale gasoline prices. We used the 
following variables in our basic models.[Footnote 87]

* MERGERS: In the mergers models we used dummy variables for each of 
the mergers (e.g., an Exxon-Mobil dummy variable for the merger between 
Exxon and Mobil) to determine the average differences in wholesale 
gasoline prices before and after the respective mergers.

* HHI (Herfindahl-Hirschman Index): In the market concentration model, 
we used an index of market concentration, the HHI, to determine the 
effects of market concentration on wholesale gasoline prices. The 
effects of market concentration incorporate mergers' effects because 
mergers increase this measure of market concentration in an amount that 
is specific to each merger. The market concentration data are based on 
refinery capacity at the refinery (or PADD) level, a higher level of 
aggregation than the rack-city level. We believe that the source of 
potential market power in the wholesale gasoline market is at the 
refinery because, as already indicated, the refinery market is 
imperfectly competitive and refiners essentially control gasoline sales 
at the racks. For instance, using market concentration data for Nevada 
(in PADD V) based on gasoline sales is less meaningful because gasoline 
sold in Nevada comes mainly from California (also in PADD V). 
Furthermore, using refinery capacity data is an appropriate measure of 
concentration in the wholesale gasoline market because refinery 
capacity captures the ability of the suppliers (refiners) to 
produce.[Footnote 88]

* CRUDE: We included the cost of crude oil (CRUDE), which has the 
largest share of input cost used in the production of gasoline, 
although for econometric and interpretative purposes we used it as part 
of the dependent variable.[Footnote 89]

* INVENTORIES RATIO: The ratio of gasoline inventories to expected 
demand (INVENTORIES RATIO)[Footnote 90] could affect the availability 
of gasoline at the wholesale level and, hence, prices--prices will 
increase if inventories are low relative to demand and decrease if 
inventories are high relative to demand.

* UTILIZATION RATES: The level of refinery capacity utilization rates 
(UTILIZATION RATES) could impact wholesale gasoline prices through 
changes in supply. Although the data for UTILIZATION RATES are 
available only at the national level and do not allow us to account for 
differences in utilization rates across the United States, the data are 
still useful because gasoline is mostly fungible, especially in the 
eastern part of the United States.

* SUPPLY DISRUPTIONS: We also included dummy variables to account for 
the supply disruptions that occurred in the Midwest in June 2000 (MW 
CRISIS) and in the West Coast during periods in 1999 and 2000 (WC 
CRISIS). These disruptions contributed to price spikes in these 
markets. We based our information on the Midwest and the West Coast 
disruptions on FTC's report[Footnote 91] and a study by FTC staff, 
respectively.[Footnote 92] In both cases, the supply disruptions were 
identified by simply comparing over time the price differences between 
the assumed affected areas and an assumed unaffected city on the Gulf 
Coast. In the case of the West Coast, the authors attributed observed 
spikes to refinery and/or pipeline problems carried in the trade press 
during the period. Although these studies seem to imply that the 
disruptions were regional in scope (Midwest--PADD II and--West Coast--
PADD V), it is difficult to determine the geographical scope of these 
disruptions or their timing and duration. In particular, the geographic 
scope of these disruptions could be smaller or bigger than the entire 
region depending on the fungibility of gasoline in the area. 
Nonetheless, as part of our sensitivity analysis, we used dummy 
variables to construct measures of the Midwest and the West Coast 
supply disruptions based on the assumption that these disruptions were 
regional in scope. We therefore consider our measures of these supply 
disruptions to be crude, at best.

Table 13 presents the expected effects of all the explanatory variables 
used in our models.[Footnote 93],,

Table 13: Expected Effects of Key Explanatory Variables on Wholesale 
Gasoline Prices: 

Explanatory variable: MERGER dummy; (e.g., EXXON-MOBIL); 
Expected effect: Uncertain; 
Explanation of expected effect on wholesale gasoline prices: Mergers 
have both market power effects and efficiency effects, which increase 
and reduce prices, respectively.[A].

Explanatory variable: HHI; 
Expected effect: Uncertain; 
Explanation of expected effect on wholesale gasoline prices: Increased 
market concentration can have both market power effects and efficiency 
effects, which increase and reduce prices, respectively.[B].

Explanatory variable: INVENTORIES RATIO; 
Expected effect: Decrease; 
Explanation of expected effect on wholesale gasoline prices: While an 
increase in the ratio of gasoline inventories to expected demand leads 
to high inventory costs, the increase provides more protection against 
unexpected or not fully anticipated supply problems, decreasing prices.
[C].

Explanatory variable: UTILIZATION RATES; 
Expected effect: Uncertain; 
Explanation of expected effect on wholesale gasoline prices: An 
increase in refinery capacity utilization rate will generally increase 
output levels, hence lower prices, but when the utilization rates are 
at an already very high level, higher utilization would increase costs 
and prices.

Explanatory variable: SUPPLY DISPRUPTIONS: MW CRISIS, WC CRISIS; 
Expected effect: Increase; 
Explanation of expected effect on wholesale gasoline prices: The supply 
disruptions, by decreasing available supply relative to demand, would 
increase prices.[D]. 

Sources: GAO analysis of previous studies on gasoline pricing.

[A] See, for example, Chouinard and Perloff (2001) and Hastings and 
Gilbert (2002); see, also Karikari et al. (2002) for railroad mergers, 
Kim and Singal (1993) for airline mergers, and Vita and Sacher (2001) 
for hospital mergers.

[B] See, for example, Borenstein and Shepard (1996b) and Hastings and 
Gilbert (2002).

[C] See, for example, Pinkse et al. (2002) who used changes in 
inventories.

[D] See EIA (2001) and Taylor and Fischer (2001).

[End of table]

It has been suggested in a previous study that spatial-price 
competition is important in U.S. wholesale gasoline markets.[Footnote 
94] Essentially, although wholesale gasoline is physically an almost 
completely homogeneous product, its geographic location could 
differentiate it from the same product in another location, implying 
that prices at the nearest neighboring rack city could influence prices 
at a rack city. We did not, however, incorporate this variable directly 
in our models because there is co-movement between the nearest price 
variable and prices since both variables are likely to be generated by 
the same set of independent variables.[Footnote 95] More importantly, 
if there are omitted regional or local variables that drive wholesale 
prices, then the nearest prices will be a strong predictor of prices, 
even if the suppliers at nearby racks do not compete. So the nearest 
prices might not actually be estimating the true spatial effects but 
simply picking up the effects of the omitted variables. However, 
dropping the nearest prices is likely to introduce correlation across 
residuals of prices at nearby racks, which could benefit from 
correction. While the distances between racks and the nearest racks 
could have helped capture the effect of spatial competition, we could 
not estimate this effect because the data do not vary across time 
within a rack city--they are time invariant. Consequently, we addressed 
the issue of spatial competition through a variance adjustment 
procedure for the error terms.[Footnote 96]

It is also likely that in markets where both branded gasoline and 
unbranded gasoline are sold, the prices of one brand could affect the 
prices of the other. However, this relationship may be less important 
than spatial-price competition for several reasons. First, the 
correlation between the prices of one brand at a rack and the nearest-
neighbor prices of the same brand was higher than the correlation 
between prices of branded and unbranded gasoline at the same 
rack.[Footnote 97] Second, a major supplier typically supplies both 
branded and unbranded gasoline at a rack but is less likely to operate 
in the nearest rack, implying that brand competition is less likely 
than spatial-price competition.

We could not estimate the effects of states' divorcement regulations, 
which restrict ownership of retail gasoline stations by gasoline 
refiners, because the data are time-invariant. The effect of this 
regulation on wholesale gasoline prices, from a theoretical 
perspective, is uncertain.[Footnote 98]

Data Sources and Sample Selection: 

In analyzing the effects of mergers and market concentration on 
wholesale gasoline prices, we used all available data from all the 
racks in the contiguous United States. Using the OPIS (Oil Price 
Information Services) rack data, we performed several data-processing 
tasks, including matching the OPIS Rack data to data from several other 
sources. The data covered three types of regular, unleaded gasoline--
conventional gasoline from 1994 through 2000, reformulated gasoline 
from 1995 through 2000, and CARB gasoline from 1996 through 2000. In 
addition, we used gasoline data from the EIA and merger data from the 
FTC and Thomson Financial.

Data Sources: 

The wholesale price data were obtained from the OPIS rack data, which 
are posted rack prices at the racks. The data are collected from more 
than 350 racks, which represent over 90 percent of the racks in the 
United States, and information on companies that supply gasoline at the 
racks, the gasoline specification, and the gasoline brand. We also 
obtained (1) data from the EIA, including crude oil prices, gasoline 
inventories, refinery capacity for the construction of market 
concentration data, and refinery capacity utilization rates and (2) 
merger data from the Federal Trade Commission and Thomson Financial. 
Table 14 lists the variables that we constructed and the data sources.

Table 14: Variables in Our Econometric Analysis of Wholesale Gasoline 
Prices: 

Variable: PRICES; 
Definition: Wholesale gasoline prices (cents per gallon, 2000 dollars); 
Variable: BRANDED; 
Definition: Branded; 
Variable: UNBRANDED; 
Definition: Unbranded; 
Data source: OPIS; ERP; 
Data frequency, level: Weekly, City.

Variable: CRUDE; 
Definition: Crude oil spot prices (cents per gallon, 2000 dollars): 
West Texas Intermediate (WTI); 
Data source: EIA; ERP; 
Data frequency, level: Weekly, National.

Variable: HHI; 
Definition: Market concentration, measured by the HHI of refinery 
capacity; 
Data source: EIA; GAO analysis; 
Data frequency, level: Yearly,[A]; PADD.

Variable: TOSCO-UNOCAL; 
Definition: Dummy variable for the Tosco-Unocal merger, equals 1 if 
postmerger period (from 4/1/1997 to 12/31/2000), 0 otherwise; 
Data source: TF[B]; OPIS; 
Data frequency, level: Weekly, City.

Variable: UDS-TOTAL; 
Definition: Dummy variable for the UDS-Total merger, equals 1 if 
postmerger period (from 10/1/1997 to 12/31/2000), 0 otherwise; 
Data source: FTC[C]; OPIS; 
Data frequency, level: Weekly, City.

Variable: MARATHON-ASHLAND; 
Definition: Dummy variable for the Marathon-Ashland merger, equals 1 if 
postmerger period (from 1/5/1998 to 12/31/2000), 0 otherwise; 
Data source: TF[B]; OPIS; 
Data frequency, level: Weekly, City.

Variable: SHELL-TEXACO I (Equilon); 
Definition: Dummy variable for the Shell-Texaco I merger, equals 1 if 
postmerger period (from 2/1/1998 to 12/31/2000), 0 otherwise; 
Data source: FTC[C]; OPIS; 
Data frequency, level: Weekly, City.

Variable: SHELL-TEXACO II (Motiva); 
Definition: Dummy variable for the Shell-Texaco II merger, equals 1 if 
postmerger period (from 7/1/1998 to 12/31/2000), 0 otherwise; 
Data source: TF[B]; OPIS; 
Data frequency, level: Weekly, City.

Variable: BP-AMOCO; 
Definition: Dummy variable for the BP-Amoco merger, equals 1 if 
postmerger period (from 12/31/1998 to 12/31/2000), 0 otherwise; 
Data source: TF[B]; OPIS; 
Data frequency, level: Weekly, City.

Variable: MAP-UDS; 
Definition: Dummy variable for the MAP-UDS merger, equals 1 if 
postmerger period (from 12/13/1999 to 12/31/2000), 0 otherwise; 
Data source: TF[B]; OPIS; 
Data frequency, level: Weekly, City.

Variable: EXXON-MOBIL; 
Definition: Dummy variable for the Exxon-Mobil merger, equals 1 if 
postmerger period (from 3/1/2000 to 12/31/2000), 0 otherwise; 
Data source: FTC[C]; OPIS; 
Data frequency, level: Weekly, City.

Variable: INVENTORIES RATIO; 
Definition: Ratio of gasoline inventories to expected demand. Gasoline 
inventories are one-period lagged levels of normalized gasoline 
inventories, and expected demand is the fitted values from a regression 
equation of a normalized volume of gasoline sales; 
Data source: EIA; GAO analysis[D]; 
Data frequency, level: Weekly, PADD.

Variable: UTILIZATION RATES; 
Definition: Refinery capacity utilization rates (in percent); 
Data source: EIA; 
Data frequency, level: Weekly, National.

Variable: MW CRISIS; 
Definition: Dummy variable for Midwest gasoline supply disruption--
equals 1if June 2000 and PADD II, 0 otherwise; 
Data source: FTC(2001a, Figure 2); 
Data frequency, level: Weekly, City.

Variable: WC CRISIS; 
Definition: Dummy variable for West Coast gasoline supply disruptions 
in 1999 and 2000--equals 1 for 3/5/99-9/10/99, 2/12/ 00-5/6/00, and 
7/10/00-12/31/00, in the West Coast, 0 otherwise; 
Data source: Taylor and Fischer (2002); EIA(2001); 
Data frequency, level: Weekly, City.

Variable: WEEKS[E]; 
Definition: Dummy variables for the 52 weeks in a year--equals 1 for 
each week of the year (e.g., Week 1), 0 otherwise; 
Data source: NA; 
Data frequency, level: Weekly, NA.

Variable: TREND[E]; 
Definition: Time trend; 
Data source: NA; 
Data frequency, level: NA.

Variable: TREND SQUARED[E]; 
Definition: Square of TREND; 
Data source: NA; 
Data frequency, level: NA.  

Legend: 

BP = British Petroleum: 

EIA = Energy Information Administration (Department of Energy): 

ERP = Economic Report of the President (February 2002, table B-66, p. 
397): 

FTC = Federal Trade Commission: 

MAP = Marathon Ashland Petroleum: 

NA = Not applicable: 

OPIS = Oil Price Information Services: 

TF = Thomson Financial: 

UDS = Ultramar Diamond Shamrock: 

Source: GAO analysis of EIA, FTC, OPIS, and Thomson Financial data.

[A] Data were not available for 1996 and 1998, and we constructed data 
for the missing years using the average of the two adjacent years.

[B] The effective date is the merger completion date.

[C] The effective date is when FTC's merger remedies became effective.

[D] Gasoline inventories were normalized using the PADD mean over the 
sample period. The demand for wholesale gasoline was based on prime 
suppliers' sales of total regular gasoline in each state. We used an 
approach similar to the Borenstein and Shepard's (1996b) study to 
estimate the demand for gasoline. A simplified demand equation, in 
reduced form, for each state was obtained using the following 
regression equation: 

[See PDF for formula]

NVOLUME is the normalized monthly demand for wholesale gasoline in each 
state--prime suppliers' sales of gasoline in each state divided by the 
state mean over the sample period. The data for prime suppliers' sales 
was obtained from the EIA. MONTHj is a monthly dummy variable, and 
TREND and TREND_SQUARED are time trend and square of time trend, 
respectively. The R2 of these predicting equations varied between 0.50 
and 0.96. The expected demand is the fitted values from estimating the 
regression equation above because it is assumed that suppliers' form 
their expectations of next-period demand based on current and past 
sales volumes observed in their markets. The expected demands for the 
states were aggregated to the PADD level to match the data for the 
inventories.

[E] The variables are instruments.

[End of table]

Selection of Geographic Markets and Gasoline Types: 

Although there is no consensus on which geographic areas across the 
United States constitute separate wholesale gasoline markets because of 
the difficulty in defining true geographic market areas, many industry 
experts generally identify a rack city as an appropriate geographic 
market. Rack cities are well defined and generally cover small 
geographic areas.[Footnote 99] Our analysis is therefore based on 
racks. The selection of the geographic areas, the gasoline 
specifications (conventional, CARB, and reformulated), and time periods 
of the analysis was based primarily on the availability of data, after 
merging and matching data from the different sources. The conventional 
gasoline contains no additive, but reformulated gasoline and CARB 
contain MTBE (methyl tertiary butyl ether) as an additive. For the 
mergers models, we used data for conventional gasoline for each of the 
mergers, except the Tosco-Unocal merger, which affected primarily 
California, where CARB gasoline is used. Data for CARB gasoline were 
used for the Shell-Texaco I and Tosco-Unocal mergers. We used 
reformulated gasoline data for the BP-Amoco, Exxon-Mobil, Marathon-
Ashland, and Shell-Texaco II mergers since they affected the East Coast 
and the Gulf Coast, the predominant markets for reformulated gasoline. 
Data for the mergers and market concentration models were based on rack 
cities that were directly affected by the mergers and rack cities not 
affected by these mergers. We had data for conventional gasoline, the 
dominant type of gasoline, from all five regional geographic regions--
data for over 280 rack cities (for branded) and over 250 rack cities 
(for unbranded) out of the over 350 rack cities in the OPIS 
database.[Footnote 100] The data for conventional gasoline were 
available from 1994 through 2000 (specifically, 2/3/94-12/31/00), 
reformulated gasoline from 1995 through 2000 (specifically, 3/2/95-12/
31/00), and CARB gasoline from 1996 through 2000 (specifically, 5/16/
96-12/31/00).

Specification of Econometric Models and Estimation Methodology: 

We used quasi reduced-form price models to analyze the effects of 
mergers and market concentration on wholesale gasoline prices because 
such models have been found to be useful in previous studies.[Footnote 
101] Two types of models were developed and estimated--one for the 
effects of the eight individual mergers and the other for the effects 
of market concentration on wholesale gasoline prices. We used 
econometric techniques appropriate for estimating our panel data--the 
fixed-effects estimator in the context of a feasible generalized least 
squares (FGLS) technique to account for contemporaneous cross-sectional 
correlations and corrections for heteroskedasticity and first-order 
autocorrelation.[Footnote 102]

Model Specifications: 

A useful methodology for estimating the effects of oil industry mergers 
on wholesale prices is to compare prices in the affected markets before 
and after the merger. One method relates wholesale gasoline prices in 
markets affected by the merger to prices in control markets that were 
not impacted by the merger or other mergers within the time frame of 
the study, after controlling for appropriate variables.[Footnote 103] 
Another method relates wholesale gasoline prices in the affected 
markets to a merger-related variable and demand and cost 
shifters.[Footnote 104] Our approach is a blend of both approaches, 
which requires that we specify a quasi reduced-form relationship for 
prices that is a function of market structure and regulatory factors, 
and other supply and demand factors, using a broad panel data of rack 
cities comprising those affected and not affected by the mergers.

Using panel data--data across markets (racks) and over time--the quasi 
reduced-form relationship for wholesale gasoline prices can generally 
be specified as follows: [Footnote 105]

[See PDF for formula]

[End of figure]

In the estimation of equation (1), ni could be treated either as fixed 
or random. We later discuss our choice of the estimation technique 
based on the context of the data, among other factors. The error 
component in equation 1 is given a first-order autoregressive error 
structure, AR(1), to help capture the dynamic effects of wholesale 
gasoline prices. More important, our statistical tests indicated that 
the error terms are AR(1), based on the estimated autocorrelation 
coefficients. Also, the statistical tests we performed indicated that 
wholesale gasoline prices (for some gasoline specifications and types) 
and crude oil prices were each nonstationary--specifically, they were 
integrated of order one, I(1), using the adjusted Dickey-Fuller (ADF) 
test for unit root.[Footnote 106] Hence, we used wholesale gross 
prices--wholesale prices less crude oil prices--to obtain stationary 
series in addition to helping to capture the market power and 
efficiency effects of mergers and market concentration.[Footnote 107]

Using the general specification provided in equation 1, we estimated 
different equations for the effects of mergers and the effects of 
market concentration on wholesale gasoline prices. We used the 
following basic equations to determine the effects of individual 
mergers and market concentration on wholesale prices of different 
specifications of gasoline (conventional, reformulated, and CARB), and 
gasoline types (branded and unbranded), using panel data of weekly data 
and racks.

Mergers Model: 

[See PDF for formula]

[End of figure]

Market Concentration Model: 

[See PDF for formula]

[End of figure]

Model Estimation Techniques: 

Our econometric analyses are based on panel data, which pool cross-
sectional and time-series data.[Footnote 109] The cross-sectional data 
are based on racks for wholesale gasoline, and the time-series data are 
weekly. Several econometric issues have to be dealt with in estimating 
the effects of mergers and market concentration on prices, using 
equations (2) and (3), respectively, and panel data. First, the 
unobserved city-specific error component could be treated as fixed or 
random. The fixed-effects estimator is preferred when observations are 
not chosen randomly and there are likely to be unobservable, site-
specific effects (see, for example, Hsiao, 2003). This estimator can be 
implemented by demeaning the data by rack city (i.e., transforming the 
data into mean-deviations). In wholesale gasoline markets such 
unobserved differences might include unmeasured supply or demand 
effects, such as different pricing strategies of the refiners at the 
different rack cities and the level of development of the 
transportation system in the different areas. A major advantage of the 
fixed-effects estimator is that there is no need to assume that the 
unobserved city-specific effects are independent of the included 
explanatory variables. Furthermore, since the selection of the rack 
cities used in our study was not randomly drawn but was based on data 
availability, we prefer the fixed-effects estimator for this study. 
This estimator allows us to account for variations in wholesale prices 
across the racks that we could not explicitly account for, such as 
transportation costs. On the other hand, the random-effects estimator 
allows one to include a time-invariant variable. Also, the random-
effects estimator allows one to make unconditional (marginal) 
inferences with respect to the population of all effects. However, one 
has to make specific assumptions about the pattern of correlation (or 
assume no correlation) between the unobserved effects and the included 
explanatory variables. The need for these assumptions is a major 
shortcoming of the random-effects estimator because there are reasons 
to believe that the assumption of no correlation may not be correct for 
wholesale gasoline markets and could bias the estimates.

Second, in both the mergers and the HHI models, we focus on their 
effects on prices, conditional on other variables in the price 
equations. Since two of the explanatory regressors in the price 
equations might be endogenous--INVENTORIES RATIO and UTILIZATION RATES-
-we test for their endogeneity using the Hausman (1978) specification 
test.[Footnote 110] In all the models, the endogenous regressors are 
instrumented using these excluded exogenous variables--time trend, time 
trend squared, and 52 seasonal weekly dummies--as well as the included 
exogenous variables in the respective models. In each case, the 
instrumented endogenous regressor is the predicted value in a 
regression of the corresponding endogenous regressor on all the 
instruments--both the excluded exogenous regressors and the included 
exogenous regressors in the respective models. Essentially, the 
instruments are used to purge the potential endogenous regressors of 
their correlations with the prices to obtain consistent estimates. If 
exogeneity of the variables is rejected, we use the instrumental 
variable method. Otherwise, we use the least squares method. 
Furthermore, if exogeneity of the variables is rejected, we check the 
appropriateness of the instruments (test of the overidentifying 
restrictions) using the Hausman (1978) test.[Footnote 111]

In a merger equation, the effect of a merger on prices is captured by 
the coefficient estimate for that merger dummy. This estimate measures 
the change in the mean of price conditional on the covariates in this 
regression equation. Technically, this is the estimate of the partial 
derivative of the conditional mean of price, where the conditioning set 
contains all regressors including inventory and capacity utilization.

It should be noted that the instrumented regressors in the instrumental 
variable estimation are not based on a true first-stage regression 
since we do not have a fully specified system of simultaneous 
equations. In particular, we do not specify a model for INVENTORY RATIO 
or for UTILIZATION RATES because our main interest is in price. Our 
estimation should, therefore, be interpreted as a single-equation 
instrumental variable estimation. A consistent estimation of the price 
equation requires that INVENTORY RATIO and UTILIZATION RATES or the 
instruments used for them be uncorrelated with the regression error, 
and we used tests for endogeneity of the regressors and exogeneity of 
the instruments to check this requirement.

Third, the regression errors are perhaps contemporaneously correlated 
across rack cities because they capture all unobservables impacting 
various rack cities at the same time. Depending on the outcome of the 
endogeneity test, we used a Feasible Generalized Least Squares/
Instrumental Variables estimation (FGLS/IV) or just the FGLS estimation 
as the proper method of inference. In either case, we accounted for 
both contemporaneous correlations and groupwise heteroskedasticity, 
and the estimation is done using panel data in a fixed-effect 
context.[Footnote 112]

Fourth, the regression errors might be serially correlated. We used the 
FGLS/IV or FGLS estimator assuming a first-order autoregressive 
structure, AR(1). We tested for the presence of AR(1) by regressing the 
residuals from the preferred estimator--FGLS/IV or FGLS--depending on 
the outcome of the endogeneity test, on one-period lagged residuals, 
and testing for significance of the coefficient.[Footnote 113]

Econometric Results: 

Our econometric results show that: 

* mergers generally increased wholesale gasoline prices,

* increased market concentration resulted in higher wholesale gasoline 
prices, and: 

* low gasoline inventories, high refinery capacity utilization rates, 
and supply disruptions increased wholesale gasoline prices.

Mergers Generally Increased Wholesale Gasoline Prices: 

Mergers, by reducing the number of suppliers of wholesale gasoline, 
affect market concentration, and hence have predicted effects on 
wholesale prices. As shown in tables 15-17, we found that wholesale 
prices generally: 

increased as result of mergers, but there were also some 
decreases.[Footnote 114] For conventional gasoline, the mergers 
resulted in increases in prices of wholesale gasoline for five of the 
seven mergers (see table 15). In particular, our model results show 
that the mergers of Marathon-Ashland, Shell-Texaco I (Equilon), BP-
Amoco, MAP-UDS, and Exxon-Mobil had increases in the prices of both 
branded and unbranded gasoline, while the mergers of UDS-Total and 
Shell-Texaco II (Motiva) resulted in decreases in prices. In table 16, 
for reformulated gasoline, the Marathon-Ashland and Exxon-Mobil mergers 
increased prices while the Shell-Texaco II (Motiva) merger decreased 
only the prices of branded gasoline. The effects of the Shell-Texaco II 
(Motiva) merger on unbranded gasoline and the BP-Amoco merger on both 
branded and unbranded gasoline prices were inconclusive. Our estimates 
in table 17 also show that for CARB gasoline, the Tosco-Unocal merger 
increased prices of branded gasoline while the Shell-Texaco I (Equilon) 
merger decreased prices of branded gasoline. The effects of the two 
mergers on unbranded gasoline were inconclusive.

The estimates in tables 15-17 are summaries of the effects of the 
individual mergers on wholesale gasoline prices (using the mergers 
models) for different gasoline specifications--conventional, 
reformulated, and CARB--and their branded and unbranded 
varieties.[Footnote 115] The full econometric estimates are provided in 
tables 21-23, and they show that all the estimated relationships are 
highly statistically significant based on the models' probability 
values (p-values).[Footnote 116] The estimates presented in tables 15-
17 are based on the estimates in tables 21-23 that include the supply 
disruptions in the Midwest in 2000 and/or in the West Coast in 1999 and 
2000, shown in column (ii) for branded gasoline and column (iv) for 
unbranded gasoline. The R-squares for these estimates range from 20 
percent to 36 percent.[Footnote 117] The autocorrelation coefficients 
indicate the presence of autocorrelation in the error terms, which we 
accounted for in the estimation process by specifying a first-order 
autoregressive structure (see the tests of autocorrelation in tables 
21-23). The Hausman (1978) specification tests indicated that the 
preferred estimator for unbranded conventional gasoline and unbranded 
CARB gasoline was the instrumental-variables (IV) technique; the other 
estimates were based on the least squares estimates.

In chapter 5, we discussed previous studies on mergers affecting 
gasoline markets, including a recent study by FTC staff, sent to us on 
March 24, 2004.[Footnote 118] Here, we provide a more detailed 
assessment of the FTC study because the study examines one of the 
mergers that we studied; it is also, to our knowledge, the first public 
retrospective analysis of mergers in the petroleum industry done by FTC 
staff. In the study, FTC staff examined the economic effects of the 
Marathon-Ashland merger and found that this merger increased wholesale 
prices of reformulated gasoline in Louisville, Kentucky, by 3 to 5 
cents per gallon during the period they analyzed--1998 and 1999. They 
argued, independent of their statistical analysis, that the increase 
was due to increased demand from St. Louis, Missouri, which switched to 
reformulated gasoline during the period of the study and not due to the 
merger. Furthermore, they found that retail prices at gasoline stations 
supplied by rack distributors did not increase, presumably due to 
competition from retailers of reformulated gasoline supplied directly 
by refiners and retailers of conventional gasoline that did not 
experience increases in their relative wholesale prices.

Although the increase in wholesale prices of reformulated gasoline 
found by the FTC is consistent with our findings, the study has 
shortcomings in several related areas, including sampling, econometric 
methodology, and interpretation of results. First, the FTC study uses 
prices in three selected control cities (Chicago, Houston, and Northern 
Virginia, which we believe includes Fairfax) to help separate the 
merger's effects from other demand and supply effects. We believe that 
all three cities fail to meet the essential requirement of a control 
unit--that the control cities and the city of interest are nearly 
identical, except for the Marathon-Ashland merger, in terms of demand 
and supply conditions of gasoline. For instance, the Marathon-Ashland 
merger affected the wholesale gasoline market in Fairfax, which would 
make Northern Virginia an inappropriate control city for this 
merger.[Footnote 119] Furthermore, other key mergers affected the 
control cities, making the control cities inappropriate. Specifically, 
the Shell-Texaco II (Motiva) merger in July 1998 affected Fairfax and 
Houston, and the BP-Amoco merger in December 1998 affected Fairfax. 
Also, the seasonal demand factors may be different between Louisville 
and the control cities.

Second, the FTC study does not take into account the potential effects 
of the BP-Amoco merger, which occurred in December 1998 and affected 
the wholesale gasoline market in Louisville. This makes it difficult to 
separate the effects of the Marathon-Ashland merger from the effects of 
the BP-Amoco merger in 1999, severely limiting the interpretation of 
the results.

Third, FTC argued that increased demand from St. Louis was solely 
responsible for the increased wholesale prices. However, the FTC study 
did not explicitly include demand from St. Louis and so it is not 
evident how much of the increase in prices was due to the Marathon-
Ashland merger and how much was due to the increased demand from St. 
Louis. Interpreting the price increase in wholesale prices as an 
artifact of St. Louis' entry into the reformulated gasoline market 
without such evidence confounds FTC's interpretation of the effects of 
the merger. Furthermore, even if the increased demand from St. Louis 
was potentially responsible for the price increase found in FTC's study 
in 1999, FTC's study fails to explain the price increase in 1998, prior 
to the switch to reformulated gasoline in St. Louis in 1999. Finally, 
using only one market (city) unnecessarily reduces the scope of 
findings for the impact of the merger.

Tables 15-17 present the results of our model showing the effects of 
each of the individual mergers on wholesale gasoline prices in the 
racks that were affected by those mergers. The cumulative effects of 
all the mergers, as well as the effects of other market structure 
factors, are estimated using market concentration, which is a 
comprehensive measure of market structure. The effects of market 
concentration on wholesale prices are presented in the next section.

Table 15: Effects of Mergers on Conventional Wholesale Gasoline Prices 
(1994-2000): 

Merger: UDS-Total; 
Geographic: region[A]: PADD II, III, IV; 
Estimates are obtained using data for: premerger period: 2/3/94-9/30/97; 
Estimates are obtained using data for: postmerger period [B]: 10/1/97-
1/31/98.

Merger: UDS-Total: Branded; 
Estimated change in price margin (cents per gallon): - 0.89[C].

Merger: UDS-Total: Unbranded; 
Estimated change in price margin (cents per gallon):  - 1.25[C].

Merger: Marathon-Ashland; 
Geographic: region[A]: PADD I, II, III; 
Estimates are obtained using data for: premerger period: 2/3/94-1/4/98; 
Estimates are obtained using data for: postmerger period[B]: 1/5/98-
6/30/98.

Merger: Marathon-Ashland: Branded; 
Estimated change in price margin (cents per gallon):  0.70[C].

Merger: Marathon-Ashland: Unbranded; 
Estimated change in price margin (cents per gallon):  0.39[C].

Merger: Shell-Texaco I; 
Geographic: region[A]: PADD II, III, IV, V; 
Estimates are obtained using data for: premerger period: 2/3/94-1/31/98; 
Estimates are obtained using data for: postmerger period[B]: 2/1/98-
12/30/98.

Merger: Shell-Texaco I: Branded; 
Estimated change in price margin (cents per gallon):  0.99[C].

Merger: Shell-Texaco I: Unbranded; 
Estimated change in price margin (cents per gallon):  1.13[C].

Merger: Shell-Texaco II; 
Geographic: region[A]: PADD I, II, III; 
Estimates are obtained using data for: premerger period: 1/5/98-
6/30/98; 
Estimates are obtained using data for: postmerger period[B]: 7/1/98-
12/30/98.

Merger: Shell-Texaco II: Branded; 
Estimated change in price margin (cents per gallon):  - 1.77[C].

Merger: Shell-Texaco II: Unbranded; 
Estimated change in price margin (cents per gallon):  - 1.24[C].

Merger: BP-Amoco; 
Geographic: region[A]: PADD I, II, III; 
Estimates are obtained using data for: premerger period: 7/1/98-
12/30/98; 
Estimates are obtained using data for: postmerger period[B]: 12/31/98-
2/29/00.

Merger: BP-Amoco: Branded; 
Estimated change in price margin (cents per gallon):  0.40[C].

Merger: BP-Amoco: Unbranded; 
Estimated change in price margin (cents per gallon):  0.97[C].

Merger: MAP-UDS; 
Geographic: region[A]: PADD II; 
Estimates are obtained using data for: premerger period: 12/31/98-
12/12/99; 
Estimates are obtained using data for: postmerger period[B]: 12/13/99- 
12/31/00.

Merger: MAP-UDS: Branded; 
Estimated change in price margin (cents per gallon):  1.38[C].

Merger: MAP-UDS: Unbranded; 
Estimated change in price margin (cents per gallon):  2.63[C].

Merger: Exxon-Mobil; 
Geographic: region[A]: PADD I, III; 
Estimates are obtained using data for: premerger period: 12/31/98-
2/29/00; 
Estimates are obtained using data for: postmerger period[B]: 3/1/00-
12/31/00.

Merger: Exxon-Mobil: Branded; 
Estimated change in price margin (cents per gallon):  3.71[C].

Merger: Exxon-Mobil: Unbranded; 
Estimated change in price margin (cents per gallon):  5.00[C].

Source: GAO econometric analysis of EIA, FTC, OPIS, and Thomson 
Financial data.

Notes: The data are from 2/3/94 to 12/31/00.

See also table 21.

