[House Hearing, 116 Congress]
[From the U.S. Government Publishing Office]


                    EXAMINING THE USE OF ALTERNATIVE
                     DATA IN UNDERWRITING AND CREDIT
                   SCORING TO EXPAND ACCESS TO CREDIT

=======================================================================

                                HEARING

                               BEFORE THE

                   TASK FORCE ON FINANCIAL TECHNOLOGY

                                 OF THE

                    COMMITTEE ON FINANCIAL SERVICES

                     U.S. HOUSE OF REPRESENTATIVES

                     ONE HUNDRED SIXTEENTH CONGRESS

                             FIRST SESSION

                               __________

                             JULY 25, 2019

                               __________

       Printed for the use of the Committee on Financial Services

                           Serial No. 116-42
                           
[GRAPHIC IS NOT AVAILABLE IN TIFF FORMAT]

                              __________
                               

                    U.S. GOVERNMENT PUBLISHING OFFICE                    
40-160 PDF                  WASHINGTON : 2020                     
          
--------------------------------------------------------------------------------------

                 HOUSE COMMITTEE ON FINANCIAL SERVICES

                 MAXINE WATERS, California, Chairwoman

CAROLYN B. MALONEY, New York         PATRICK McHENRY, North Carolina, 
NYDIA M. VELAZQUEZ, New York             Ranking Member
BRAD SHERMAN, California             PETER T. KING, New York
GREGORY W. MEEKS, New York           FRANK D. LUCAS, Oklahoma
WM. LACY CLAY, Missouri              BILL POSEY, Florida
DAVID SCOTT, Georgia                 BLAINE LUETKEMEYER, Missouri
AL GREEN, Texas                      BILL HUIZENGA, Michigan
EMANUEL CLEAVER, Missouri            SEAN P. DUFFY, Wisconsin
ED PERLMUTTER, Colorado              STEVE STIVERS, Ohio
JIM A. HIMES, Connecticut            ANN WAGNER, Missouri
BILL FOSTER, Illinois                ANDY BARR, Kentucky
JOYCE BEATTY, Ohio                   SCOTT TIPTON, Colorado
DENNY HECK, Washington               ROGER WILLIAMS, Texas
JUAN VARGAS, California              FRENCH HILL, Arkansas
JOSH GOTTHEIMER, New Jersey          TOM EMMER, Minnesota
VICENTE GONZALEZ, Texas              LEE M. ZELDIN, New York
AL LAWSON, Florida                   BARRY LOUDERMILK, Georgia
MICHAEL SAN NICOLAS, Guam            ALEXANDER X. MOONEY, West Virginia
RASHIDA TLAIB, Michigan              WARREN DAVIDSON, Ohio
KATIE PORTER, California             TED BUDD, North Carolina
CINDY AXNE, Iowa                     DAVID KUSTOFF, Tennessee
SEAN CASTEN, Illinois                TREY HOLLINGSWORTH, Indiana
AYANNA PRESSLEY, Massachusetts       ANTHONY GONZALEZ, Ohio
BEN McADAMS, Utah                    JOHN ROSE, Tennessee
ALEXANDRIA OCASIO-CORTEZ, New York   BRYAN STEIL, Wisconsin
JENNIFER WEXTON, Virginia            LANCE GOODEN, Texas
STEPHEN F. LYNCH, Massachusetts      DENVER RIGGLEMAN, Virginia
TULSI GABBARD, Hawaii
ALMA ADAMS, North Carolina
MADELEINE DEAN, Pennsylvania
JESUS ``CHUY'' GARCIA, Illinois
SYLVIA GARCIA, Texas
DEAN PHILLIPS, Minnesota

                   Charla Ouertatani, Staff Director
                   
                   
                   TASK FORCE ON FINANCIAL TECHNOLOGY

               STEPHEN F. LYNCH, Massachusetts, Chairman

DAVID SCOTT, Georgia                 FRENCH HILL, Arkansas, Ranking 
JOSH GOTTHEIMER, New Jersey              Member
AL LAWSON, Florida                   BLAINE LUETKEMEYER, Missouri
CINDY AXNE, Iowa                     TOM EMMER, Minnesota
BEN McADAMS, Utah                    WARREN DAVIDSON, Ohio
JENNIFER WEXTON, Virginia            BRYAN STEIL, Wisconsin
                            
                            
                            C O N T E N T S

                              ----------                              
                                                                   Page
Hearing held on:
    July 25, 2019................................................     1
Appendix:
    July 25, 2019................................................    41

                               WITNESSES
                        Thursday, July 25, 2019

Evans, Lawrance L., Managing Director, Financial Markets and 
  Community Investment, U.S. Government Accountability Office 
  (GAO)..........................................................    10
Girouard, Dave, CEO and Co-Founder, Upstart Network, Inc.........    11
Johnson, Kristin N., McGlinchey Stafford Professor of Law, Tulane 
  University Law School..........................................     8
Rieke, Aaron, Managing Director, Upturn..........................     6
Wu, Chi Chi, Staff Attorney, National Consumer Law Center (NCLC).     5

                                APPENDIX

Prepared statements:
    Evans, Lawrance L............................................    42
    Girouard, Dave,..............................................    54
    Johnson, Kristin N...........................................    57
    Rieke, Aaron.................................................    74
    Wu, Chi Chi..................................................    80

              Additional Material Submitted for the Record

Lynch, Hon. Stephen F.:
    Written statement of the Cato Institute's Center for Monetary 
      and Financial Alternatives.................................    96
    Domino: A Blog About Student Debt............................    99
    Written statement of the Financial Data and Technology 
      Association of North America (FDATA North America).........   102
    Written statement of FICO....................................   105
    Written statement of VantageScore............................   107
McHenry, Hon. Patrick:
    Written statement of the Cato Institute's Center for Monetary 
      and Financial Alternatives.................................   111

 
                    EXAMINING THE USE OF ALTERNATIVE
                    DATA IN UNDERWRITING AND CREDIT
                   SCORING TO EXPAND ACCESS TO CREDIT