[A] PADD I=East Coast, PADD II=Midwest, PADD III=Gulf Coast, PADD IV=
Rocky Mountain, and PADD V=West Coast.

[B] The effective date, which is the first date in the postmerger 
period, is based on either the merger completion date or the date when 
FTC's merger remedies became effective. As shown in the table, when 
mergers closely followed each other, they tended to shorten the before- 
merger and after-merger time periods that we could model, especially 
when more than one merger affected the same rack cities. Nonetheless, 
we believe we had sufficient data for the analysis.

[C] The estimated changes in prices are statistically significant at 
the 1 percent level or lower.

[End of table]

Table 16: Effects of Mergers on Reformulated Wholesale Gasoline Prices 
(1995-2000): 

Merger[A]: Marathon-Ashland; 
Geographic: region[B]: PADD I, II; 
Estimates are obtained using data for: premerger period: 3/2/95-1/4/98; 
Estimates are obtained using data for: postmerger period[C]: 
1/5/98-6/30/98.

Merger[A]: Marathon-Ashland: Branded; 
Estimated change in price margin (cents per gallon): 0.71[D].

Merger[A]: Marathon-Ashland: Unbranded; 
Estimated change in price margin (cents per gallon): 0.86[D].

Merger[A]: Shell-Texaco II; 
Geographic: region[B]: PADD I, III; 
Estimates are obtained using data for: premerger period: 
1/5/98-6/30/98; 
Estimates are obtained using data for: postmerger period[C]: 
7/1/98-12/30/98.

Merger[A]: Shell-Texaco II: Branded; 
Estimated change in price margin (cents per gallon): - 0.39[E].

Merger[A]: Shell-Texaco II: Unbranded; 
Estimated change in price margin (cents per gallon): 0.09.

Merger[A]: BP-Amoco; 
Geographic: region[B]: PADD I, II; 
Estimates are obtained using data for: premerger period: 
7/1/98-12/30/98; 
Estimates are obtained using data for: postmerger period[C]: 
12/31/98-2/29/00.

Merger[A]: BP-Amoco: Branded; 
Estimated change in price margin (cents per gallon): 0.55.

Merger[A]: BP-Amoco: Unbranded; 
Estimated change in price margin (cents per gallon): 0.40.

Merger[A]: Exxon-Mobil; 
Geographic: region[B]: PADD I, III; 
Estimates are obtained using data for: premerger period: 
12/31/98-2/29/00; 
Estimates are obtained using data for: postmerger period[C]: 
3/1/00-12/31/00.

Merger[A]: Exxon-Mobil: Branded; 
Estimated change in price margin (cents per gallon): 1.61[D].

Merger[A]: Exxon-Mobil: Unbranded; 
Estimated change in price margin (cents per gallon): 1.01[E].

Source: GAO econometric analysis of EIA, FTC, OPIS, and Thomson 
Financial data.

Notes: The data are from 3/2/95 to 12/31/00.

See also table 22.

[A] No estimates are reported for the UDS-Total merger because data are 
available for only one rack city. See table 22 for details.

[B] PADD I=East Coast, PADD II=Midwest, and PADD III=Gulf Coast. PADD 
II had data for only one rack city.

[C] The effective date, which is the first date in the postmerger 
period, is based on either the merger completion date or the date when 
FTC's merger remedies became effective. As shown in the table, when 
mergers closely followed each other, they tended to shorten the before- 
merger and after-merger time periods that we could model, especially 
when more than one merger affected the same rack cities. Nonetheless, 
we believe we had sufficient data for the analysis.

[D] The estimated changes in prices are statistically significant at 
the 1 percent level or lower.

[E] The estimated changes in prices are statistically significant at 
the 5 percent level or lower.

[End of table]

Table 17: Effects of Mergers on CARB Wholesale Gasoline Prices (1996-
2000): 

Merger: Tosco-Unocal; 
Geographic: region[A]: PADD V; 
Estimates are obtained using data for: premerger period: 
5/16/96-4/10/97; 
Estimates are obtained using data for: postmerger period[B]: 
4/11/97- 1/31/98.

Merger: Tosco-Unocal: Branded; 
Estimated change in price margin (cents per gallon): 6.87[C].

Merger: Tosco-Unocal: Unbranded; 
Estimated change in price margin (cents per gallon):  Estimated change 
in price margin (cents per gallon): -1.58.

Merger: Shell-Texaco I; 
Geographic: region[A]: PADD V; 
Estimates are obtained using data for: premerger period: 
4/11/97-1/31/98; 
Estimates are obtained using data for: postmerger period[B]: 
2/1/98-12/31/00.

Merger: Shell-Texaco I: Branded; 
Estimated change in price margin (cents per gallon): - 0.69[C].

Merger: Shell-Texaco I: Unbranded; 
Estimated change in price margin (cents per gallon): -0.24. 

Source: GAO econometric analysis of EIA, FTC, OPIS, and Thomson 
Financial data.

Notes: The data are from 5/16/96 to 12/31/00.

See also table 23.

[A] PADD V=West Coast (only California).

[B] The effective date, which is the first date in the postmerger 
period, is based on either the merger completion date or the date when 
FTC's merger remedies became effective. As shown in the table, when 
mergers closely followed each other, they tended to shorten the before-
merger and after-merger time periods that we could model, especially 
when more than one merger affected the same rack cities. Nonetheless, 
we believe we had sufficient data for the analysis.

[C] The estimated changes in prices are statistically significant at 
the 5 percent level or lower.

[End of table]

Increased Market Concentration Resulted in Higher Wholesale Gasoline 
Prices: 

We show in tables 18 and 19 that increased market concentration in 
wholesale gasoline markets resulted in price increases for conventional 
gasoline, as well as for boutique fuels--reformulated gasoline and CARB 
gasoline. This finding is partly attributed to the mergers' reducing 
the number of suppliers in the wholesale gasoline markets. The changes 
in wholesale prices of conventional gasoline, however, varied across 
broad geographic regions partly because of differences in access to 
gasoline supplies from other refining centers of the country or from 
abroad. As shown in table 18, the increases in prices were larger in 
the western part of the United States (PADDs IV and V) than the eastern 
part (PADDs I, II, and III) for branded gasoline. In table 19, the 
wholesale prices of CARB gasoline (sold only in California) were 
substantially larger as a result of increased market concentration. 
This is partly due to the unique requirement in California as well as 
the state's relative isolation from the major refining centers in the 
Gulf Coast.[Footnote 120]

Tables 18 and 19 summarize our econometric estimates of the effects of 
market concentration on different gasoline specifications 
(conventional, reformulated, and CARB) and their branded and unbranded 
varieties, based primarily on the econometric results in tables 24-28. 
All the estimated models are highly statistically significant based on 
the models' probability values (or p-values). The estimates presented 
in tables 18 and 19 are based on the estimates in tables 24-28 that 
include the supply disruptions in the Midwest in 2000 and/or on the 
West Coast in 1999 and 2000 in column (ii) for branded gasoline and 
column (iv) for unbranded gasoline. Similar to the estimates for the 
mergers, the R-squares for these estimates range from 16 percent to 36 
percent.[Footnote 121] Also, the autocorrelation coefficients indicate 
the presence of first-order autoregressive error structure; see the 
tests of autocorrelation in tables 24-28. Furthermore, the Hausman 
(1978) specification tests indicated that the preferred estimator for 
unbranded conventional gasoline, broadly, and for unbranded CARB 
gasoline was the instrumental-variables (IV) technique; the other 
estimates were based on the least squares estimates.

Table 18: Effects of Market Concentration on Conventional Wholesale 
Gasoline Prices (1994-2000): 

All regions[A]: Branded; 
Market concentration (HHI): 1994: 803; 
Market concentration (HHI): 2000: 1101; 
Market concentration (HHI): Increase in HHI: 298; 
Estimated change in wholesale price margin due to increase in HHI 
(cents per gallon)[B]: 0.15[C].

All regions[A]: Unbranded; 
Market concentration (HHI): 1994: 803; 
Market concentration (HHI): 2000: 1101; 
Market concentration (HHI): Increase in HHI: 298; 
Estimated change in wholesale price margin due to increase in HHI 
(cents per gallon)[B]: 0.33[C].

Geographic area: Eastern United States (PADDs I, II, III): Branded; 
Market concentration (HHI): 1994: 773; 
Market concentration (HHI): 2000: 1090; 
Market concentration (HHI): Increase in HHI: 317; 
Estimated change in wholesale price margin due to increase in HHI 
(cents per gallon)[B]: 0.25[C].

Geographic area: Eastern United States (PADDs I, II, III): Unbranded; 
Market concentration (HHI): 1994: 773; 
Market concentration (HHI): 2000: 1090; 
Market concentration (HHI): Increase in HHI: 317; 
Estimated change in wholesale price margin due to increase in HHI 
(cents per gallon)[B]: 0.10.

Geographic area: Western United States (PADDs IV, V): Branded; 
Market concentration (HHI): 1994: 1032; 
Market concentration (HHI): 2000: 1180; 
Market concentration (HHI): Increase in HHI: 148; 
Estimated change in wholesale price margin due to increase in HHI 
(cents per gallon)[B]: 0.56[C].

Geographic area: Western United States (PADDs IV, V): Unbranded; 
Market concentration (HHI): 1994: 1032; 
Market concentration (HHI): 2000: 1180; 
Market concentration (HHI): Increase in HHI: 148; 
Estimated change in wholesale price margin due to increase in HHI 
(cents per gallon)[B]: 1.29[D].

Source: GAO econometric analysis of EIA, FTC, and OPIS data.

Notes: The data are from 2/3/94 to 12/31/00.

See also tables 24-26.

[A] All states except Alaska, California, Connecticut, Delaware, the 
District of Columbia, Hawaii, Massachusetts, New Hampshire, New Jersey, 
and Rhode Island. These states were excluded due to the lack of 
sufficient data.

[B] The changes in prices were obtained by multiplying the increases in 
HHI by the marginal effects (coefficients) of HHI in tables 24-26, 
columns (ii) and (iv) for branded and unbranded, respectively.

[C] The estimated changes in prices are statistically significant at 
the 1 percent level or lower.

[D] The estimated changes in prices are statistically significant at 
the 5 percent level or lower.

[End of table]

Table 19: Effects of Market Concentration on Wholesale Prices of 
Boutique Fuels (1995-2000): 

Reformulated wholesale gasoline: 1995-2000[A]: Branded; 
Market concentration (HHI): 1995: 1,237; 
Market concentration (HHI): 2000: 1,477; 
Market concentration (HHI): Increase in HHI: 240; 
Estimated change in price margin due to increase in HHI 
(cents per gallon)[B]: 0.98[C].

Reformulated wholesale gasoline: 1995-2000[A]: Unbranded; 
Market concentration (HHI): 1995: 1,237; 
Market concentration (HHI): 2000: 1,477; 
Market concentration (HHI): Increase in HHI: 240; 
Estimated change in price margin due to increase in HHI 
(cents per gallon)[B]: 0.89[C].

CARB reformulated wholesale gasoline: 1996-2000[D]: Branded; 
Market concentration (HHI): 1995: 965; 
Market concentration (HHI): 2000: 1,267; 
Market concentration (HHI): Increase in HHI: 302; 
Estimated change in price margin due to increase in HHI 
(cents per gallon)[B]: 7.19[C].

CARB reformulated wholesale gasoline: 1996-2000[D]: Unbranded; 
Market concentration (HHI): 1995: 965; 
Market concentration (HHI): 2000: 1,267; 
Market concentration (HHI): Increase in HHI: 302; 
Estimated change in price margin due to increase in HHI 
(cents per gallon)[B]: 7.94[E]. 

Source: GAO econometric analysis of EIA, FTC, and OPIS data.

Note: 

See also tables 27 and 28.

[A] The data are from 3/2/95 to 12/31/00 for Connecticut, Delaware, 
Maryland, Massachusetts, New Jersey, New York, Pennsylvania, Rhode 
Island, and Virginia in PADD I; Kentucky in PADD II; and Texas in PADD 
III.

[B] The changes in prices were obtained by multiplying the increases in 
HHI by the marginal effects (coefficients) of HHI in table 27 for 
reformulated and table 28 for CARB reformulated, columns (ii) and (iv) 
for branded and unbranded wholesale gasoline, respectively.

[C] The estimated changes in prices are statistically significant at 
the 1 percent level or lower.

[D] The data are from 5/16/96 to 12/31/00 for California.

[E] The estimated changes in prices are statistically significant at 
the 10 percent level or lower.

[End of table]

Low Gasoline Inventories, High Refinery Capacity Utilization Rates, and 
Supply Disruptions Increased Wholesale Gasoline Prices: 

We found that the effects of other factors on wholesale gasoline prices 
during the second half of the 1990s are generally consistent with 
expectations. We discuss the effects of the gasoline inventories and 
refinery capacity utilization rates on prices based on the regression 
results for conventional wholesale gasoline (the dominant gasoline type 
used in most geographic regions) presented in table 24 for branded and 
unbranded gasoline. The results for the supply disruptions are based on 
the estimates for conventional gasoline in table 24 and for CARB in 
table 28.[Footnote 122] Also, we used the market concentration model 
because market concentration better represents overall market structure  
conditions than mergers. Using results from the econometric models of 
pricing of wholesale conventional gasoline, we found that wholesale 
prices were higher when inventories were low relative to expected 
demand and when refinery capacity utilization rates were high. Also, 
both supply disruptions in the Midwest and in the West Coast regions 
were associated with higher gasoline prices.[Footnote 123]

Summary statistics for selected variables used in the econometric 
analysis are presented in table 20 based on data for conventional 
gasoline. The results show that wholesale prices (wholesale gasoline 
prices less crude oil prices) of branded gasoline exceeded those of 
unbranded gasoline by about 2 cents for conventional and reformulated 
gasoline, but by more for CARB. Also, the wholesale gasoline markets 
used in our study were, on average, close to moderately concentrated 
based on the HHI, with wide variations across states. The refinery 
capacity utilization rates averaged 94 percent. And there were wide 
variations in gasoline inventories and demand that generally reflect a 
seasonal pattern.

The estimates in table 24 show that the other explanatory factors used 
in the models generally have the expected effects. The subsequent 
discussions are based mainly on the estimates in columns (ii) and (iv) 
of the tables. The effect of INVENTORIES RATIO is unambiguously 
negative, which indicates that lower gasoline inventories (relative to 
demand) had the expected effect of increasing prices.[Footnote 124] In 
particular, prices are about 1 cent higher from about May to October, 
the summer driving months, when inventories are low relative to 
expected demand, compared to the period from about November to April 
when inventories are high relative to expected demand.[Footnote 125] We 
included refinery capacity utilization rates--a variable that has not 
been used in previous studies but has been suggested as influencing 
gasoline prices by industry experts--in the model to assess its impact 
on wholesale gasoline prices. In table 24, the results indicate that 
higher utilization rates are associated with higher prices--
particularly for the estimates for unbranded gasoline. A 1 percent 
increase in refinery capacity utilization rates resulted in about 0.10 
to 0.20 cent per gallon increase in prices. We found that prices were 
higher because high refinery capacity utilization rates in the oil 
refining industry leave little room for error in predicting short-run 
demand.

As shown in tables 24 and 28, both supply disruptions in the Midwest 
and the West Coast were associated with higher gasoline prices of 
branded and unbranded gasoline. The effects of both the Midwest and 
West Coast supply disruptions on prices ranged from about 4 to 5 cents 
per gallon for conventional gasoline, and the effects of the West Coast 
supply disruptions on CARB gasoline ranged from about 4 to 8 cents per 
gallon. Also, the price increases were slightly larger for unbranded 
gasoline than for branded, consistent with the fact that disruptions 
would reduce the supply of unbranded gasoline more than branded as 
refiners meet the demand from their branded distributors first.

Table 20: Selected Summary Statistics for Conventional Wholesale 
Gasoline Markets: 

Variable: BRANDED PRICES[A]: Conventional; 
Mean: 18.57; 
Standard deviation: 7.04; 
Minimum: -0.59;
Maximum: 72.98. 

Variable: BRANDED PRICES[A]: Reformulated; 
Mean: 19.70; 
Standard deviation: 5.74; 
Minimum: 4.06; 
Maximum: 55.86. 

Variable: BRANDED PRICES[A]: CARB; 
Mean: 35.53; 
Standard deviation: 13.15; 
Minimum: 2.59; 
Maximum: 96.51. 

Variable: UNBRANDED PRICES[B]: Conventional; 
Mean: 16.96; 
Standard deviation: 7.11; 
Minimum: -3.45; 
Maximum: 85.12. 

Variable: UNBRANDED PRICES[B]: Reformulated; 
Mean: 18.31; 
Standard deviation: 6.12;
Minimum: 3.19;
Maximum: 68.85. 

Variable: UNBRANDED PRICES[B]: CARB; 
Mean: 30.66; 
Standard deviation: 14.36; 
Minimum: 1.45; 
Maximum: 103.40.

Variable: HHI; 
Mean: 950; 
Standard deviation: 387; 
Minimum: 520; 
Maximum: 1827.

Variable: NUMBER OF SUPPLIERS; 
Mean: 10; 
Standard deviation: 5; 
Minimum: 1; 
Maximum: 26.

Variable: INVENTORIES[C]; 
Mean: 1.00; 
Standard deviation: 0.07; 
Minimum: 0.72; 
Maximum: 1.33.

Variable: DEMAND[D]; 
Mean: 1.004; 
Standard deviation: 0.15; 
Minimum: 0.77; 
Maximum: 1.30.

Variable: UTILIZATION RATES; 
Mean: 93.5%; 
Standard deviation: 3.2%; 
Minimum: 84.6%; 
Maximum: 100.5%.

Variable: DISTANCE (in miles); 
Mean: 48; 
Standard deviation: 29; 
Minimum: 2; 
Maximum: 208.

Variable: TERMINALS[E]; 
Mean: 12; 
Standard deviation: 8; 
Minimum: 1; 
Maximum: 34. 

Source: GAO analysis of Census Bureau, EIA, and OPIS data.

Note: Branded and unbranded prices are in cents per gallon.

[A] BRANDED PRICES are branded wholesale gasoline prices less crude oil 
prices.

[B] UNBRANDED PRICES are unbranded wholesale gasoline prices less crude 
oil prices.

[C] INVENTORIES are normalized inventories of wholesale gasoline at the 
PADD level.

[D] DEMAND is normalized expected sales for wholesale gasoline at the 
PADD level.

[E] TERMINALS is the number of racks in a state.

[End of table]

Our Econometric Methodology Had Some Limitations: 

There are some limitations to our methods for estimating the effects of 
individual mergers and market concentration on wholesale gasoline 
prices. First, the timing of a merger is based on the effective date of 
the merger provided by FTC, which is either the merger completion date 
or the date when FTC's merger remedies became effective if the merger 
was subject to remedies. Although the true effective dates of some 
mergers could be some time after these dates, we could not perform 
sensitivity tests on the timing of the mergers since changing the 
timing of one merger could coincide with the timing of another merger, 
as the mergers typically occurred very close to each other and there 
were overlaps in certain rack cities. In any case, the effective date 
is what most experts use to date mergers, and it is expected that using 
these dates would generally underestimate the effects of the mergers.

Second, the market concentration variable, measured by the HHI, was 
measured at the PADD level using refinery capacity. While we believe 
that in a vertically integrated gasoline market, market power is better 
captured by production of gasoline at the refinery level, the data for 
refinery capacity include the production of other products in addition 
to gasoline. Also, data were not available for two years (1996 and 
1998).[Footnote 126]

Third, some variables were only available at higher levels of 
aggregation than we would have preferred or were not publicly 
available. The gasoline inventories were available at the regional 
(PADD) level, and refining capacity utilization rates were available at 
the national level, instead of the city or even state level; however, 
these limitations are less important since gasoline is mostly fungible, 
particularly in the regions in the eastern half of the country (PADDs 
I, II, and III).

Fourth, to estimate the effects of mergers on prices, we would have 
preferred to use market shares of the merged companies. However, these 
data are not usually available because they are proprietary. We 
therefore determined the effects of the mergers by estimating the 
difference in average prices before and after the effective dates of 
the mergers. Also, because of the closeness of the timing of the oil 
industry mergers in the second half of the 1990s as well as the 
overlapping nature of the mergers, estimates from our econometric 
models captured the mergers' effects on price margins over shorter time 
periods.

Fifth, we could not obtain data that would directly capture possible 
vertical relationships between the refiners and marketers of gasoline, 
and the role of independent refiners and retailers. However, we 
attempted to capture some of these effects indirectly by performing 
separate analyses for gasoline types (branded and unbranded) since 
integrated refiners sell primarily branded gasoline, and independent 
refiners are dominant in the unbranded market nationwide.

Although there are limitations to our methodology of estimating the 
effects of mergers and market concentration on wholesale gasoline 
prices, our model specifications and results are generally consistent 
with previous studies.[Footnote 127]

Table 21: Econometric Estimates of Mergers' Effects on Conventional 
Wholesale Gasoline Prices: 

Independent variable: UDS-TOTAL; 
Branded: FGLS: (i): -1.0240[B] (0.0249); 
Branded: FGLS: (ii): - 0.8894[B] (0.0202); 
Unbranded: FGLS/IV: (iii): -1.2519[B] (0.0726); 
Unbranded: FGLS/ IV[A]: (iv): -1.2466[B] (0.0667).

Independent variable: MARATHON-ASHLAND; 
Branded: FGLS: (i): 0.9218[B] (0.0192); 
Branded: FGLS: (ii): 0.6995[B] (0.0149); 
Unbranded: FGLS/IV: (iii): 0.5859[B] (0.0573); 
Unbranded: FGLS/IV[A]: (iv): 0.3850[B] (0.0523).

Independent variable: SHELL-TEXACO I; 
Branded: FGLS: (i): 1.9289[B] (0.0362); 
Branded: FGLS: (ii): 0.9920[B] (0.0261); 
Unbranded: FGLS/IV: (iii): 2.0012[B] (0.0681); 
Unbranded: FGLS/IV[A]: (iv): 1.1345[B] (0.0569).

Independent variable: SHELL-TEXACO II; 
Branded: FGLS: (i): -1.7556[B] (0.0384); 
Branded: FGLS: (ii): -1.7686[B] (0.0332); 
Unbranded: FGLS/IV: (iii): -1.2156[B] (0.0912); 
Unbranded: FGLS/ IV[A]: (iv): -1.2406[B] (0.0919).

Independent variable: BP-AMOCO; 
Branded: FGLS: (i): 0.3303[B] (0.0176); 
Branded: FGLS: (ii): 0.4007[B] (0.0158); 
Unbranded: FGLS/IV: (iii): 0.7236[B] (0.0519); 
Unbranded: FGLS/IV[A]: (iv): 0.9679[B] (0.0836).

Independent variable: EXXON-MOBIL; 
Branded: FGLS: (i): 3.8154[B] (0.0788); 
Branded: FGLS: (ii): 3.7107[B] (0.0687); 
Unbranded: FGLS/IV: (iii): 5.0514[B] (0.1301); 
Unbranded: FGLS/IV[A]: (iv): 5.0005[B] (0.1007).

Independent variable: MAP-UDS; 
Branded: FGLS: (i): 0.9339[B] (0.0772); 
Branded: FGLS: (ii): 1.3846[B] (0.0703); 
Unbranded: FGLS/IV: (iii): 1.9756[B] (0.1479); 
Unbranded: FGLS/IV[A]: (iv): 2.6333[B] (0.1507).

Independent variable: INVENTORIES RATIO; 
Branded: FGLS: (i): - 8.3756[B] (0.1306); 
Branded: FGLS: (ii): - 8.5344[B] (0.1346); 
Unbranded: FGLS/IV: (iii): - 5.9058[B] (1.4575); 
Unbranded: FGLS/IV[A]: (iv): - 6.6552[B] (1.3967).

Independent variable: UTILIZATION RATES; 
Branded: FGLS: (i): 0.0873[B] (0.0346); 
Branded: FGLS: (ii): 0.0975[B] (0.0335); 
Unbranded: FGLS/IV: (iii): 0.2274[B] (0.0685); 
Unbranded: FGLS/IV[A]: (iv): 0.2678[B] (0.0687).

Independent variable: MW CRISIS; 
Branded: FGLS: (i): NA; 
Branded: FGLS: (ii): 4.2460[B] (0.1076); 
Unbranded: FGLS/IV: (iii): NA; 
Unbranded: FGLS/IV[A]: (iv): 5.4164[B] (0.1243).

Independent variable: WC CRISIS; 
Branded: FGLS: (i): NA; 
Branded: FGLS: (ii): 4.7384[B] (0.1946); 
Unbranded: FGLS/IV: (iii): NA; 
Unbranded: FGLS/IV[A]: (iv): 5.2531[B] (0.3339).

Independent variable: Constant; 
Branded: FGLS: (i): - 0.0090 (0.2677); 
Branded: FGLS: (ii): - 0.0044 (0.2234); 
Unbranded: FGLS/IV: (iii): 0.0228 (0.2959); 
Unbranded: FGLS/IV[A]: (iv): 0.0113 (0.2495).

Independent variable: Model prob-value; 
Branded: FGLS: (i): 0.0000[B]; 
Branded: FGLS: (ii): 0.0000[B]; 
Unbranded: FGLS/IV: (iii): 0.0000[B]; 
Unbranded: FGLS/IV[A]: (iv): 0.0000[B].

Independent variable: R-squared[C]; 
Branded: FGLS: (i): 0.18; 
Branded: FGLS: (ii): 0.24; 
Unbranded: FGLS/IV: (iii): 0.11; 
Unbranded: FGLS/IV[A]: (iv): 0.20.

Independent variable: Hausman 1 (x^2 , df)[D]; 
Branded: FGLS: (i): (3.10, 2); 
Branded: FGLS: (ii): (2.70, 2); 
Unbranded: FGLS/IV: (iii): (6.02, 2)[E]; 
Unbranded: FGLS/IV[A]: (iv): (9.20, 2)[E].

Independent variable: Hausman 2 (x^2 , df)[F]; 
Branded: FGLS: (i): NA; 
Branded: FGLS: (ii): NA; 
Unbranded: FGLS/IV: (iii): (-52.06, 51); 
Unbranded: FGLS/IV[A]: (iv): (1.10, 51).

Independent variable: AR(1) coefficient[G]; 
Branded: FGLS: (i): 0.8352[B]; 
Branded: FGLS: (ii): 0.8259[B]; 
Unbranded: FGLS/IV: (iii): 0.8305[B]; 
Unbranded: FGLS/IV[A]: (iv): 0.8149[B].

Independent variable: Rack cities; 
Branded: FGLS: (i): 282; 
Branded: FGLS: (ii): 282; 
Unbranded: FGLS/IV: (iii): 256; 
Unbranded: FGLS/IV[A]: (iv): 256.

Independent variable: Weeks; 
Branded: FGLS: (i): 361; 
Branded: FGLS: (ii): 361; 
Unbranded: FGLS/IV: (iii): 361; 
Unbranded: FGLS/ IV[A]: (iv): 361. 

Legend: 

FGLS=Feasible generalized least squares.

FGLS/IV=FGLS using instrumental variables.

NA=Not available.

Source: GAO econometric analysis of EIA, FTC, OPIS, and Thomson 
Financial data.

Note: The values in parentheses are standard errors.

[A] The instruments excluded the squared time trend variable to obtain 
valid instruments. The effects of the mergers were however similar.

[B] The estimates are significant at the 1 percent level or lower.

[C] R-squared is based on a regression of the dependent variable on its 
predicted values.

[D] Hausman 1: The null hypothesis is INVENTORIES RATIO and UTILIZATION 
RATES are exogenous. The test statistic is based on Hausman's (1978) 
specification test.

residuals on one-period lagged values.

[E] The estimates are significant at the 5 percent level or lower.

[F] Hausman 2: The null hypothesis is the instruments are exogenous or 
valid (no overidentifying restrictions). The test statistic is based on 
Hausman's (1978) specification test. A negative c2 is interpreted as 
lack of evidence to reject the null hypothesis; see Stata 7, Reference 
H-P, (2001), vol. 2, p. 13.

[G] A test of first-order autocorrelation, AR(1), using a test of 
significance of the coefficient from a regression of the residuals on 
one-period lagged values.

[End of table]

Table 22: Econometric Estimates of Mergers' Effects on Reformulated 
Wholesale Gasoline Prices: 

Independent variable: UDS-TOTAL; 
Branded: FGLS (i): - 0.3848[A] (0.0757); 
Branded: FGLS (ii): - 0.3875[A] (0.0745); 
Unbranded: FGLS (iii): - 0.2260 (0.1720); 
Unbranded: FGLS (iv): - 0.2237 (0.1679).

Independent variable: MARATHON-ASHLAND; 
Branded: FGLS (i): 0.7042[A] (0.2237); 
Branded: FGLS (ii): 0.7131[A] (0.2221); 
Unbranded: FGLS (iii): 0.8493[A] (0.3127); 
Unbranded: FGLS (iv): 0.8558[A] (0.3060).

Independent variable: SHELL-TEXACO II; 
Branded: FGLS (i): - 0.3770[B] (0.1844); 
Branded: FGLS (ii): - 0.3896[B] (0.1825); 
Unbranded: FGLS (iii): 0.1117 (0.3643); 
Unbranded: FGLS (iv): 0.0862 (0.3531).

Independent variable: BP-AMOCO; 
Branded: FGLS (i): 0.5641[B] (0.2324); 
Branded: FGLS (ii): 0.5500[B] (0.2309); 
Unbranded: FGLS (iii): 0.3790 (0.3252); 
Unbranded: FGLS (iv): 0.3976 (0.3185).

Independent variable: EXXON-MOBIL; 
Branded: FGLS (i): 1.5718[A] (0.3023); 
Branded: FGLS (ii): 1.6080[A] (0.3010); 
Unbranded: FGLS (iii): 0.9613[B] (0.4546); 
Unbranded: FGLS (iv): 1.0118[B] (0.4503).

Independent variable: INVENTORIES RATIO; 
Branded: FGLS (i): - 3.4738[A] (0.8283); 
Branded: FGLS (ii): - 3.4529[A] (0.8275); 
Unbranded: FGLS (iii): - 3.8467[A] (0.9472); 
Unbranded: FGLS (iv): - 3.8524[A] (0.9432).

Independent variable: UTILIZATION RATES; 
Branded: FGLS (i): 0.1898[C] (0.0972); 
Branded: FGLS (ii): 0.1905[B] (0.0971); 
Unbranded: FGLS (iii): 0.0812 (0.1051); 
Unbranded: FGLS (iv): 0.0835 (0.1048).

Independent variable: MW CRISIS; 
Branded: FGLS (i): NA; 
Branded: FGLS (ii): 2.8199[A] (1.0261); 
Unbranded: FGLS (iii): NA; 
Unbranded: FGLS (iv): 5.2124[A] (1.4006).

Independent variable: Constant; 
Branded: FGLS (i): 0.0588 (0.6665); 
Branded: FGLS (ii): 0.0565 (0.6561); 
Unbranded: FGLS (iii): 0.0048 (0.7107); 
Unbranded: FGLS (iv): 0.0042 (0.6908).

Independent variable: Model prob-value; 
Branded: FGLS (i): 0.0000[A]; 
Branded: FGLS (ii): 0.0000[A]; 
Unbranded: FGLS (iii): 0.0000[A]; 
Unbranded: FGLS (iv): 0.0000[A].

Independent variable: R-squared[D]; 
Branded: FGLS (i): 0.23; 
Branded: FGLS (ii): 0.24; 
Unbranded: FGLS (iii): 0.23; 
Unbranded: FGLS (iv): 0.24.

Independent variable: Hausman 1 (x^2, df)[E]; 
Branded: FGLS (i): (1.87, 2); 
Branded: FGLS (ii): (1.99, 2); 
Unbranded: FGLS (iii): (0.93, 2); 
Unbranded: FGLS (iv): (0.97, 2).

Independent variable: AR(1) coefficient[F]; 
Branded: FGLS (i): 0.8382[A]; 
Branded: FGLS (ii): 0.8375[A]; 
Unbranded: FGLS (iii): 0.8365[A]; 
Unbranded: FGLS (iv): 0.8347[A].

Independent variable: Rack cities; 
Branded: FGLS (i): 22; 
Branded: FGLS (ii): 22; 
Unbranded: FGLS (iii): 19; 
Unbranded: FGLS (iv): 19.

Independent variable: Weeks; 
Branded: FGLS (i): 305; 
Branded: FGLS (ii): 305; 
Unbranded: FGLS (iii): 305; 
Unbranded: FGLS (iv): 305.

Legend: 

FGLS=Feasible generalized least squares.

FGLS/IV=FGLS using instrumental variables.

NA=Not available.

Source: GAO econometric analysis of EIA, FTC, OPIS, and Thomson 
Financial data.

Note: The values in parentheses are standard errors.

[A] The estimates are significant at the 1 percent level or lower.

[B] The estimates are significant at the 5 percent level or lower.

[C] The estimates are significant at the 10 percent level or lower.

[D] R-squared is based on a regression of the dependent variable on its 
predicted values.

[E] Hausman 1: The null hypothesis is INVENTORIES RATIO and UTILIZATION 
RATES are exogenous. The test statistic is based on Hausman's (1978) 
specification test.

[F] A test of first-order autocorrelation, AR(1), using a test of 
significance of the coefficient from a regression of the residuals on 
one-period lagged values.

[End of table]

Table 23: Econometric Estimates of Mergers' Effects on CARB Wholesale 
Gasoline Prices: 

Independent variable: SHELL-TEXACO I; 
Branded: FGLS (i): - 0.2365 (0.3976); 
Branded: FGLS (ii): - 0.6933[A] (0.3167); 
Unbranded: FGLS/IV (iii): - 0.0143 (0.6401); 
Unbranded: FGLS/IV (iv): - 0.2440 (0.4619).

Independent variable: TOSCO-UNOCAL; 
Branded: FGLS (i): 7.3136[B] (3.8245); 
Branded: FGLS (ii): 6.8685[A] (3.3136); 
Unbranded: FGLS/IV (iii): -1.2480 (1.4079); 
Unbranded: FGLS/IV (iv): -1.5767 (1.2388).

Independent variable: INVENTORIES RATIO; 
Branded: FGLS (i): -20.5206[C] (6.1944); 
Branded: FGLS (ii): -20.9206[C] (5.9529); 
Unbranded: FGLS/IV (iii): -11.8892 (9.8474); 
Unbranded: FGLS/IV (iv): - 9.7019 (9.2235).

Independent variable: UTILIZATION RATES; 
Branded: FGLS (i): 0.3336 (0.2187); 
Branded: FGLS (ii): 0.3625[B] (0.2186); 
Unbranded: FGLS/IV (iii): 0.4464 (0.4928); 
Unbranded: FGLS/IV (iv): 0.5667 (0.4812).

Independent variable: WC CRISIS; 
Branded: FGLS (i): NA; 
Branded: FGLS (ii): 4.8834[A] (2.0148); 
Unbranded: FGLS/IV (iii): NA; 
Unbranded: FGLS/IV (iv): 10.5541[C] (2.5493).

Independent variable: Constant; 
Branded: FGLS (i): 0.6609 (2.3521); 
Branded: FGLS (ii): 0.3891 (1.6817); 
Unbranded: FGLS/IV (iii): 0.0437 (2.0216); 
Unbranded: FGLS/IV (iv): -0.0171 (1.4937).

Independent variable: Model prob-value; 
Branded: FGLS (i): 0.0011[C]; 
Branded: FGLS (ii): 0.0000[C]; 
Unbranded: FGLS/IV (iii): 0.4093; 
Unbranded: FGLS/IV (iv): 0.0002[C].

Independent variable: R-squared[D]; 
Branded: FGLS (i): 0.21; 
Branded: FGLS (ii): 0.36; 
Unbranded: FGLS/IV (iii): 0.03; 
Unbranded: FGLS/IV (iv): 0.34.

Independent variable: Hausman 1 (x^2, df)[E]; 
Branded: FGLS (i): (1.27, 2); 
Branded: FGLS (ii): (1.77, 2); 
Unbranded: FGLS/IV (iii): (5.39, 2)[B]; 
Unbranded: FGLS/IV (iv): (7.43, 2)[A].

Independent variable: Hausman 2 (x^2, df)[F]; 
Branded: FGLS (i): NA; 
Branded: FGLS (ii): NA; 
Unbranded: FGLS/IV (iii): (20.47, 51); 
Unbranded: FGLS/IV (iv): (9.02, 51).

Independent variable: AR(1) coefficient[G]; 
Branded: FGLS (i): 0.8863[C]; 
Branded: FGLS (ii): 0.8647[C]; 
Unbranded: FGLS/IV (iii): 0.8240[C]; 
Unbranded: FGLS/IV (iv): 0.7510[C].

Independent variable: Rack cities; 
Branded: FGLS (i): 6; 
Branded: FGLS (ii): 6; 
Unbranded: FGLS/IV (iii): 7; 
Unbranded: FGLS/ IV: (iv): 7.

Independent variable: Weeks; 
Branded: FGLS (i): 242; 
Branded: FGLS (ii): 242; 
Unbranded: FGLS/IV (iii): 242; 
Unbranded: FGLS/IV (iv): 242.

Legend: 

FGLS=Feasible generalized least squares.

FGLS/IV=FGLS using instrumental variables.

NA=Not available.

Source: GAO econometric analysis of EIA, FTC, OPIS, and Thomson 
Financial data.

Note: The values in parentheses are standard errors.

[A] The estimates are significant at the 5 percent level or lower.

[B] The estimates are significant at the 10 percent level or lower.

[C] The estimates are significant at the 1 percent level or lower.

[D] R-squared is based on a regression of the dependent variable on its 
predicted values.

[E] Hausman 1: The null hypothesis is INVENTORIES RATIO and UTILIZATION 
RATES are exogenous. The test statistic is based on Hausman's (1978) 
specification test.

[F] Hausman 2: The null hypothesis is the instruments are exogenous or 
valid (no overidentifying restrictions). The test statistic is based on 
Hausman's (1978) specification test.

[G] A test of first-order autocorrelation, AR(1), using a test of 
significance of the coefficient from a regression of the residuals on 
one-period lagged values.

[End of table]

Table 24: Econometric Estimates of Market Concentration on Conventional 
Wholesale Gasoline Prices: 

Independent variable: HHI; 
Branded: FGLS (i): 0.0015[A] (0.0002); 
Branded: FGLS (ii): 0.0005[A] (0.0002); 
Unbranded: FGLS/IV (iii): 0.0019[A] (0.0001); 
Unbranded: FGLS/IV (iv): 0.0011[A] (0.0001).

Independent variable: INVENTORIES RATIO; 
Branded: FGLS (i): - 8.3367[A] (0.1314); 
Branded: FGLS (ii): - 8.5415[A] (0.1344); 
Unbranded: FGLS/IV (iii): - 6.2377[A] (1.5351); 
Unbranded: FGLS/IV (iv): - 6.6088[A] (1.4540).

Independent variable: UTILIZATION RATES; 
Branded: FGLS (i): 0.1012[A] (0.0359); 
Branded: FGLS (ii): 0.1113[A] (0.0348); 
Unbranded: FGLS/IV (iii): 0.2352[A] (0.0717); 
Unbranded: FGLS/IV (iv): 0.2374[A] (0.0710).

Independent variable: MW CRISIS; 
Branded: FGLS (i): NA; 
Branded: FGLS (ii): 4.2808[A] (0.1041); 
Unbranded: FGLS/IV (iii): NA; 
Unbranded: FGLS/IV (iv): 5.0059[A] (0.1409).

Independent variable: WC CRISIS; 
Branded: FGLS (i): NA; 
Branded: FGLS (ii): 4.9552[A] (0.2065); 
Unbranded: FGLS/IV (iii): NA; 
Unbranded: FGLS/IV (iv): 5.2203[A] (0.3327).

Independent variable: Constant; 
Branded: FGLS (i): - 0.0091 (0.2797); 
Branded: FGLS (ii): - 0.0044 (0.2319); 
Unbranded: FGLS/IV (iii): 0.0225 (0.3124); 
Unbranded: FGLS/IV (iv): 0.0130 (0.2575).

Independent variable: Model prob-value; 
Branded: FGLS (i): 0.0000[A]; 
Branded: FGLS (ii): 0.0000[A]; 
Unbranded: FGLS/IV (iii): 0.0000[A]; 
Unbranded: FGLS/IV (iv): 0.0000[A].

Independent variable: R-squared[B]; 
Branded: FGLS (i): 0.18; 
Branded: FGLS (ii): 0.23; 
Unbranded: FGLS/IV (iii): 0.10; 
Unbranded: FGLS/IV (iv): 0.17.

Independent variable: Hausman 1 (x^2, df)[C]; 
Branded: FGLS (i): (4.41, 2); 
Branded: FGLS (ii): (3.59, 2); 
Unbranded: FGLS/IV (iii): (5.74, 2)[D]; 
Unbranded: FGLS/IV (iv): (5.44, 2)[D].

Independent variable: Hausman 2 (x^2, df)[E]; 
Branded: FGLS (i): NA; 
Branded: FGLS (ii): NA; 
Unbranded: FGLS/IV (iii): (-7.90, 51); 
Unbranded: FGLS/IV (iv): (-21.59, 51).

Independent variable: AR(1) coefficient[F]; 
Branded: FGLS (i): 0.8364[A]; 
Branded: FGLS (ii): 0.8269[A]; 
Unbranded: FGLS/IV (iii): 0.8265[A]; 
Unbranded: FGLS/IV (iv): 0.8139[A].

Independent variable: Rack cities; 
Branded: FGLS (i): 282; 
Branded: FGLS (ii): 282; 
Unbranded: FGLS/IV (iii): 256; 
Unbranded: FGLS/IV (iv): 256.

Independent variable: Weeks; 
Branded: FGLS (i): 361; 
Branded: FGLS (ii): 361; 
Unbranded: FGLS/IV (iii): 361; 
Unbranded: FGLS/IV (iv): 361.

Legend: 

FGLS=Feasible generalized least squares.

FGLS/IV=FGLS using instrumental variables.

NA=Not available.

Source: GAO econometric analysis of EIA, FTC, and OPIS data.

Note: The values in parentheses are standard errors.

[A] The estimates are significant at the 1 percent level or lower.

[B] R-squared is based on a regression of the dependent variable on its 
predicted values.

[C] Hausman 1: The null hypothesis is INVENTORIES RATIO and UTILIZATION 
RATES are exogenous. The test statistic is based on Hausman's (1978) 
specification test.

[D] The estimates are significant at the 10 percent level or lower.

[E] Hausman 2: The null hypothesis is the instruments are exogenous or 
valid (no overidentifying restrictions). The test statistic is based on 
Hausman's (1978) specification test. A negative c2 is interpreted as 
lack of evidence to reject the null hypothesis; see Stata 7, Reference 
H-P, (2001), vol. 2, p. 13.

[F] A test of first-order autocorrelation, AR(1), using a test of 
significance of the coefficient from a regression of the residuals on 
one-period lagged values.

[End of table]

Table 25: Econometric Estimates of Market Concentration on Conventional 
Wholesale Gasoline Prices: Eastern Region (PADDs I-III): 

Independent variable: HHI; 
Branded: FGLS/IV (i): 0.0013[A] (0.0001); 
Branded: FGLS/IV (ii): 0.0008[A] (0.0001); 
Unbranded: FGLS (iii): 0.0008[A] (0.0002); 
Unbranded: FGLS (iv): 0.0003 (0.0002).

Independent variable: INVENTORIES RATIO; 
Branded: FGLS/IV (i): - 5.2729[A] (1.6506); 
Branded: FGLS/IV (ii): - 5.5154[A] (1.5890); 
Unbranded: FGLS (iii): -7.2808[A] (0.2110); 
Unbranded: FGLS (iv): - 7.3751[A] (0.2265).

Independent variable: UTILIZATION RATES; 
Branded: FGLS/IV (i): 0.2260[A] (0.0752); 
Branded: FGLS/IV (ii): 0.2219[A] (0.0751); 
Unbranded: FGLS (iii): 0.0045 (0.0476); 
Unbranded: FGLS (iv): 0.0084 (0.0465).

Independent variable: MW CRISIS; 
Branded: FGLS/IV (i): NA; 
Branded: FGLS/IV (ii): 4.1027[A] (0.1614); 
Unbranded: FGLS (iii): NA; 
Unbranded: FGLS (iv): 5.3168[A] (0.1488).

Independent variable: Constant; 
Branded: FGLS/IV (i): 0.0477 (0.3400); 
Branded: FGLS/IV (ii): 0.0340 (0.2869); 
Unbranded: FGLS (iii): - 0.0133 (0.3430); 
Unbranded: FGLS (iv): - 0.0139 (0.2804).

Independent variable: Model prob-value; 
Branded: FGLS/IV (i): 0.0000[A]; 
Branded: FGLS/IV (ii): 0.0000[A]; 
Unbranded: FGLS (iii): 0.0000[A]; 
Unbranded: FGLS (iv): 0.0000[A].

Independent variable: R-squared[B]; 
Branded: FGLS/IV (i): 0.11; 
Branded: FGLS/IV (ii): 0.16; 
Unbranded: FGLS (iii): 0.14; 
Unbranded: FGLS (iv): 0.22.

Independent variable: Hausman 1 (x^2, df)[C]; 
Branded: FGLS/ IV: (i): (7.20, 2)[D]; 
Branded: FGLS/IV (ii): (6.38, 2)[D]; 
Unbranded: FGLS (iii): (3.73, 2); 
Unbranded: FGLS (iv): (3.51, 2).

Independent variable: Hausman 2 (x^2, df)[E]; 
Branded: FGLS/ IV: (i): (-2.62, 51); 
Branded: FGLS/IV (ii): (-13.52, 51); 
Unbranded: FGLS (iii): NA; 
Unbranded: FGLS (iv): NA.

Independent variable: AR(1) coefficient[F]; 
Branded: FGLS/ IV: (i): 0.8274[A]; 
Branded: FGLS/IV (ii): 0.8191[A]; 
Unbranded: FGLS (iii): 0.8176[A]; 
Unbranded: FGLS (iv): 0.8064[A].

Independent variable: Rack cities; 
Branded: FGLS/IV (i): 250; 
Branded: FGLS/IV (ii): 250; 
Unbranded: FGLS (iii): 235; 
Unbranded: FGLS (iv): 235.

Independent variable: Weeks; 
Branded: FGLS/IV (i): 361; 
Branded: FGLS/IV (ii): 361; 
Unbranded: FGLS (iii): 361; 
Unbranded: FGLS (iv): 361.

Legend: 

FGLS=Feasible generalized least squares.

FGLS/IV=FGLS using instrumental variables.

NA=Not available.

Source: GAO econometric analysis of EIA, FTC, and OPIS data.

Note: The values in parentheses are standard errors.

[A] The estimates are significant at the 1 percent level or lower.

[B] R-squared is based on a regression of the dependent variable on its 
predicted values.

[C] Hausman 1: The null hypothesis is INVENTORIES RATIO and UTILIZATION 
RATES are exogenous. The test statistic is based on Hausman's (1978) 
specification test.

[D] The estimates are significant at the 5 percent level or lower.

[E] Hausman 2: The null hypothesis is the instruments are exogenous or 
valid (no overidentifying restrictions). The test statistic is based on 
Hausman's (1978) specification test. A negative c2 is interpreted as 
lack of evidence to reject the null hypothesis; see Stata 7, Reference 
H-P, (2001), vol. 2, p. 13.

[F] A test of first-order autocorrelation, AR(1), using a test of 
significance of the coefficient from a regression of the residuals on 
one-period lagged values.

[End of table]

Table 26: Econometric Estimates of Market Concentration on Conventional 
Wholesale Gasoline Prices: Western Region (PADDs IV-V): 

Independent variable: HHI; 
Branded: FGLS/IV (i): 0.0051[A] (0.0021); 
Branded: FGLS/IV (ii): 0.0038[B] (0.0023); 
Unbranded: FGLS (iii): 0.0088[A] (0.0040); 
Unbranded: FGLS (iv): 0.0087[A] (0.0038).

Independent variable: INVENTORIES RATIO; 
Branded: FGLS/IV (i): - 12.0385[C] (2.9641); 
Branded: FGLS/IV (ii): -12.2358[C] (2.8805); 
Unbranded: FGLS (iii): - 6.6587[C] (1.4067); 
Unbranded: FGLS (iv): -7.2434[C] (1.4044).

Independent variable: UTILIZATION RATES; 
Branded: FGLS/IV (i): 0.2166 (0.1644); 
Branded: FGLS/IV (ii): 0.2136 (0.1641); 
Unbranded: FGLS (iii): 0.1899[A] (0.0936); 
Unbranded: FGLS (iv): 0.2032[A] (0.0928).

Independent variable: WC CRISIS; 
Branded: FGLS/IV (i): NA; 
Branded: FGLS/IV (ii): 1.4228[C] (0.4406); 
Unbranded: FGLS (iii): NA; 
Unbranded: FGLS (iv): 0.9352 (0.7404).

Independent variable: Constant; 
Branded: FGLS/IV (i): 0.1191 (0.8819); 
Branded: FGLS/IV (ii): 0.1059 (0.8074); 
Unbranded: FGLS (iii): 0.0050 (0.9221); 
Unbranded: FGLS (iv): 0.0059 (0.7684).

Independent variable: Model prob-value; 
Branded: FGLS/IV (i): 0.0000[C]; 
Branded: FGLS/IV (ii): 0.0000[C]; 
Unbranded: FGLS (iii): 0.0000[C]; 
Unbranded: FGLS (iv): 0.0000[C].

Independent variable: R-squared[D]; 
Branded: FGLS/IV (i): 0.29; 
Branded: FGLS/IV (ii): 0.32; 
Unbranded: FGLS (iii): 0.28; 
Unbranded: FGLS (iv): 0.31.

Independent variable: Hausman 1(x^2, df)[E]; 
Branded: FGLS/IV (i): (9.47, 2)[C]; 
Branded: FGLS/IV (ii): (9.48, 2)[C]; 
Unbranded: FGLS (iii): (3.76, 2); 
Unbranded: FGLS (iv): (4.31, 2).

Independent variable: Hausman 2 (x^2, df)[F]; 
Branded: FGLS/IV (i): (1.00, 51)[G]; 
Branded: FGLS/IV (ii): (1.90, 51)[G]; 
Unbranded: FGLS (iii): NA; 
Unbranded: FGLS (iv): NA.

Independent variable: AR(1) coefficient[H]; 
Branded: FGLS/IV (i): 0.8737[C]; 
Branded: FGLS/IV (ii): 0.8855[C]; 
Unbranded: FGLS (iii): 0.8814[C]; 
Unbranded: FGLS (iv): 0.8709[C].

Independent variable: Rack cities; 
Branded: FGLS/IV (i): 32; 
Branded: FGLS/IV (ii): 32; 
Unbranded: FGLS (iii): 21; 
Unbranded: FGLS (iv): 21.

Independent variable: Weeks; 
Branded: FGLS/IV (i): 361; 
Branded: FGLS/IV (ii): 361; 
Unbranded: FGLS (iii): 361; 
Unbranded: FGLS (iv): 361.

Legend: 

FGLS=Feasible generalized least squares.

FGLS/IV=FGLS using instrumental variables.

NA=Not available.

Source: GAO econometric analysis of EIA, FTC, and OPIS data.

Note: The values in parentheses are standard errors.

[A] The estimates are significant at the 5 percent level or lower.

[B] The estimates are significant at the 10 percent level or lower.

[C] The estimates are significant at the 1 percent level or lower.

[D] R-squared is based on a regression of the dependent variable on its 
predicted values.

[E] Hausman 1: The null hypothesis is INVENTORIES RATIO and UTILIZATION 
RATES are exogenous. The test statistic is based on Hausman's (1978) 
specification test.

[F] Hausman 2: The null hypothesis is the instruments are exogenous or 
valid (no overidentifying restrictions). The test statistic is based on 
Hausman's (1978) specification test.

[G] The difference in variances of the estimators is not positive 
definite, and the value was obtained using a generalized inverse--the 
test is interpreted as lack of evidence to reject the null hypothesis; 
see Greene (2000), p. 386.

[H] A test of first-order autocorrelation, AR(1), using a test of 
significance of the coefficient from a regression of the residuals on 
one-period lagged values.

[End of table]

Table 27: Econometric Estimates of Market Concentration on Reformulated 
Wholesale Gasoline Prices: 

Independent variable: HHI; 
Branded: FGLS (i): 0.0041[B] (0.0016); 
Branded: FGLS (ii): 0.0041[B] (0.0016); 
Unbranded: FGLS (iii): 0.0037[A] (0.0019); 
Unbranded: FGLS (iv): 0.0037[A] (0.0019).

Independent variable: INVENTORIES RATIO; 
Branded: FGLS (i): -3.5124[B] (0.8145); 
Branded: FGLS (ii): -3.4990[B] (0.8147); 
Unbranded: FGLS (iii): -3.7669[B] (0.9561); 
Unbranded: FGLS (iv): -3.7742[B] (0.9543).

Independent variable: UTILIZATION RATES; 
Branded: FGLS (i): 0.1827[C] (0.1006); 
Branded: FGLS (ii): 0.1830[C] (0.1005); 
Unbranded: FGLS (iii): 0.0770 (0.1098); 
Unbranded: FGLS (iv): 0.0797 (0.1096).

Independent variable: MW CRISIS; 
Branded: FGLS (i): NA; 
Branded: FGLS (ii): 2.6429[B] (1.0268); 
Unbranded: FGLS (iii): NA; 
Unbranded: FGLS (iv): 4.8318[B] (1.3905).

Independent variable: Constant; 
Branded: FGLS (i): 0.0815 (0.7560); 
Branded: FGLS (ii): 0.0790 (0.7432); 
Unbranded: FGLS (iii): 0.0091 (0.8223); 
Unbranded: FGLS (iv): 0.0088 (0.7980).

Independent variable: Model prob-value; 
Branded: FGLS (i): 0.0000[B]; 
Branded: FGLS (ii): 0.0000[B]; 
Unbranded: FGLS (iii): 0.0003[B]; 
Unbranded: FGLS (iv): 0.0000[B].

Independent variable: R-squared[D]; 
Branded: FGLS (i): 0.15; 
Branded: FGLS (ii): 0.16; 
Unbranded: FGLS (iii): 0.16; 
Unbranded: FGLS (iv): 0.17.

Independent variable: Hausman 1 (x^2, df)[E]; 
Branded: FGLS (i): (1.96, 2); 
Branded: FGLS (ii): (2.08, 2); 
Unbranded: FGLS (iii): (1.75, 2); 
Unbranded: FGLS (iv): (1.81, 2).

Independent variable: AR(1) coefficient[F]; 
Branded: FGLS (i): 0.8451[B]; 
Branded: FGLS (ii): 0.8447[B]; 
Unbranded: FGLS (iii): 0.8414[B]; 
Unbranded: FGLS (iv): 0.8401[B].

Independent variable: Rack cities; 
Branded: FGLS (i): 22; 
Branded: FGLS (ii): 22; 
Unbranded: FGLS (iii): 19; 
Unbranded: FGLS (iv): 19.

Independent variable: Weeks; 
Branded: FGLS (i): 305; 
Branded: FGLS (ii): 305; 
Unbranded: FGLS (iii): 305; 
Unbranded: FGLS (iv): 305.

Legend: 

FGLS=Feasible generalized least squares.

FGLS/IV=FGLS using instrumental variables.

NA=Not available.

Source: GAO econometric analysis of EIA, FTC, and OPIS data.

Note: The values in parentheses are standard errors.

[A] The estimates are significant at the 5 percent level or lower.

[B] The estimates are significant at the 1 percent level or lower.

[C] The estimates are significant at the 10 percent level or lower.

[D] R-squared is based on a regression of the dependent variable on its 
predicted values.

[E] Hausman 1: The null hypothesis is INVENTORIES RATIO and UTILIZATION 
RATES are exogenous. The test statistic is based on Hausman's (1978) 
specification test.

[F] A test of first-order autocorrelation, AR(1), using a test of 
significance of the coefficient from a regression of the residuals on 
one-period lagged values.

[End of table]

Table 28: Econometric Estimates of Market Concentration on CARB 
Wholesale Gasoline Prices: 

Independent variable: HHI; 
Branded: FGLS (i): 0.0283[B] (0.0157); 
Branded: FGLS (ii): 0.0238[B] (0.0132); 
Unbranded: FGLS/IV (iii): 0.0390[C] (0.0154); 
Unbranded: FGLS/IV[A] (iv): 0.0263[B] (0.0142).

Independent variable: INVENTORIES RATIO; 
Branded: FGLS (i): -22.3141[D] (6.2573); 
Branded: FGLS (ii): -22.6641[D] (6.0101); 
Unbranded: FGLS/IV (iii): -10.1821 (9.7460); 
Unbranded: FGLS/IV[A] (iv): -3.8253 (9.5057).

Independent variable: UTILIZATION RATES; 
Branded: FGLS (i): 0.3526 (0.2200); 
Branded: FGLS (ii): 0.4020[B] (0.2194); 
Unbranded: FGLS/IV (iii): 0.9110[B] (0.4970); 
Unbranded: FGLS/IV[A] (iv): 0.9707[B] (0.5460).

Independent variable: WC CRISIS; 
Branded: FGLS (i): NA; 
Branded: FGLS (ii): 4.0592[B] (2.1198); 
Unbranded: FGLS/IV (iii): NA; 
Unbranded: FGLS/IV[A] (iv): 7.9664[D] (2.8164).

Independent variable: Constant; 
Branded: FGLS (i): 0.5462 (2.2909); 
Branded: FGLS (ii): 0.3394 (1.6786); 
Unbranded: FGLS/IV (iii): 0.0838 (1.9545); 
Unbranded: FGLS/IV[A] (iv): 0.0104 (1.4941).

Independent variable: Model prob-value; 
Branded: FGLS (i): 0.0005[D]; 
Branded: FGLS (ii): 0.0000[D]; 
Unbranded: FGLS/IV (iii): 0.0146[C]; 
Unbranded: FGLS/IV[A] (iv): 0.0001[D].

Independent variable: R-squared[E]; 
Branded: FGLS (i): 0.28; 
Branded: FGLS (ii): 0.36; 
Unbranded: FGLS/IV (iii): 0.18; 
Unbranded: FGLS/IV[A] (iv): 0.32.

Independent variable: Hausman 1 (x^2, df)[F]; 
Branded: FGLS (i): (1.33, 2); 
Branded: FGLS (ii): (1.67, 2); 
Unbranded: FGLS/IV (iii): (7.83, 2)[C]; 
Unbranded: FGLS/IV[A] (iv): (21.65, 2)[D].

Independent variable: Hausman 2 (x^2, df)g; 
Branded: FGLS (i): NA; 
Branded: FGLS (ii): NA; 
Unbranded: FGLS/IV (iii): (0.58, 51); 
Unbranded: FGLS/IV[A] (iv): (-2165, 51).

Independent variable: AR(1) coefficient[H]; 
Branded: FGLS (i): 0.8789[D]; 
Branded: FGLS (ii): 0.8648[D]; 
Unbranded: FGLS/IV (iii): 0.8045[D]; 
Unbranded: FGLS/IV[A] (iv): 0.7504[D].

Independent variable: Rack cities; 
Branded: FGLS (i): 6; 
Branded: FGLS (ii): 6; 
Unbranded: FGLS/IV (iii): 7; 
Unbranded: FGLS/IV[A] (iv): 7.

Independent variable: Weeks; 
Branded: FGLS (i): 242; 
Branded: FGLS (ii): 242; 
Unbranded: FGLS/IV (iii): 242; 
Unbranded: FGLS/IV[A] (iv): 242. 

Legend: 

FGLS=Feasible generalized least squares.

FGLS/IV=FGLS using instrumental variables.

NA=Not available.

Source: GAO econometric analysis of EIA, FTC, and OPIS data.Note: The 
values in parentheses are standard errors.

[A] Monthly (seasonal) dummies were used as instruments instead of 
weekly (seasonal) dummies to obtain valid instruments. The effects of 
HHI were however similar.

[B] The estimates are significant at the 10 percent level or lower.

[C] The estimates are significant at the 5 percent level or lower.

[D] The estimates are significant at the 1 percent level or lower.

[E] R-squared is based on a regression of the dependent variable on its 
predicted values.

[F] Hausman 1: The null hypothesis is INVENTORIES RATIO and UTILIZATION 
RATES are exogenous. The test statistic is based on Hausman's (1978) 
specification test.

[G] Hausman 2: The null hypothesis is the instruments are exogenous or 
valid (no overidentifying restrictions). The test statistic is based on 
Hausman's (1978) specification test. A negative c2 is interpreted as 
lack of evidence to reject the null hypothesis; see Stata 7, Reference 
H-P, (2001), vol. 2, p. 13.

[H] A test of first-order autocorrelation, AR(1), using a test of 
significance of the coefficient from a regression of the residuals on 
one-period lagged values.

[End of table]

[End of section]

Appendix V: Comments from the Federal Trade Commission's Commissioners: 

UNITED STATES OF AMERICA 
FEDERAL TRADE COMMISSION 
WASHINGTON, D.C. 20580:

Office of the Chairman:

James E. Wells: 
Director:
Natural Resources & Environment:
U.S. General Accounting Office: 
441 G. St. N.W.

Washington, DC 20548:

August 25, 2003:

Dear Mr. Wells:

This letter submits the preliminary view of the Federal Trade 
Commission ("Commission") on the General Accounting Office ("GAO") 
report entitled "Effects of Mergers and Market Concentration in the 
U.S. Petroleum Industry in the 1990s" ("Report"). Our 
response consists of this letter and the three enclosures. The Report 
purports to examine the effects of recent mergers on several aspects of 
the petroleum industry, including wholesale gasoline prices, 
concentration, vertical integration, and barriers to entry. We 
understand that the GAO will publish it shortly. Regrettably, the 
Commission has had only a limited opportunity to review the report, 
receiving it in early August just a few weeks before the Report was to 
be published. [NOTE 1]

The subject of this Report is important and timely, and warrants 
careful and reliable analysis for Congress to be able to make informed 
policy determinations. Unfortunately, the Report in its present form is 
so flawed that reliable judgments cannot be formed regarding the 
competitive effects of mergers in the petroleum industry. These flaws 
include Methodological mistakes that make the Report's quantitative 
analyses wholly unreliable. For example, the Report does not use 
obvious controls for isolating the effect of a merger. It does not 
properly compare supposedly affected areas with unaffected areas. It 
also does not include non-merger factors that almost certainly will 
affect price, like seasonality, supply disruptions, and temperature.

Critical factual assumptions that are both unstated and unjustified. 
For example, the Report simply assumes that state boundaries delimit 
meaningful geographic markets - an assumption that in most cases is 
devoid, to our knowledge, of any empirical basis or support. These 
assumptions often then are combined in the Report with further 
methodological flaws that do not meaningfully distinguish correlation 
from causation.

Conclusions that lack any quantitative foundation. For example, there 
appears to be no quantitative basis for the Report's conclusion that 
unbranded gasoline has become less available. At the same time, the 
Report makes no effort to assess the (major) regional differences 
regarding the availability of unbranded gasoline, making the Report's 
treatment of regional differences inconsistent as well as arbitrary.

The Commission has spent significant resources investigating 
consummated mergers, both to determine whether past enforcement actions 
were correct, and to identify anticompetitive mergers the effects of 
which could be attenuated by future Agency action. As a result, we have 
accumulated substantial methodological expertise and have applied that 
expertise to the oil industry as part of our enforcement mission. Based 
on this expertise and our initial review of the analyses in the report, 
we find that the event study and the price-concentration regression are 
fundamentally flawed. [NOTE 2]

1. THE REPORT'S ECONOMETRIC ANALYSES ARE FUNDAMENTALLY FLAWED:

The heart of the Report consists of two econometric analyses. The first 
performs what is sometimes called an "event study." The analysis 
attempts to isolate the impact of eight petroleum mergers that occurred 
in the late 1990s on the price of wholesale gasoline (adjusting for 
crude oil costs). The results purport to show that six of the eight 
mergers in question were 
associated with statistically significant price increases, ranging from 
less than one cent per gallon to over five cents per gallon for at 
least one type of gasoline. (According to the event study, the other 
two mergers were associated with price decreases.) The second 
econometric analysis contained in the Report seeks to describe the 
relationship between wholesale gasoline prices (adjusting for crude oil 
costs) and wholesale concentration measured at the state level. This 
price-concentration analysis purports to show a positive and 
significant relationship between higher wholesale gasoline prices and 
industry concentration. Depending upon a variety of factors, such as 
fuel type and region of the country, the analysis estimates that an 
increase in the Herfindahl-Hirschman Index (HHI) of 100 points may lead 
to an increase in the price of wholesale gasoline of as much as four 
cents per gallon.

Based on our initial review of these econometric analyses, the 
methodologies underlying both analyses are fundamentally flawed. Five 
primary reasons support this conclusion.

First, the models used do not control for the many factors that could 
cause prices to increase. Isolating the effect on price from a merger 
necessarily requires the correct and comprehensive identification of 
factors that might influence demand (seasonality, temperature, income) 
as well as those that might influence supply (supply disruptions, 
changes in gasoline formulation). The Report is conspicuous in its 
failure to control for any of these factors. For example, the period at 
issue was characterized by several supply shortages, which can cause 
short-term price spikes entirely unrelated to the mergers under 
investigation. Similarly, seasonality is a crucial factor in analyzing 
this market: gasoline prices tend to increase in the summer in response 
to increased demand. Not controlling for seasonal effects is especially 
problematic because in some cases the post-merger period contains only 
a short time period, encompassing just one season, while the pre-merger 
period includes at least an entire year. By not controlling for such 
factors, the Report fails to provide meaningful information regarding 
whether price changes were merger-related or not.

An approach superior to that of the study would be to compare price 
changes in the affected markets with price changes in carefully 
selected comparable non-merger markets. If post-merger prices in non-
merger markets went up as much as those in the merger market, then 
there is no rational basis for concluding that the merger caused the 
price increases. No such carefully defined "natural experiment" was 
conducted by GAO staff; at least none is included in the Report. 
Instead, the preferred estimates in the Report simply compare the post-
merger prices in the areas affected by the merger to the pre-merger 
prices in those areas.

Second, the price-concentration methodology used by the GAO is subject 
to several well known problems that make it unacceptable as an 
alternative to a well-conducted event study. The most important of 
these problems is the difficulty in distinguishing between correlation 
and causation. Simply because two factors move together does not mean 
that one caused the other.

Third, any reliable price-concentration analysis necessarily requires 
that concentration be calculated in an economically well-defined market 
- that is, an area in which a particular merger or other increase in 
concentration is likely to have an economic effect. The Report's 
assumptions 
of state-wide geographic markets are unjustified. We are not aware of 
any supporting empirical data that markets generally coincide with 
state boundaries. Indeed, all of the data with which we are familiar 
point to the conclusion that wholesale markets in this industry rarely 
coincide with state limits. Accordingly, while price-concentration 
analyses may provide some useful information on general industry trends 
in concentration, they cannot be used to determine if an economically 
meaningful relationship exists between price and concentration.

Fourth, the results in the Report are, in many cases, not robust. 
Economists usually consider various approaches to estimating a model to 
determine whether the results from one approach (or "specification") 
are consistent with those using alternative approaches. This procedure 
is known as checking the "robustness" of the results. If results differ 
substantially for different methods or approaches, the reliability of 
the results is questionable. The Report's results in fact differ 
substantially across models. For example, in some cases, when 
estimating the effect of a particular merger on wholesale gasoline 
prices, the Report finds positive effects with some specifications and 
negative effects in other specifications.

Finally, documentation of the technical work on the econometric models 
is incomplete. For example, there is no discussion regarding how 
divestitures were treated, or which terminal racks were used in which 
regressions, or how price observations were constructed. The Report's 
approach to these issues, had they been better documented, might raise 
more concerns about the methodology used to reach the reported results. 
Moreover, given the failure to provide the underlying data, it is 
impossible to replicate independently the Report's results or to 
perform more rigorous robustness tests (including taking into account 
the missing factors that influence price changes as discussed above). 
Results that cannot be replicated or thoroughly analyzed for robustness 
are of little scientific value.

II. THE REPORT'S ASSERTIONS ABOUT STRUCTURAL CHANGES IN THE PETROLEUM 
INDUSTRY, AND THEIR COMPETITIVE EFFECT ON GASOLINE MARKETS, ARE ALSO 
FLAWED:

In addition to the flaws in the Report's quantitative analyses, there 
are several conclusions in the Report that appear to be without 
quantitative support. Other observations appear to overlook important 
factual issues, or invite unwarranted conclusions about the effect of 
particular facts on the extent of competition in the market.