                              ----------                              


                        Thursday, July 25, 2019

             U.S. House of Representatives,
                Task Force on Financial Technology,
                           Committee on Financial Services,
                                                   Washington, D.C.
    The task force met, pursuant to notice, at 10:02 a.m., in 
room 2128, Rayburn House Office Building, Hon. Stephen F. Lynch 
[chairman of the task force] presiding.
    Members present: Representatives Lynch, Scott, Gottheimer, 
Lawson, Axne, McAdams, Wexton; Hill, Luetkemeyer, Emmer, 
Davidson, and Steil.
    Ex officio present: Representative McHenry.
    Also present: Representatives Green, Himes, Porter; 
Gonzalez of Ohio, and Hollingsworth.
    Chairman Lynch. Good morning. The Task Force on Financial 
Technology will come to order.
    Without objection, the Chair is authorized to declare a 
recess of the task force at any time.
    Also, without objection, members of the full Financial 
Services Committee who are not members of this task force are 
authorized to participate in today's hearing consistent with 
the committee's practice.
    Today's hearing is entitled, ``Examining the Use of 
Alternative Data in Underwriting and Credit Scoring to Expand 
Access to Credit.'' I now recognize myself for 4 minutes to 
give an opening statement.
    I want to thank everyone for being here at our second 
Financial Technology Task Force hearing. Today's hearing will 
focus on the use of alternative data, the financial and 
nonfinancial data that is not traditionally used by national 
consumer reporting agencies in credit underwriting. With an 
estimated 26 million consumers lacking in any credit history, 
and another 19 million with an outdated or short credit 
history, lenders have looked to other means of assessing the 
creditworthiness of applicants.
    As a result, alternative data has become a hot topic. It 
has the potential to expand credit access but also raises 
concerns over the nature and sources of its data points. There 
is also significant regulatory uncertainty surrounding its use. 
Today, we will hear testimony and discuss questions on all of 
these issues. The promise of fintech lending has been to lower 
costs and bring new consumers into the market. This promise has 
been fueled by data points outside of the traditional factors 
used by underwriters like payment history and credit 
utilization. Today, lenders use an array of financial and 
nonfinancial data in their decision-making. Some factors, such 
as utility bill or rent payments, resemble traditional factors. 
Others, such as living in public housing, who your friends are, 
and what their credit scores are, your ZIP Code, your reading 
choices, educational attainment, educational institutions, and 
driving habits or online shopping habits, are a significant 
departure from traditional factors.
    We know that Facebook has up to 52,000 data points on each 
of its 2.7 billion daily users, and they sell access to that 
data to its advertisers. Use of this and other data can 
potentially help 45 million Americans who might have trouble 
accessing credit with traditional factors alone. Take for 
instance, a 28-year-old woman in a modest-paying job, or maybe 
with 2 very modest paying jobs, who has never had a credit card 
or taken out a personal loan or mortgage loan. She might find 
herself denied access to credit based on traditional factors, 
even though she is working 12 hours a day. But a lender using 
alternative data might take into account that she went to a 
reputable school, had a job with a reputable employer, and 
always pays her rent and utility bills on time. In that case, 
they might approve her application for credit. It is very 
likely we have one or two staffers working here on Capitol Hill 
who fit that exact description.
    However, it is not hard to imagine a similar scenario with 
much different results. Say, a young man with a decent but 
short credit history might be right on the cusp of being deemed 
creditworthy by traditional factors. However, a lender using 
alternative data sees in his rental history that he moves 
frequently, moves around a lot. In the last few years, he has 
had several domiciles. They may also see he doesn't have a 
college degree and that his Facebook friends have below average 
credit scores. So, they deny him access to credit. 
Unfortunately, this probably describes a number of our military 
personnel as they repeatedly move domiciles as a result of 
multiple redeployments during their careers.
    Without question, there are instances when using 
alternative data in credit underwriting has potential positive 
impacts. However, right now, oversight of its use is either 
highly fragmented or completely nonexistent, leading to 
uncertainty for lenders and potential harm for consumers. That 
is why we are here today, to better understand how to harness 
the benefits and mitigate the harms of using alternative data.
    I look forward to the testimony of our witnesses and the 
discussion of our Members. With that, I now recognize the 
ranking member of the task force, the gentleman from Arkansas, 
Mr. Hill, for 5 minutes for an opening statement.
    Mr. Hill. I thank the chairman. I appreciate you convening 
this hearing and I appreciate our witnesses appearing today. We 
are grateful for your advice and counsel today. Analyzing the 
use of alternative data in the marketplace lending industry is 
an important sector within our broader study of the fintech 
ecosystem. I am pleased that we were able to bring everybody 
together and do a deeper dive on this topic.
    Marketplace or fintech lenders are categorized through 
their digital or online focus and have recently emerged and 
grown quite a bit over the last decade. According to S&P 
Global, marketplace lending grew by 30 percent in 2017. They 
provide unsecured credit to individuals and working capital to 
small businesses. They have unique funding models with 
financing provided by investors, credit facilities, 
securitization, and, of course, balance sheet cash.
    Typically, these lenders currently lend through two primary 
models: a bank partnership model, in which a bank originates 
the loan, which is generally sourced and served by the 
marketplace lender; or a direct lender model, in which a 
marketplace lender acquires the applicable regulatory licensing 
in all of the States of our country.
    To help determine a borrower's creditworthiness, 
marketplace lenders often use some form of alternative data, 
hence the topic today. Traditional lenders typically have used 
FICO scores, 3 years of tax returns, payment history for credit 
cards, mortgages, or student loans in order to establish a risk 
profile for their borrowers.
    However, marketplace lenders robustly combine FICO scores 
with alternative data points to better gauge a borrower's 
character and economic situation. Examples of these data points 
include education level, employment status, utility and rent 
payments, et cetera. Analyzing these data points has the 
potential to widen the universe of borrowers and provide 
greater access to affordable credit.
    Importantly, a report by TransUnion outlined that lenders 
that utilized alternative data were able to lend to an 
additional 66 percent of borrowers in current markets and 56 
percent in new markets.
    Today, we will explore concerns about how alternative data 
can best comply with critical fair lending requirements, which 
will be discussed in more depth. However, I do want to remind 
my colleagues that we don't want overregulation to stifle 
innovation and prevent the American consumer from now being 
able to access affordable credit through this new methodology.
    As to compliance obligations, obviously, I want to 
highlight some of the ongoing issues that have been evolving 
within the marketplace lending industry. The Treasury report--
which I regularly reference in these hearings--released a year 
ago now, provides a comprehensive review of the fintech sector. 
It has a robust analysis of this industry. It generally favors 
innovation, but identified certain important policies that need 
to be highlighted and discussed today, such as codify the 
valid-when-made doctrine, codify the role of the bank as the 
true lender of a loan that is made, allowing the testing of new 
credit models and data sources for financial institutions, and 
think through this issue of special charters or harmonization 
of this process across our States.
    The report also mentioned the third-party lender guidance. 
I know the FDIC and the OCC have been focused on this due to 
the rise of marketplace lenders and strong bank partnerships. 
As a former community banker, I well understand the compliance 
responsibilities around vendor partner, due diligence, 
onboarding of new partnerships, and board of director reviews.
    Also, as a result, as a banker, I understand the importance 
of banks maintaining a robust level of safety and soundness and 
constantly facing changing technology but assuring a vigor in 
compliance on both data security and privacy.
    I look forward to hearing the thoughts of the panel today, 
and over the years, I think this is going to be a fascinating 
way for Mr. Lynch and I to make recommendations to our full 
Financial Services Committee on how we can broaden marketplace 
lending. So, with that, I would like to yield the balance of my 
time to my friend, the ranking member of the Full Committee, 
Mr. McHenry of North Carolina.
    Mr. McHenry. Thank you. And, look, technology is creating 
new pathways for more consumers to access the financial system. 
That is a good thing. We are talking about people who are 
otherwise credit invisible or lack enough credit history to 
finance things like a mortgage, credit cards, or other loans. 
Alternative data draws on nontraditional sources of financial 
history, including bill payment history in areas like mobile 
phones, that are now essential ingredients, with 81 percent of 
Americans owning a smartphone at this point or using a 
smartphone, and rent. And by harvesting this type of data about 
the consumer, lenders have a more holistic picture about the 
consumer to whom they are lending.
    Yet, this new era is not without its challenges. We need to 
ensure that alternative data remains nondiscriminatory and that 
consumer data and privacy are protected. So, it is our job to 
ensure responsible innovation continues to be a driving force 
of the American economy, but in particular, in financial 
services. I yield back.
    Chairman Lynch. The gentleman yields back.
    The Chair now recognizes the gentleman from Georgia, Mr. 
Scott, for 1 minute for an opening statement.
    Mr. Scott. Thank you very much, Chairman Lynch, and let me 
commend you and Mr. Hill for providing this excellent 
bipartisan leadership on what I refer to as the thrilling new 
frontier. We are in a situation now where technology is moving 
at such a rapid pace, and where we need to look at where we 
need to adjust the sails and make sure everybody has an 
opportunity to be able to participate in this. And, of course, 
as we look at this scoring technology, we look at how it is 
impacting our financial system. There is no group that needs 
our help more than the 26 million Americans who have no credit 
history. There are also 19 million Americans who have a very 
limited credit history. And when you put the totality of the 
unbanked and the underbanked in there, we can see that we must 
not leave these parts of our population behind.
    So, I am looking forward to this, to making sure that we 
determine effectively how data is used in lending decisions and 
credit scoring, so all the American people can participate in 
this glorious new frontier.
    Thank you, Mr. Chairman.
    Chairman Lynch. I thank the gentleman. On behalf of this 
bipartisan task force, I want to welcome our distinguished 
panel. I would like to welcome the testimony of: Chi Chi Wu, a 
staff attorney with the National Consumer Law Center, based in 
Boston, my hometown; Aaron Rieke, managing director at Upturn, 
which is a nonprofit focused on promoting equity and digital 
technology through research and advocacy; Kristin Johnson, 
McGlinchey Stafford Professor of Law at Tulane University Law 
School; Lawrance Evans, Managing Director of Financial Markets 
and Community Investment at the Government Accountability 
Office; and Dave Girouard, founder and CEO at Upstart, which is 
a fintech lender focused on direct-to-consumer loans.
    Witnesses are reminded that your oral testimony will be 
limited to 5 minutes. And without objection, your written 
statements will be made a part of the record.
    Ms. Wu, you are now recognized for 5 minutes to give an 
oral presentation of your testimony.