For example, the Report suggests that the mergers in question have 
raised barriers to entry, while acknowledging that the effect of these 
mergers on entry barriers could not be quantified. The Report further 
observes that mergers may have made it more difficult for smaller firms 
to compete, or for new competitors to enter these markets. These 
observations, however, even if true, do not mean that competition in 
the petroleum industry has been harmed or eliminated. For example, to 
the extent that mergers confer cost-reducing scale advantages (as the 
Report suggests), consumers will benefit when cost savings are passed 
on through lower prices. Complaints from small competitors that 
competition with larger-scale entities is putting them in 
jeopardy therefore may well suggest enhanced competition, as all firms 
feel pressured by competition to reduce costs by whatever means 
possible and thereby reduce retail prices to consumers. Similarly, 
other structural factors detailed in the Report - such as minimum 
volume requirements and the alleged preferences of refiners to deal 
with larger distributors - are as consistent with a theory of enhanced 
competition as they are with a theory that petroleum industry 
consolidation has adversely affected consumers.

Another finding in the Report without quantitative support is the 
conclusion that vertical integration between refining and marketing has 
increased. Characterizing the degree of vertical integration between 
functional levels is more complicated than the Report suggests. EIA 
data on volume of gasoline distributed by channel of distribution 
indicate that, for the nation as a whole, the sale of gasoline through 
independent distributors - by far the leading channel of gasoline 
distribution - has increased in recent years. These data indicate that 
overall vertical integration between marketing and refining has not 
increased in recent years. These data, and our own experience, also 
reveal that vertical integration between refining and marketing differs 
significantly across different geographic areas within the United 
States. Moreover, the competitive implications of vertical integration 
are complex, with the potential for procompetitive as well as 
anticompetitive effects.

Finally, the Report finds that unbranded gasoline has become less 
available. This conclusion appears to have no firm quantitative 
foundation, but is instead based on interviews with various industry 
participants. In fact, the availability of unbranded gasoline varies 
significantly across geographic areas. The Report specifically notes 
that hypermarkets almost always supply unbranded gasoline and are 
growing significantly. The success of hypermarkets and other unbranded 
marketers in some areas of the country raises important questions about 
the competitive significance of "branded gasoline" in attracting 
consumers, and suggests that generalizations about possible impediments 
to the expansion of unbranded marketers are unwarranted.

III. CONCLUSION:

The Commission and its staff stand ready to provide further assistance 
to the GAO. The Report deals with a timely and important topic, and its 
findings have potentially important implications for public policy 
regarding petroleum mergers. In the Commission's view, however, this 
report does not meet the high standards of "accountability, integrity, 
and reliability" [NOTE 3] we would expect from GAO's reports and 
publications. We, too, are governed by similar standards for protecting 
the public interest. [NOTE 4] Accordingly, we remain willing and eager 
to assist in the production of a more accurate report.

By direction of the Commission.

Signed by: 

Timothy J. Muris: 
Chairman

Enclosures:

(1) Discussion of Deficiencies in Chapter 5 of the GAO Report "Effects 
of Mergers and Market Concentration in the U.S. Petroleum Industry in 
the 1990's":

(2) FTC Staff Comments on the GAO methodology for "Econometric Analysis 
of Effects of Market Concentration and Mergers on U.S. Wholesale 
Gasoline Prices in the 1990's," December 20, 2002:

(3) Professor John Geweke, "Empirical Evidence on the Competitive 
Effects of Mergers in the Gasoline Industry," unpublished draft, July 
16, 2003:

NOTES

[1] Commission staff were not provided with a draft of the Report until 
August 4, 2003 and were not permitted to make or retain copies of the 
Report, despite the fact that it is roughly 200 pages long and includes 
complex econometric analyses that took the GAO most of a year to 
complete. The Commissioners received copies on August 12 but were not 
allowed to share these copies with the FTC staff. GAO also declined to 
provide the Commission with the underlying data used for the Report, 
and did not supply it with a detailed description of the Report's final 
methodology. It has therefore effectively been impossible for the 
Commission and its staff to analyze, replicate, or test fully the 
Report's methodology. A more detailed commentary on the econometric 
analysis in the report is attached at Enclosure 1.

[2] The Commission staff previously provided GAO staff with preliminary 
oral and written comments in December 2002 on issues, among others, 
relating to the data and methodology encompassed in the Report's 
econometric analyses. (See Enclosure 2 to this letter.) GAO staff 
appears to have ignored most of the comments provided by Commission 
staff about the basic methodology that the GAO staff proposed to use at 
that time. GAO staff also failed to apprise Commission staff of 
methodological changes made subsequent to December 2002, which changes 
further undermine the reliability of the conclusions in the Report's 
econometric analyses. FTC staff also forwarded the attached report by 
Professor John Geweke on "Empirical Evidence on the Competitive Effects 
of Mergers in the Gasoline Industry," unpublished draft, July 16, 2003, 
to GAO staff. (See Enclosure 3) Professor Geweke is one of the most 
widely respected econometricians in the United States.

[3] "Accountability describes the nature of GAO's work. GAO helps the 
Congress oversee federal programs and operations to ensure 
accountability to the American people." See www.gao.gov. "Integrity 
describes the high standards that GAO sets for itself in the conduct of 
its work. GAO takes a professional, objective, fact-based, nonpartisan, 
nonideological, fair, and balanced approach to all of its activities. 
Integrity is the foundation of reputation, and GAO's approach to its 
work assures both." Id. "Reliability describes GAO's goal for how its 
work is viewed by the Congress and the American public. GAO produces 
high quality reports, testimony, briefings, legal opinions, and other 
products and services that are timely, accurate, useful, clear and 
candid." Id.

[4] The Federal Trade Commission Act provides that the Commission take 
action when it determines that such action would be in the public 
interest. 15 U.S.C. ï¿½ 45(b). For provisions specifically addressing 
data quality, see also Data Quality Act, i.e., Treasury and General 
Government Appropriations Act for Fiscal 2001, Pub. L. No. 106-554, ï¿½ 
515 (Dec. 21, 2000); 67 Fed. Reg. 8452 (Feb. 22, 2002) (Office of 
Management and Budget guidance); http://www.ftc.gov/ppa/2002/08/
fyi0242.htm (FTC guidelines implementing the Data Quality Act).

The following are GAO's comments on the Federal Trade Commission's 
letter dated August 25, 2003: 

GAO's Comments: 

1. We agree that the issues addressed in this report are important and 
timely, particularly since no comprehensive study has been done on the 
effects of the recent merger wave in the petroleum industry in the 
second half of the 1990s. We disagree, however, with FTC's assertion 
that the methodology we used in our study is flawed. Our methodologies 
incorporate state of the art techniques in econometrics and are 
consistent with existing literature and the comments of industry 
experts. In developing our empirical approach, we relied on GAO 
economists and obtained comments from economists outside GAO, including 
our consultant/peer reviewer, who is a recognized expert in the 
modeling of gasoline markets. As stated in a paper by FTC's (former) 
Director and the Deputy Director of the Bureau of Economics,[Footnote 
128] "Analyses can lead to different conclusions because of different 
data, different economic modeling, different econometric techniques, 
and /or fundamental mistakes." Furthermore, they stated that, "there is 
no 'perfect' econometric studyï¿½ Lack of unnoticeable perfection should 
not be a bar to an econometric study being given weight." We agree with 
these statements, especially given the complexity of our study. 
Nonetheless, partly in response to FTC comments, we re-estimated our 
models to account for the effects of gasoline supply disruptions that 
occurred in some parts of the West Coast and Midwest regions.

2. We provided an opportunity for FTC to review a draft of this report 
on August 5, 2003, consistent with GAO's policy. Copies of the draft 
report were delivered to FTC staff, who retained them for the period of 
their review. Copies were subsequently delivered to FTC's Commissioners 
on August 12, 2003, and they retained the copies. The copies provided 
to the Commissioners were the same drafts shared with FTC staff earlier 
on August 5, 2003. GAO's policy does not prevent the Commissioners from 
sharing their copies with FTC staff. We obtained all the data used in 
this report from publicly available sources, including a substantial 
purchase of data from OPIS, Thomson Financial, and J.S. Herold, Inc., 
which we have no obligation to share. We provided a complete and 
detailed description of the data and their sources in the draft report 
that FTC reviewed.

3. We disagree. In developing our econometric models, we considered and 
discussed the importance of merger variables, market concentration 
variables, and other supply and demand variables, and we controlled for 
such variables when we believed it was appropriate. We specifically 
considered and discussed the following variables in our models: crude 
oil prices, the ratio of gasoline inventories to expected demand, 
refinery capacity utilization rates, and supply disruptions in the 
Midwest and West Coast regions. The ratio of gasoline inventories to 
expected demand captures the behavioral response to seasonality and 
temperature (see below). In the draft report (but not in the final 
report), we also considered income, population density, prices in 
nearby rack cities, distances between nearby rack cities, divorcement 
laws--which could help capture the effects of vertical relationships 
between refining and retail gasoline marketing--year-specific effects, 
week-specific effects, and city-specific effects. While we used 
wholesale gasoline prices minus crude oil prices as the dependent 
variable for economic and statistical reasons, we also estimated the 
models with the crude costs as an explanatory variable and the results 
were generally similar.

Although no econometric model perfectly depicts reality, we believe 
that our current models are methodologically sound and produce 
reasonable estimates. FTC's suggestion that we use seasonality, 
temperature, and supply disruptions in our merger regressions means 
resorting to proxies when we have more direct measures of demand and 
supply shocks. FTC's suggestion is contrary to accepted econometric 
practice. Seasons and temperature affect gasoline prices by changing 
demand and supply. Supply disruptions affect gasoline price through 
changes in inventory. Since we used measures of gasoline inventories 
and demand, resorting to proxies is not necessary.

Our overall methodology is consistent with previous studies of gasoline 
markets, and our findings are fact-based and objective. External 
experts, including those who have conducted empirical studies in the 
petroleum industry, reviewed our econometric model outline and provided 
comments that we incorporated in our analysis. In addition, we 
consulted with a well-known and respected expert on economic modeling 
of the petroleum industry, who reviewed the methodology and the models' 
results. We disagree with FTC's assertion that we did not use 
appropriate variables in our models or did not appropriately control 
for the following variables.

Seasonality and temperature: As stated above, in our models, 
seasonality and temperature are captured by the variable for gasoline 
inventories relative to demand. (See figure 22 in the report, which 
clearly shows the seasonal variations in gasoline inventories and 
demand).

Supply disruptions: We acknowledged the potential effects of the West 
Coast disruptions and the Midwest disruptions on wholesale prices of 
gasoline in our draft report. Nonetheless, in responding to FTC's 
comments, we have subsequently included proxies to account for the 
effects of these disruptions, and they did not significantly change our 
underlying results about the effects of mergers and market 
concentration on wholesale gasoline prices. This is not surprising 
because we believed we had indirectly captured some of the effects of 
the supply disruptions through the inventory variable. Moreover, we 
believe that the proxies used for the supply disruptions were crude and 
imprecise for the following reasons. First, the supply disruptions are 
not identified as affecting many of the mergers and areas that we 
modeled--the supply disruptions affected the Midwest (PADD II) and West 
Coast (PADD V, excluding California) for conventional gasoline and CARB 
in the West Coast. Second, as indicated above, we believe that the 
behavior of gasoline inventories and demand, which we included in our 
models, is useful in capturing some effects of the supply disruptions. 
FTC itself, in its investigation of the Midwest supply disruptions (see 
Midwest Gasoline Price Investigation--FTC, 2001a), determined that low 
gasoline inventories were a primary factor affecting the disruptions. 
In fact, we found that the relationship between gasoline price margins 
and inventories, as measured by the correlation coefficient, nearly 
doubled during the supply disruptions compared to the whole sample 
period. (The correlation coefficients increased from between - 0.16 and 
- 0.17 for the whole sample period to between - 0.27 and - 0.32 during 
the disruptions). In addition, FTC indicated that demand in the Midwest 
increased significantly relative to the average increase for the 
nation. Third, it was difficult to construct variables for the supply 
disruptions because of the lack of appropriate and comprehensive data. 
In particular, there is no accurate information on the timing, 
duration, and specific geographic areas that were affected by these 
supply disruptions. For instance, with respect to the West Coast 
disruptions, one of the authors of a paper by FTC staff on these 
disruptions (see Taylor and Fischer, 2002), indicated in an email to 
GAO staff, "While this paper does a fairly good job of identifying West 
Coast Supply disruptions for the years it looks at, I would not want to 
claim it is totally comprehensive." We believe that the assumption that 
the West Coast disruptions affected the whole of the West Coast (PADD 
V) overstates the coverage and impact of the disruptions. Also, for the 
Midwest disruption, FTC did not identify the specific geographic 
markets for conventional gasoline that were affected by the disruption. 
Using the whole Midwest geographic region would tend to overstate the 
coverage and impact of the disruption, because it would attribute to 
the disruption the effects that may be attributed to mergers.

Nonetheless, we constructed two different measures for the Midwest and 
the West Coast disruptions, using the available limited information. We 
constructed a Midwest disruption indicator variable based on FTC's 
(2001a) report, which suggested that the supply disruption occurred 
roughly in June 2000 in the Midwest (PADD II). Similarly, we 
constructed a West Coast disruptions indicator variable based on the 
study by FTC staff (Taylor and Fischer, 2002), suggesting that the 
supply disruptions occurred in some periods of 1999 and 2000 in the 
West Coast (PADD V).

4. We recognized the importance and difficulty of defining appropriate 
geographic markets for gasoline, especially at the wholesale levels. We 
discussed the issue of defining meaningful geographic gasoline markets 
(including wholesale) with FTC and other oil industry experts. FTC 
indicated to us that it could not provide specific evidence on actual 
geographic markets for wholesale gasoline across the United States 
because, when performing analysis of potential mergers, FTC focuses on 
a limited geographic area and relies substantially on proprietary 
company data, which are not publicly available. Like other industry 
experts that we contacted, FTC agreed in our December 2002 meeting that 
it was appropriate to use terminal cities and even states, in some 
cases, as geographic markets for wholesale gasoline. We therefore used 
rack cities as the geographic unit.

In the draft report, we used data for gasoline prime suppliers provided 
by the Department of Energy's Energy Information Administration (EIA), 
available only at the state level, in measuring market concentration 
(HHI) at the wholesale level. We believed that using the state-level 
HHI was reasonable, and FTC has not provided any reason or evidence for 
why doing so would bias our results. In the final report, however, we 
have used yearly HHI based on refinery capacity because we believe, 
after consultation with our expert/consultant, that market 
concentration at this level captures more effectively the ability of 
refiners to control gasoline sales (or their market power). As stated 
in the draft report, we noted the limitation of using HHI data because 
of potential problems with geographic market delineation and indignity. 
In the final report, our use of HHI data at the refinery level is 
likely to reduce the potential endogeneity problem because the HHI at 
that level would likely be exogenous to rack prices.

A study by Vita (2000), an FTC staff member, used the state as the 
geographic unit in analyzing retail gasoline prices, even though it is 
generally agreed that the geographic gasoline market at the retail 
level is smaller than at the wholesale and that it is therefore less 
meaningful to use the state as the geographic unit for retail markets. 
Professor Geweke, an econometrician whom FTC cites in its comments and 
whom FTC has asked to review research on the effects of petroleum 
mergers, commented in his review of GAO's 1986 study of wholesale 
gasoline prices that using the geographic unit of the state was 
inappropriate. However, Geweke apparently found nothing inappropriate 
with Vita's use of the state as a geographic unit.

We disagree with FTC's assertion that we did not meaningfully 
distinguish between correlation and causation. In fact, the use of 
appropriate economic structure for modeling is a common basis for 
inferring causation.

5. We disagree. Economic findings can be qualitative or quantitative. 
We clearly indicated in chapter 4 that we based our finding that 
unbranded gasoline has become less available on extensive interviews of 
industry participants in different regions of the country, who 
consistently indicated to us that that was the case. While it would be 
desirable to ascertain this finding quantitatively, according to the 
EIA there are currently no systematic and comprehensive data available 
on unbranded gasoline supply. We stated in the draft and final reports 
that we could not statistically quantify the extent to which unbranded 
gasoline supply has decreased because the data required for such an 
analysis do not currently exist. We also stated in the draft and final 
reports that EIA--the federal agency mandated by Congress to collect 
energy data, including gasoline supply--told us that "the agency does 
not require petroleum companies to report gasoline data in the form 
that would permit the identification of branded and unbranded sales." 
We also disagree with FTC's assertion about our treatment of regional 
differences (we assume FTC is referring to our estimates for the 
effects of market concentration on conventional gasoline in the eastern 
versus the western part of the United States in chapter 5). We did not 
imply any relationship between our discussion of less availability of 
unbranded gasoline in chapter 4 and the separate econometric estimates 
for the east and west in chapter 5. The east-west distinction was based 
primarily on the degree of integration of refining markets within these 
broad geographic regions.

6. Given FTC's mission to protect the public interest in mergers 
affecting the petroleum industry, we expected FTC to have considerable 
expertise in this industry. However, FTC has not provided evidence to 
support its criticisms of our analysis, even though FTC's officials 
have stated, "In most circumstances a technically-based critique should 
be supported by an empirical analysis that shows that dealing 
appropriately with the technical issue makes a meaningful difference in 
the results" (see Scheffman and Coleman (undated), p. 3). The only FTC 
study on the competitive effects of mergers in the petroleum industry 
that we are aware of is a recently released study on the effects of the 
Marathon-Ashland merger. The FTC study looked at the effects of this 
merger in only one rack city, Louisville. We believe that FTC's study 
has several shortcomings, including the econometric methodology and the 
interpretation of the results.[Footnote 129]

7. As part of our peer review process, we provided an outline of our 
econometric methodology, which was a roadmap of our methodology, to 
many experts in the petroleum industry, and FTC. While we provided no 
specific econometric equation(s), we included a list of potential 
variables and proposed estimation techniques. FTC provided us written 
comments on the preliminary outline of our econometric model that 
included data and methodology issues. At FTC's request, we met and 
discussed each of the issues raised in their written comments, 
particularly the issues that they deemed to be crucial. We discussed 
issues FTC felt might be addressed, but some of the issues FTC raised 
were so complex and theoretical that they themselves could not offer 
feasible solutions. In instances where it was reasonable and possible 
to make changes, we did so. In particular, a major point of concern FTC 
expressed after our December 2002 meeting was the limitation of the HHI 
data EIA provided to us--the mergers were not reflected in the HHI data 
until the merged firms began to file a combined report with EIA, which 
could be months or even years after a merger is completed. We 
subsequently contacted EIA, who provided us with revised HHI data, 
adjusted properly for the timing of the merger, as well as monthly data 
(instead of the annual data that EIA had provided to us earlier). FTC 
also provided information on possible sources of data on the West Coast 
disruptions. We also made changes to our model based on comments we 
received from experts in industry and academia. A few days before we 
delivered the draft report to FTC for their review, FTC sent to us the 
paper by Professor Geweke, and the accompanying letter stated that it 
was subject to revision. We made no changes to our draft based on 
Professor Geweke's paper.

8. Our statistical results do show the results questioned by FTC. We 
believe that our results are reasonable and consistent with the 
findings of the few previous studies that have been done on this issue. 
Our responses to FTC's specific comments follow.

9. We believe that our models appropriately control for the many 
variables that could affect gasoline prices. We have fully discussed 
these issues in comment 3 above, including our preferred methods for 
addressing seasonality and temperature and our incorporation of 
alternative measures of supply disruptions. Our analysis would not 
likely be affected by changes in gasoline formulations. For instance, 
for CARB, the change from Phase I to Phase II occurred in 1996, and our 
analysis reflects this change because the data used start from 1996.

10. We disagree. The approach suggested by FTC attempts to match the 
many diverse merger cities to a representative nonmerger city, and 
FTC's plain words suggest an unconditional comparison that assumes that 
all differences between the cities are due solely to differences caused 
by the merger. While this matching process might be useful in theory, 
it is almost impossible to find a control city that has the same demand 
and supply characteristics, except for the merger, when one has to use 
all the available cities that were affected by the mergers. The merger 
affected cities are generally diverse because in most cases a merger 
affected more than one broad geographic area, and the affected racks 
generally include large as well as small cities. Furthermore, even if 
one could select a nonmerger control city, that city is likely to be 
near the merger cities. In that case, the mergers could indirectly 
affect prices in the selected nonmerger control city as well, violating 
the requirement that the merger should not affect the selected control 
city. We note that despite FTC's experience in reviewing most of the 
mergers that we modeled, FTC did not provide any examples of 
appropriate control cities in our discussions with them concerning our 
proposed methodology and the mergers we were analyzing.

We also disagree with FTC's assertion that we did not appropriately 
perform the pre-and postanalysis for the mergers. In fact, we clearly 
identified the pre-and postperiods of the mergers to determine the 
effects of the mergers in the merger affected cities. Furthermore, in 
constructing the data, we used due diligence to ensure that there were 
enough data both before and after the mergers to estimate the mergers' 
effects.

11. While we agree that two factors moving together do not imply 
causation, we disagree with the remainder of FTC's comment. We note 
that our market concentration analysis was a complement to the event 
type study of the mergers. FTC's comment suggests that FTC is confusing 
the link between market concentration and mergers. A merger constitutes 
a single event, and therefore could be modeled as an event study, as we 
did. Market concentration captures a number of events that occur over 
time, in particular mergers, but also other factors such as entry and 
exit, which are often difficult to date. Therefore, we disagree that 
one could model the effects of market concentration as an event, as FTC 
suggests. We found it more appropriate to model the effects of market 
concentration as a regression of prices on market concentration, among 
other variables, measured over time.

We disagree with FTC's characterization of our price-concentration 
study. In the draft and final reports, we have recognized and discussed 
the methodological issues associated with price-concentration studies, 
including a citation of the study by Evans, Froeb (the current Director 
of FTC's Bureau of Economics), and Werden (1993). We note that FTC's 
horizontal merger guidelines are premised on links between 
concentration and market power effects, such as price increases. We 
believe that an econometric estimation based on economic theory, and 
controlling for other extraneous factors, generally allows meaningful 
estimates that can be interpreted as causal.

12. We disagree. See comments 3 and 9 above.

13. We disagree. The consistency of the results we obtained from the 
different specifications and estimations of our models, as well as 
consistency in the results for the merger effects and the market 
concentration effects, supports the robustness of our results. In 
particular, we provide the following evidence for the robustness of our 
results in the draft report. First, because market concentration is the 
cumulative effect of the mergers and other competitive factors, one 
would expect that the results from the market concentration models and 
mergers models would be similar, but not exactly comparable, if mergers 
are the predominant contributing factor to market concentration. In our 
study, the majority of the results for the two approaches were similar. 
In the draft report, we also estimated the effects of the mergers using 
two approaches--using data for all rack cities and using data for only 
the merger cities. Both approaches yielded similar results. Second, in 
the draft report, in cases where the estimated results from using 
different approaches had different signs, we discuss the possible 
reasons for such differences and why we chose the preferred approach. 
In particular, when we included the refinery capacity utilization 
rates, in addition to the ratio of gasoline inventories to demand, only 
the estimated effects of market concentration for conventional gasoline 
changed signs. We explained why we preferred the models that excluded 
these variables. Specifically, the expected signs were inconsistent for 
the utilization rates but not for the ratio of gasoline inventories to 
demand. Furthermore, when the ratio of gasoline inventories to demand 
was excluded, the utilization rates variable had the expected sign. 
Also, the data for utilization rates are available nationally, while 
the data for the ratio of gasoline inventories to demand are available 
regionally, which better captures differences in prices across markets. 
Nonetheless, in the final report, both the ratio of gasoline 
inventories to demand and refinery capacity utilization rates variables 
are included in all the models that were estimated and have the 
economically expected signs in all our models.

FTC supports its claim of lack of robustness by citing a few examples 
in its technical comments on the draft report. For example, FTC stated 
that, "In Table 16 [in the draft report], the estimated price effect of 
the BP-Amoco merger ranges from no price effect to 3.5 cents a gallon 
among the three regression specifications." But the results in the 
draft report were 2.03 (and highly significant) using all rack cities 
and without year effects, 1.14 (and not significant) using all rack 
cities and without year effects, and 3.54 (and highly significant) 
using only the merger affected cities and without year effects. 
Although one of the results is not statistically significant, they all 
have the same sign. Furthermore, the results without the year effects 
are reasonably similar. Also, FTC stated that in a few cases the 
estimates with instrumental variables were different from those without 
the variables. But, econometrically, robustness of the estimates does 
not require these two estimates to be the same. In fact, as noted in 
Evans, Froeb, and Werden (1993), "With panel data, fixed-effects 
procedures can be combined with instrumental variables to eliminate 
bias." 

14. We provided a detailed and complete description of the basis of our 
econometric models, data sources, sample selection process (including 
tables detailing the list of variables, definitions, sources, data 
frequency, and level--see table14), and specification of the 
econometric models and estimation techniques (see appendix IV). As 
detailed in table 14, the effective dates for the mergers were based on 
dates FTC recommended when merger remedies (divestitures) became 
effective. These dates, upon FTC's recommendation to EIA, were also 
used by EIA to compute the market concentration (HHI) data, based on 
prime suppliers' sales used in the draft report. We stated in the draft 
and final reports that we used all available data on the rack cities 
from the Oil Price Information Services (OPIS)--data on over 280 rack 
cities for branded and over 250 for unbranded conventional gasoline out 
of the 350 rack cities in the OPIS database, representing over 90 of 
the racks in the United States. For the merger analysis, we indicated 
in the draft report that we also used only rack cities where the 
merging companies operated before they merged. We also stated in the 
draft and final reports that we used the average prices at a rack. We 
believe that we have provided a full and complete documentation of our 
econometric methodology.

15. We disagree. As stated in comment (4), we believe that findings can 
be qualitative or quantitative. Throughout the report, we believe our 
findings have been well supported, both quantitatively and 
qualitatively.

16. We disagree with FTC's characterization of our discussion of 
barriers to entry. While we stated in the draft and final reports that 
mergers have had an impact on barriers to entry, based on information 
from industry officials, we stated that we could not quantify the 
extent of this impact because of a lack of data as well as a lack of 
consensus on an appropriate measure. We discussed the overall 
importance of barriers to entry in a market; FTC recognizes the 
importance of barriers to entry in its horizontal merger guidelines. 
Nonetheless, nowhere in the draft report do we say that barriers to 
entry have harmed or eliminated competition in the petroleum industry.

17. We disagree with FTC's assertion that a finding must only be based 
on quantitative analysis, especially given that, as FTC stated, 
vertical integration between functional levels is complex. Our report 
presents examples of mergers that were vertical in nature (that is, the 
mergers involved different functional levels of the merging companies), 
which would contribute to increased vertical integration (see table 2). 
FTC's use of the term "independent distributors" may be misleading. In 
the draft report, distributors (or jobbers) are generally independent 
middlemen and their activities, most often, do not decouple the 
vertical chain between refining and retail. While it is true that the 
distributors are the largest channel for distributing gasoline to the 
retail level, this distribution function does not affect the 
contractual relationship between the refiner and a retailer for branded 
gasoline, which represents the largest volume of gasoline sold. In 
fact, as noted by Royer (1998, p. 95), vertical coordination involving 
contractual supply relationships could increase the downstream 
company's costs because unlike market transactions for the intermediate 
good, a downstream company can no longer turn to alternative suppliers 
for its inputs. Moreover, as we state in our report, according to an 
EIA report and discussions with EIA officials, there have been a 
substantial number of vertical mergers in the downstream market between 
refiners and marketers over the decade of the 1990s. However, as per a 
close-out meeting with EIA, we did add language to our report to 
recognize that there has been a shift during this period toward the 
divestiture of certain downstream assets, such as refineries, by fully 
integrated companies. We do not know the basis of FTC's assertion that 
vertical integration between refining and marketing differs 
significantly across different geographic areas within the United 
States, given the complicated nature of vertical integration between 
functional levels in the petroleum industry. In the draft and final 
reports, as in the case of the potential effects of (horizontal) 
mergers and market concentration, we discussed the procompetitive as 
well as the anticompetitive effects of vertical integration.

18. We disagree. As we stated in comment 5 above, economic findings can 
be based on qualitative or quantitative information.

19. We appreciate FTC's offer of further assistance but have chosen 
instead to respond to its concerns. In particular, we have taken their 
suggestion to investigate in more detail the supply disruptions in the 
Midwest and West Coast, have reestimated our models incorporating these 
supply disruptions, and have discussed the implications of these supply 
disruptions in our response. Incorporating these supply disruptions did 
not change our finding that mergers generally led to price increases.

The report uses data and information from a wide range of reliable 
sources. Our methodology is sound, transparent, and consistent with the 
economic literature on mergers and market concentration. Our report's 
results are presented in a balanced, fact-based, and objective manner 
and have undergone external peer review. Moreover, the report provides 
valuable information about the overall effects of the mergers in the 
petroleum industry, information that was critically lacking.

We believe that this report meets our core values of accountability, 
integrity, and reliability. We welcome continuing public scrutiny or 
discourse on such an important issue that impacts public policy.

20. We are honoring the restriction imposed on the release on Professor 
Geweke's unpublished draft paper and are not publishing this 
attachment.

[End of section]

Appendix VI: Comments from the Federal Trade Commission's Bureau of 
Economics Staff: 

Bureau of Economics:

August 25, 2003:

Enclosure 1:

Discussion of Deficiencies in Chapter 5 of the GAO Report "Effects of 
Mergers and Market Concentration in the U.S. Petroleum Industry in the 
1990's" [NOTE 1]

1. Background:

These are comments of the Federal Trade Commission's (FTC) Bureau of 
Economics (BE) Staff on Chapter 5 and the related appendix of the 
Report. These comments reflect the limited ability that FTC staff had 
to review the draft. [NOTE 2] Additional time would have allowed a 
more comprehensive set of staff comments. Nevertheless, even within the 
short time we have had to review the report, we have identified 
fundamental methodological flaws with the econometric analyses that we 
will discuss in detail. These fundamental flaws mean that the Report 
cannot provide a reliable basis for addressing the issues it claims to 
study.

BE staff provided six pages of comments to the GAO in a December 20, 
2002 letter as well as verbal comments at a subsequent meeting with GAO 
staff on the draft methodology for the GAO study, dated December 2, 
2002. [NOTE 3] Most of the FTC staff comments on the draft econometric 
model were not incorporated into the Report. In addition, changes in 
the methodology in the Report and the methodology provided in December 
2002 raise additional concerns. [NOTE 4] This comment includes the 
comments that the FTC staff previously gave to the GAO as well as 
comments on the changes made from the December 2002 methodology and 
problems that were not evident in that methodology.

II. Summary:

Chapter 5 of the Report contains two econometric analyses. The first 
attempts to estimate the wholesale price effects of eight petroleum 
mergers on conventional, reformulated (RFG), and CARB gasoline 
(separately for branded and unbranded). The second attempts to estimate 
the relationship between wholesale prices and concentration (HHI) for 
the three main formulations of gasoline. [NOTE 5] The Report purports 
to show: 
(1) significant price increases for at least one gasoline formulation 
for six of the eight mergers reviewed; and (2) a significant positive 
relationship between price and concentration. As discussed below, there 
are fundamental flaws with both the merger event analysis and the 
price/margin concentration analysis. Given the severity of the flaws, 
the results of the statistical analysis cannot be used to make reliable 
inferences about the price effects from the mergers analyzed or the 
relationship between price and concentration.

The flaws with the analyses stem both from the underlying theoretical 
models and the implementation of the models used by the GAO. Of key 
importance is the lack of controls for important factors that affect 
the price of gasoline. Without such controls, the analyses cannot be 
used to reliably isolate the impact of the mergers or concentration on 
price. There are also 
serious problems with the interpretation and characterization of the 
results. In several instances the results in the Report are not as 
robust as represented in the discussion of the results. Also, the 
results of some of the specifications reveal some of the underlying 
methodological problems. Finally, the size of the estimated price 
effects, both positive and negative, of the mergers is implausibly 
large in the context of this industry. [NOTE 6]

An additional problem is that the documentation of the regressions and 
methodology is incomplete. The discussion in the Report should be 
sufficient so that a researcher with access to the same data set could 
replicate the results. Unfortunately, the discussion does not satisfy 
this basic requirement.

III. Merger Event Study Methodology:

There are several generally recognized methodologies that might be used 
to analyze the price effects of a merger through an event study. These 
approaches emphasize the need to control for factors unrelated to a 
merger that may influence prices. Controlling for such factors is 
important because the goal of an event study is to isolate the effect 
of the event on the variable of interest such that any changes in the 
variable after the event can be attributed to the event. GAO's 
preferred model, however, excludes important control variables, and as 
a result, can not reliably isolate any price effect-be it positive or 
negative-that can be attributed to a merger.

Among the factors not adequately controlled for in GAO's specifications 
are supply shocks, changes in fuel specifications, and seasonal 
effects. The Report's use of instrumental variables to deal with 
endogeneity issues is also problematic. Finally, it is well known that 
achieving reliable results in an event study requires that the event's 
pre-and post-periods be appropriately specified: unfortunately, the 
Report is flawed on this dimension as well.

The remainder of this section discusses: (1) general methodologies 
employed in event studies regarding the effects of mergers; (2) why the 
approach chosen by the GAO is inferior to potential alternatives; and 
(3) why the implementation of the approach chosen by the GAO is 
fundamentally flawed.

A. General Approaches in Merger Event Studies:

The GAO Report correctly cites a number of published economic studies 
that examine the price effects of consummated mergers. [NOTE 7] In most 
merger event studies that examine the price of products before and 
after a merger, one of two types of regressions has been estimated. In 
the first type of regression (see Barton and Sherman (1984) and Kim and 
Singal (1993)), the price of the product affected by the merger is 
compared to a substitute product or the same product in another market 
that faces similar demand and cost conditions before and after the 
merger. Specifically, the analysis is a reduced form regression of the 
price of the product of the merged firm relative to control product(s) 
on various time trends and a merger dummy variable. To implement this 
approach for oil mergers, the dependent variable would be the price of 
gasoline in 
a city where the merger reduced the number of competitors and the 
independent variables would include the price of gasoline in a nearby 
city (or set of cities) that arguably has the same supply and demand 
characteristics but is not affected by the merger. The choice of 
control cities, i.e., the cities where the merger should not affect 
prices, has to be made carefully and should be subjected to sensitivity 
analysis.

In the second type of regression (see Schumann et al, (1992) and 
(1997)), the price of the merged firm's product (or market price) is 
regressed on demand and supply/cost shifters plus a merger dummy. The 
researcher is trying to model how prices are deternlined in the markets 
at issue, and the merger is one of the factors potentially affecting 
price. This approach can be problematic due to the lack of available 
demand and supply variables that have sufficient variation over time 
and over geography to capture adequately the factors impacting price, 
thereby isolating the effect of the merger. This problem is 
particularly acute in gasoline markets because there are few variables 
that are available on a weekly basis at the city level to help explain 
rack price variation.

A third approach, used in Vita and Sacher (2001), combines elements of 
both of these approaches. In their study of a hospital merger, they 
examined the price of the merged firm relative to the price of a 
control group of firms unaffected by the merger that should be affected 
by the same demand and supply factors and regressed these relative 
prices on demand shifters, cost shifters, and the merger event to gauge 
the effect of the merger.

B. Approach Used By GAO:

The methodology described in the GAO December 2002 draft suggested that 
the regressions would use control cities as well as supply and demand 
shifters. This would have been similar to Vita and Sacher (2001). The 
Report, however, does not use a control city methodology. This is 
surprising given the following quote from page 123 of the Report and 
the related footnote 90, "... prices at the nearest rack could 
influence prices at the rack city. We did not however, incorporate this 
variable directly in our model because there is co-movement between the 
nearest price variable and prices since both variables are likely to be 
generated by 
the same set of independent variables." Footnote 90 describes, "In fact 
in our preliminary estimation we found that the estimated coefficient 
on the nearest prices were not statistically different from 1.":

The GAO appears to use these explanations as a reason for not including 
control cities in the analysis. On the contrary, these findings 
strongly suggests that prices from nearby racks (not affected by the 
merger) are an important, if not the important, control variable. As 
the quotation from the Report suggests, the price at the nearest rack 
and the price at the merger rack are determined by the same independent 
variables, i.e., the demand and supply variables. Using the price at an 
appropriate nearby (control) rack is particularly important because 
demand and supply variables that are specific to control racks are not 
readily available and therefore unobservable. All of these unobservable 
demand and supply effects are measured jointly in the control rack 
price. We thus believe that, assuming control racks were selected 
appropriately [NOTE 8] and additional supply and demand parameters were 
included to measure any price changes between the control racks and the 
merger racks unrelated to the merger as appropriate, the control rack 
approach would be superior. The GAO, however, chose not to use the 
control rack approach but rather to try to control for supply and 
demand factors directly.

C. Methodological Problems with the GAO's Approach:

1. GAO's Model is Under Specified:

When using the approach of controlling for supply and demand factors 
directly without using control cities, the independent variables in the 
regressions, as noted in the Report, should consist of market structure 
and regulatory factors, cost/supply factors, and demand factors. 
(P.130-31) This analysis is most similar to the approach taken in the 
Schumann et al studies. If 
the GAO staff had been able to develop control variables for all the 
market structure, regulatory factors, cost/supply factors and demand 
factors, this approach would have been useful. The Report's regressions 
contain no measures of market structure factors, other than the merger 
dummy, no measures of regulatory factors, minimal cost/supply factors 
and no demand factors. The relatively low amount of variation explained 
by the regressions estimated in the Report (with R^2 of less than 20%) 
suggests that important factors explaining pricing were excluded from 
the regression. As a result, the approach employed by the GAO is not 
viable for estimating merger effects. Thus, the results are of no value 
for studying the effects of the mergers analyzed. [NOTE 9]

The basic equation estimated by the GAO is as follows[NOTE 10]:

[See PDF for formula]

[End of figure]

For the GAO preferred specification, the above equation is estimated 
using data for just the racks directly affected by the merger. In 
addition, the refinery utilization variable is typically dropped in the 
preferred specification. The merger effect, (3,, is simply a comparison 
of the average price after the merger compared to the average price 
before the merger, controlling for the inventory and in some cases 
utilization. This model is under-specified. There are any 
number of additional supply, demand, and regulatory variables that will 
differ before and after the merger. We discuss these further below.

Given all of these issues, the regression estimates clearly suffer from 
omitted variable bias. The effects of omitted variable bias are well 
documented in the economics literature. [NOTE 11] Symptoms of omitted 
variable bias are clearly evident in the results presented in the 
Report. For example, footnote 87 states, "Also, the estimates with 
years variable (dummy variable for years) appear unreasonable. 
Furthermore, we did not have a good economic reason for including the 
years variable." The reason for including this variable is there are 
omitted variables that are correlated with years. It is likely that 
this was a highly significant variable in the regressions and was 
serving as a proxy for other variables. Events correlated with years 
include formulation changes, supply disruptions, refinery closures and, 
in some parts of the country, changes in imports. [NOTE 12] The effect 
of including the years variable can be seen in Tables 15-17. As 
discussed more fully below, the inclusion of the years variable changes 
the estimated results for the two transactions affecting CARB gasoline, 
Tosco-Unocal and Shell-Texaco I, from finding positive effects on 
prices from the merger to negative effects. While not as dramatic as 
the change in the CARB results, many of the other estimated mergers 
effects change, either to increase or decrease the effect, when the 
years variable is added to the regression.

The rationale provided by the GAO for dropping the years variable 
raises another concern. Classical statistical inference - like that 
used in the GAO report - assumes that the regression specification is 
chosen independently of the results it generates. If not, then the 
researcher injects his beliefs into the estimation process. The 
consequence of such "specification searching," or, less charitably, 
"regression fishing" is that the estimated coefficients are biased, as 
are the standard errors. [NOTE 13] It appears that, in at least one 
instance, the GAO report did indeed discard a specification because
 the "estimates appear unreasonable" (Footnote 87).

In what is deemed a sensitivity check, the prices of all the racks 
having the same gasoline specification are pooled, those racks affected 
by the merger and those presumed to be unaffected by the merger, and 
the effect of the merger is estimated using equation (1). When the 
regression is estimated using data from the merger and non-merger 
affected racks, the merger effect, Beta 1 is calculated as a comparison 
of the average price in the merger racks post-merger compared to the 
average rack price of the non-merger racks pre-and post-merger and the 
merger racks pre-merger. Thus, the control group price is an average of 
the rack price in merger cities pre-merger and the rack price in non-
merger cities pre-and post-merger. This is not an appropriate control 
price. Since the merger effect is being calculated based on a control 
price equal to the average price of all the cities during different 
time periods in the data, this analysis provides neither a meaningful 
pre-and post-merger comparison nor a good sensitivity check.

Both merger affected and non-affected racks could have been used to 
calculate a difference in difference estimator, if implemented 
correctly. [NOTE 14] One way for the estimator to be implemented 
correctly would be to add an additional variable to equation (1) as 
follows:

[See PDF for formula]

[End of figure]

The estimate of Beta 1 would be the merger effect controlling for the 
time-specific effects of gasoline prices unrelated to the merger.

Even with the difference-in-difference estimation, it is again crucial 
to pick the non-affected racks carefully. The control racks should have 
the same supply, demand and regulatory characteristics. Given that 
finding control racks can be difficult it is important to also include 
additional variables measuring supply, demand and regulatory changes to 
measure possible differences in the merger and control racks.

2. Key Control Variables Important to Wholesale Gasoline Prices Were 
Excluded:

There are a number of factors that affect price not currently included 
in the model. The FTC staff comment in December stated:

"In addition to the number of supply outages that need to be included 
in the model, all areas of the country do not have the same 
reformulated gasoline, the formulation changes between winter and 
summer happen at different times of the year in various parts of the 
country, there have been a large number of changes in gasoline 
formulations in addition to RFG (reformulated gasoline) and CARB 
(California Air Resources Board gasoline) and phase II of RFG began in 
2000. In addition, the price of conventional gasoline in any state or 
city may be affected by the existence of reformulated gasoline. For 
example, conventional gasoline on the West Coast may be higher priced 
than conventional in the rest of the country since conventional 
gasoline in the west is a substitute in production for CARB gasoline.":

FTC staff also gave the GAO staff copies of a map showing all the 
various formulations of gasoline within the country, including CARB and 
RFG as well as low Reid vapor pressure conventional gasoline that must 
be sold in various parts of the country, usually during the summer 
months. There is no mention in the Report of these other formulations 
and whether changes in these formulations possibly affected the 
regression results. All of these issues suggest the need for additional 
variables in the analysis.

In addition, there are several important variables that were removed 
from the GAO December methodology. The original December 2002 
methodology included a number of dummy variables that may have removed 
most of the meaningful variation from the data.

Therefore, excluding some of these variables may have been warranted. 
[NOTE 15] But completely omitting measures of supply disruptions, 
seasonal variables, or year dummies is not appropriate. Consider first 
the impact of supply disruptions on the estimation of price effects. 
Clearly the effect of the supply shocks that caused Midwest Gasoline 
price increases of the summer of 2000 and West Coast outages in various 
years will influence the results of the merger effect regressions 
because these outages were in the post-merger period for some of the 
mergers. [NOTE 16] To the extent that prices were higher in the after 
period as a result of these outages, not controlling for these 
variables would result in an observed "effect" from the merger where 
there may have been a smaller effect or none at all.

While the inventory ratio variable may control for these effects to 
some degree as the Report suggests in footnote 87, preliminary work by 
FTC staff suggest that inventories do not change dramatically when 
there are supply shocks. Our regressions of rack prices in Midwest 
cities on the PADD II inventory level and a supply disruption variable 
for the Midwest Gasoline episode shows that the supply disruption 
variable has a large positive coefficient and is highly significant. In 
December, FTC staff gave the GAO a copy of a recently published paper 
documenting supply outages on the West Coast and discussed the need for 
the GAO staff to research other supply outages. [NOTE 17]

The results in the current Report do not incorporate any information on 
supply outages either in the regressions (where they should be) or in 
interpreting the results. Given the high number of supply outages in 
the year 2000 and that the GAO data set ends in December of 2000, 
estimating the effects of the Exxon-Mobil merger and the Marathon-UDS 
transaction, which occurred in 2000 and late 1999 respectively, will be 
very difficult even with a control methodology. Isolating the effect of 
the merger from the effect of supply disruptions will be difficult 
because both occurred during the same time period. As FTC staff noted 
in its December 2002 comments on the GAO's draft model (footnote 8):

"Some of the mergers will be especially difficult to model. The 
Marathon Ashland purchase of Ultramar Diamond Shamrock assets in 
Michigan occurred in late 1999 and the data ends in 2000. The summer of 
2000 includes the Midwest gasoline episode as well as major pipeline 
problems in Michigan. In addition a refinery closed in Michigan right 
before the purchase. The same lack of sufficient length of the data set 
applies to other mergers as well depending on which areas of the 
country are being examined."

Another type of supply shock involves changes in fuel specifications. 
In 2000 the implementation of RFG phase II (one of the prime causes of 
the Midwest gasoline spike) began, which likely increased costs for the 
industry. As mentioned above, multiple fuel specifications, such as RFG 
and CARB, or changes in fuel specifications, such as RFG phase I to RFG 
phase II can cause the price of conventional gasoline to be higher in a 
given region because all of these products are substitutes in 
production. [NOTE 18]

Another glaring omission in the Report is the lack of controls for 
seasonal effects. A simple graph of the difference between the spot or 
rack price of gasoline and the price of crude oil will show that this 
margin is highly seasonal: the margin is generally wider in the summer 
than in the winter. [NOTE 19] Given the short merger windows used by 
the GAO and the lack of controls for seasonal affects, some of the 
merger results (such as Marathon-Ashland, Shell-Texaco II (Motiva) and 
UDS-Total) are likely being driven by seasonal effects, not by the 
mergers themselves. For example, the time periods examined for the 
Marathon-Ashland joint venture 
were four years before the merger and six months after. The six months 
after included the summer of 1998. The finding that prices were higher 
during this period is not surprising because prices generally rise in 
the summer. The time periods examined for the Shell-Texaco II joint 
venture (Motiva) were six months before the merger and six months 
after. In this case the comparison was between the summer of 1998, 
before the merger, and the winter of 1998 after. Finding that prices 
decreased during this time period also is not surprising. The time 
periods and effects for the UDS-Total merger are similar.

3. Report's Use of Instrumental Variables is Problematic and 
Incomplete:

Another methodological problem involves the use and choice of 
instrumental variables in the Report. There is no discussion in the 
Report of why the instruments used [NOTE 20] are valid instruments for 
the given endogeneity problem. [NOTE 21] Footnote 11 of the December 
FTC staff comments said:

"If the controls are correlated with the HHI, then using weak (or 
inappropriate) instruments may give worse estimates than not using 
instruments. The final Report needs to explain why instruments for 
control variables are needed, show that the instruments are 
sufficiently powerful as to improve the results and discuss the changes 
in the results when using instrumental variables."

While this remark was made in the context of endogeneity in the price-
concentration analysis, it also applies to endogeneity issues in the 
merger event analysis.

The current Report does test for endogeneity, i.e., whether there is 
the need for instruments, but does not discuss why these are valid 
instruments or show that the instruments are sufficiently powerful to 
improve the results. There is a well known economic literature on the 
impact of using inappropriate or weak instruments in instrumental 
variables regression. [NOTE 22] As a general matter, the first stage 
instrumental variable results should be reported to show the effects of 
using the instruments. The goodness of fit of the first stage 
regressions as well as the coefficient estimates and their significance 
are important in evaluating the use of instrumental variables. The 
Report does not provide these results, making it difficult to assess 
the validity of these instruments. (The suggestions that the 
instrumental variables estimation is appropriate because the results do 
not change very much when estimating the regression with and without 
instrumental variables is discussed in Section D below.):

4. Specification of Before and After Periods of Merger Events is 
Problematic:

Another issue with the "event" study methodology used by the GAO 
involves the specification of the before and after periods when 
multiple mergers affect a rack in close succession. The point of an 
event study is to isolate the effect of a given event, in this case a 
merger, from all the other events that have occurred. The best way to 
isolate the effect in this case would have been to concentrate on the 
racks that did not have mergers in rapid succession. The second best 
way to isolate the events would be to specify overlapping merger dummy 
variables. In effect, the approach would be to assess whether the first 
merger has a price effect and then test whether the price effect of the 
second merger would be a price increase or decrease on top of the first 
merger effect. Instead, in the Report, the effect of the second merger 
is calculated by comparing the prices after the second merger to the 
prices before the first merger. This procedure will give misleading 
results, because the price effects of the second merger will reflect 
competitive conditions at the time it occurs, not competition prior to 
the first merger.

The specification of the merger dummy variable is of crucial 
importance. The effect of the merger is estimated by comparing the 
average price before and after the merger. As mentioned earlier, in 
some cases the before-merger period is multiple years and the after-
merger period is only six months. It is an important question whether 
six months of data is sufficient to reasonably calculate a merger 
effect. With respect to this issue, the Report states: "When mergers 
closely followed each other, it tended to shorten the before and after 
merger time periods that we could model, especially when more than one 
merger affected the same rack cities. Nonetheless we believe we had 
enough data." There is no discussion of how this conclusion was 
reached. The ability to do sensitivity testing on the size of the 
merger windows is not as problematic as the Report suggests. Additional 
data could have been used in the post-merger period to see if the 
results changed. In addition, the pre-merger period could have been 
shortened to see if that had any effect as well. Sensitivity tests on 
the duration of the pre-and post-merger periods should have been 
conducted; unfortunately, they were not.

D. Reports Results are Not Robust in Many Cases:

Because the basic methodology is fundamentally flawed, exhaustively 
discussing the results at length is unwarranted. One important point, 
however, that further undermines the reliability of the results is that 
many results are not robust to the different estimations used in the 
Report. Different specifications and estimation methodologies can 
frequently be used to estimate a given relationship. To the extent 
estimates vary significantly across specifications, assessing what is 
the "true" relationship is difficult (unless one has good reasons to 
pick a particular 
specification or group of specifications as more reliable and clearly 
explains the choice). There are several examples of lack in robustness 
in the results. For example, consider Tables 15-17 of the Report, which 
summarize the estimated merger effects from various specifications. In 
Table 16, the estimated price effect of the BP-Amoco merger ranges from 
no statistically significant price effect to 3.5 cents a gallon among 
the three regression specifications. In Table 17, the estimated price 
effects of both the Tosco-Unocal and the Shell-Texaco I transactions 
shift from being large, negative and statistically significant to being 
large, positive and statistically significant among the regression 
specifications. These results show that small changes in specification 
lead to large changes in results. The inclusion of year effects and 
using data on all the racks, albeit incorrectly, [NOTE 23] has sizeable 
effects on the regression results for a number of the mergers.

Other examples concerning the lack of robustness of the regression 
results include the results with and without instrumental variables. 
The Report mentions that the results using the instrumental variable 
estimation techniques are not very different from the fixed and random 
effects regression results. Examination of the regression results shows 
that this is not true. Tables 27-32 show the individual merger 
regression results. The first few columns of these tables show the 
results without instrumental variables and the last few columns show 
the results with instrumental variables. The regressions estimating the 
effects of the MAP-UDS merger on the price of conventional gasoline 
show that without instrumental variables there would have been no 
estimated price effect. The same is true for the estimates of the 
effect of the BP-Amoco merger on conventional gasoline, the BP-Amoco 
merger on reformulated gasoline, and Shell-Texaco I joint venture on 
branded CARB gasoline.

Moreover, when the instrumental variables approach is used, the 
relationship between price and refinery capacity utilization goes from 
positive and significant to negative and significant. The relationship 
between refinery utilization and prices is strongly expected to be 
positive. This is yet another reason to suspect that the use of these 
particular instruments was problematic. [NOTE 24]

The current results are also not robust across different racks. 
Comparing Tables 37 and 38 shows that the Tosco-Unocal merger did not 
have a statistically significant effect in the instrumental variables 
specification with the three racks used in the estimation in Table 38, 
but had a very large effect when using the six racks in California. 
Because the effects estimated for all six racks include the three racks 
where the merger had no effects, the Tosco-Unocal merger had a very 
large effect on those additional three other racks. The sensitivity of 
the results to which racks are included is an important robustness 
check and helps the reader to judge the quality of the results. The 
Report fails to report results for subsets of the racks or indicate 
whether such sensitivity checks were conducted.

Another problem is that the very strong statement in the Report that 
all known variables that affect the wholesale price of gasoline have 
been included in the regressions cannot be supported. Few of the 
regressions explain more than 15-20 percent of the variation in the 
dependent variable. If any additional control variables that would 
increase the explanatory power of the regressions are correlated with 
the merger time periods and/or cities, which is highly likely, the 
merger results would change. [NOTE 25]

IV. Price Concentration Methodology:

The Report's second econometric analysis seeks to describe the 
relationship between wholesale gasoline prices (adjusting for crude oil 
costs) and state-level concentration. While price-concentration 
studies were once a focus in the economics literature on market 
structure and industry competitiveness, these studies have been largely 
abandoned in favor of analyses like merger event studies that attempt 
to model more directly and with more precision the effects of 
structural change (such as mergers) upon prices. There are a number of 
widely recognized methodological issues with price concentration 
studies. These issues were highlighted by the 
FTC staff in their communications with GAO staff last December. 
Surprisingly, the Report does not acknowledge these problems. The 
results of the Report's price-concentration analysis also suffers from 
a lack of robustness, and there is little discussion comparing or 
reconciling the results from the price-concentration study and those of 
the merger event study.

A. Methodological Issues in Price Concentration Studies:

There is a large literature on the problems with obtaining meaningful 
estimates from price or margin on concentration regressions. [NOTE 26] 
Meaningful estimates are estimates that can be interpreted as causal. 
[NOTE 27] In other words, the estimated relationship can show that 
increases in concentration result in higher prices or margins. The key 
issues were summarized in our December 2002 letter and are still 
relevant to the Report:

The wholesale antitrust market(s) for gasoline are not likely at the 
state level; some markets are smaller and some larger. Unless the GAO 
has evidence that the changes in state level HHI's are closely 
correlated with changes in the concentration of relevant markets, the 
analysis is unlikely to provide meaningful results. The effect of 
aggregation on the estimated relationship is not easily predicted.

The reduced form model of price/margin as a function of HHI has a 
number of theoretical problems. The major problem is that the 
coefficient on HHI can not be estimated consistently. Articles in the 
Handbook of Industrial Organization (1989) by Schmalensee and 
Breshnahan discuss this issue. If the GAO is going to estimate this 
type of relationship, the large literature on the problems with this 
type of model needs to be acknowledged and addressed and/or the results 
should be given with appropriate caveats acknowledging these problems. 
The problems with estimating this type of relationship include:

(1) HHI is a function of individual firm price and quantity decisions 
which are affected by supply and demand shocks. The error term in the 
regression is also a function of these shocks. Therefore the Ml and the 
error term are likely to be correlated. While the GAO staff are using 
an instrument, number of suppliers in a state, for HHI to mitigate 
endogeneity issues, it is unclear why this instrument will allow the 
identification of the effects of competition or efficiency but is not 
correlated with other variables of interest. While HHI in a state and 
the number of suppliers will be correlated, it is likely that the 
number of suppliers will be correlated with barriers to entry, supply 
shocks, exit, etc.

(2) HHI may also be correlated with omitted variables that affect price 
such as various measures of fixed and variable costs and barriers to 
entry.

It is important to understand the source(s) of variation in the HHI 
both in formulating and in interpreting the model. The annual or 
monthly HHI's may be changing due to mergers, entry, exit, or relative 
price changes caused by supply disruptions or other factors. Few of the 
changes in the HM's will be caused by mergers. This point needs to made 
clear in interpreting the results. In addition, there is evidence that 
HHI may only matter past a critical point. The model as currently 
written is testing for a linear relationship. Alternative 
specifications should be used to test other functional forms.

There is no basis for the suggestion in the Report that, to the extent 
racks are close to each other and would tend to have similar market 
characteristics, the available state level data on market concentration 
is a reasonable variable. Although the GAO staff have added additional 
instruments to the December 2002 methodology, the same criticisms 
apply. These instruments - the unemployment rate and lags of the 
dependent variables - are unlikely to solve the multiple endogeneity 
problems nor do they solve the market mismeasurement problem. In 
addition, the criticisms about omitted variable bias and instrumental 
variable issues outlined when discussing the problems with the event 
study methodology are applicable to the price-concentration regressions 
as well. No functional form other than a linear relationship between 
price and concentration was tested. Given this linear relationship, a 
100 point change in the HHI between 900 and 1000 is treated the same as 
between 4900 and 5000. Because any relationship between 
price and concentration may not be linear (for instance, there may be 
only a relationship above a certain threshold), forcing a linear 
relationship may give misleading results. In addition, the use of year 
and HHI interaction terms should be discussed and justified. There is 
no discussion of why an interaction between the concentration and year 
variables for years when there was a sizeable change in the HHI is 
appropriate. No interpretation of these estimates is offered.

B. The Report's Price-Concentration Results are Not Robust:

Abstracting from the general problems with estimating a price-
concentration relationship, the price-concentration relationship as 
estimated in the Report is not robust in many circumstances. Tables 39 
and 40 show that the relationship between price and concentration for 
conventional gasoline, both branded and unbranded, is negative and 
significant in the first fixed effect instrumental variable 
regressions. In each of the next two estimations a variable is dropped 
and the relationship becomes positive and significant. There is no 
reason to prefer the last two specifications. As discussed earlier, 
there is no reason to drop the utilization or the inventory variable 
because of multicollinearity. The Report could just as easily have 
reached the conclusion that the relationship between the price of 
conventional gasoline and concentration is negative based on the 
results presented in the Report.

While many of the specifications showed a positive and significant 
relationship between price and concentration, the range of the effects 
is very large. The effects shown on Tables 39 to 46 give a range of the 
price effect of a 100 point change in the HHI of 0 to approximately 3 
cents per gallon. The results in Tables 18 and 19 show much larger 
effects. By adding seasonal dummy variables the price effect of a 100 
point change in the HHI ranges from 0 to 4 cents per gallon. [NOTE 28] 
With year interaction terms, the rationale for which is not clearly 
explained, the effect of a 100 point HHI change can be as large as 10 
cents per gallon.

The results presented in Tables 41 and 42 show estimates of the effects 
of market concentration on branded and unbranded conventional prices by 
regions. Results are not reported for the specifications of fixed or 
random effects instrumental variables regressions with both control 
variables, utilization and inventory ratio. The results on these two 
tables do show that without using the instruments there would be no 
relationship between concentration and price in conventional gasoline 
in the eastern half of the United States.

There is also no discussion in the Report comparing the results of the 
merger analysis and the price-concentration analysis. Because 
concentration is affected by factors other than mergers, such as entry, 
exit, and expansion, the sum of the estimated merger effects should be 
less than the effect from the change in concentration. [NOTE 29] 
In at least some cases, this is not true. Table 19 
shows that the aggregate effect of increased concentration in 
California has raised the price of branded GARB by 5 cents per gallon. 
The combined estimated effect of Shell-Texaco I (Equilon) and Tosco-
Unocal mergers on the price of branded CARB given in Tables 37 and 38 
is 8.5 cents a gallon. Because these two mergers caused only a portion 
of the change in concentration in California, a price change associated 
with the mergers larger than that associated with the change in 
concentration is puzzling. The same comparison can be made from other 
mergers and regions. The GAO preferred estimate of the price effect of 
the Exxon-Mobil merger is approximately a nickel. The entire change in 
concentration in PADD's I-III is calculated to have a similar effect on 
price. [NOTE 30]

V. The Report's Documentation is Poor:

In any well-performed study, the descriptions of the data work and the 
econometrics are sufficient to allow the reader to understand fully how 
to calculations were done. Such descriptions enable interpretation and 
replication of the results. The current Report does not include 
complete descriptions of a number of calculations and regressions. For 
example, while the merger retrospective regressions do list the number 
of racks included in each regression there is no list of which racks 
are included in each regression. By itself, this omission makes 
interpreting the results and replication impossible.

Moreover, the Report contains no discussion of how divestitures were 
handled in the estimation. For example, it is unclear from the text 
whether the racks affected by the Exxon-Mobil divestitures are included 
in the regressions analyzing the impact of the Exxon-Mobil merger. The 
description in the Report about the Exxon-Mobil merger is simply 
unclear. On the one hand, there are suggestions that there was an 
estimated increased price from the merger in markets in PADD I where 
divestitures of wholesale and retail assets eliminated the overlaps. On 
the other hand, the report does not estimate the effects of the Exxon-
Mobil merger in California where there were also divestitures. It is 
similarly unclear what procedure was followed for the BP-Amoco merger 
with divestitures in the Eastern United States or the Shell-Texaco 
merger with divestitures in California.

Another problem is that the information in the text or in the tables on 
occasion seems to be contradictory. One example is the description in 
Table 14 that suggests that any given merger variable has a value of 
one at the time of the merger and stays one throughout the rest of the 
data set. This procedure would lead to overlapping merger variables. 
The text, however, discusses the need to avoid overlapping merger 
variables.

Finally, there is no discussion of how price is measured. Although the 
prices are rack prices from the Oil Price Information Service, there is 
no discussion of the exact calculation of the rack averages and which 
formulations are used, such as reformulated with MTBE or ethanol. How 
the racks were used to calculate state level averages for the price 
concentration analysis is also omitted.

VI. Conclusion:

Omitted control variables in both of the Report's econometric analyses 
are fundamental flaws. Both studies: (1) fail to control adequately for 
exogenous factors that impact wholesale gasoline prices; (2) suffer 
from endogeneity problems that are not adequately addressed; and (3) 
have results that are not robust.

There are other problems specific to each study such as questionable 
pre-and post-event periods in the merger event study and the assumption 
that state-level wholesale markets are economically meaningful in the 
price-concentration study. The Report's documentation of its 
methodology is inadequate, particularly in view of the potential 
significance of the Report's findings for public policy in the 
petroleum industry. These flaws make the Report unable to isolate 
reliably either the effects of mergers or of concentration on wholesale 
gasoline prices. Thus, the Report cannot be used to inform public 
policy.

NOTES: 

[1] In the memorandum, we will refer to the GAO report as the "Report." 

[2] FTC staff was only given access to the Report on August 4, 2003 (a 
few weeks prior to when we understand that GAO plans to publish the 
report) and was not allowed to retain copies of the report.

[3] FTC staff also forwarded the attached report by Professor John 
Geweke on "Empirical Evidence on the Competitive Effects of Mergers in 
the Gasoline Industry," unpublished draft, July 16, 2003, to GAO staff. 
Professor Geweke is one of the most widely respected econometricians in 
the United States.

[4] Given the important differences between the December 2002 
methodology and the methodology used in the current Report, the 
opinions of external reviewers may not be the same for the Report as 
for the December methodology. If the external reviewers have not 
reviewed the final Report, they should be given that opportunity or the 
representation of their views should state that fact.

[5] Both of these analyses use posted rack (wholesale) pricing data, 
which will not reflect any discounts distributors may receive off 
posted prices. Gasoline is also sold to dealers at a dealer tank wagon 
price or is transferred to refiner owned and operated outlets at some 
internal transfer price. The portion of transactions that are sold on a 
rack basis varies considerably across the country, and as a result, any 
inferences about effects on retail prices from predictions regarding 
rack prices must be carefully qualified.

[6] Many of the estimated merger effects from the preferred estimations 
are over a nickel a gallon. When determining what is the relevant 
antitrust market in which to assess a proposed merger, the FTC asks the 
question: what is the smallest relevant market in which a hypothetical 
monopolist could impose a small but significant and non-transitory 
price increase? See U. S. Department of Justice and the Federal Trade 
Commission, Horizontal Merger Guidelines, April 1997. The FTC has 
generally applied a one cent per gallon price increase as the standard 
for evaluating market definition in refined petroleum products, 
including gasoline. While a one cent per gallon increase is smaller 
than that generated under the 5% price increase rule that the FTC 
typically uses in evaluating mergers, a one cent per gallon increase is 
extremely significant in this industry. The petroleum industry is 
characterized by large volumes and relatively thin margins. For 
example, data from the Energy Information Administration indicate that 
the net refinery margins (which reflect crude costs as well as refinery 
operating costs) were on the order of four cents per gallon during the 
late 1990's.

[7] Relevant economic papers include Barton, D.M., and R. Sherman,"The 
Price and Profit Effects of Horizontal Merger: A Case Study", Journal 
of Industrial Economics, 33(2), December 1984, pp. 165-77. Kim,E.H, and 
V. Singal,"Mergers and Market Power: Evidence from the Airline 
Industry,"American Economic Review, 83(3), June 1993, pp. 549-69. 
Schumann, L., J. Reitzes, and R. Rogers, "In the Matter of Weyerhaeuser 
Company: The Use of a Hold-Separate Order in a Merger with Horizontal 
and Vertical Effects", Journal of Regulatory Economics, 11(3), May 
1997, pp. 271-89. Schumann, L., R. Rogers, and J. Reitzes, Case Studies 
of The Price Effects of Horizontal Mergers, Federal Trade Commission, 
April 1992. Vita, M. and S. Sacher, "The Competitive Effects of Not-
for-Profit Hospital Mergers: A Case Study," Journal ofIndustrial 
Economics, 49(1), March 2001, pp. 63-84.

[8] The control rack should not be directly or indirectly affected by 
the merger. Both firms should not be posting at the control rack and the 
control rack should not be so close to the merger rack that arbitrage 
is likely. A control rack would have the same demand, bulk supply and 
fuel specifications as the merger rack.

[9] The following quotation from page 85 of the Report is incorrect 
with respect to the economics literature and information on the 
petroleum industry, and it even conflicts with other parts of the 
Report, demonstrated by the lack of explanatory power of the variables 
used in the regressions: "However, we believe that our model 
specification have captured all the relevant variables that could 
affect wholesale gasoline prices. Moreover, we believe that our 
economic methodology is sound and generally consistent with previous 
studies."

[10] The preferred specification used by GAO included city fixed 
effects or random effects as well as a correction for autocorrelation.

[11] For a discussion of omitted variable bias and its effects see the 
literature review on returns to education in Ehrenberg, R.G. and Smith, 
R.S., Modern Labor Economics, Fourth Edition, Harper Collins, (1991), 
pp. 320--330.

[12] In California, an additional complication is the need to model or 
include a variable for a new CARB specification in 1996.

[13] See E.E. Leamer, Specification searches: ad hoc inference with 
nonexperimental data, Wiley, New York (1978).

[14] The differences-in-differences model is discussed in depth in 
Angrist, J.D. and Krueger, A.B, "Chapter 23: Empirical Strategies in 
Labor Economics," Handbook of Labor Economics, Vol. 3, 1999, pp. 1296-
-1299.

[15] Including week, season, year and PADD level dummy variables in the 
same regression may remove much of the meaningful variation while 
controlling for seasonal effects. If the year dummies are to proxy for 
supply disruptions such as refinery outages, a change in the level of 
gasoline imports, gasoline formulation changes, or a demand change, it 
would be better to measure these effects directly.

[16] For at least some of the West Coast mergers, the after-merger 
period ends before the major supply disruptions in 1999-2000.

[17] Taylor, C., and J. Fischer, "A Review of West Coast Gasoline 
Pricing and the Impact of Regulations," International Journal of the 
Economics of Business, Vol. 10(2), 2003, pp. 225--243.

[18] In order to produce additional reformulated gasoline a refiner has 
to produce less conventional gasoline, all else being equal. A refiner 
will decide on how much reformulated and conventional gasoline to make 
based on the relative margins. Since making reformulated gasoline 
requires additional capital investment, the margin on reformulated, and 
hence conventional, must be higher to cover the capital investment.

[19] The price-concentration analysis in the Report did use seasonal 
dummy variables.

[20] The instrumental variables include: number of suppliers at the 
rack, state level unemployment rate, the previous period's state level 
unemployment rate, the previous period's inventory ratio, the previous 
period's utilization, level, a time trend, and a time trend squared. 
The number of suppliers at the rack changes for a number of reasons 
including mergers. There is also no discussion in the Report of how the 
number of suppliers at the rack was determined. If the number of 
suppliers at the rack is detennined by counting firms posting at the 
rack, this raises other issues, including: one firm posting multiple 
brands at a rack and traders who post infrequently at the rack and do 
not have their own supply but are merely reselling gasoline from 
another supplier to that rack.

[21] The endogeneity problem in this case is that prices, inventories, 
and utilization (a gross measure of quantity) may be jointly 
(simultaneously) determined. Nevertheless, the test for endogeneity 
assumes that the instruments being used in the test are valid.