STATEMENT OF CHI CHI WU, STAFF ATTORNEY, NATIONAL CONSUMER LAW 
                         CENTER (NCLC)

    Ms. Wu. Mr. Chairman, Ranking Member Hill, and members of 
the task force, thank you for inviting me to testify today. I 
am testifying on behalf of the low-income clients of the 
National Consumer Law Center. We have heard several times today 
that there are tens of millions of consumers who are credit 
invisible. The topic of this hearing, alternative data, is 
often promoted as the solution. The thing is, alternative data 
includes lots of different types of data used in lots of 
different ways. Some types of data and uses can be helpful; 
others can hurt. As we say, the devil is in the details.
    The number one consideration for alternative data should be 
consumer choice. That should be the touchstone for all data 
collection. Now, we have heard with respect to the Equifax data 
breach a repeated complaint: Hey, none of us gave Equifax 
permission to collect our data.
    Let's get this right with respect to alternative data. 
Let's make sure it is the consumer's choice, that consumers 
make knowing and affirmative decisions to allow the use of this 
data, and the data is only used in the ways that consumers give 
permission for and expect. Another consideration for 
alternative data is whether it is used to create second-chance 
scores for just credit-invisible consumers or whether it is 
dumped wholesale into traditional credit reports where it might 
damage the records of consumers who already have a score. We 
want to give credit-invisible consumers a chance to be seen 
without hurting any of the nearly 200 million consumers who are 
already visible.
    As for types of data, bank account transaction data has 
shown a lot of promise, but it is also a juicy target. Debt 
collectors would love to get ahold of it. And bank account data 
can include sensitive information, such as where a consumer 
shops. There should be appropriate guardrails for sharing bank 
account data.
    Rent payment information is another type of data looks 
promising, specifically when no additional late payments are 
reported. But we don't want to penalize tenants who invoke 
their rights to withhold rent over poor conditions.
    Payday loan information, in contrast, is probably harmful. 
It is designed to lead to a cycle of debt, and just reporting 
it can hurt a consumer. And it is probably not necessary 
because most payday borrowers actually have credit records.
    Gas and electric utility data can be potentially harmful if 
added in the wrong way. If reported monthly without giving 
consumers a choice, it has the potential to hurt tens of 
millions of low-income consumers by adding new reports of 30- 
or 60-day late payments. In contrast, efforts to include 
utility data on a voluntary basis could be useful, and new 
voluntary products show there is no need for utility credit 
reporting where the consumer has no choice.
    And then, of course, there is Big Data--things like social 
media profiles, web browsing history, and behavioral data. 
There are a lot of unanswered questions about the 
predictiveness and the accuracy of Big Data. Some of it is also 
troubling because it strongly reinforces inequality. For 
example, education, that is, what kind of degree a consumer 
has, is highly correlated with the income and education of 
one's parents. And using social media profiles, particularly 
friend networks, raises concerns about racial disparities, 
given who most people's friends and families are likely to be.
    Speaking of racial disparities, we know there are 
tremendous racial disparities with respect to traditional 
credit scores. It is the result of centuries of slavery and 
discrimination which led to the huge racial wealth gap. 
Alternative financial data is also likely to have racial 
disparities for the same reasons. The critical question is 
whether the alternative data or algorithms lessen or increase 
racial disparities and whether it is more predictive or less 
than traditional models.
    These two questions are closely tied to the test for 
disparate impact under the Equal Credit Opportunity Act. If the 
alternative data is less predictive, there is less of a 
business justification for it, under the disparate impact test. 
On the other hand, if it creates less of a racial disparity, it 
could be a less discriminatory alternative than traditional 
scoring.
    In terms of regulation, all third-party alternative data 
used for credit should be considered a consumer report under 
the Fair Credit Reporting Act (FCRA). Unfortunately, several 
courts of appeals haven't respected the plain language of the 
FCRA and its broad coverage. We urge Congress to reaffirm this 
broad coverage, because the FCRA has critical protections. One 
of the key issues with alternative data is accuracy, the FCRA 
addresses accuracy, and it gives consumers the right to dispute 
errors.
    The FCRA, as well as the ECOA, also requires notices for 
the purpose of transparency, requiring lenders to disclose the 
source and type of information so consumers aren't left in the 
dark as to the reasons for credit decisions. Having black boxes 
to evaluate creditworthiness should be a thing of the past. I 
thank you for the opportunity to testify and I look forward to 
your questions.
    [The prepared statement of Ms. Wu can be found on page 80 
of the appendix.]
    Chairman Lynch. Very good, thank you.
    Mr. Rieke, you are now recognized for 5 minutes.

      STATEMENT OF AARON RIEKE, MANAGING DIRECTOR, UPTURN

    Mr. Rieke. Chairman Lynch, Ranking Member Hill, and 
distinguished members of the task force, thank you for the 
opportunity to testify today. We are here because approximately 
45 million Americans do not have access to credit because there 
is a lack of quality data with which to underwrite them. 
Alternative data can certainly help. I want to echo Ms. Wu and 
say that the devil is in the details, and to suggest that we 
are really talking about two categories of data here: 
conventional data; and fringe data.
    Conventional data consists of things like various payment 
histories, bank account balances, information about an 
individual person's financial capacity. Fringe data consists of 
things like social media data, information that may be 
correlated with this financial capacity but is much further 
removed.
    Conventional data is promising; fringe data raises 
concerns. To understand why, think about traditional FICO 
credit scores. These credit scores are not conceptually 
complex. Most of their predictive value comes from people's 
payment histories. That is really the number one factor in the 
recipe of FICO scores. The logic is simple. If a consumer is 
keeping up with their current financial obligations, it is 
reasonable to predict that they can take on new financial 
obligations. As it turns out, the same basic logic applies to 
many kinds of conventional data. The best available evidence 
suggests that bill payment histories are similarly predictive 
and can help otherwise unscoreable consumers access credit.
    Another example, cash-flow data obtained from a consumer's 
bank account with their express permission, can provide an 
immediate high-quality picture of that person's ability to 
repay a loan, even without a credit bureau being involved at 
all. That is conventional data.
    The story gets murkier when we talk about fringe data. 
Expansive data sets about people's social connections, the 
kinds of websites they visit, where they shop, and how they 
talk do not have the same simple, intuitive connection to each 
individual's ability to repay a loan. These can yield blunt 
stereotypes that might be predictive, but for the wrong 
reasons.
    Let me offer you an analog analogy. Imagine I offered to 
build you a credit-scoring model that relied on a person's ZIP 
Code. That should feel intuitively wrong. I want to unpack why. 
First, we know that geography reflects deep-seated social 
inequities. The result would almost certainly be textbook 
disparate impact.
    Second, judging from ZIP Codes would paint with too broad a 
brush. It would do little to help many of the unscoreable 
consumers we seek to help most who already live in low-income 
neighborhoods. Latching on to traditional markers of wealth and 
privilege aren't going to get us to where we want to be. My 
point is that thousands of behavioral data points thrown into a 
complicated, machine-learning, artificial intelligence model 
can actually act and behave just like a ZIP Code. In the 
absence of rigorous public scrutiny, we should be skeptical of 
fringe data.
    I want to note, because Facebook was brought up in opening 
remarks, that Facebook has for a number of years had a policy 
that prohibits third parties from using Facebook users' data 
for any kind of eligibility purpose. So, if you see a start-up 
company touting their use of Facebook data, ask them why they 
are violating Facebook's policies. That may not be the case 
forever, but I think today that indicates that we are not ready 
to embrace this new data set. In short, this task force should 
focus its efforts on encouraging the use of alternative data 
that is closely related to loan performance, has an 
understandable relationship with an individual applicant's 
creditworthiness, and has been evaluated for compliance with 
anti-discrimination laws.
    Fortunately, this is all doable. More collection and use of 
alternative data makes the Fair Credit Reporting Act and the 
Equal Credit Opportunity Act more important than ever before. I 
would urge Congress to ensure that new kinds of alternative 
data are only used for credit underwriting, where we have 
researched and understood their role, and not for things like 
employment and insurance.
    Finally, as you are all aware, thanks to the advocacy of 
Ms. Wu and her colleagues, any new policies around alternative 
data must respect important State and local consumer 
protections. Thank you again for the opportunity to testify, 
and I welcome your questions.
    [The prepared statement of Mr. Rieke can be found on page 
74 of the appendix.]
    Chairman Lynch. I thank the gentleman.
    Ms. Johnson, you are now recognized for 5 minutes for a 
summary of your testimony.