[22] For an example see, Staiger, R.W. and J.H. Stock, "Instrumental 
Variables Regressions with Weak Instruments," Econometrica, Vol. 65, 
1997, pp. 557--586 and Bound, J., D. Jaeger and R. Baker, "The Cure Can 
Be Worse Than the Disease: A Cautionary Tale Regarding Instrumental 
Variables," NBER Technical Paper 137, June 1993.

[23] As discussed on page 9, the pooling of all racks was incorrectly 
done. The merger effect is being calculated based on a control price 
that is an average of non-merger affected racks before and after the 
merger period and the merger racks before the merger.

[24] The rationale for dropping the utilization variable from the 
preferred regression specifications does not make sense. The fact that 
utilization is correlated with the inventory ratio variable is not 
relevant. Eliminating variables from a regressions because of 
multicollinearity when both variables are independently significant is 
not appropriate. There are a number of significant changes in the 
results when the utilization variable is excluded.

[25] As mentioned earlier, variables measuring seasonality, supply 
outages and formulation changes will likely be correlated with the 
merger variable.

[26] Examples of this literature include: Evans, W., L. Froeb, and G. 
Werden, "Endogeneity in the Concentration-Price Relationship: Causes, 
Consequences, and Cures" Journal ofIndustrial Economics, Vol. 41, 1993, 
pp. 431--438 and Breshnahan, T.F. "Empirical Studies of Industries with 
Market Power," in Schmalensee R. and Willig R. (Eds), Handbook of 
Industrial Organization, Vol. II, Ch. 17, 1989, (North-Holland, 
Amsterdam).

[27] Given the number of well documented theoretical problems with 
estimating and interpreting price concentration regressions, the 
following statement on p. 146 of the Report cannot be supported: "Also 
we used the market concentration model because market concentration 
better represents overall market conditions than mergers."

[28] It is not clear why seasonal dummy variables are included in the 
price-concentration regressions but not in the merger event studies. 
Any variable that explains variation in the price variable should be 
included in both studies. If seasonal effects were significant in the 
price-concentration analysis (and they must have had sizeable effects 
given the change in the results), then they should have been included 
in the merger event analysis.

[29] This would be true unless the merger effects were very short 
lived.

[30] Even as a purely descriptive matter, the Report does a poor job 
in linking concentration changes to mergers. In Chapter 3, however, the 
Report offers a statistical correlation analysis to associate the 
degree or connections between merger activity and concentration. Data 
used for this correlation are the HHI estimates for domestic crude oil, 
refining capacity, and the Herold data set on merger and acquisitions. 
This analysis shows positive correlations between HHI and merger 
activity. Correlation analysis does not establish causation, however, 
and we suspect that similar results would have been obtained had HHI 
estimates been correlated with the overall merger activity in the 
economy or stock market indices. As far as we can tell, merger 
transaction value from the Herold data set are not separated out by 
industry segment or geographic area: thus it appears that any 
functional level (crude, refining, or wholesaling) and geographic area 
that might be affected by a given merger appears to have the same 
weight in the correlation against-the relevant BM. The data for crude 
oil and refining does permit estimation of the concentration changes 
associated with individual mergers because the data is reported by 
firm. This would be the usual approach in connecting mergers with 
changes in HHI. While the wholesale data does not identify firm-
specific data because of confidentiality restrictions imposed by EIA, 
this data is available monthly. Because the month when transactions are 
consummated is known, fairly strong inferences can be made about the 
effect of particular mergers on HHI. It is not clear why such an 
analysis was not done in the Report. 

The following are GAO's comments on the Federal Trade Commission's 
Bureau of Economics staff letter dated August 25, 2003.

GAO's Comments: 

1. FTC's claim that the staff had limited ability to review the draft 
is inconsistent with FTC's statement that they have "spent significant 
resources investigating consummated mergersï¿½" and have "accumulated 
substantial methodological expertise and have applied that expertise to 
the oil industry as part of our enforcement mission." We provided the 
draft report to FTC for review as a courtesy because GAO is not 
required to obtain formal agency comments for a report that did not 
specifically audit that agency's actions--in this case, FTC's merger 
enforcement actions. We first delivered the draft to FTC on August 5, 
2003, and received their written comments on August 25, 2003, totaling 
29 pages, excluding other enclosures. We had discussed our study's 
approach and overall methodology with FTC staff from the beginning of 
our study, including our meeting in December 2002, when we discussed 
FTC's written comments to the outline of our preliminary methodology.

2. We disagree that our econometric analyses have fundamental 
methodological flaws because we used sound econometric analysis that is 
consistent with the existing literature. We also solicited and obtained 
comments from experts who reviewed the econometric approach and we 
incorporated these comments into our model development, as appropriate. 
In addition, we consulted with a recognized expert in econometric 
modeling of petroleum markets, who peer reviewed our detailed 
econometric analysis and results and provided comments, which we 
incorporated as appropriate.

3. FTC provided us with written comments on our preliminary model 
outline. At the request of FTC, we met to discuss the issues in the 
written comments that they deemed to be crucial. At this meeting we 
discussed issues FTC felt might be addressed, but some of the issues 
FTC staff raised were so complex and theoretical that they themselves 
could not offer feasible solutions. In instances where it was 
reasonable and possible to make changes, we did so. In particular, a 
major point of concern FTC expressed after our December 2002 meeting 
was the limitation of the HHI data, based on prime suppliers' sales, 
that EIA provided to us--the mergers were not reflected in the HHI data 
until the merged firms began to file a combined report with EIA, 
possibly months or even years after a merger was completed. We 
subsequently contacted EIA, who provided us with revised HHI data, 
adjusted properly for the timing of the mergers. Nonetheless, we have 
replaced the monthly HHI based on prime suppliers' sales with yearly 
HHI based on refinery capacity because we believe the latter measure 
better captures the ability of suppliers (refiners) in wholesale 
markets to control production.

4. It is not unusual for a GAO report to be published within a few 
weeks of giving an agency access to the report. Also, where there is 
concern about premature disclosure, it is not unusual for GAO to demand 
the return of a draft report. All drafts remain the property of GAO.

5. FTC provided us with Professor John Geweke's unpublished review of 
previous studies on competitive effects in the gasoline industry. We 
were already aware of these studies, having read them and even cited 
them in the model outline that we provided FTC for review in December 
2002. Although we respect Geweke's scholarship and contributions in 
econometrics, he has not done or published any research work on the 
gasoline industry (based on his vita that FTC attached to his paper) to 
gain a thorough understanding of the gasoline market. We disagree with 
his overall assessment of the competitive effects of mergers in the 
gasoline industry. In fact, FTC's use of Geweke's expertise is 
inconsistent with remarks of FTC Chairman Timothy Muris in a speech 
entitled, "Improving the Economic Foundations of Competition Policy," 
dated January 15, 2003, stating that "antitrust analysis, if done 
correctly, uses the NIE (New Institutional Economics) approach--that 
is, a careful, fact-based economic analysis grounded in a thorough 
understanding of the relevant institutions" (p. 1).

6. We disagree with FTC's assertion that there are fundamental flaws in 
our models and that the results cannot be used to make inferences about 
the price effects from the mergers analyzed and from the effects of 
increased market concentration. First, our methodology was based on 
sound and reasonable approaches to analyzing the effects of mergers and 
market concentration on prices, as found in the economic literature. We 
appreciate FTC's concern and acknowledge that, like all econometric 
studies, ours is not perfect. Contrary to what FTC purports, we also 
believe the limitations are not "severe" because our methodology is 
consistent with previous studies and is in accordance with accepted 
econometric practice. Indeed, FTC has recognized in a speech by its 
(former) Director and Deputy Director of the Bureau of Economics that 
"analyses can lead to different conclusions because of different data, 
different economic modeling, different econometric techniques, and/or 
fundamental mistakes" (Scheffman and Coleman, undated, p. 2). 
Furthermore, they stated that "there is no 'perfect' econometric studyï¿½ 
Lack of unachievable perfection should not be a bar to an econometric 
study being given weight" (p. 3). FTC criticized our draft report 
without offering any empirical support that those criticisms are valid 
or providing reasons why potentially different conclusions could be 
obtained. FTC's comments may also be contrary to the statement by its 
(former) Director and Deputy Director of the Bureau of Economics that 
"a technically-based critique should be supported by an empirical 
analysis that shows that dealing appropriately with the technical issue 
makes a meaningful difference in the results" (Scheffman and Coleman, 
undated, p. 3).

7. We disagree with FTC's key comment on our econometric methodology 
that we do not control for important factors that affect gasoline 
prices. The reason given by FTC is that the amount of variation 
explained by our models is relatively low (reported R-squares are less 
than 20 percent). See comment 24. There are several reasons why FTC's 
characterization of the explanatory power of our models of gasoline 
prices is flawed. We specified the dependent variable as a price-crude 
cost margin (crude costs represents over 60 to 70 percent of refining 
costs). This approach is a generally accepted and statistically 
preferred technique for assessing the market power and/or efficiency 
effects of mergers and market concentration. Our specifications 
generally resulted in R-squares that ranged from about 16 to 36 
percent. It is not unusual to obtain low R-squares for models that 
explain price-cost margins. For example, Collins and Preston[Footnote 
130] obtained R-squares of about 20 percent. To demonstrate that the 
apparently low R-squares are primarily due to having the crude costs as 
part of the dependent variable, we reestimated the models with the 
crude costs as an explanatory variable, instead of the specification 
that we presented in the draft report, and the R-squares for the merger 
and market concentration models were very high; they generally exceeded 
80 percent. Second, the relatively low R-squares in the draft report 
were probably due to the lack of data for capital costs for refining, 
which FTC did not list as one of the key omitted variables, presumably 
because FTC also recognized the lack of appropriate data for this 
variable. With wholesale gasoline prices less crude oil prices as the 
dependent variable, as much as 60 percent of the variation could be due 
to capital costs, which we do not have data for. Therefore, we believe 
our models, which explain about 15 to 35 percent of the variation in 
gasoline prices after accounting for crude oil costs, are sound and 
reasonable. Third, as noted in Kennedy,[Footnote 131] "In general, 
econometricians are interested in obtaining 'good' parameter estimates 
where 'good' is not defined in terms of R2. Consequently, the measure 
of R2is not of much importance in econometrics.": 

Furthermore, contrary to FTC's claim, our model specifically 
incorporates important factors that affect the price of (wholesale) 
gasoline. We provide the following additional explanations for why our 
econometric specifications do not lack important controls.

a. First, our models, apart from including the effects of mergers or 
market concentration, account for key demand and supply variables, 
including crude oil prices (which account for about 60 to 70 percent of 
refining costs), gasoline inventories relative to demand, and refinery 
capacity utilization. In the economic literature and in suggestions of 
experts, all these factors have been found to be important in 
determining wholesale gasoline prices.

b. Second, certain factors, some suggested by FTC, are being captured 
by other factors that are already in the model. For instance, 
seasonality, which FTC suggested that we include in our models (see 
comments 14 and 44), is captured by the variable for gasoline 
inventories relative to demand. The issue of seasonality of gasoline 
prices is primarily related to the behavior of gasoline inventories and 
demand. Therefore, we prefer to use the actual data that represent 
seasonality rather than an indicator variable for seasonality, a proxy, 
which FTC seems to prefer. This same reasoning applies to FTC's 
suggestion in the Commission's comments that we include temperature, 
which is also intended to reflect seasonality. In addition, in the 
Commission's comments, FTC suggested that we include income in our 
models. However, as we stated in our draft report, the available data 
for income by city does not vary over time (time-invariant) and could 
not be estimated because we use a fixed-effects estimator, which makes 
it impossible to estimate time-invariant variables, including income.

c. Third, we constructed measures of the supply disruptions in the 
Midwest and West Coast--albeit crude and imperfect measures--that we 
included in the models. The majority of results of the models changed 
little when the supply disruptions were included. See comment 3 in 
appendix V for a detailed response.

d. Fourth, FTC indicates that we failed to account for another type of 
supply shock involving changes in fuel specifications that affect 
reformulated gasoline (RFG) and CARB (see comments 14 and 37). As noted 
by FTC, there was a change from Phase I to Phase II RFG in the Midwest 
in 2000. Due to insufficient data on RFG in the Midwest, our analysis 
of the Midwest (PADD II) market focused only on conventional gasoline, 
and we believe that our estimated results are reasonable and valid. 
Furthermore, the RFG market in the Midwest over the period of our study 
was very small relative to conventional gasoline. For CARB, the change 
from Phase I to Phase II occurred in 1996, and our analysis reflects 
this change because our analysis covered the period from 1996 through 
2000.

e. Fifth, the variables incorporated in the models depend on the 
econometric technique used for the type of data involved in our 
analysis, namely, panel data--data for multiple rack cities over a 
period of time. There are two common approaches for estimating panel 
data: the fixed-effects estimator and the random-effects estimator. The 
former is preferred when observations are not chosen randomly and there 
are likely to be unobservable, site-specific effects. This estimator is 
implemented by including an indicator variable for each rack city 
(city-specific effects). In wholesale gasoline markets, such unobserved 
differences might include (1) unmeasured supply or demand effects, such 
as different pricing strategies of the refiners at the different rack 
cities, and (2) the level of development of the transportation system 
in the different areas. A major advantage of the fixed-effects 
estimator is that there is no need to assume that the unobserved city-
specific effects are independent of the included explanatory variables. 
Furthermore, the selection of the rack cities used in our study was not 
random but was based on data availability. We therefore prefer the 
fixed-effects estimator in the final report. On the other hand, the 
random-effects estimator allows one to include a time-invariant 
variable. Also, the random-effects estimator allows one to make 
unconditional (marginal) inferences with respect to the population of 
all effects. However, one has to make specific assumptions about the 
pattern of correlation (or assume no correlation) between the 
unobserved effects and the included explanatory variables. The need for 
these assumptions is a major shortcoming of the random-effects models 
because there are reasons to believe that the assumption of no 
correlation may not be correct for wholesale gasoline markets and could 
bias the estimates. (See, for example, Hsiao).[Footnote 132]

8. We disagree that our current methodology differs significantly, in 
substance, from the December 2002 preliminary methodology that we 
provided to the external reviewers. Furthermore, we have indicated in 
the report that some of the reviewers had the opportunity to review 
only the preliminary model outline while others reviewed the full and 
complete draft report.

9. FTC's criticism of our use of rack price data is unwarranted, given 
that FTC used the same data from OPIS (Oil Price Information Services) 
in its report to the Congress, Midwest Gasoline Price Investigation, 
dated March 29, 2001, on wholesale gasoline prices in the Midwest. 
While there are different gasoline price series at the wholesale level, 
as we have laid out in chapter 4 of both the draft and final reports, 
the rack market is still the predominant market in the United States as 
a whole, and there are no data to verify the extent to which wholesale 
gasoline markets vary geographically. Also, we stated in the draft and 
final reports that we did not infer from the econometric analysis what 
our findings on wholesale prices imply for retail prices.

10. We disagree with FTC's characterization of the interpretation of 
our results. We used sound econometric methodology and we carefully 
estimated and interpreted the econometric results.

11. We disagree. The consistency of the results we obtained from the 
different specifications and estimations of a particular model, as well 
as consistency in the results for the two different model types--the 
mergers' effects and the market concentration effects--support the 
robustness of our results. In particular, we provide the following 
evidence for the robustness of our results.

a. First, the results with and without the supply disruptions were 
generally similar. We also used the HHI based on prime suppliers' sales 
and the majority of the results were similar qualitatively to those 
obtained using the HHI based on refinery capacity.

b. Second, because market concentration reflects the cumulative effect 
of the mergers and other competitive factors, one would expect that the 
results from the market concentration models and mergers models would 
be similar if mergers are the predominant contributing factor to market 
concentration. In our study, the two approaches yielded qualitatively 
similar results. In the draft report, we also estimated the effects of 
the mergers using two approaches--using data for all racks and using 
data for only the merger affected cities. Both approaches yielded 
qualitatively similar results.

12. We disagree. While there is no a priori basis for what the 
magnitude of the effects of the mergers and market concentration should 
be, we think that our estimated effects, which are generally below 5 
cents per gallon (cpg) for conventional and reformulated gasoline, are 
reasonable given that the average levels of the wholesale price margins 
(wholesale gasoline prices less crude oil prices) ranged from about 20 
cpg for conventional and RFG gasoline to about 30 cpg for CARB 
gasoline.

13. We do not agree. We provided detailed descriptions of each variable 
that we used, including the frequency, time period covered, gasoline 
specifications and brands, and sources of the data. We specified 
completely our basic econometric equations for both the mergers models 
and market concentration models, and any modifications that we made to 
the basic equations. We also indicated the estimation techniques used 
and why we used them and the various statistical tests that were 
performed, which are all common in the econometric literature.

14. We disagree with FTC's assertion that our econometric methodology 
excludes important control variables. We included key control 
variables, including consideration of these specific issues, given data 
availability. See comments 7(b)(c)(d)(e).

15. Most of the estimated effects of the mergers and market 
concentration are below 5 cpg, especially for conventional and RFG 
gasoline prices, which both averaged about 20 cpg over the sample 
period of our study. The estimates are above 5 cpg for both the Exxon-
Mobil merger in the case of reformulated gasoline and the Tosco-Unocal 
merger in the case of CARB gasoline, for which the wholesale price 
margin averaged about 30 cpg.

16. The FTC's test may be useful in analyzing the effects of specific 
mergers and may not apply to our study, which looks at the effects of 
multiple overlapping mergers. In fact, the apparent "one-cent per 
gallon increase" rule for gasoline markets used by FTC is ad hoc--it 
does not come from economic theory or from empirical information about 
gasoline markets. FTC's statement that EIA found that net refinery 
margins were about 4 cents per gallon is not inconsistent with our 
overall findings, where our price margins are only net of crude costs. 
Furthermore, FTC's consultant, Professor John Geweke, in his review of 
the study by Hastings (2002), apparently did not find any objection to 
the result that prices increased by about 5 cpg after the Atlantic 
Richfield Oil Company (ARCO) announced the long-term lease of service 
stations from Thrifty in 1997.

17. We disagree with FTC's characterization of how we dealt with the 
issue of endogenous variables in the draft report. As indicated in a 
previous study FTC cited, the instruments for endogenous variables may 
not meet the requirements for ideal instruments, but they serve to deal 
with the problem of endogeneity of the variables. (See Evans, Froeb, 
and Werden (1993).) Also, even when the instruments are possibly 
correlated with the stochastic disturbance term (i.e., the instruments 
are not ideal), the instrumental variables (IV) estimates may be 
preferred to the non-IV estimates when the R-squares between the 
endogenous regressors and the instruments are not low (exceed 0.1); see 
Hahn and Hausman (2003). In other words, even when the instruments are 
not perfect (or ideal), it is still preferable to use the instruments 
if they are not weak, as in our case, rather than not instrumenting at 
all.

As indicated in the draft report, we dealt with the problem of 
endogenous variables as follows. First, we ran regressions of the 
endogenous variables on the selected instruments to ensure that the 
instruments were not weak. The results indicated that the estimates 
were highly statistically significant. Second, our tests for exogeneity 
of the endogenous variables generally rejected the null hypothesis that 
the variables were exogenous--indicating that the instruments would be 
preferred. Third, our tests indicated that the overidentifying 
restrictions of the instruments were not ideal, using standard 
econometric tests. Finding ideal instruments for these endogenous 
variables is difficult, especially for market concentration, when 
modeling gasoline prices. None of the previous studies on price-
concentration performed this test, including the 1993 study by Evans, 
Froeb, and Werden, which we, as well as FTC, cited. We also note that 
FTC, while criticizing us for the instruments that we used failed to 
provide any suggestions on what instruments would be more appropriate. 
Furthermore, we discussed the issue of the instruments with FTC staff 
during our meetings. In fact, we stated in our draft report that the 
issue of finding ideal instrument(s) for market concentration, if any, 
in a price relationship was discussed extensively in FTC's (2001b) 
Empirical Industrial Organization Roundtable (see for, example, pp. 17-
18, 28, 36-37), where it was agreed that no commonly accepted solution 
existed and that this issue was problematic.

Nonetheless, in the final report, we have used a modified set of 
instruments to account for the potential endogenous regressors--ratio 
of inventories to demand and refinery capacity utilization rates. The 
instruments were 52 weekly (seasonal) dummies, time trend, and squared 
time trend. Our tests results of exogeneity of the endogenous 
regressors and overidentification of the instruments, based on the 
Hausman (1978) specification test, indicated that the two regressors 
were generally exogenous, although endogenous in some models; in those 
cases the instruments were found to be appropriate or valid. 
Furthermore, the HHI measure based on refinery capacity would more 
likely be exogenous to rack prices, unlike the actual flow of gasoline 
sales, which are more reactive to actual current gasoline prices.

18. We disagree with FTC's assertion that we did not appropriately 
specify the pre-and postperiods for the mergers. In fact, in the draft 
report, we used two approaches, where we clearly identified the pre-and 
postperiods of the mergers, to determine the effects of the mergers in 
the merger affected cities, and both approaches yielded similar 
results. Furthermore, in constructing the data, we used due diligence 
to ensure that there were enough data both before and after the mergers 
to estimate the mergers' effects.

The approach suggested by FTC attempts to match the diverse merger 
cities to a representative nonmerger city. While this matching process 
might be useful in concept, it is seldom pragmatic to find a control 
city that has the same demand and supply characteristics, except for 
the merger, when one wants to consider all the available cities that 
were affected by the mergers. These cities are generally diverse 
because in most cases a merger affected more than one broad geographic 
area, and the affected racks generally include large as well as small 
cities. Furthermore, even if one could select a nonmerger control city, 
that city is likely to be near the merger cities. In this case, the 
mergers could indirectly affect prices in the selected nonmerger 
control city as well, violating the requirement that the mergers should 
not affect the selected control city. In our study we use a statistical 
technique to adjust for contemporaneous error effects across the rack 
cities.

19. The first approach would not be feasible for analyzing the effect 
of mergers on gasoline markets because it was virtually impossible to 
select an appropriate control city for each merger because each merger 
affected multiple cities that are generally diverse. Although we 
believe the approach might be appropriate for other products--for 
example, in the case of hospital mergers, where control cities might be 
easily identified because nearby cities might have similar demand and 
supply characteristics, and a merger that affects one city is not 
likely to affect the other city. For wholesale gasoline markets, while 
a nearby city is likely to have similar supply and demand 
characteristics as the merger affected cities, the merger would very 
likely affect prices in the nearby city due to spatial-price 
competition, violating a key requirement for a control city. 
Furthermore, we note that despite FTC's experience in reviewing most of 
the mergers that we modeled, FTC did not provide any examples of 
control cities that would be appropriate for the mergers when we 
discussed with FTC staff, earlier on, the mergers that we were 
analyzing and our proposed methodology. Moreover, our statistical 
approach, the fixed-effects estimator, uses a methodology that allows 
us to control for the unique characteristics of each city and 
contemporaneous error effects across rack cities.

For the second approach, although there are potential problems in 
obtaining demand and supply variables that vary over time and space, we 
do not believe this problem is as acute in gasoline markets. As 
indicated in comment 7, our models generally do a good job in 
explaining the variations in gasoline prices. Therefore, we believe 
that FTC's criticisms of these models are exaggerated. Also, based on 
our review of the literature, modeling, and discussions with experts, 
many of the supply and demand factors that FTC claims that we excluded 
can be captured by the variables that we included in the models, 
especially gasoline inventories relative to demand.

While our approach is in the spirit of the second methodology because 
we control for key demand and supply factors, it also reflects the 
first methodology since we use data for all available rack cities 
(merger and nonmerger affected rack cities), with the nonmerger 
affected rack cities serving as control cities.

20. In order to show that our methodology has been applied to many 
industries, we cited in the draft report all the studies FTC cited, 
including a study on railroad mergers published by GAO staff--Karikari, 
Brown, and Nadji (2002)--not cited by FTC.

21. We disagree with FTC's claim that we indicated in the December 2002 
model outline that the regressions would include control cities. We 
rather indicated that we would include the prices in nearby cities as 
one of the explanatory variables to capture spatial-price competition 
in gasoline markets. In fact, FTC stated in its comment on our December 
2002 model outline that "the use of nearby margins as a control factor 
is complicated by the number of gasoline specifications and other 
factors. The choice of the nearby margins will be difficult." A nearby 
city is not likely to be an appropriate control city. We explained in 
both the draft and final reports that we did not eventually include the 
variable for the nearby prices. This is because if regional and local 
variables that drive wholesale prices, such as transportation costs, 
are omitted, then prices in the nearby cities will be a strong 
predictor of prices in the other cities, even if the suppliers at 
nearby racks do not compete; this, therefore, creates statistical 
problems due to correlation with the error term. Furthermore, because 
the mergers and market concentration are likely to affect prices in the 
nearby cities; the nearby prices should not be included in the models 
when the mergers or market concentration variable is already included.

In the final report, we have incorporated the possible effects of 
nearby cities through the estimation procedure--we handled this 
potential problem by accounting for contemporaneous cross-sectional 
(rack city) correlations.

22. We considered using the nearby prices to capture the spatial-price 
competition in gasoline markets, contrary to FTC's assertions that 
prices at the nearby cities would be used as control cities. The nearby 
cities are not likely to be appropriate control cities for the reasons 
we stated in comments 19 and 21, specifically the potential for 
indirect effects of mergers on the control cities. Based on FTC's 
comment on the requirements needed to qualify as a control city, it is 
nearly impossible to find cities that would completely fulfill these 
requirements. We therefore chose an approach that did not depend upon a 
matching of merger cities with appropriately defined control cities.

23. While we agree with FTC that these conditions should be included in 
the list of requirements needed to qualify as a control city, these 
conditions, augmented by the concerns above, led us to conclude that it 
is not practical to identify an appropriate control city for the 
purposes of our multimerger analysis. Furthermore, the nearby rack city 
was based solely on distances between cities and would have served a 
different function in our model, as opposed to being used as a control 
city in the sense used by FTC.

24. We disagree. The models that we developed included regulatory 
factors, such as divorcement regulations, and several other supply and 
demand factors. However, in the final estimation, while we could not 
directly include all possible demand and supply factors, due to data 
limitations and estimation techniques, our estimated results show that 
our models are not under-specified. See comment 7.

25. We do not agree with FTC's characterization. As we indicate in 
comment 7 above, our models generally explain a high proportion of the 
variations in gasoline prices. FTC's comment about low R-squares is 
unreasonable and unwarranted because higher R-squares are obtainable 
but those models have less preferred statistical properties. 
Furthermore, experts in gasoline markets reviewed our methodology.

26. We disagree. The model specified by FTC did not represent our 
preferred model in the draft and final reports. In fact, we stated that 
"we used data for all the racks, and the specific merger dummy 
variables (MERGERki) are applicable only in the rack cities where the 
merging companies operated." Also, because we used data for all the 
rack cities, we estimated the effects of each of the multiple mergers 
in the same model.

27. As FTC economists have stated there is no perfect econometric model 
and we have replaced the word "all" with "key" in the final report. See 
comment 7(a).

28. We do not agree with FTC's claim that our models are under-
specified. In the draft report, the refinery capacity utilization rates 
variable was dropped because its expected sign was inconsistent when 
included with the ratio of gasoline inventories to demand variable. 
Furthermore, when the ratio of gasoline inventories to demand was 
excluded, the utilization rates variable had the expected sign. Also, 
the data for utilization rates are available nationally, while the data 
for the ratio of gasoline inventories to demand are available 
regionally, which better captures differences in prices across markets. 
Nonetheless, we have included the refinery capacity utilization rates 
variable in all our models. For the effect of a merger, any measure of 
its effects using an indicator variable is a comparison of average 
prices before and after the merger. Our models are not under-specified 
as discussed in comment 7 above.

29.While we agree that if relevant variables are omitted and those 
variables are correlated with the included variables, the parameter 
estimates would be statistically biased, we disagree that our models 
suffer from omitted variable bias. See comment 7.

30. We disagree with these assertions by FTC. First, our study was not 
affected by the changes in gasoline formulations (see comment 7(d)). 
Second, the year effects would be poor proxies for the supply 
disruptions, which included disruptions and refinery closures, because 
the supply disruptions typically did not span the whole annual period 
from January to December of a year or they affected certain time 
periods in one year and other time periods in another year. In this 
case, using year dummies would overstate the impact of the supply 
disruptions. Third, we estimated separate models for different regions 
(the East and the West) of the United States to account for regional 
differences such as imports. In general, while the year effects would 
control for cyclical patterns common to all rack cities if they 
existed, we do not believe there is an annual cyclical phenomenon in 
the gasoline markets that we studied. Furthermore, the year effects 
were excluded because of their correlations with the key variables--
mergers and market concentration. The mergers or market concentration 
variables are correlated with the year effects merely because these 
effects appeared in certain years, and the merger has a stronger 
conceptual basis for inclusion. FTC's inference about the effect of 
omitted variables on the estimates is speculative and has no 
econometric evidence.

31. In the draft report, the year effects were included in some of our 
models as part of our sensitivity analysis. However, as indicated in 
comment 30, there is no convincing reason to include the year effects 
in the models. Furthermore, we disagree with FTC's characterization of 
our estimates that take into account year effects. Only the results in 
table 17 of the draft report were affected qualitatively with the 
inclusion of the year effects.

32. We disagree with FTC that our econometric methodology was based on 
some sort of "regression fishing." In an attempt to develop a 
reasonable model, it is not unusual to try other specifications and use 
reasonable judgment based on the institutions of the market being 
modeled to make model selections. Indeed, the methodology suggested by 
FTC--selection of control cities--could be subject to FTC's critique 
that it involves a selection bias by the researcher.

33. While it is true that omitted variables could bias the estimates, 
this does not apply to our models. Also, we believe that a more 
relevant discussion of the statistical effects of omitted variables is 
captured in Greene,[Footnote 133] which uses an example for gasoline 
markets.

34. We disagree. Our data for CARB starts in 1996 and therefore 
incorporates the CARB Phase II. See comment 7(d).

35. We do not agree with FTC's characterization. In the draft report, 
our preferred model used data for all the racks because it enabled us 
to control for systematic variations across all the racks, both merger 
affected and nonmerger affected cities. We performed a sensitivity 
check estimating the mergers models with data for only the merger 
cities. Also, we did not use a control group price (see comments 19, 
21, and 22). As indicated, the nonmerger cities were included for a 
different purpose.

36. FTC suggests that we estimate a difference-in-difference model of 
the mergers' effects instead of the event-regression estimates. And, as 
emphasized in FTC's comments, this approach must be implemented 
correctly. However, FTC's proposed difference-in-difference equation 
is flawed. Adding "Dmergtime" (time dummy) to the merger dummy (event 
dummy) does not produce a difference-in-difference estimator in the 
present analysis that covers more than two time periods. This point is 
presented in standard econometric textbooks that discuss the 
difference-in-difference estimation (see, for example, Wooldridge, 
Econometric Analysis of Cross Section and Panel Data, 2002, pp. 129-
130). The correct way to implement a difference-in-difference estimator 
when the sample covers multiple time periods is to include the merger 
dummy variable used in our study as well as group-specific and time-
specific dummies (see, for example, Bertrand, Duflo, and 
Mullainathan).[Footnote 134] The models that included the year dummies 
incorporate the "correct" difference-in-difference estimation. FTC's 
criticism is unfounded and their suggested alternative is flawed.

Furthermore, as FTC notes above, the estimation would depend crucially 
on how the control cities are picked. As indicated in comments 19 and 
21, it is nearly impossible to have appropriate control cities for each 
of the mergers, as reflected in FTC's comments, where they acknowledge 
difficulty in choosing appropriate control cities. Also, as noted in 
Angrist and Krueger (1999, p.1299), the difference-in-difference 
technique, like other techniques, is not guaranteed to identify the 
merger effects.

37. We disagree with FTC that the formulation changes were not 
incorporated in our models because either (1) we did not model those 
fuel specifications because of a lack of data or (2) our data started 
in the same period when the changes went into effect. See also comment 
7(d). We agree with FTC that the existence of other formulations could 
affect the prices of the formulations that we modeled. Nonetheless, as 
FTC agreed during our December 2002 meeting, the effects of other 
formulations could be minimal because they are typically a small 
percentage of the total volume of gasoline in the areas that we 
modeled. Furthermore, there were not sufficient data on rack prices for 
the other formulations to incorporate their effects.

38. We disagree with FTC that the gasoline formulation changes were not 
incorporated in our models. While we agree that there are various 
formulations of gasoline within the country, we did not model every 
formulation because most of them are limited to very small areas, and 
there are not enough systematic rack price data for them. We stated in 
our draft and final reports that we modeled the three dominant gasoline 
formulations--conventional, reformulated, and CARB--because there were 
enough data in the OPIS database for these formulations. Although we 
did not model specifically the Reid vapor pressure, this factor is 
generally accounted for by seasonality, which we accounted for in our 
study.

39. We disagree with FTC's characterization of the scope of our 
December 2002 model outline and the draft report. The estimates in the 
draft report are based contextually on the December 2002 outline. We 
provided detailed discussions of why some of the variables could not be 
directly estimated. We did not simply decide to exclude those 
variables. See comments 7 and 30-32 regarding our consideration of 
supply disruptions, seasonal variables, and year indicator variables.

40. While we agree that these effects are important, we do not agree 
that they could be easily incorporated in the models due to the lack of 
accurate data. Nonetheless, as indicated above, we included these 
effects in our models (see comment 7(c)).

41. FTC appears to be inconsistent in the methodology that it thinks 
would be more appropriate, given the available data. FTC had indicated 
earlier (see comment 30) that year effects might be used as a proxy for 
supply disruptions, changes in imports, and gasoline formulations, but 
now seems to suggest that it is better to measure these effects 
directly. The latter reasoning leads to our preference for using the 
ratio of inventory to demand variable. We believe that our models have 
reasonably accounted for these effects, albeit indirectly, because they 
could not be measured appropriately with the available data.

42. We disagree with FTC's overall assessment of the relationship 
between inventories and supply shocks. We also disagree with FTC's 
claim that inventories do not change when there are supply shocks 
because inventories are more likely to be drawn down during supply 
shocks to help make up for disrupted supplies. While the ratio of 
gasoline inventories to demand variable might not completely account 
for the effects of the supply disruptions, we found that gasoline 
prices and inventories are inversely related. We infer also that the 
inventory levels variable likely used by FTC in its statement is not a 
useful measure of the role of inventory during supply shocks because it 
does not account for demand. Furthermore, we obtained a significant and 
positive effect for the Midwest disruption variable when we estimated 
its impact using a crude and imperfect measure. We used an indicator 
variable for the whole of the Midwest over a certain time period when 
the supply disruption might have been in effect, while the disruption 
might have affected only specific areas with differential price 
effects. For the West Coast, we also constructed a crude and imperfect 
measure of the disruptions and incorporated it in other specifications 
of our models. As already indicated, the disruptions did not affect all 
of the mergers or all geographic markets. In particular, our estimates 
of the Exxon-Mobil merger effects are not affected directly because the 
merger did not affect rack cities in the Midwest (PADD II) for 
conventional or reformulated gasoline, or racks in the West Coast (PADD 
V) for CARB gasoline. See comment 2 in appendix V for a detailed 
account.

43. We disagree. We believe that we have fully addressed these issues. 
See comment 7(d).

44. FTC's suggestion that we use seasonality, temperature, and supply 
disruptions in our merger regressions means resorting to proxies when 
we have more direct measures of demand and supply shocks. This 
suggestion is inconsistent with accepted econometric practice and FTC's 
suggestion above. Seasons and temperature affect gasoline prices by 
changing demand and supply. See figures 22 and 23, which show the 
seasonal nature of the variable--ratio of gasoline inventories to 
demand--that we used to capture seasonality. Furthermore, we estimated 
that the price (margins) were higher in the summer driving months than 
in other months.

45. We do not agree with FTC's characterization of the merger results. 
First, the merger windows were determined by the nature of the mergers 
variable and the estimation. Furthermore, we believe that 6 months or 
more of weekly data before or after a merger are enough data for the 
analysis. More importantly, the idea of using the control variable for 
seasonality in our model (ratio of gasoline inventories to demand) is 
to be able to isolate the effects of the mergers. The effects of the 
mergers, as captured by our models, are not likely to be affected by 
whether the post merger period coincides with summer or winter months. 
Contrary to FTC's claims, as indicated in tables 15 and 16 of our draft 
and final reports, the months following the Marathon-Ashland merger 
were from January to June, and the months following the Shell-Texaco II 
(Motiva) merger were from July to December--these periods included only 
a few summer months.

46. We disagree with FTC. We stated in the draft report that the 
instruments were relevant to the endogenous variables. Furthermore, we 
performed tests of overidentifying restrictions of the instruments, 
which most of the previous studies FTC cited as examples of studies on 
price and market concentration did not perform. Although our test 
indicated that the restrictions for the overidentifying restrictions 
for the instruments were not ideal, we stated in the draft report why 
we believe the estimates are still sound and reasonable. Nonetheless, 
using the modified instruments in the final report, our tests indicated 
that all the instruments used in our report were appropriate and valid.

47. See comment 7(d).

48. In some of our model specifications in the draft report, we used 
weekly (seasonal) dummy variables in the market concentration model, 
but not seasonal (quarterly) dummy variables, which is likely what FTC 
is referring to.

49. We disagree. We stated in the draft report that the data for the 
suppliers were measured by the number of refiners/suppliers (posting 
prices) at the rack. We disagree with FTC that this is not an 
appropriate way to account for competition among suppliers because the 
number of suppliers is a key determinant of market concentration. While 
the number of suppliers may include traders who infrequently post 
prices at the rack, we think that what matters to the distributor 
(buyer) at the rack is the number of independently posted prices, which 
is reflected in the suppliers' data. We note that FTC did not indicate 
what would be an appropriate measurement of the number of suppliers at 
the racks. Nonetheless, in the final report, we use the following 
variables as instruments (excluded exogenous variables) for the ratio 
of inventories to demand and refinery capacity utilization rates--52 
weekly (seasonal) dummies, time trend, and time trend squared. Our 
tests indicated that the instruments were appropriate.

50. We disagree. As stated in comment 17, we indicated in the draft 
report that the regressions for the relationship between each of the 
endogenous variables and the instruments were highly significant, 
implying that the instruments are not weak. These estimates are not 
necessarily reported in the economic literature (see, for example, the 
studies by Evans, Froeb, and Werden (1993) and Staiger and Stock 
(1997)). On the other hand, in the draft report, we reported the 
estimates with and without the instruments, where necessary. As in 
indicated in the report, the instruments are used to purge the 
potential endogenous regressors of their correlations with the prices 
to obtain consistent estimates. And given that we do not have a fully 
specified system of simultaneous equations, we do not see the merit of 
reporting the "first-stage" estimates in this instrumental variables 
approach.

51. We agree with FTC that the endogeneity problem with inventories and 
utilization may be because these variables may be jointly determined 
with prices. We stated in the draft report that we also added market 
concentration to the list of potential endogenous variables. It is 
standard in endogeneity tests to assume that the instruments are ideal 
(or valid), as was done by Evans, Froeb, and Werden (1993).

Nonetheless, in the report, the HHI data based on refinery capacity are 
assumed to be exogenous, and the instruments used for inventories and 
utilization are appropriate based on the tests performed.

52. We cited in the draft report a study by Hahn and Hausman (2003), 
which also deals with the issue of instruments that are not ideal, an 
issue more relevant to our study. FTC fails to recognize that while the 
issues of weak instruments and instruments that do not meet the 
requirements for ideal instruments are related, our results suggest 
that the instruments are not weak, even if not ideal. Nonetheless, in 
the final report, all the instruments used are appropriate and valid.

53. We disagree with FTC that the best way to isolate the effects of 
the multiple mergers, which FTC reviewed prior to their occurring, is 
to concentrate on racks that did not experience the wave of mergers. 
First, there is no known definition of what series of mergers would 
qualify as having occurred in "rapid succession," and FTC did not 
provide one. Second, since it is generally better to use more rather 
than less information, we prefer our methodology, which uses all 
available useful information.

54. We do not agree with the approach suggested by FTC. FTC's approach 
would be feasible if the rack cities affected by the mergers were the 
same. Again, we think this could bias the results because the sample 
used might not be representative.

55. We do not agree with FTC's characterization of how we estimated the 
merger effects. Our methodology accounts for the previous mergers by 
including merger dummies for all the mergers that affected the racks. 
More importantly, as reflected in the draft report, for racks that had 
multiple mergers, the estimation of the effects of a second merger 
compares the prices in the premerger period (which coincides with when 
the first merger was in effect) to the prices in the postmerger period. 
(See, for example, the time modeled for the Marathon-Ashland merger, 
which occurred prior to the Shell-Texaco II (Motiva), and the Shell-
Texaco II (Motiva) merger in table 5 of the final report).

56. While we agree that the merger dummy variable is of crucial 
importance, we disagree with FTC's characterization of the sufficiency 
of the data that we used, especially for the postmerger period. First, 
the length of time over which we estimated the effects was not 
arbitrarily determined. As stated in the draft and final reports, some 
of the postmerger periods were relatively short because of the wave of 
the mergers that FTC reviewed (see tables 15-17). Second, while FTC 
does not provide any evidence why 6 months of data are not enough to 
determine the effects of the merger, we particularly note that 6 months 
(actually approximately 24 observations) provide a reasonable duration 
to estimate merger impacts in wholesale gasoline markets where price 
changes can be frequent.

57. We disagree. We could not extend the study beyond 2000 because of 
data limitations and the scope of the study. Furthermore, in an 
overlapping mergers framework, given the merger variable that we used, 
extending the data might benefit only the latest merger and not the 
prior mergers. We also disagree with FTC that the premerger period 
could be arbitrarily shortened. It is generally better to include all 
useful information rather than less information. As stated in the final 
report, we could not reasonably perform sensitivity tests given the 
wave of the mergers and the relatively short merger windows that were 
available. Again, in our meetings, FTC did not provide an example of a 
merger that could benefit from such sensitivity test or suggest what 
alternative merger dates could be used.

58. We disagree with FTC's assertion that our methodology is 
fundamentally flawed, because we used sound econometric methodology 
that is supported by the approaches used in previous studies. Also, the 
outline of our preliminary methodology was reviewed by experts who 
provided us comments that we incorporated, as appropriate. Our expert 
consultant/peer reviewer also reviewed and provided comments to our 
estimation and interpretation of results, which we incorporated, as 
appropriate. We disagree that "many" results are not robust to the 
different estimations used in the draft report as well as in the 
report--in fact, most of the estimates for the merger effects and for 
the market concentration effects were qualitatively and quantitatively 
similar. (See, for example, tables 15-17). We discussed in the draft 
report why a certain estimation technique was used and why a certain 
variable was excluded.

59. FTC incorrectly represents these results in the draft report. We 
provided separate results for branded and unbranded gasoline. For 
branded, the estimates were 1.01, 0.25, and 2.14 cents per gallon (only 
the last value was statistically significant), and for unbranded, the 
estimates were 2.03, 1.14, and 3.54 cents per gallon (only the second 
value was not statistically significant). Nonetheless, in the report 
the effects of the BP-Amoco merger are not statistically significant.

60. As already indicated, there is no specific economic rationale for 
including year dummies in the models, which have been dropped from our 
models in the final report. In the report, the effects of the Tosco-
Unocal are positive for branded gasoline, and the effects of the Shell-
Texaco I are negative. The effects of these mergers on the prices of 
unbranded CARB gasoline are not statistically significant.

61. We believe that FTC incorrectly assumed that in the draft report, 
only the use of the instrumental variable estimation changes between 
the results reported in tables 27-32 for each type of gasoline. We did 
not indicate that the results with the fixed-and random-effect 
regressions are not very different from the results using the 
instrumental variable techniques. The specifications for the fixed and 
random effects estimates included the variable for refinery capacity 
utilization rates, while the estimates with the instruments generally 
excluded this variable, particularly for our preferred models. 
Therefore, these two sets of results are not directly comparable. Also, 
we disagree with FTC's characterization of our results in tables 27-38 
of the draft report. The results for the individual mergers, which are 
based on data for only the merger affected racks, were in tables 27-38, 
and not only in tables 27-32 as claimed by FTC. Apart from the few 
estimates cited by FTC, the estimates were similar qualitatively for 
the estimates of the mergers' effects on branded and unbranded 
gasoline. Also in the draft report, we presented the results of our 
preferred model for the effects of the mergers, based on data for all 
the racks. Although the results for using data for only the rack cities 
affected by a specific merger and using data for all the rack cities 
were qualitatively similar, the results based on all the rack cities 
are preferred because of the importance of spatial competition in 
gasoline markets.

In the report, the estimates of the mergers and market concentration 
variables for the models are generally positive for the different 
specifications, including the models that used instrumental variables 
and those that did not.

62. We agree with FTC that the effect of high refinery capacity 
utilization rates is likely to be positive, as we stated in the draft 
and final reports (see table 13). However, we disagree with FTC's claim 
that the inconsistency in the effects of this variable is due to the 
instruments used; rather, we discussed in the draft report that the 
inconsistency is more likely due to its correlation with the ratio of 
gasoline inventories to demand variable. In fact, as we indicated in 
the draft report, the adjusted R-squares increased marginally when the 
refinery capacity utilization rates variable was excluded, which is one 
of the reasons we excluded that variable in our preferred models. 
Nonetheless, in the final report, we did not exclude the utilization 
variable in any of our models, and all the instruments used were valid.

63. We do not agree with FTC's characterization that the results are 
not robust across different racks. In the draft report, as we indicated 
in the titles for tables 37 and 38, the results are primarily for the 
effects of the Tosco-Unocal and Shell-Texaco I (Equilon), respectively. 
It was therefore inappropriate to infer the complete results of the 
effects of the Tosco-Unocal merger from table 38 in the draft report. 
Furthermore, the results in tables 25 and 26 of the draft report, based 
on all the racks, were generally consistent with the estimates in table 
37 of the draft report for the Tosco-Unocal merger, which were based on 
only the merger affected cities. These results were also consistent 
with the findings from previous studies. (See Hastings and Gilbert, 
2002).

64. We disagree. See comments 19, 21, 35 and 36.

65.As we indicated in comment 63, like FTC, we believe that the 
utilization variable would have a positive effect. Nonetheless, the 
utilization variable is not excluded from any of our models in the 
final report.

66. We disagree with FTC that our models exclude important relevant 
variables, although since "all" in this context might be overstated, we 
have dropped it from the final report. Our models, as specified and 
estimated, are not under-specified and do not exclude key variables, as 
we explained in detail in comment 7. Low R-squares are not unusual in 
studies of price-cost margins like ours. We have fully explained the 
reasons for the low R-squares and have also explained how higher R-
squares can be produced for our models but at the cost of creating 
statistical problems. See comment 7.

67. We disagree with FTC's characterizations of price-concentration 
studies. Contrary to what FTC stated, the studies that both we and FTC 
cited are price-concentration and not merger event studies. See for 
example, the studies by Evans, Froeb, and Werden (1993) and Kim and 
Singal (1993). It is our understanding that such studies continue to 
underlie FTC's 1992 Horizontal Merger Guidelines, which imply a causal 
link between concentration and market power. In addition, alternative 
methodological approaches to the same question can provide a type of 
robustness analysis.

68. We disagree. In the draft report, we acknowledged and discussed the 
methodological issues associated with price-concentration studies, 
including a citation of the study by Evans, Froeb, and Werden (1993). 
In addition, we stated, "Generally, there are potential problems in 
estimating a relationship between prices and market concentration.": 

69. We disagree. Our results are sound and reasonably robust. 
Furthermore, in the draft report, we explained why we used certain 
specifications and why some results differed. We disagree with FTC's 
assertion that we did not compare or reconcile the results from the 
price-concentration study with those of the merger event study. First, 
throughout the draft and final reports, we stated that the market 
concentration effects would capture the effects of the mergers as well 
as other competitive conditions. Second, while we did not directly 
compare the two results because of possible intervening factors that we 
could not measure--such as entry and exit into the market--our results 
for the mergers and market concentration are broadly consistent. We 
found that most of the mergers were associated with price increases and 
that increased market concentration was generally associated with price 
increases.

70. While we are aware of the potential problems with price-
concentration studies, and therefore reported the analysis as 
supportive of the results of the merger-event studies, we disagree with 
FTC's characterization that we did not discuss these issues. In the 
draft report we recognized and discussed the limitations of modeling 
price-concentration. See comments 68 and 69.

71. As we have explained earlier, we have directly or indirectly 
accounted for the effects of seasonality, supply outages and 
formulation changes in our models. See comment 7(d).

72. In the draft report, we cited the study by Evans, Froeb, and Werden 
(1993), which is the most recent study. This study cites and even 
quotes statements from the study by Bresnahan regarding price-
concentration studies. We also cited FTC's Empirical Industrial 
Organization Roundtable (2001b), which discussed some of these issues. 
Nonetheless, these problems do not negate the use of market 
concentration in price studies.

73. We agree that it is generally difficult to identify the "true" 
markets for wholesale gasoline and that some wholesale markets could be 
larger or smaller than a state. However, we believe that using the 
state level HHI in the draft report was reasonable. During our 
meetings, FTC, despite its claimed expertise, did not suggest any 
feasible alternatives for determining where it makes sense to use state 
level data and where it does not. We noted in the draft report the 
limitations of using HHI at the state level, which was the only market 
concentration data on wholesale gasoline markets available to us and 
which we obtained only by working closely with EIA. FTC has not 
provided any reason or evidence for why our results would be biased. 
More importantly, the merger effects are generally consistent with the 
market concentration effects. Nonetheless, in the final report, we use 
refinery capacity data rather than state level data for prime suppliers 
because we believe this is a better indicator of market power in 
wholesale gasoline markets. See also comment 3.

74. We recognized the limitations of using the HHI in our models. See 
comments 68 and 69.

75. We disagree. In the draft report, our choice of instruments was 
fully evaluated, as discussed in comment 17. We appropriately 
recognized and cited the study by Evans, Froeb, and Werden (1993), 
which provides a detailed account of the issues that FTC outlines here. 
Nonetheless, in the final report, all the instruments are appropriate, 
based on the tests performed.

76. While changes in HHI may reflect other factors in addition to 
mergers--such as entry, exit, and relative price changes--we believe, 
and FTC has not provided evidence to the contrary, that the numerous 
mergers that occurred in this industry during the period that we 
modeled significantly increased market concentration. We disagree that 
mergers will cause "few" of the changes in the HHIs. We show in chapter 
3 of the final report that the changes in HHI are associated with 
mergers in many geographic regions and market segments. Furthermore, 
the HHI is the core data of FTC/DOJ horizontal merger guidelines, and 
we are surprised at the criticism that FTC has focused on the use of 
this variable in our analysis.

77. We disagree. See comment 70.

78. We stated in the draft report that our results are generally 
consistent with other specifications, including squared HHI. 
Furthermore, it is common to use linear functions in price-
concentration studies; see, for example, Evans, Froeb, and Werden 
(1993). In the report, using the preferred HHI measure based on 
refinery capacity, the preferred specification for our models was 
linear and consistent with previous studies.

79. We believe it is reasonable to assume that racks that are nearest 
to each other would tend to have similar prices that move together due 
to spatial-price competition. As we indicated in 17, we used the 
instruments to deal with the potential endogeneity problems, albeit 
imperfectly, and conducted tests that we report. In the final report we 
deal with the issue of nearby racks in the estimation technique--we 
accounted for contemporaneous cross-city correlations.

80. We believe the criticisms are unwarranted in this instance as 
discussed in comments 7 and 17.

81. In the draft report, we used other functional forms and found 
similar results. Consequently, we disagree with FTC. See comment 79. As 
indicated in comment 79, the linear specification is consistent with 
previous studies.

82. In the draft report, in exploring why the year dummies were 
significant, we observed a significant shift in the HHI in a certain 
year, which seems to be related to the mergers. We therefore used an 
interaction term to determine if the effects of the HHI before and 
after the shift in HHI were different. The results were generally 
consistent with the results presented in the draft report. A similar 
shift was found with our preferred HHI measure, but the findings were 
generally unchanged and are not reported.

83. As already indicated in 17, we preferred the instrumental variables 
estimates that include the utilization rates on econometric grounds. 
Also, we disagree with FTC's characterization that the price 
concentration results are not robust.

84. We disagree. We believe that the estimated effects are reasonable 
in terms of the average levels of the prices as well as the estimated 
effects found in previous studies of gasoline markets. For instance, 
Hendricks and McAfee (2000) simulated that the Exxon-Mobil merger would 
have resulted in price increases for CARB gasoline. These estimates are 
generally consistent with our findings.

85. We disagree. As indicated in the draft report, we provided reasons 
why we believe that the specifications excluding national refinery 
capacity utilization rates were better. Also, as indicated in comment 
17, the results with the instruments are statistically preferred. We 
note the inconsistency in FTC's criticisms--FTC had indicated that some 
of the variables could be endogenous (see, for example, comments 75 and 
76), which implies that instruments should be used, but they now seem 
to prefer the estimates without the instruments. Nonetheless, in the 
final report, we have provided results that include both utilization 
and inventory ratio and used instruments for these variables, where 
appropriate.

86. We disagree. Generally, the results for the merger effects and 
market concentration cannot be systematically compared. Specifically, 
FTC's assertion that the merger effects should be less than the effects 
from changes in concentration is not necessarily correct, since the 
other factors affecting concentration could have both positive and 
negative directional effects. However, there is evidence that these 
results are generally consistent because market concentration is 
closely related to the mergers, as shown in chapter 3, and the increase 
in concentration implies an increase in the price cost margins, other 
factors held constant.

87. We disagree with FTC's assessment of the relationships between the 
effects of mergers and market concentration. FTC erroneously implies 
that the market concentration effects are a simple summation of the 
mergers effects. They are not. This is because, as FTC noted, there are 
intervening factors, such as entry and exit, between the merger and 
market concentration effects. However, because market concentration is 
closely related to the mergers, we found the two effects to be 
generally consistent (see comment 70).

88. We disagree with FTC's characterization of the variable that we 
used. In the draft report, we used weekly (seasonal) dummies, not 
quarterly seasonal dummies as FTC claims. The variable was tried to see 
if it would make a difference in our results, even though we already 
had the ratio of gasoline inventories to demand variable, which was 
available weekly. Given that it did not make a significant difference 
to our results, we did not include it in our mergers models. In the 
final report, the weekly (seasonal) dummies are used as instruments. 
See comment 17.

89. FTC's questioning of any relationship between mergers and market 
concentration is puzzling, given that the 1992 Horizontal Merger 
Guidelines, jointly issued by FTC and DOJ, and FTC's merger review 
process prominently highlight the link between mergers and market 
concentration. As we noted in the draft report, the correlation 
coefficient is a statistical measure of the strength of association or 
relationship between two variables. While we believe that there is a 
logical and foundational link between mergers and market concentration 
based on economic theory, we did not state in the draft report that our 
correlation analysis establishes causation. Also, because of a lack of 
detailed data on mergers by segment or geographic area, we used 
correlation analysis to determine, at broad levels, the association 
between overall merger activity and market concentration for the 
various petroleum market segments. While correlation does not infer 
causation, it is an acceptable statistical method to determine the 
direction and extent of relationship between two variables. Because 
many large mergers during this period involved firms that were highly 
vertically integrated, we believe that the correlation between 
concentration and overall merger activity reflects market realities. 
Overall, while we did not infer causality from this analysis, the 
results indicate that mergers and market concentration are broadly 
related in the segments that we analyzed.

90. It is not clear what FTC is saying here about documentation in a 
"well performed study." If FTC is taking issue with our documentation, 
our draft report fully provided and discussed characteristics of the 
data used, including data sources, construction of the data, frequency, 
gasoline formulations and brands, and time periods. See, for example, 
tables 13 and 14 of the final report (which were also in the draft 
report). For the regressions, we provided in the draft and final 
reports the basic model specifications, the specific variables used in 
each equation, as well as the estimation techniques. We do not believe 
it is worth listing the almost 300 racks that were used in the analysis 
because FTC should be able to identify all racks that the mergers they 
reviewed affected. We believe that with the data in hand and our 
descriptions, a researcher should be able to replicate the results.

91. We disagree with FTC that our analysis does not indicate how we 
handled FTC's divestitures. In the draft and final reports we stated 
that our study did not assess the appropriateness of FTC's review and 
actions they took regarding the mergers, including divestiture 
requirements. All the estimates of the mergers' effects--including the 
Exxon-Mobil, BP-Amoco and Shell-Texaco I (Equilon) mergers--are 
conditioned on any divestitures required by FTC. We note that 
apparently to account for the effect of divestitures, FTC suggested 
that the effective date of the Exxon-Mobil merger be changed from 11/
30/99, which was the merger completion date, to 3/1/00. Accordingly, in 
the draft report where the market concentration data were based on 
prime suppliers' sales, we used the EIA's revised HHI calculations to 
reflect the change in the merger effective date. Our analysis was based 
on wholesale markets (racks) where the mergers overlapped based on the 
OPIS data--there were no data for the Exxon-Mobil merger for the racks 
in California.

92. This procedure would not appropriately measure the size of the 
merger, particularly in the case of refining, because as correctly 
indicated by FTC, the Herold data are not separated out by geographic 
area. For crude oil, there were many mergers for which there are no 
production data to construct market concentration.

93. We do not believe that this is an appropriate way to determine the 
links between mergers and market concentration. For instance, the data 
would not appropriately measure the size of the merger in each 
geographic market, nor would it be useful in capturing the effects of 
the intervening factors, such as entry and exit, on market 
concentration over time. Nonetheless, in the final report, we used data 
on refinery capacity to measure market concentration--these data are 
available only annually.

94. We disagree with FTC's characterization of our econometric 
methodology and results. While each merger dummy variable is turned on 
throughout the postmerger period in racks affected by the merger, the 
estimation procedure generates a parameter estimate for the postmerger 
period up to the onset of the next merger that affected the same racks 
as the previous merger. See the postmerger periods identified for each 
merger in tables 15-17.

95. We disagree with FTC's assertions about the price data we used. We 
stated in the draft report and final reports that the prices are the 
average prices at the racks. We also indicated that the three 
formulations used are conventional, reformulated, and CARB gasoline, 
and the data were available for regular gasoline, branded and 
unbranded, as well as the relevant time periods. We state in the report 
that based on the available data, the product type used for 
conventional gasoline is clear and the type for reformulated and CARB 
gasoline is MTBE. In the draft report, we did not state that we used 
state-level rack prices in the price-concentration analysis. We used 
the average prices at the rack city level.

96. The underlying modeling issues raised by FTC in its discussion of 
omitted variables were addressed. Our models address this specific 
conceptual concern about omitted variables and adequately account for 
the key variables that affect wholesale gasoline prices. See comment 7 
for more details.

97. We disagree. Our models adequately control for exogenous factors 
that impact wholesale gasoline prices using variables that directly, 
instead of indirectly, address behavioral issues. See comment 7 for 
more detail.

98. We disagree. Our models treat the issue of endogeneity extensively 
and are consistent with previous studies, including the study by Evans, 
Froeb, and Werden (1993). See comment 17 for more detail.

99. We disagree. Our models are generally robust, and we carefully 
explain any differences for the specifications. See comments 58-64 and 
67 for more detail.

100. We disagree. We clearly specified the pre-and postevent periods 
for the merger event studies, estimated the mergers' effects and 
provided the appropriate interpretation. FTC's approach of seemingly 
"matching" rack cities to nonrack cities would not be appropriate for 
this study. See comments 18 and 19 for more detail.

101. We disagree. We carefully discussed the problems with market 
concentration in the price-concentration studies, including the data 
used for market concentration. We believe the results are sound and 
reasonable and are consistent with the results of the mergers' effects. 
See comment 74 for more detail.

102. We disagree. We have provided sufficient documentation of our 
methodology that, we believe, experts in the gasoline markets, 
including FTC, would find useful in undertaking similar studies. Given 
the potential significance of our findings for public policy, we 
believe that FTC, the agency that reviews mergers in the gasoline 
markets, should undertake an independent and public study of the 
effects of the wave of mergers that it has reviewed in the second half 
of the 1990s.

103. We disagree. Our results are based on sound econometric 
methodology--they are consistent with previous studies, and external 
peer review experts who reviewed various stages of the report generally 
approve of it. Our models have reasonably isolated the effects of the 
mergers, as well as the effects of markets concentration, which 
captures the cumulative effects of the mergers and other competitive 
factors. The effects of the mergers and market concentration on 
wholesale gasoline prices are generally consistent. The debate is about 
differences in approach and data, and does not involve fundamental 
flaws, and therefore can be useful for public policy. Other approaches 
may be informative, and we encourage independent analysis of this 
important policy issue by FTC or other other parties.

[End of section]

Appendix VII: GAO Contacts and Staff Acknowledgments: 

GAO Contacts: 

Jim Wells (202) 512-3841 Godwin M. Agbara (202) 512-3841: 

Acknowledgments: 

In addition to those named above, Albert Abuliak, Hashem Dezhbakhsh, 
Barbara El-Osta, Scott Farrow, John A. Karikari, Cynthia Norris, 
Barbara Timmerman, and Lynn Wasielewski made key contributions to this 
report.

[End of section]

Bibliography: 

Angrist, J. and A. Kruger, "Empirical Strategies in Labor Economics," 
in Handbook of Labor Economics, edited by O. Ashenfelter and D. Card, 
vol. 3A, (New York: Elsevier, 1999): 1277-1366.

Arellano, M. and S. Bond, "Some Tests of Specification for Panel Data: 
Monte Carlo Evidence and An Application to Employment Equations," The 
Review of Economic Studies," vol. 58, no. 2 (1991): 277-297.

Azzeddine, A. and D. Rosenbaum, "Differential Efficiency, Market 
Structure and Price," Applied Economics, vol. 33 (2001): 1351-1357.

Baffes, J., "Explaining Stationary Variables with Non-Stationary 
Regressors," Applied Economic Letters, vol. 4, no. 1 (1997): 69-75.

Barton, D. and R. Sherman, "The Price and Profit Effects of Horizontal 
Merger: A Case Study," The Journal of Industrial Economics, vol. 33, 
no. 2 (1984): 165-177.

Bertrand, M., E. Duflo, and S. Mullainathan, "How Much Should We Trust 
Difference-in-Difference Estimates?" unpublished paper, June 2003.

Bhargava, A., L. Franzini, and W. Narendranathan, "Serial Correlation 
and the Fixed Effects Model," The Review of Economic Studies, vol. 49, 
no. 4 (1982): 533-549.

Borenstein S., A. Cameron, and R. Gilbert, "Do Gasoline Prices Respond 
Asymmetrically to Crude Oil Price Changes?" The Quarterly Journal of 
Economics, vol. 112, no. 1 (1997): 305-339.

Borenstein, S. and A. Shepard, "Sticky Prices, Inventories, and Market 
Power in Wholesale Gasoline Markets," National Bureau of Economic 
Research, Inc., Working Paper 5468, 1996a.

Borenstein, S. and A. Shepard, "Dynamic Pricing in Retail Gasoline 
Markets," RAND Journal of Economics, vol. 27, no. 3 (1996b): 429-451.

Chouinard H., and J. Perloff, "Gasoline Price Differences: Taxes, 
Pollution Regulations, Mergers, Market Power, and Market Conditions," 
unpublished paper, 2001.

CCollins, N. and L. Preston, "Price-Cost Margins and Industry 
Structure," Review of Economics and Statistics, vol. 51, no. 3 (1969): 
271-286.

Considine, T., "Inventories and Market Power in the World Crude Oil 
Market," unpublished paper, 2002.

Creswell, Jr., J., S. Harvey, and L. Silvia, "Mergers in the U.S. 
Petroleum Industry 1971-1984: An Updated Comparative Analysis," FTC 
Economic Report (May 1989).

Davidson, R. and J. MacKinnon, Estimation and Inference in Econometrics 
(New York: Oxford University Press, 1993).

Dickey, D., and W. Fuller, "Distribution of the Estimators for 
Autoregressive Time Series With a Unit Root," Journal of the American 
Statistical Association, vol. 74 (1979): 427-431. Evans, W., L. Froeb, 
and G. Werden, "Endogeneity in the Concentration Price Relationship: 
Causes, Consequences, and Cures," The Journal of Industrial Economics, 
vol. XLI (1993): 431-438.

Federal Trade Commission, Midwest Gasoline Price Investigation: Final 
Report of the Federal Trade Commission (Washington, D.C., March 29, 
2001a).

Federal Trade Commission, Empirical Industrial Organization Roundtable 
(Washington, D.C., September 11, 2001b).

Focarelli, D. and F. Panetta, "Are Mergers Beneficial to Consumers? 
Evidence from the Market for Bank Deposits, American Economic Review 
(September 2003): 1152-1172.

Geweke, John, "Empirical Evidence on the Competitive Effects of Mergers 
in the Gasoline Industry," unpublished paper, July 16, 2003.

Greene, W., Econometric Analysis, Fourth Edition (Upper Saddle River, 
New Jersey: Prentice Hall, 2000).

Hahn, J. and J. Hausman, "Weak Instruments: Diagnosis and Cures in 
Empirical Econometrics," American Economic Review (May 2003): 118-125.

Hansen, L. "Large Sample Properties of Generalized Method of Moments 
Estimators," Econometrica, vol. 50, no. 4 (1982): 1029-1054.

Hastings, J. and R. Gilbert, "Vertical Integration in Gasoline Supply: 
An Empirical Test of Raising Rivals' Costs," Program on Workable Energy 
Regulation (POWER), PWP-084, 2002.

Hastings, J. "Vertical Relationships and Competition in Retail Gasoline 
Markets: Empirical Evidence from Contract Changes in Southern 
California," Program on Workable Energy Regulation (POWER), PWP-075, 
2002.

Hausman, J. "Specification Tests in Econometrics," Econometrica, vol. 
46, no. 6 (1978): 1251-1271.

Hausman, J. and W. Taylor, "Panel Data and Unobservable Individual 
Effects," Econometrica, vol. 49, no. 6 (1981): 1377-1398.

Hendricks, K. and R. Preston McAfee, "A Theory of Bilateral Oligopoly, 
With Applications to Vertical Mergers," unpublished paper, 2000.

Hsiao, C., Analysis of Panel Data, Second Edition (New York: Cambridge 
University Press, 2003).

Im, K., M. Pesaran, and Y. Shin, "Testing for Unit Roots in 
Heterogeneous Panels," Journal of Econometrics, vol. 115, no. 1 (2003): 
53-74.

Karikari, J., S. Brown, and M. Nadji, "The Union Pacific/Southern 
Pacific Railroads Merger: Effect of Trackage Rights on Rates," Journal 
of Regulatory Economics, vol. 22, no. 3 (2002): 271-285.

Kennedy, P., A Guide to Econometrics, Fourth Edition (Cambridge, MA: 
The MIT Press, 1998).

Kim, E. and V. Singal, "Mergers and Market Power: Evidence from the 
Airline Industry," The American Economic Review, vol. 83, no. 3 (1993): 
549-569.

Manuszak, M., "The Impact of Upstream Mergers on Retail Gasoline 
Markets," unpublished paper, 2001.

Muris, T., "Improving the Economic Foundations of Competition Policy," 
Remarks at George Mason University Law Review's Winter Antitrust 
Symposium, Washington, D.C., January 15, 2003.

Newey, W. and K. West, "A Simple, Positive Semi-definite, 
Heteroskedasticity and Autocorrelation Consistent Covariance Matrix," 
Econometrica, vol. 55 (1987): 703-708.

Pinkse, J., M. Slade, and C. Brett, "Spatial Price Competition: A 
Semiparametric Approach," Econometrica, vol. 70, no. 3 (2002): 1111-
1153.

Royer, J., "Market Structure, Vertical Integration, and Contract 
Coordination," in The Industrialization of Agriculture: Vertical 
Coordination in the U.S. Food System, edited by J. Royer and R. Rogers 
(Aldershot, England: Ashgate, 1998).

Sargan, J., " The Estimation of Economic Relationships Using 
Instrumental Variables," Econometrica, vol. 26 (1958): 393-415.

Scheffman, D. and M. Coleman, "Current Economic Issues at the FTC," 
Review of Industrial Organization, vol. 21 (2002): 357-371.

Scheffman, D. and M. Coleman, "FTC Perspectives on the Use of 
Econometric Analyses in Antitrust Cases," unpublished paper, undated,  
[Hyperlink, http://www.ftc.gov/be/ftcperspectivesoneconometrics.pdf] 
(September 5, 2003).

Scheffman, D. and P. Spiller, "Geographic Market Definition Under the 
U.S. Department of Justice Merger Guidelines," Journal of Law and 
Economics, vol. 30, no. 1 (1987): 123-147.