STATEMENT OF KRISTIN N. JOHNSON, MCGLINCHEY STAFFORD PROFESSOR 
              OF LAW, TULANE UNIVERSITY LAW SCHOOL

    Ms. Johnson. Good morning, Chairman Lynch, Ranking Member 
Hill, Ranking Member McHenry, members of the committee, and 
members of the task force. Thank you for inviting me to 
participate in this hearing to discuss the use of alternative 
data in credit underwriting and credit scoring. I am a 
professor of law and associate dean of faculty research at 
Tulane University Law School, but I have previously worn other 
hats. I was an analyst at Goldman Sachs, a vice president and 
associate general counsel at JPMorgan, and an associate at a 
New York law firm with a globally recognized transactional 
practice. During my tenure in financial services and as an 
academic, I have learned a few things about financial markets, 
including the lesson that credit is a critical resource.
    Individuals and families increasingly rely on credit to 
finance household purchases and overcome significant 
unanticipated expenses. Without access to credit on fair and 
reasonable terms, it can be extraordinarily expensive to be 
poor. For families with fragile financial circumstances, credit 
may serve as a lifeline, enabling consumers to meet short-term 
debt obligations and to pay for education, transportation, 
housing, medicine, childcare, and even food.
    Two critical developments create promise for the 26 million 
Americans referenced earlier as credit invisible, those without 
credit histories, and the 19 million Americans who have thin, 
impaired, or stale credit histories described as unscoreable. 
First, the birth of Big Data. The collection, storage, and 
analysis of vast volumes of consumer data fuels artificial 
intelligence or automated decision-making platforms. Similar to 
the proliferation of AI in health care, employment, criminal 
law, surveillance, and communications, the rise of AI in 
finance monetizes consumer data. Consumers' web browsing, 
click-stream data, and social media networking, which we could 
describe as consumers' digital interface, is matched with or 
paired with consumers' financial transactions, checking and 
saving account cash flows, and credit and debit card 
transactions, fueling data mining and engendering a new set of 
behavioral criteria we can describe as alternative data.
    While fintech firms integrating alternative data offer 
great promise, it is very much worth noting that this new 
species of financial market intermediaries also presents great 
concerns. In my limited time this morning, I note three 
challenges that arise when we integrate and endeavor to 
regulate alternative data.
    First, alternative data may, as mentioned earlier, 
disadvantage vulnerable, marginalized consumers, particularly 
those who are members of legally protected classes. Under the 
behavioral scoring model, your friends on Facebook, the people 
in the pictures you post on Instagram, and those you chat with 
on WhatsApp--I am happy to deconstruct that later for those 
unfamiliar--may signal more than whether or not you have street 
cred. These connections may determine the interest rate on your 
next mortgage.
    It is not yet clear how these new sources of data will 
impact credit invisibles and unscoreables, groups often 
disproportionately comprised of women and people of color. 
Unsavory lending practices, detestable marketing tactics, and 
usurious interest rates have too often plagued these 
marginalized consumers.
    Second, learning algorithms evaluate facially neutral, 
alternative data, yet may result in variables that function as 
proxies for protected traits or result in decisions that may 
have a disparate impact on members of legally protected 
classes.
    Consider, for example, Amazon's recent experiment with an 
algorithm tasked with reviewing resumes for a software 
programmer position. Armed with the resumes of previous hires 
and general instructions regarding qualifications, the 
algorithm went rogue. Because previous hires were predominantly 
men, the algorithm began to discount references to women, 
including references to women's chess club captain or all-
women's colleges. Unknowingly, the algorithm replicated 
historic discriminatory hiring biases. In credit decisions, 
these results may be actional, as noted earlier, under the 
Equal Credit Opportunity Act and fair lending and fair housing 
regulations.
    Finally, alternative data raises concerns regarding 
consumer privacy and cybersecurity concerns. Beyond Equifax's 
settlement this week, there is more breaking cybersecurity 
news. A 20-year-old computer programmer successfully launched a 
cyber attack against another nation's national revenue agency, 
signaling that it is imperative to ensure that any entities 
that collect, store, and transfer consumer data have developed 
sufficient security mechanisms.
    CRAs may also struggle with respect to the obligation to 
describe and explain adverse credit decisions. Because of the 
inscrutable nature of learning algorithms, they are non-
intuitive, opaque, and their operations are not often easily 
explained.
    Finally, in my written testimony I note as well that there 
is an even newer class of emerging financial intermediaries 
within the fintech ecosphere, or ecosystem--blockchain-based 
CRAs. I reference in my written testimony Bloom, one example of 
a blockchain-based credit reporting agency or an entity that 
will operate in a manner similar to a credit reporting agency, 
that is also presumably to rely on alternative data.
    For these reasons, I encourage and urge Congress to think 
carefully about comprehensive legislation outlining the 
appropriate uses for alternative data and data governance, 
storage, transfer, and cybersecurity protections, as well as 
enforcement of antidiscrimination norms.
    [The prepared statement of Ms. Johnson can be found on page 
57 of the appendix.]
    Chairman Lynch. Mr. Evans, you are now recognized to give 
us a 5-minute summary of your testimony.

 STATEMENT OF LAWRANCE L. EVANS, MANAGING DIRECTOR, FINANCIAL 
       MARKETS AND COMMUNITY INVESTMENT, U.S. GOVERNMENT 
                  ACCOUNTABILITY OFFICE (GAO)

    Mr. Evans. Thank you very much, Chairman Lynch. I am 
pleased to appear before you, Ranking Member Hill, Ranking 
Member McHenry, and the members of the task force to discuss 
the use of alternative data in underwriting. My testimony is 
largely based on our December 2018 report, which covered 
several fintech lending issues. The problem with the current 
credit-granting ecosystem has been well-articulated, namely its 
limits in its ability to reach certain borrowers. We know that 
alternative data provides an opportunity to improve the status 
quo by expanding access to credit, improving prices, speeding 
up decision-making, and preventing fraud, but it is also 
important to know that some of what we refer to as alternative 
data is not new.
    However, the types of alternative data available have 
expanded significantly due to the ability to secure large 
volumes of consumer and behavioral information, including data 
on consumer spending and shopping habits, internet browsing 
history, online social media networks, educational 
affiliations, and other factors that may not have a clear nexus 
with creditworthiness.
    In combination with analytic techniques like machine-
learning, these factors provide predictive power for fintech 
companies looking to enhance their ability to determine who is 
eligible for credit. But alternative data is not a panacea. 
Depending on the specifics of these data and the analytical 
techniques used to extract information from them, these 
innovative approaches can bring significant risk. One of the 
major concerns is that usage of that data may produce lending 
outcomes that result in disparate impacts or violations of fair 
lending laws, unintentionally in some cases.
    For example, according to a Federal Reserve newsletter, it 
has been reported that some lenders consider whether a person's 
online social network includes people with poor credit 
histories, which can raise concerns about discrimination 
against those living in disadvantaged areas.
    Another concern is that there may be a lack of transparency 
about what alternative data are being used and how they 
ultimately factor into credit decisions. This potential opacity 
could raise issues, not only for consumers, but for fintech 
firms themselves looking to comply with fair lending 
requirements. It may also be unclear whether a borrower has the 
ability to dispute the accuracy of the information used.
    The great challenge ahead is to manage the risk-reward 
balance of innovation and ensure our experience with 
alternative data does not mimic our experience with alternative 
mortgage products leading up to the financial crisis. To better 
ensure the risks are managed without stifling innovation, which 
is extremely important, policymakers and regulators will need 
to sort through a number of different tradeoffs and 
considerations.
    In the meantime, implementing key recommendations that GAO 
has offered to regulators would assist them in addressing some 
important deficiencies as we see them. Fintech lenders and 
their banking partners we spoke to indicated they face 
challenges due to regulatory uncertainty about the appropriate 
use of alternative data. Representatives of one bank said that 
a fintech partner's use of alternative data may be attractive 
from an innovation and business perspective, but the bank would 
likely hesitate to use this data due to regulatory uncertainty. 
While Federal agencies monitor the use of alternative data, 
they have not provided firms with the types of communication 
that they need to really think through the appropriate use of 
this data in the underwriting process.
    We believe coordinated guidance from the regulators may 
better position fintech lenders and their bank partners to 
responsibly use alternative data. In our prior work, we have 
also recommended that agencies formally evaluate the 
feasibility and benefits of adopting knowledge-building 
initiatives. We believe these initiatives will help firms 
understand the applicable regulations, improve regulators' 
knowledge of fintech products, and facilitate interactions 
between all parties.
    Chairman Lynch, Ranking Member Hill, Ranking Member 
McHenry, and members of the task force, this concludes my 
opening statement. I look forward to any questions you may 
have.
    [The prepared statement of Mr. Evans can be found on page 
42 of the appendix.]
    Chairman Lynch. Thank you, Mr. Evans.
    Mr. Girouard, you are now recognized for 5 minutes. 
Welcome.

    STATEMENT OF DAVE GIROUARD, CEO AND CO-FOUNDER, UPSTART 
                         NETWORK, INC.

    Mr. Girouard. Chairman Lynch, Ranking Member Hill, Ranking 
Member McHenry, and members of the Task Force on Financial 
Technology, thank you for the opportunity to participate in 
today's conversation. My name is Dave Girouard, and I am co-
founder and CEO of Upstart, which is a leading artificial 
intelligence lending platform. I founded Upstart more than 7 
years ago, in order to improve access to affordable credit. In 
the last 5 years, almost $4 billion in bank quality consumer 
loans have been originated on our platform, using a model that 
combines alternative data with AI and machine-learning 
algorithms to determine a borrower's creditworthiness.
    Concerns about fairness in algorithmic lending, 
particularly in the use of alternative data, are well-founded. 
As a company focused entirely on reducing the price of credit 
for the American consumer, fairness is an issue we care about 
deeply. In the early days at Upstart, we conducted a 
retroactive study with a large credit bureau, and we uncovered 
a jarring pair of statistics: Just 45 percent of Americans have 
access to bank quality credit, yet 83 percent of Americans have 
never actually defaulted on a loan. This is not what we would 
call fair lending.
    The FICO score was introduced in 1989 and has since become 
the default way that banks judge a loan applicant, but in 
reality, FICO is extremely limited in its ability to predict 
credit performance because it is narrow in scope and inherently 
backward-looking.
    At Upstart, we decided to use modern technology and data 
science to find more ways to prove that consumers are indeed 
creditworthy, to bridge that 45 percent versus 83 percent gap. 
We believe that consumers are more than their credit scores, 
and by going beyond the FICO score and including a wide variety 
of other information such as a consumer's employment history 
and educational background, we have built a significantly more 
accurate credit model.
    While most people believe a better credit model means 
saying no to more applicants, the truth is just the opposite. 
Because Upstart's model is more accurate, we have significantly 
higher approval rates and lower interest rates than a 
traditional model.
    But we also understood that consumer protection laws 
weren't to be taken lightly. Thus, we proactively met with the 
appropriate regulator, the Consumer Financial Protection Bureau 
(CFPB), before launching our lending program. After several 
years of good-faith efforts between Upstart and the CFPB to 
determine the proper way to measure bias, we demonstrated that 
our AI-driven model doesn't result in unlawful disparate impact 
against protected classes of consumers.
    Because AI models change and improve over time, we 
developed automated tests with the regulators' input, in order 
to report on the impact of our credit decisions across 
underserved groups on a quarterly basis. We have been providing 
this information to the CFPB for the last 18 months. Moreover, 
we were able to report to the CFPB that our AI-based system 
improved access to affordable credit; specifically, our model 
approves 27 percent more consumers and lowers interest rates by 
3.57 percentage points compared to a traditional lending model. 
For near-prime consumers in the 620 to 660 FICO range, our 
model approves 95 percent more consumers and reduces interest 
rates by 5.42 percentage points compared to a traditional 
model. And, most importantly, Upstart's model provides higher 
approval rates and lower interest rates for every traditionally 
underserved demographic. That is the type of consumer benefit 
we should all get excited about.
    In September 2017, Upstart received the first ever no-
action letter from the CFPB, recognizing that Upstart's 
platform improves access to affordable credit without 
introducing unlawful bias. The concern that use of alternative 
data and algorithmic decisioning can replicate or even amplify 
human bias in lending is well-founded. However, in Upstart's 
experience, the fair-lending laws enacted in the 1970s and the 
substance of fair-lending enforcement, that is, monitoring and 
testing the impact on actual consumers who apply for loans, 
translates very well to the AI-driven world of today.
    But in reality, the path we walked at Upstart is 
insufficient to create a robust and competitive market that 
will maximize financial inclusion and credit access. In our 
early days at Upstart, we couldn't know for certain whether our 
model would be biased. It wasn't until loans were originated 
that we were able to demonstrate that our platform was fair. As 
an early-stage startup, this was a risk worth taking, but it is 
not a risk a large bank would have considered.
    If broader and deeper financial inclusion among American 
consumers is important to this committee, it is worth 
considering rulemaking or legislation that will provide some 
type of limited sandbox for model development and testing. By 
combining regulatory support with model innovation, with 
rigorous and standardized testing, we can ensure that we don't 
forego the clear and obvious benefits that AI and alternative 
data can offer to the American consumer. Thank you.
    [The prepared statement of Mr. Girouard can be found on 
page 54 of the appendix.]
    Chairman Lynch. Thank you. I now yield myself 5 minutes for 
questioning. Thank you all. This is a great group. One of the 
nicer things about this task force is that it is bipartisan, 
and we are here for the same purpose: We are looking for 
guidance. We have an assortment of issues that we are 
confronting. Obviously, the banking industry is transformed, I 
think, because of technology so that we have an old banking 
culture that is very much rule-based, and it seems to be 
merging or morphing into this sort of tech hybrid where you 
move fast and break things. And so, there is that clash of 
cultures.
    But I can generally group our concerns into four areas. One 
is the whole issue of companies vacuuming up this personal 
data, this behavioral surplus, as Shoshana Zuboff describes in 
her book, ``The Age of Surveillance Capitalism.'' And under 
what conditions do consumers have a choice in terms of what 
gets vacuumed up and what gets used in terms of the algorithms 
that are employed to judge their creditworthiness or on other 
matters. There is that whole permission aspect that Ms. Wu and 
Ms. Johnson both raised. Actually, all of you, I think, 
addressed that in some regard.
    Then the data use, how that gets used, what data is 
permissible to use and what is not. Then, one of the concerns 
that this committee has is regarding the security of that data. 
We had Facebook in, and Mr. Marcus, who is heading up their 
Libra cryptocurrency project. And it is obvious from our 
history with Facebook, that Facebook does not do privacy well, 
and so we worry about that. If you look at the terms of service 
agreement, the one that is on your phone with Facebook, it is 
about 20 pages long. And if you look at it closely, it 
basically is the opposite of a privacy agreement. It basically 
gives Facebook the ability to gather all your data and then 
sell it to their advertisers. And if you don't agree, if you 
don't click, ``I agree,'' you don't get Facebook. So, I am 
worried about fintechs using that same sort of adhesion 
contract to get people to surrender their data, in order to get 
the value of what Mr. Girouard has described, which is perhaps 
lower rates, better access to credit, all the benefits that 
might flow from one of the fintech lenders.
    And then, lastly, we are struggling with how to hold people 
accountable with financial data. Should there be--I asked Mr. 
Marcus, but he wasn't forthcoming with an answer--I asked him, 
I said, would you accept fiduciary liability for the 
mishandling of consumers' personal financial data because of 
the consequences that can occur because of that mishandling?
    So, Ms. Wu, let's talk about, how do we get into this? How 
do we introduce this permission regime where people can--and, 
Ms. Johnson, I will go to you on this as well--how do we 
introduce this? Right now, it is a permissionless vacuuming up 
of data. How do we change the paradigm and the model from what 
we have now to a more rule-based, if you will, structure with 
some of the fintech that is emerging?
    Ms. Wu. Thank you for the question, Chairman Lynch, it is 
an excellent one. There is the sort of limited issue of 
alternative data for credit purposes, where we would urge that 
any legislation always be on an opt-in basis, that consumers 
have a choice, and that that choice be real and meaningful, 
that it not be in mice type of 20 pages of fine print that you 
mentioned. From a broader perspective of privacy in general, 
yes, we all should have more control over our own data, the 
right to opt in, opt out, or even have our data deleted.
    Chairman Lynch. Very good.
    Ms. Johnson?
    Ms. Johnson. I think that Ms. Wu's point is absolutely 
consistent with what our expectations ought to be. I think the 
challenges are two-part, one part technical, so I might defer 
to Mr. Girouard to respond as to how their model might address 
this very specific and technical point. But for AI to be 
effective, as I describe in my written testimony, there has to 
be a certain quantity or volume of observations available. They 
have to be uniform to a certain extent, and that facilitates 
the learning algorithm's ability to work through the data in a 
manner that is exceptionally efficient and reduces operating 
costs, thereby enabling fintech lending platforms to reduce the 
cost of borrowing for consumers.
    One of the challenges I am very curious about how we will 
navigate is the extent to which we are rightly asking that 
consumer's consent prior to their data being used, and how we 
reconcile that with how machine-learning algorithms operate. 
So, I think there is a gap there that we have to have enough of 
a conversation about, to be successful in crafting regulation.
    The other thing I just mentioned really quickly about 
consent is that the extent that the data is being gathered 
really may be the point of departure for some of our concerns. 
In many instances, consumers are completely unaware that the 
data is being gathered. And in some instances, they are 
voluntarily giving the data over for the better credit 
opportunities or reduced price credit opportunities, which is 
disconcerting, to be quite honest, because it suggests those 
who are most vulnerable might be exposed to--or exploited, in 
fact, by arrangements whereby they share the most intimate 
details of their financial lives or their personal lives for 
the purpose of getting better access to credit.
    Chairman Lynch. Thank you very much. I now yield to my 
friend, the gentleman from Arkansas, Mr. Hill, for 5 minutes.
    Mr. Hill. Thank you, Mr. Chairman.
    And, again, thanks to the panel. This is another really 
excellent panel that has been assembled for the task force 
work, and I think all of you bring a great perspective.
    Certainly, this issue of customer choice is an important 
one, and we all are frustrated, I think, with moving away from 
passwords into a more robust authentication, which is critical 
to a digital world, critical to fintech being successful, 
whether you are working at the biggest bank in the country or a 
great startup. We need to get beyond ``password1'' and our name 
as authentication, and we have been talking about that a lot.
    Secondly, this issue of, I own the data, I am the consumer, 
and I am allocating you some data for a project we are working 
on together, and so broadening that transparency in access to 
my data for the purpose of taking a decision that I want to 
have with an online partner. These are really important areas 
and thank you for bringing those up.
    Mr. Girouard, I want to talk a little bit about your model 
and the alternative use or, as was described by Mr. Rieke, your 
expansion, I would say, of conventional data. I will ask if you 
use ``fringe data'' or not, as he defined it, but we will find 
out. But I am very impressed that you have been working 18 
months with the CFPB, which is a beloved institution in 
Washington and certainly in this committee and, therefore, has 
imminent authority over that relationship. And congratulations 
for having a no-action letter. We think that is a great 
improvement for CFPB operations as an absolutely serious 
comment and a great way for them to demonstrate the ability for 
fintech innovation in a mini sandbox if you want to call that a 
derivative of that.
    My first question is, the conventional data you expand 
beyond FICO, what is the nature of that in your business?
    Mr. Girouard. Sure. I want to say first thatthe data we use 
in our models comes entirely from two sources: one, is a credit 
reporting agency; and two, is directly from the consumer 
themselves. So we aren't ``hoovering data in many places.'' We 
don't take data from Facebook, et cetera. What we to do is 
include information--and I had mentioned a couple of them--
somebody's work history, where do they work, are they a nurse, 
are they a policeman, et cetera, their educational history, the 
degree of education obtained, their area of study. These are 
things that are unique to our model. We also look at some 
behavioral things when they interact with us, what sort of--how 
much--what size loan do they ask for, how did they find us, 
things of this nature. These all end up being helpful and 
predictive toward our model.
    Mr. Hill. Do you consider that--of course, it is provided 
by the customer. They are seeking the loan, so they have 
granted you permission to do that. Are you also seeing their 
cash-flow data for a period of months by access to their bank 
account in making your determination?
    Mr. Girouard. Today, that is not something we do. We do 
request and, with consent, get access to a bank account really 
for verification purposes and to avoid fraud and such. But it 
is, as of today, not part of our credit decisioning.
    Mr. Hill. You talked about how you are doing that, and, of 
course, the CFPB is learning, too, and you keep some of your 
credit, and your partner bank has some of your credit 
originated by you on their books, and then you securitize 
credit. So, for the loans that you keep and for the loans that 
are on the bank's books, of course, those are being reviewed by 
compliance officials for compliance with all fair lending laws 
and the like? Isn't that right?
    Mr. Girouard. Sure. There are many layers of oversight and 
governance over what we do. The vast majority, almost all 
Upstart loans are originated through bank partners--some of 
which are FDIC-regulated, and some of which are OCC-regulated. 
So, we are beholden to all of them and go through very regular 
audits and such.
    Mr. Hill. What is your view of what statutorily ought to 
change about the creation of a sandbox at our bank regulatory 
agencies? What does that mean to you? I see it in your 
testimony. You don't really explain what you mean by that. How 
do you define it?
    Mr. Girouard. Our belief, as I said, is that the right way 
to handle regulation for alternative data, and the use of 
alternative data is actually to measure the outcome, to look at 
its impact on consumers and whether there is bias in the 
outcome. The challenge with that, and the way the world works 
today, is, you don't know until you originate the loans. So, 
you are taking on some risk that, during that period of 
evaluation of building and testing that model, you could be in 
violation of the law, of fair-lending laws. The sandbox concept 
is, how do you actually make progress there? How do you 
actually build a better model that is both more effective and 
more accurate, but also fair and unbiased without testing and 
moving? And the notion of a sandbox is to provide some freedom, 
not just for a startup like we were 5 years ago, but to a large 
financial institution, a bank, to do the same thing.
    Mr. Hill. This is like a phase one or a phase two clinical 
trial in the drug research industry. How long do you think that 
would take and how much of a Big Data set would that be, in 
just your world of personal lending, do you think would be 
necessary to prove out a concept like that, analytically? 18 
months?
    Mr. Girouard. Yes. That is--
    Mr. Hill. Do you look at it in time, or do you look at it 
in total data set, or both?
    Mr. Girouard. It is a little of both.
    Mr. Hill. Because you have to go through the economic cycle 
of these borrowers, to some degree, some seasoning of these 
borrowers.
    Mr. Girouard. That is really about the efficacy question, 
meaning, does this model work well? But the fairness question 
actually is answered quite quickly because you know right away 
who you are approving andwho you are not approving.
    Mr. Hill. Thank you. I yield back, Mr. Chairman.
    Chairman Lynch. The Chair now recognizes the gentleman from 
Georgia, Mr. Scott, for 5 minutes.
    Mr. Scott. Mr. Evans, let me start with you, because in 
your testimony you provide a very good survey of the literature 
of the potential benefits of alternative data, but you also 
mention the risks. First of all, I think it would be helpful if 
you gave us some examples. What are we talking about when we 
say alternative data? What would that be?
    Mr. Evans. This could range from data that we have had 
significant experience with, like on-time rental payments, 
mobile payments, and the like. But it could also be data that 
we glean from your digital footprint online or your browser 
history.
    Mr. Scott. But these data points also must uphold the fair 
lending laws and standards that we have in place. I think the 
critical question is, how do we strike the balance? How do we 
strike the necessary balance, particularly given the innovative 
nature, the rapidity of our technology moving?
    Mr. Evans. Excellent question. And there are two things 
that I would point out from our body of work. One, we looked 
across the globe, and we looked at some of the innovative 
things other countries were doing, and they were things like 
the regulatory sandboxes, and innovation offices. We have to 
understand the technology, and the way to understand the 
technology is to engage. We have recommendations that are open 
to regulators to make sure they are carefully thinking through 
whether these innovation offices and other types of knowledge-
sharing initiatives would be appropriate here in the United 
States.
    Also, guidance is extremely important because it sets the 
rules of the road. It sets parameters. And if the fintech firms 
aren't getting that kind of guidance, they are not--
    Mr. Scott. And do you think the regulators are living up to 
that? Do you think they are giving this guidance properly now?
    Mr. Evans. I would say no. There are certainly places where 
you can find good information from the Federal Reserve and 
others, but they haven't communicated this guidance in a 
written, formal way, so that people understand that this is 
relevant guidance for firms to follow. When you get many 
touches across the fragmented regulatory system, it is helpful 
to know that the guidance is coordinated; it is not coming from 
just one regulator.
    Mr. Scott. Ms. Johnson, you said something in your 
statement that I agree wholeheartedly with: you said that 
credit is a critical choice. It is almost a life-and-death 
choice. Can you imagine not having a checking account? Not 
having a savings account? Not having a credit card? Not having 
any history in this time? And yet, we have almost 60 million 
Americans in that shape. How critical, in your words, is this, 
at this point, with our unbanked, and if we fail in this 
ability to make the alternatives work, what would that look 
like? How serious is this situation facing hese 60 million 
unbanked, or what you refer to as invisibles, and making them 
visible?
    Ms. Johnson. This is a great question. Thank you. I think I 
might dissect or sort of bifurcate the question into two parts, 
one part just being thoughtful at the outset about the idea 
that credit, as we are describing it, originates from--or the 
decision-making process, or determinations about credit, 
originates from an evaluation of eligibility, right? The notion 
that credit reports are used merely for credit is mistaken. We 
know that credit reports might be used in other processes to 
determine employment and access to other resources. So, in some 
instances, we are talking about credit and the data that is 
evaluated to determine whether or not someone has access to 
credit, as a gateway. This is a sort of a gateway to a variety 
of critical access, to a variety of critical and important 
resources in our society. Credit is a critical resource and 
credit reports are a critical factor in the lives of 
individuals because it may impact their ability to access other 
resources beyond credit, right?
    Mr. Scott. Yes.
    Ms. Johnson. That is the first point, just to segregate out 
the ideas that what we are evaluating here, the data that is 
being gathered, there are many important impacts with respect 
to that data, that are beyond just simply whether or not one 
qualifies for a credit card or a home mortgage loan. Although, 
access to those resources is important as well.
    I would also underscore--Congresswoman Porter was one of my 
colleagues in the academy before joining you all here on the 
committee and in Congress, and her work has historically, along 
with others, underscored the significance of the financial 
status of individuals as impacting a variety of elements of 
their lives, and your point underscores that as well. I just 
suggest that credit and the data that is being gathered for the 
purposes of evaluating credit will impact access to financing, 
but it impacts access to a number of other things, including 
education.
    Mr. Scott. And, Mr. Chairman, may I just ask this--one of 
the values of--
    Chairman Lynch. The gentleman has gone a minute-and-a-half 
over. Go ahead, though.
    Mr. Scott. Thank you. One of the values of the fintechs is 
that they are now providing help and services to the unbanked 
that our traditional banks are not doing, will not do. And I am 
not going to ask you to answer that, but I am sure you will 
agree that that is an area we can develop more of, to use our 
emerging fintechs to be a valuable asset, because many of the 
existing actors in the financial services industry are not 
going to touch these unbanked and underbanked. But, anyway, 
thank you.
    And thank you, Mr. Chairman. I'm sorry.
    Chairman Lynch. Quite all right.
    The Chair now recognizes the ranking member of the full 
Financial Services Committee, the gentleman from North 
Carolina, Mr. McHenry, for however much time he may consume.
    Mr. McHenry. I will respect the Chair. Thank you, Mr. 
Lynch. And thank you, Mr. Hill, for your leadership.
    It is my hope that this task force can--we can build some 
consensus around financial technology. This is a nonideological 
space in an otherwise highly polarized Washington. And I think 
it shows that we can use technology to get better societal 
outcome--well, the same or better societal outcomes that we 
seek in current law.
    We have very important provisions of law that have been put 
in place through a massive amount of work to ensure that we 
don't discriminate against people based off of what I would 
describe as superficial reasons. And that work, where you are 
located, what you look like, who your parents were, any of that 
stuff, right?
    And what we see now in China is that you have this--you 
have a social score as well. And it is political connections 
and all of this stuff. And I hear this underlying the whole 
panel, we don't want that. Just because you tweet and you are a 
jerk on Twitter doesn't mean you are uncreditworthy. Or if you 
follow nuts on the left or the right on Twitter, that should 
not make you more or less creditworthy.
    Getting into the fundamentals of this, how you use 
alternative data, Mr. Girouard, you brought this up. Let's talk 
about the sandbox approach that Mr. Hill brought up in his 
question.
    So, the question of innovation and financial inclusion, I 
think, should go hand-in-hand. What are the benefits of a 
sandbox approach, Mr. Girouard, in your view?
    Mr. Girouard. As someone who has gone through the process, 
as we did over 4 years, frankly, with the CFPB, the sandbox 
isn't to our advantage. We already walked the walk and walked 
over the coals.
    But honestly, in the interest of the American consumer, you 
want a robust environment where not just small companies but 
the largest banks have an opportunity to innovate in modeling 
and in credit decisioning, because it can only benefit the 
consumer.
    A sandbox is necessary because--let me just give an 
example. In the very early days of our lending, I met with the 
CEO of one of the top banks in the country, one of the largest 
card issuers in the country, and his words to me were, ``I love 
what you are doing. I am really glad you are doing it, because 
we will never be able to do that.''
    And I think honestly, it may be to my business advantage 
that that is the case, but it is not to the American consumer's 
advantage. We need innovation across the industry, not just in 
unsecured personal loans, but in mortgages, in auto lending, in 
HELOCs, in all flavors of credit.
    Mr. McHenry. What will the benefit be if you use 
alternative data and somebody has, under a traditional score, 
less than A-plus credit, but you see through alternative data 
that they actually pay their rent, they pay their cellphone 
bill, and they have never missed those payments, it enhances 
that credit score, right?
    Others, it would actually say that that credit score is not 
as good, because they are not paying or they continue to have 
issues.
    There is this picking and choosing, when you say, we only 
want to use good stuff, if it is alternative data. Well, that 
is not representative that everyone is a good credit risk, 
right? How do you prove that out in terms of ensuring it is not 
discriminatory based off of our traditional metrics under 
Federal law?
    Mr. Girouard. Let me just say, the important background is 
that FICO and income, which are the two anchors of almost any 
lender, are terribly biased. And they are so biased that the 
additional of alternative data, whether that is education, 
whether that is the name of the company you work for--there are 
a variety of other things--actually reduces the bias and the 
credit decisioning, because the baseline is so biased itself. 
That is why it represents such an opportunity.
    The other really important--
    Mr. McHenry. Okay. Across the panel, does anybody disagree 
with that statement?
    Ms. Johnson. I would add something.
    Mr. McHenry. But any disagreement with the contents in the 
last 5 sentences of what Mr. Girouard said?
    Anyone on the panel?
    Ms. Johnson. There is bias certainly in the existing data, 
because it is the result of systemic--we just talk about data 
collection for algorithms generally. We have to acknowledge 
that at the outset, the data that is being collected is biased.
    One of the best and easiest, most accessible examples, 
would be in criminal law enforcement. To the extent that an 
area is overpoliced by police in a particular city or area, 
there will be more arrests in that area--
    Mr. McHenry. No, but I am talking about consumer credit, 
and I am talking about the specifics of this. That is a larger 
societal issue. We are the Financial Services Committee and not 
the Judiciary Committee. That is a major issue; I certainly 
understand that. And I appreciate that.
    But let's talk about what we are going to fix here in the 
Financial Services Committee. When you say that alternative 
data can be an enhancement--and I understand all of the caveats 
that all of you in a very loyal sort of way, if I would say, 
say, yes, it has great opportunities but there are risks. Of 
course there are, right?
    But when we are talking about getting unbanked or credit 
invisible people and making them visible, I think that is a 
proper societal tradeoff in order to get more people into the 
world of being banked, rather than underbanked or unbanked.
    And so, I appreciate the hearing. And with that, Mr. 
Chairman, I yield back.
    Mr. Scott [presiding]. Thank you, Mr. McHenry.
    The gentleman from New Jersey, Mr. Gottheimer, is 
recognized now for 5 minutes.
    Mr. Gottheimer. Thank you, Mr. Chairman, and thank you to 
all of the witnesses for being here today. I appreciate it.
    Traditional information used to make lending decisions and 
establish credit scores often does not account for the 26 
million customers and consumers without a credit history or the 
19 million consumers with a short or outdated credit history to 
form a credit score, groups that are often labeled as thin file 
or credit invisible.
    Thankfully, lenders and CRAs have started using alternative 
data to make lending decisions, determine credit scores, and 
expand consumers' access to data.
    I personally believe that this is the future in the era of 
renting and Venmo and Uber, that we need to give the next 
generation of consumers the ability to build a stronger credit 
file through nontraditional data sources. That is why I am 
working on the Credit Access and Inclusion Act, legislation 
that would allow the reporting of certain alternative data like 
rent and telecom payments to consumer reporting agencies to 
help thin-file consumers build their credit scores and 
hopefully access credit.
    We also must ensure traditional credit bureaus and those 
using alternative financial service data still comply with the 
Fair Credit Reporting Act, also known as FCRA.
    Ms. Wu, if I can start with you, how can we ensure that 
alternative data sources comply with FCRA data furnishing 
requirements?
    Ms. Wu. Thank you for the question, Congressman Gottheimer.
    One of the things we need to clarify is that any time 
third-party data is used for credit decisioning, it should be 
covered by the Fair Credit Reporting Act.
    The example of Facebook, for example. Facebook may have a 
disclaimer in its website saying you are not supposed to use it 
for credit. But if they are doing it wink, wink, nudge, nudge, 
and lenders are using it for credit, it should be covered by 
the Fair Credit Reporting Act.
    And so Congress should clarify that, but I also want to say 
in the area of sandboxes, the devil is also in the details. 
Sandboxes shouldn't be a license to ignore things like the Fair 
Credit Reporting Act and the requirements for accuracy, 
predictiveness, and notices.
    Mr. Gottheimer. Thanks for your answer.
    Just a follow-up to that, what kinds of alternative 
information would you seek to use that is not already shared by 
applicants or regularly requested as part of loan applications, 
rental payments, bank statements, and, of course, under the 
Fair Credit Reporting Act?
    Ms. Wu. First of all, the most important aspect is consumer 
choice. The consumer should be allowed the option of sharing it 
or not. So if they want to share their bank account data, if 
they want to share their utility payment or rent payment data, 
they should be permitted to. But if they don't want to, if they 
want to say, hands off my data, that also should be respected. 
And then the lender should consider that in the same way they 
consider credit data.
    The other side of this equation of alternative data is, are 
the lenders actually going to use it? We have seen lenders who 
won't even upgrade to the latest FICO model, let alone use an 
alternative score. So, one of the tough parts is actually 
getting the lenders to look at it.
    And I think one of the things that this committee has done 
that is useful is passing Chairman Lynch's bill giving the CFPB 
authority to regulate the scoring models. We have heard from 
Mr. Evans that there needs to be guidance from the regulators. 
The best thing to do is have the experts at the CFPB review 
these models and ask, is this predictive, is this accurate, 
does this create disparate impact? And the bill that this 
committee passed does that.
    Mr. Gottheimer. Do you see that changing with some of the 
financial institutions? I know that many aren't considering 
other datasets. Do you see that changing? Is there a desire 
to--how is the trend line on that? What do you think would 
really spur that along?
    Ms. Wu. I think the things that will spur it along are 
things like, Fannie Mae and Freddie Mac are going to be needing 
to update their scoring models, and we have actually encouraged 
the use of pilots, limited pilots with alternative scores.
    Mr. Gottheimer. And are we seeing good news out of that? 
Are we getting more access to credit for people? I really am 
grateful for your leadership in this space, because I think it 
is very, very important that more people have access who should 
get it, who qualify for it, but just because of traditional, 
the way we have done things forever, they are not getting 
access to it, or because it is so black box that you don't know 
what is in it. And I think that lack of transparency also has a 
big impact.
    Ms. Wu. Fannie and Freddie have not adopted the new scoring 
models yet, but some of the other testing that has gone on has 
shown some promise.
    Again, the devil is in the details. We need to be careful. 
There is going to be some disparate impact. But the thing about 
the disparate impact test is, it doesn't say, okay, there are 
some racial disparities you have to stop. Are there more racial 
disparities or less? Is it predictive? Predictiveness is so key 
here. And if it is not predictive, you shouldn't be using it.
    Mr. Gottheimer. Thank you so much. I yield back.
    Mr. Scott. Thank you. The gentleman from Ohio, Mr. 
Davidson, is recognized for 5 minutes.
    Mr. Davidson. Thank you, Mr. Chairman. I thank our 
witnesses, and I thank all of my colleagues for thoughtful 
questions and good dialogue. And hopefully, this will yield 
some progress in this really important space.
    Mr. Girouard, I want to follow up where Mr. Hill left off 
when he was talking with you about how much time would this 
take and how would a sandbox work in a regulatory framework 
where we have maybe provided certainty for this path with 
legislation.
    And in your response to him, you said, well, we don't 
really need 18 months; you can know pretty quickly whether it 
is discriminatory or not; i.e., is it working? And I just want 
to pick up from there, because it seems incomplete.
    Because if you give credit to everyone at low rates or, 
say, free, it is not discriminatory; it is all free to 
everyone, whomever shows up, or it is a fixed rate for 
everyone, no matter what, it is not discriminatory.
    But if there is a massive default rate, it really doesn't 
work, right? You do care about defaults, correct? How far into 
that process could we know is it both nondiscriminatory and 
actually effective in the sense that it provides a useful tool?
    Mr. Girouard. That is a good question. It certainly varies 
based on the nature of the product. A mortgage, for example, 
plays out over many, many more years. But you do need enough 
data, you do need to understand both fairness and efficacy. 
Fairness can be sorted out fairly quickly. Efficacy takes time. 
You need to see how a loan performs.
    Mr. Davidson. Is it really fair to give money to somebody 
who has no hope of repaying it?
    Mr. Girouard. No, it is actually against everybody's best 
interest to do that. Ability to repay--
    Mr. Davidson. Efficacy is inherently linked to fairness is, 
I guess, the point. And so, I am just curious. If you look at 
probabilistic models and you look at the statistics and say, 
hey, if you have this pattern, is there a dataset that shows 
what the--95 percent certainty, 99 percent certainty, what 
range of probability of payment history in the early years, 
could you say the sandbox has produced an effective tool so 
that it is both nondiscriminatory and it is efficacious?
    Mr. Girouard. Congressman, you are asking exactly the right 
questions. The sandbox has to be defined in a way that allows 
the lender to decide if this new model works. And it won't be 
the same sandbox for every type of credit product for a variety 
of reasons--
    Mr. Davidson. Okay. That gives me concern, because there is 
no real hope to pass a law that could provide certainty. It is 
essentially like, go negotiate your own deal with a regulator.
    Mr. Girouard. With all due respect, I think rulemaking 
could absolutely define a sandbox in terms of number of loans, 
how long the sandbox can operate for, the total dollars in it. 
There is no question in my mind that a reasonable process could 
define rules that put a sandbox in place for the major areas of 
credit for consumers. That would make a significant improvement 
in the ability to see innovation in this area.
    Mr. Davidson. Yes. Thanks for your expertise, and I 
appreciate your experience in the matter.
    Mr. Rieke, your background in privacy at the FTC is 
interesting, because so much of this links on privacy. And in 
the United States, particularly in banking, with Gramm-Leach-
Bliley, financial institutions have a carveout where they treat 
data differently.
    In a way, financial institutions, and frankly all sorts of 
institutions, if they were looking at their balance sheet, they 
might treat their dataset as a valuable asset. Consumers, 
however, don't necessarily realize that some places they are 
considered to have a property right in their data. Is it an 
asset for both?
    And as people give up this data, one of the concerns is, 
how do we reconcile the de facto impact of GDPR and the looming 
patchwork of privacy laws coming in the United States and 
Congress' failure to act on privacy with that framework so that 
consumers can control their data some and not find themselves, 
well, wait, I was denied credit. Well, yes, you blocked all 
access to your background, if you go to the far end. And on the 
other hand, the idea that, gee, if you click these terms and 
conditions, anything that is in it is fair game.
    How do we regulate privacy in this space with respect to 
credit?
    Mr. Rieke. That is a great question. I think the FCRA is a 
strong start. If you squint at the text of the FCRA, what comes 
out of that is if your data is used for important eligibility 
purposes, certain rights and protections attach.
    Now, the FCRA is pretty old now. And as Ms. Wu said, if I 
am giving permission to Facebook to hand my data over to a 
lender, it is questionable whether that framework would attach. 
But I think looking at the spirit of the FCRA, which was 
created especially for these concerns and were some expansion 
so that statute might make sense for the digital age, would be 
where I would start.
    Mr. Davidson. All right. Thank you. My time has expired and 
I yield back.
    Mr. Scott. Thank you.
    And now the gentlewoman from Virginia, Ms. Wexton, is 
recognized for 5 minutes.
    Ms. Wexton. Thank you very much.
    And thank you to the panelists for coming today. This is 
really fascinating, and you are giving us all a lot to wrap our 
heads around.
    Mr. Girouard, I am really interested in your model and 
especially the fact point--the datapoint that it reduces 
interest rates by 5.42 percentage points and approves 95 
percent more consumers in that near-prime area.
    What kind of response are you getting from lenders about 
your model? Are they enthusiastic about it?
    Mr. Girouard. By lenders, do you mean banks we partner or 
mean to partner with?
    Ms. Wexton. Yes.
    Mr. Girouard. Thank you, Congresswoman. I would generally 
say there is a lot of excitement about the potential for a 
model like this to be able to serve more customers, to be able 
to build on their side, lower the risk of lending. A more 
accurate model is intuitively compelling to a bank officer.
    Having said that, there certainly remains a lot of concern 
about regulatory uncertainty. And there is not in any sense a 
clear-eyed statement or a sense from the regulators how to 
think about this area of technology to a bank. A no-action 
letter that we received from the CFPB is a great start. It is 
not by any sense a panacea, because there are many other 
regulators. There are many limits to a no-action letter, so 
there is plenty of room for either regulatory action or 
rulemaking to provide more clarity.
    Ms. Wexton. I understand that there is some question about 
regulatory certainty. But are the lenders willing to accept 
that your model is a more accurate credit reporting model?
    Mr. Girouard. I think I can comfortably say yes. I am 
almost universally seen acknowledgment that our model is more 
accurate and more inclusive.
    Ms. Wexton. Okay. And Ms. Wu had indicated that one of the 
things that we should consider is making any of these 
alternative datapoints that are being used for credit to be 
considered as a report under the FCRA.
    Would that impact your ability to create this algorithm, or 
is that something that would not be an issue for you?
    Mr. Girouard. FCRA is to cover third-party data, data 
reported to--and then can be shared with a lender. And that is 
one part of our data. The other part, which is important to us, 
is the data that a consumer, with our consent, with their 
consent, submits to us.
    And again, that can be--if they are stating their income to 
you. That is not something generally a credit reporting agency 
has information on.
    There will always, at least in my mind, be two paths for 
data to come to a bank and a lender, one through FCRA-related 
data, through credit reporting agencies, and the other provided 
by the consumer themselves. And they are both important.
    Ms. Wexton. Okay. Thank you.
    And, Ms. Wu, you had also indicated that there should be an 
opportunity for consumers to opt out of these alternative 
datasets being used for credit purposes, is that correct?
    Ms. Wu. Thank you for the question, Congresswoman.
    I actually would urge that it would be an opt-in process, 
that any time you are creating these large new datasets, 
consumers give their written authorization to have their 
utility or their bank account information included, to be 
considered.
    Ms. Wexton. So, they would have to affirmatively opt in--
    Ms. Wu. Yes.
    Ms. Wexton. --and then get it used.
    Okay. And I guess a part of that would be a declination or 
a refusal to opt in could not be used against them, right? It 
wouldn't factor into the algorithm, but it wouldn't be down 
counted for not--
    Ms. Wu. If they already have a traditional credit file and 
score and they decline to opt in to alternative data, we would 
say the lender should go ahead and use the traditional credit 
score. If they don't opt in, then the data can't be used, 
obviously.
    Ms. Wexton. All right.
    Ms. Johnson, as a law professor, I know that you are 
familiar with the difference between de jour discrimination and 
de facto.
    Is there a way to be proactive in this space and make sure 
that we don't end up with de facto discrimination in these 
algorithms, or is it always going to be retrospective, looking 
back and seeing what the analysis provides us?
    Ms. Johnson. Thank you for the question, Congresswoman.
    I think that there is a way for us to be thoughtful in 
advance of the release of these types of products in financial 
markets.
    I think earlier, Chairman Lynch referenced the ``