Scherer, F., Industrial Market Structure and Economic Performance, 
Second Edition (Chicago: Rand McNally, 1980).

Schumann, L., J. Reitzes, and R. Rogers, "In the Matter of Weyerhaeuser 
Company: The Use of a Hold-Separate Order in a Merger with Horizontal 
and Vertical Effects," Journal of Regulatory Economics, vol. 11, no. 3 
(1997): 271-289.

Spiller, P. and C. Huang, "On the Extent of the Market: Wholesale 
Gasoline in the Northeastern United States," The Journal of Industrial 
Economics, vol. 35, no. 2 (1986): 13-145.

Staiger, D. and J. Stock, "Instrumental Variables Regression with Weak 
Instruments," Econometrica, vol. 65, no. 3 (1997): 557-586.

Stock, J., and M. Watson, Introduction to Econometrics (Boston, MA: 
Addision-Wesley, 2003).

Taylor, C. and J. Fischer, "A Review of West Coast Gasoline Pricing and 
the Impact of Regulations," unpublished paper, 2002.

Taylor, C. and D. Hosken, "The Economic Effects of the Marathon-Ashland 
Joint Venture: The Importance of Industry Supply Shocks and Vertical 
Market Structure," FTC Bureau of Economics Working Paper 270, March 
2004.

U.S. Department of Energy, Energy Information Administration, 
"Petroleum Product Prices and Supply Disruptions," in Electricity 
Shortage in California: Issues for Petroleum and Natural Gas Supply, 
June 2001.

U.S. Department of Justice and Federal Trade Commission 1992 Horizontal 
Merger Guidelines, 
[Hyperlink, http://www.ftc.gov/bc/docs/horizmer.htm]

U.S. General Accounting Office, Energy Prices: Gasoline Price Increases 
in Early 1985 Interrupted Previous Trend. [Hyperlink, http://
www.gao.gov/cgi-bin/getrpt?GAO/RCED-86-165BR], 
Washington, D.C., September 25, 1986.

U.S. General Accounting Office, Energy Security and Policy: Analysis of 
the Pricing of Crude Oil and Petroleum Products. [Hyperlink, http://
www.gao.gov/cgi-bin/getrpt?GAO/RCED-93-17], Washington, 
D.C., March 19, 1993.

U.S. Senate, Gas Prices: How Are They Really Set? Report prepared by 
the Majority Staff of the Permanent Subcommittee on Investigations, 
released in conjunction with the Permanent Subcommittee on 
Investigations' Hearings on April 30, 2002 and May 2, 2002.

Vita, M., "Regulatory Restrictions on Vertical Integration and Control: 
The Competitive Impact of Gasoline Divorcement Policies," Journal of 
Regulatory Economics, vol. 18, no. 3 (2000): 217-233.

Vita, M., and S. Sacher, "The Competitive Effects of Not-for-Profit 
Hospital Mergers: A Case Study," The Journal of Industrial Economics, 
vol. XLIX (2001): 63-84.

White, H., Asymptotic Theory for Econometricians (San Diego, CA: 
Academic Press, 2001).

Wooldridge, J., Econometric Analysis of Cross Section and Panel Data 
(Cambridge, MA: MIT Press, 2002).

Wu, D. "Alternative Tests of Independence Between Stochastic Regressors 
and Disturbances," Econometrica, vol. 41, no. 4 (1973): 733-750.

(360103): 

FOOTNOTES

[1] Dr. Borenstein is E.T. Grether Professor of Business Administration 
and Public Policy at the Haas School of Business, University of 
California, Berkeley. He is also the Director of the University of 
California Energy Institute.

[2] There does not appear to be consensus on what to call the different 
classes of companies within the industry. In this study, we adopt the 
term "fully vertically integrated" to refer to companies that operate 
in all segments. The partially vertically integrated or independents 
include independent producers that operate only in the upstream. For 
the downstream, we use "independent refiners" to refer to companies 
that operate only in refining and marketing, "jobbers" to refer to 
those that buy gasoline at wholesale and resell it at wholesale and/or 
retail, and "retailers" to refer to those that operate only at the 
retail level. While some companies, especially the fully vertically 
integrated companies, own and operate transportation facilities 
(midstream), we use the term "pipeline companies," when applicable, to 
refer to companies that only provide pipeline services. 

[3] Although natural gas is an important product of upstream activities 
and crude oil and natural gas are often jointly produced, our study 
focuses on petroleum, and we will not discuss impact of mergers on the 
natural gas market. 

[4] In general, product yields from a barrel of crude oil depend on the 
quality of the crude input and/or the configuration of the refinery. In 
general, light and"sweet" (i.e., high gravity and low sulfur) crudes, 
such as the West Texas Intermediate (WTI), yield a greater proportion 
of products such as gasoline, distillate, and jet fuel. Also, more 
sophisticated refineries generally yield higher gasoline and other 
lighter products. Many U.S. refineries are sophisticated because U.S. 
refiners have made substantial investments to upgrade their refineries 
to allow them to maximize the yield of gasoline and other light 
products. 

[5] Since January 1, 2002, two of the refining firms--Phillips and 
Conoco--have merged and another, Equilon, has become part of Shell. 
Also, four others are joint ventures between two existing refiners that 
are already counted as separate companies. 

[6] The latter three are no longer members. 

[7] 15 U.S.C. 18a, as amended.

[8] Fifteen days for cash tender offers and bankruptcy filings.

[9] If the review indicates a need for further investigations, a second 
request may be issued to the merging parties for an additional waiting 
period of 30 days (20 days prior to February 1, 2001, and 10 days for 
cash tender offers).

[10] Federal Trade Commission and Department of Justice, 1992 
Horizontal Merger Guidelines (with April 8, 1997, Revisions to Section 
4 on Efficiencies), http://www.ftc.gov/bc/docs/horizmer.htm. The 
guidelines were originally developed by DOJ in 1968 and updated in 1982 
and 1984 prior to joint FTC and DOJ issuance in 1992. 

[11] Federal Trade Commission: Study Needed to Assess the Effects of 
Recent Divestures on Competition in Retail Markets (GAO-02-793, 
September 25, 2002).

[12] To calculate the HHI, FTC must define the relevant product market 
and geographic market likely to be affected by the proposed merger. HHI 
is equal to the sum of the squares of the market shares of each firm in 
the market. Thus, a market consisting of four firms, each with a 25 
percent share, would have an HHI of 2,500. The measure ranges between 0 
and 10,000.

[13] Because of concerns about confidentiality of individual company 
data at the wholesale level, EIA could not provide us with the market 
share data for individual wholesale gasoline suppliers. Instead, the 
agency calculated the HHIs and concentration ratios for us. 

[14] John S. Herold, Inc., data were available starting in 1991. A 
company official told us that to the best of his knowledge, the data 
includes all known merger transactions during this time period. 

[15] Midstream assets can include trucks, tankers, and pipelines, but 
for the purpose of this report, it only includes pipelines, which 
constitute the bulk of petroleum transportation.

[16] The mergers depicted in figure 8 involved firms in which one or 
both belonged to EIA's Financial Reporting System (FRS) companies at 
the time of the merger or became an FRS company after the merger 
occurred. FRS companies are U.S.-based major energy producers that 
report financial statistics to the EIA used by the agency to prepare 
its annual Performance Profiles of Major Energy Producers. According to 
EIA, as of 2002, criteria for selecting FRS companies include a company 
that accounts for (1) at least 1 percent of U.S. crude oil or natural 
gas liquids reserves or production, (2) at least 1 percent of U.S. 
natural gas reserves or production, or (3) at least 1 percent of U.S. 
crude oil distillation capacity.

[17] Both types of mergers could have implications for the industry's 
market structure because they both could affect horizontal market 
concentration and/or vertical integration. Chapter 3 of this report 
examines mergers and market structure in more detail. 

[18] As indicated in chapter 1, the FTC may require one or both merging 
firms to divest some assets to a third party as an anticompetitive 
remedy.

[19] According to John. S. Herold officials, mergers among the large 
fully integrated oil companies, such as the Exxon-Mobil and the BP-
Amoco mergers, were typically counted only in the upstream segment, 
although these mergers also involved downstream assets.

[20] The desire to enhance stock values, like maximizing profits, is an 
ultimate goal of companies. However, we discuss it since many industry 
officials cited it as a reason for mergers. 

[21] According to the 1992 Horizontal Merger Guidelines, market power 
is defined as the seller's ability to profitably maintain prices above 
competitive levels for a significant period of time.

[22] This contrasts with a 1989 study by the FTC that examined mergers 
in the U.S. petroleum industry from 1971 to 1984 and concluded that 
mergers had little impact on industry concentration. 

[23] We did not determine midstream pipeline market concentration 
because of data availability issues, complications in pipeline 
ownership, and difficulties in defining the relevant geographic market. 
However, FTC officials told us that they are working on concentration 
issues for oil pipelines. 

[24] FTC and DOJ have defined market power for a seller as the ability 
profitably to maintain prices above competitive levels for a 
significant period of time.

[25] HHI is equal to the sum of the squares of the market shares of 
each firm in the market. Thus, a market consisting of four firms, each 
with a 25 percent share of the market, would have an HHI of 2,500. The 
measure ranges between 0 and 10,000.

[26] FTC and DOJ, 1992 Horizontal Merger Guidelines.

[27] We used the HHI because it is the most comprehensive measure of 
market concentration available. Other measures of market concentration 
include the share of the market controlled by the four or eight largest 
firms (known as four-firm or eight-firm concentration ratios, CR4 or 
CR8, respectively).

[28] The Department of Energy (DOE) has divided the United States into 
five regions known as Petroleum Administration for Defense Districts 
(PADD). See figure 11 for PADDs and states in each PADD.

[29] We could not analyze concentration at the retail level because 
there are no comprehensive data at this level.

[30] Correlation coefficients, which range from -1 to +1, are commonly 
converted into and discussed in terms of percentages.

[31] John S. Herold, Inc., tracks the transaction values of mergers at 
the time of the offer and bases this value on the seller's assets or 
the offer from the buyer. We calculated an average value of 
transactions by dividing the reported total value of yearly 
transactions by the number of mergers for that year. We adjusted the 
total yearly value of transactions for inflation using the Producer 
Price Index for Energy from the 2002 Economic Report of the President. 
While the total transaction value of the mergers reflects both the 
number and the size of the mergers, the average transaction value 
primarily captures the size of the mergers. In our correlations, we 
also used the total transaction value, and the results were similar.

[32] We had some data limitations in performing these correlation 
analyses. First, transaction values were not reported for all mergers 
in our merger database. Merger transaction values were reported for 
about 57 percent of the mergers overall. More importantly, the 
transaction values were reported for all of the major mergers--i.e., 
mergers with transaction values exceeding $1 billion. Second, merger 
transaction values were not separated by segment to allow for the 
correlation of merger transaction values for each segment or level with 
the corresponding segment's/level's HHI. Nonetheless, we believe that 
our correlation analyses provide a broad indication of the potential 
statistical association between mergers and market concentration. We 
believe that the use of merger transaction values for the overall 
industry to estimate the statistical correlation with concentration at 
the segment or other operating level is reasonable because many of the 
mergers for which transactions were reported involved vertically 
integrated oil companies whose mergers could potentially affect 
concentration throughout the industry spectrum.

[33] While we are aware that other factors--such as entry and exit--may 
affect concentration, we focus our examination on the linkage between 
merger activity, as measured by the average yearly transaction values 
of mergers, and market concentration.

[34] However, if the same PADD I refiners are also mostly responsible 
for importing gasoline into the PADD, it could have implications for 
the PADD's wholesale gasoline market concentration. In addition, the 
extent to which these companies control vital infrastructure, such as 
terminals and pipelines, within the region could impact competitive 
conditions.

[35] Some industry officials and experts believe that the California 
refining market, which is a part of PADD V, is more concentrated than 
the PADD as a whole because a unique (CARB) gasoline is consumed in the 
state and the production of the gasoline is dominated by a few large 
refiners. 

[36] Many analysts believe that the relevant market for wholesale 
gasoline may be defined at the state or possibly the terminal level. 
Here, we are using states as the definition of the market, even though 
in some cases this definition may be too large or too small, depending 
upon the particular geographic market.

[37] The state is the smallest geographic level for which EIA computes 
HHI for wholesale gasoline markets. We performed our analysis at the 
state level but grouped the states according to their respective PADDs. 
EIA computed the HHIs for wholesale gasoline for us using data 
submitted by wholesale gasoline suppliers that the agency calls "prime 
suppliers." The agency performed the calculation, rather than give us 
the data to do the calculation, to protect the confidentiality of 
individual companies.

[38] Our correlations for wholesale gasoline supply were between the 
lag of the average yearly transaction values of mergers and market 
concentration, as measured by the state monthly HHIs. This was 
especially necessary because the HHIs were monthly while the 
transaction values of mergers were measured on an annual basis. (See 
appendix III for details.)

[39] Justine Hastings and Richard Gilbert, "Vertical Integration in 
Gasoline Supply: An Empirical Test of Raising Rivals' Costs," Working 
Paper Series of the Program on Workable Energy Regulation (POWER), 
University of California Energy Institute, Berkeley, California, July, 
2001. See also, Zava Aydemir and Stefan Buehler, "Estimating Vertical 
Foreclosure in U.S. Gasoline Supply," Working Paper No. 0212, 
Socioeconomic Institute, University of Zurich, November 2002. This 
study specifically found evidence of both market foreclosure and 
efficiency effects in vertical integration in U.S. refining, but the 
foreclosure effect dominated the efficiency effect and led to increased 
wholesale gasoline prices. 

[40] U.S. Department of Energy, Energy Information Administration, The 
U.S. Petroleum Refining and Gasoline Marketing Industry, June 1999. 

[41] However, according to EIA analysts, even though vertical 
integration may have increased--especially in the downstream segment 
between refining and marketing--there has been a shift toward 
divestiture of downstream assets (particularly refineries) by fully 
vertically integrated oil companies since the 1990s, giving independent 
refiners opportunities to acquire and grow their refining assets. 

[42] These mergers are also documented in Justine Hastings and Richard 
Gilbert, "Vertical Integration in Gasoline Supply: An Empirical Test of 
Raising Rivals' Costs," Working Paper Series of the Program on Workable 
Energy Regulation (POWER), University of California Energy Institute, 
Berkeley, California, July, 2001. Tosco also acquired some retail 
assets from Exxon and Mobil on the East Coast. This acquisition was a 
result of a divesture FTC mandated as a condition for the merger of 
Exxon and Mobil.

[43] The term offshore indicates a portion of open sea and the 
petroleum exploration and production activities carried out in such 
areas, while onshore refers to land operations.

[44] Production exhibits economies of scale if average (per unit) costs 
fall as output increases, and diseconomies of scale if average cost 
increases as output increases. Scale economies can occur at the level 
of the individual plant, in which case they generally reflect elements 
of the production process. Scale economies may be distinguished as 
technological, reflecting changes in input use as output expands, or 
pecuniary, reflecting changes in prices paid for inputs as output 
expands. At the firm level, they may also reflect elements of marketing 
and distribution costs. 

[45] John J. Coyle, Edward J. Bardi, and Robert A. Novack, 
Transportation, 5TH Ed., (Mason, Ohio: South-Western College 
Publishing, 2000).

[46] As discussed in chapter 4, however, hypermarkets such as Wal-Mart 
and Costco, have entered the retail gasoline market using the marketing 
strategy of high volume and low prices. They have the advantage of 
already owning the land for the gasoline retail site.

[47] Unbranded gasoline purchasers have traditionally been able to shop 
around for the best available price in the marketplace without any 
binding contractual arrangement. However, this situation may be 
changing because, many distributors told us, some suppliers of 
unbranded gasoline are now requiring buyers to sign a binding contract 
to guarantee their supply. 

[48] In a time of low gasoline supply, branded companies will 
accommodate their branded contracts first, shifting gasoline supply 
from unbranded to branded, causing unbranded gasoline prices to rise.

[49] According to EIA officials, FRS companies accounted for about 85 
percent of the total U.S. gasoline supply.

[50] EIA's data combined sales by lessee dealers and open dealers. 

[51] Fully vertically integrated oil companies who merged with other 
companies also divested some refineries as mandated by the FTC as part 
of its remedy for potential anticompetitive effects of such mergers.

[52] As discussed earlier, data are not available on unbranded gasoline 
supply to determine the percentage of the unbranded gasoline sales that 
hypermarkets represent.

[53] The small, unbranded open dealer usually becomes branded by 
entering into a contract with a distributor to supply branded gasoline, 
which must be retailed under the brand's trademark.

[54] Energy Analysis International, Inc, U.S. Hypermart Petroleum 
Market Study, 2001 Edition.

[55] Our analysis is based on regular, unleaded gasoline, which is the 
predominant type of gasoline sold. CARB is California Air Resources 
Board's requirement to have reformulated gasoline for lower pollution. 
Conventional gasoline contains no additive, but reformulated and CARB 
gasoline contain MTBE (methyl tertiary butyl ether) as an additive. 

[56] Wholesale gasoline prices are measured by the average prices at 
the terminals or racks. To help isolate the effects of mergers and 
market concentration on gasoline prices at the wholesale level, it was 
necessary to account for the effect of changes in crude oil prices. 
Henceforth, we refer to wholesale gasoline prices minus crude oil 
prices simply as wholesale gasoline prices. 

[57] The other contributors to the marginal wholesale costs are labor 
costs, capital costs, energy, and purchased services. We could not 
subtract capital costs or labor costs from the wholesale prices because 
the available data for these inputs are annual price indices. See 
appendix IV for details. 

[58] The size and duration of the disruptions would depend on several 
conditions, including the preexisting market conditions and how the 
refining industry chooses to respond to the disruptions. See appendix 
IV for a complete discussion of these supply disruptions. 

[59] We refer to all the transactions collectively as mergers, since 
they led to the consolidation of assets. We could not analyze the BP-
Amoco merger with ARCO in April 2000 or the Chevron-Texaco merger in 
October 2000 because of data limitations.

[60] While some of these mergers were included in our study because of 
FTC's review of them, our study did not assess the appropriateness of 
FTC's review and actions they took regarding these selected mergers. 

[61] The industry participants we interviewed include petroleum 
marketing associations, independent refiners, and fully vertically 
integrated oil companies. 

[62] Unless otherwise noted, all the estimated changes in prices 
(increases or decreases) are statistically significant. In other words, 
the estimated changes are statistically different from zero at the 10 
percent significance level or less. See appendix IV for more discussion 
of the econometric results of the effects of individual oil industry 
mergers.

[63] See appendix IV for complete details of these results.

[64] Pinkse et al. obtained similar results using data on percentage 
changes in gasoline inventories.

[65] See appendix IV for complete details of these results.

[66] See Justine Hastings and Richard Gilbert, "Vertical Integration in 
Gasoline Supply: An Empirical Test of Raising Rivals' Costs," Program 
on Workable Energy Regulation (POWER), PWP-084, 2002.

[67] For unbranded gasoline, our results were not statistically 
significant. Hastings also found that ARCO's purchase of Thrifty's 
retail gasoline stations in California in 1997, which decreased the 
market share of independent retailers and increased retail market 
concentration, raised retail prices. See Justine Hastings, "Vertical 
Relationships and Competition in Retail Gasoline Markets: Empirical 
Evidence from Contract Changes in Southern California," Program on 
Workable Energy Regulation (POWER), PWP-075 (2001). We did not examine 
the effect of mergers on retail prices of gasoline because that is 
beyond the scope of this study. 

[68] See Kenneth Hendricks and R. Preston McAfee, "A Theory of 
Bilateral Oligopoly, With Applications to Vertical Mergers," 
unpublished paper (2000). The authors did not analyze the actual 
effects of the merger between the two companies that occurred in 
December 1999. 

[69] The authors measured prices as wholesale gasoline prices less 
marginal cost ("price-cost margin"). 

[70] See Hayley Chouinard and Jeffrey Perloff, "Gasoline Price 
Differences: Taxes, Pollution, Regulations, Mergers, Market Power, and 
Market Conditions," unpublished paper (2001).

[71] See Christopher Taylor and Daniel Hosken, "The Economic Effects of 
the Marathon-Ashland Joint Venture: The Importance of Industry Supply 
Shocks and Vertical Market Structure," FTC Bureau of Economics Working 
Paper 270, March 17, 2004. The FTC notified GAO of this study on March 
24, 2004. As we discuss in appendix IV of our report, the FTC study has 
shortcomings in several areas, including the econometric methodology 
and interpretation of the results. 

[72] The statistical properties of price-crude cost margins provided 
another motivation for their use.

[73] See appendix IV for a complete discussion of the limitations of 
our econometric methodology.

[74] The guidelines categorize markets with concentration levels, as 
measured by the Herfindahl-Hirschman Index, of less than 1,000 as 
unconcentrated, from 1,000 to 1,800 as moderately concentrated, and 
markets above 1,800 as highly concentrated.

[75] While we are aware that other factors may affect market 
concentration, such as growth of the market, entry, and exit, we 
focused our examination on the linkage between merger activity, as 
measured by the average yearly transaction values of mergers, and 
market concentration.

[76] For the correlations, we also used the yearly total value of 
mergers as it captures both the size and the number of mergers, and the 
results were similar to the average annual transaction value of 
mergers.

[77] Wholesale gasoline sales occur at terminals or racks that are near 
or in cities, sometimes referred to as rack cities.

[78] In order to preserve competition at racks that would have been 
affected significantly by the mergers, the FTC provided remedies, 
particularly in the form of divestitures, in order to replace the 
competition that would be lost as a result of the mergers. A 
divestiture requires that one or both of the merging parties sell some 
of its assets to restore or maintain competition where it might be 
harmed. 

[79] The United States is divided into five regions: PADD I, the East 
Coast region; PADD II, the Midwest region; PADD III, the Gulf Coast 
region; PADD IV, the Rocky Mountain region; and PADD V, the West Coast 
region.

[80] Downstream businesses include refining and marketing (wholesale 
and retail). 

[81] The affiliated distributors buy only from their parent companies, 
which are typically the large integrated oil companies, while the 
independents typically buy from the lowest-priced seller. Both buyers 
sell to their own retail stations as well as to other retail stations. 

[82] The integrated refiners are large companies that typically sell 
branded gasoline that bears their trademarks (e.g., Exxon and BP), 
while the independents are small, tend to be less integrated, and sell 
a higher proportion of unbranded gasoline. Branded gasoline contains an 
additive associated with the brand, but unbranded gasoline need not, 
and often does not, contain the same additive package. Integrated 
refiners use exchange agreements to get gasoline in locations where 
they do not have refineries or rack space. 

[83] See, for example, Hendricks and McAfee (2000).

[84] See Pinkse et. al (2002).

[85] The rack prices can be contract or noncontract. Transfer prices 
are implicit prices at which integrated refiners supply their company-
owned and company-operated retail stations. Dealer-tankwagon prices are 
contract prices charged to lessee dealers (dealers that operate retail 
stations leased from an integrated refiner) and open dealers (dealers 
that own a retail station but contract with a refiner to sell its 
branded gasoline). See chapter 4 for a description of the wholesale 
gasoline marketing structure.

[86] See Hastings and Gilbert (2002) and Hendricks and McAfee (2000) 
for a similar approach using price margins (wholesale prices less crude 
costs or prices in other rack cities).

[87] See for example, studies in the oil industry by Borenstein and 
Shepard (1996a, 1996b), Chouinard and Perloff (2001), GAO (1993), 
Hastings and Gilbert (2002), Hastings (2002), Pinkse et al. (2002), and 
Vita (2000).

[88] Hendricks and McAfee (2000) advocated using capacity to measure 
the effect of mergers on gasoline markets. We constructed the market 
concentration using the Department of Energy's EIA (Energy Information 
Administration) data on refinery capacity, which are annual. A 
limitation of the refinery capacity data is that they do not give the 
exact yield for gasoline--some refineries can yield about 55 to 60 
percent and others can yield only about 45 to 50 percent of gasoline. 
However, the refinery capacity data are generally fixed for a long 
period of time.

[89] We did not include CRUDE directly as an explanatory variable in 
the price-margin equation, because the dependent variable is defined as 
wholesale prices less crude oil costs. Although we did not have price 
data for other inputs, such as labor and capital costs, at the rack 
city level, we do not expect this to significantly affect our findings 
because these inputs comprise a small share of the inputs used to 
produce gasoline. Crude oil costs constitute about 66 percent of total 
refining costs. The other costs are capital costs (20 percent), labor 
costs (6 percent), purchased services costs (6 percent), and energy 
costs (2 percent). 

[90] The gasoline inventories, based on EIA data, included gasoline 
inventories at bulk racks and refineries and in pipelines at the PADD 
level. The data used were aggregated for finished gasoline, including 
conventional and reformulated. However, the aggregate data reflect the 
dominant type of gasoline in the region.

[91] See FTC (2001a) for more details.

[92] See Taylor and Fischer (2001) and EIA (2001). 

[93] We did not include year effects because while the year effects 
would control for cyclical patterns that are common to all rack cities, 
we do not believe there are annual cyclical phenomena in the gasoline 
markets that we studied. Also, we could not estimate the effects of 
income and population density--demand-related variables--because the 
data do not vary across time within a rack city (they are time 
invariant).

[94] See Pinkse et al. for details.

[95] In fact, in our preliminary estimations we found that the 
estimated coefficients on the nearest prices were not statistically 
different from one. 

[96] See the section on estimation methodology below for a discussion 
of how we handled this potential problem by accounting for 
contemporaneous cross-sectional (rack city) correlations.

[97] We should note that branded and unbranded gasoline might compete 
to some extent in a market.

[98] See Vita (2000) for a detailed discussion of the possible effects 
of divorcement regulations.

[99] Most of the studies on wholesale gasoline markets have used a rack 
city as the unit of analysis; see, for example, Borenstein and Shepard 
(1996a, 1996b), Hastings and Gilbert (2002), and Pinkse et al. (2002). 

[100] There are no rack data for Hawaii and the District of Columbia.

[101] The equations we estimated are single-equation or limited 
information models because we do not specify the complete structural 
equations for the other potential endogenous variables. The regressions 
used to obtain the estimated values of the other endogenous variables 
are computational devises used to purge the prices from potential 
correlation with the error term. A reduced-form price model is useful 
for analyzing the total impact of a policy-relevant event, such as a 
merger, on prices. In addition, a reduced-form model may provide more 
robust and reliable estimates; see, for example, Schmidt (2001). See 
also FTC (2001b, p. 24).

[102] See Greene (2000) for the FGLS technique.

[103] See, for example, Barton and Sherman (1984) and Kim and Singal 
(1993).

[104] See, for example, Karikari et al. (2002).

[105] See, for example, Chouinard and Perloff (2001), Hastings and 
Gilbert (2002), and Pinkse et al. (2002).

[106] See Dickey and Fuller (1979).

[107] The ADF unit root test indicated that the HHI is stationary only 
in PADD I. Nonetheless, we regard the unit root tests to be weak 
because the HHI is bounded (ranges from 0 to 10,000). Furthermore, it 
would not be appropriate to first-difference the HHI to obtain 
stationarity since the data are generally constant over some relatively 
long periods of time.

[108] The eastern half of the United States consists of PADDs I, II, 
and III, which are generally areas to the east of the Mississippi 
River, and the western half consists of PADDs IV and V.

[109] An important purpose in combining cross-sectional and time-series 
data is to control for individual city-specific unobservable effects, 
which may be correlated with explanatory variables in the model.

[110] The Hausman (1978) specification test can be used to test for 
endogeneity of regressors; see, for example, Wooldridge (2002, pp. 118-
119).

[111] The Hausman (1978) specification test can be used to test for 
overidentifying restrictions; see, for example, Wooldridge (2002, pp. 
122-123). Our tests indicated that the two regressors were exogenous in 
some models. In the cases where the variables were endogenous, the 
tests indicated that the instruments were appropriate or valid. (Also, 
the instruments were relevant for both the INVENTORY RATIO and 
UTILIZATION RATES--the R2s for the first-stage regressions ranged from 
85 to 90 percent, and from 58 to 66 percent, respectively). See the 
regression estimates in tables 21-28 for details. 

[112] In cases where the estimation method is FGLS/IV, we used two 
Stata software programs to carry out the testing and estimation. The 
programs are a panel-data instrumental variable estimator (IVREG2) and 
the feasible generalized least squares (XTGLS) estimator. To be able to 
use the two programs in an integrated fashion, we had to slightly 
modify XTGLS so that it takes as input the residuals of the IVREG2 
estimator for calculating autocorrelation and contemporaneous 
correlation parameters. Without this modification, a two-stage XTGLS 
would calculate the instrumental variables (IV) residuals using the 
instrument rather than the endogenous regressors causing biased 
estimation. (See, for example, Davidson and Mackinnon (1993, p. 221). 

[113] See pp. 282-283 of Wooldridge. Our tests indicated the presence 
of AR(1) in all the models. See the regression estimates in tables 21-
28 for details.

[114] A statistically significant change means that the estimated 
changes in prices are statistically different from zero. 

[115] A complete discussion of the effects of each merger on prices is 
provided in chapter 5. 

[116] The only exception is column (iii) of table 23. All the 
econometric estimates were obtained using Stata (Version SE 8.0), 
College Station, Texas. 

[117] When we estimated the models for conventional gasoline by 
including crude oil prices as a regressor, instead of as part of the 
dependent variable, the R-squares exceeded 80 percent.

[118] See Taylor and Hosken (2004).

[119] From the OPIS rack database, both Ashland and Marathon were 
important participants in the wholesale gasoline market in Chicago from 
1994 until 1997, when Ashland left. The merger also affected the 
markets in Norfolk and Richmond, both in Virginia.

[120] A complete discussion of the effects of market concentration on 
prices is provided in chapter 5. 

[121] When we estimated the models for conventional gasoline by 
including crude oil prices as a regressor, instead of as part of the 
dependent variable, the R-squares exceeded 80 percent. We also 
considered other possible relationships between the HHI and prices, 
including the squared HHI. The results for the HHI were not 
statistically significant, or the estimates were not inconsistent with 
our results.

[122] The results were generally similar to those for the mergers 
models. The Midwest supply disruptions affected only one rack city in 
the data for reformulated gasoline.

[123] We could not obtain estimates for demographic factors such as 
income and population density or for competitive conditions such as 
distance, number of terminals, and divorcement regulations because the 
data are time-invariant.

[124] Pinkse et al. (2002) also obtained a negative effect in their 
model, even though they used percentage changes in inventories. 

[125] The coefficient for INVENTORIES RATIO represents the change in 
prices when the ratio of inventories to expected demand increases by 
100 percent. Therefore the estimated coefficients are about 7 to 9 
cents per gallon (see columns (ii) and (iv) of table 24). Assuming an 
increase of about 14 percent (which is about 2 standard deviations for 
INVENTORIES, see table 20) from May to October, it implies that prices 
are typically about 1 cent per gallon higher from May through October 
compared to the other months.

[126] Another possible measure of market concentration for our study is 
using HHI data that (1) are based on gasoline sales by prime suppliers 
who are not all refiners and (2) exclude small refiners. When we used 
these data in our models, the results were generally similar but not as 
robust compared to those reported.

[127] See chapter 5 for a discussion of results from previous studies. 
See also, for example, Borenstein and Shepard (1996b), Chouinard and 
Perloff (2001), and Pinkse et al. (2002). 

[128] Scheffman and Coleman, FTC Perspectives on the Use of Econometric 
Analyses in Antitrust Cases, unpublished paper, undated.

[129] See Taylor and Hosken (2004). See appendix IV for a detailed 
assessment of the FTC study. The other two FTC studies of consummated 
mergers that we know of were not for the petroleum industry. (See a 
study by Scheffman and Coleman (2002), p. 364, FTC's former Director 
and Deputy Director of the Bureau of Economics, respectively). 

[130] "Price-Cost Margins and Industry Structure," Review of Economics 
and Statistics," vol. 51 (August 1969): 277, table 3.

[131] A Guide to Econometrics, Fourth Edition (The MIT Press: 
Cambridge, MA, 1998): 27.

[132] Analysis of Panel Data, Second Edition (Cambridge University 
Press: New York, 2003).

[133] Econometric Analysis, Fourth Edition (Upper Saddle River, New 
Jersey: Prentice Hall, 2000): 334-337.

[134] How Much Should We Trust Difference-in-Difference Estimates? MIT 
Economics Department Working Paper 2003.

GAO's Mission: 

The General Accounting Office, the investigative arm of Congress, 
exists to support Congress in meeting its constitutional 
responsibilities and to help improve the performance and accountability 
of the federal government for the American people. GAO examines the use 
of public funds; evaluates federal programs and policies; and provides 
analyses, recommendations, and other assistance to help Congress make 
informed oversight, policy, and funding decisions. GAO's commitment to 
good government is reflected in its core values of accountability, 
integrity, and reliability.

Obtaining Copies of GAO Reports and Testimony: 

The fastest and easiest way to obtain copies of GAO documents at no 
cost is through the Internet. GAO's Web site ( www.gao.gov ) contains 
abstracts and full-text files of current reports and testimony and an 
expanding archive of older products. The Web site features a search 
engine to help you locate documents using key words and phrases. You 
can print these documents in their entirety, including charts and other 
graphics.

Each day, GAO issues a list of newly released reports, testimony, and 
correspondence. GAO posts this list, known as "Today's Reports," on its 
Web site daily. The list contains links to the full-text document 
files. To have GAO e-mail this list to you every afternoon, go to 
www.gao.gov and select "Subscribe to e-mail alerts" under the "Order 
GAO Products" heading.

Order by Mail or Phone: 

The first copy of each printed report is free. Additional copies are $2 
each. A check or money order should be made out to the Superintendent 
of Documents. GAO also accepts VISA and Mastercard. Orders for 100 or 
more copies mailed to a single address are discounted 25 percent. 
Orders should be sent to: 

U.S. General Accounting Office

441 G Street NW,

Room LM Washington,

D.C. 20548: 

To order by Phone: 

Voice: (202) 512-6000: 

TDD: (202) 512-2537: 

Fax: (202) 512-6061: 

To Report Fraud, Waste, and Abuse in Federal Programs: 

Contact: 

Web site: www.gao.gov/fraudnet/fraudnet.htm E-mail: [email protected]

Automated answering system: (800) 424-5454 or (202) 512-7470: 

Public Affairs: 

Jeff Nelligan, managing director, [email protected] (202) 512-4800 U.S.

General Accounting Office, 441 G Street NW, Room 7149 Washington, D.C.

20548: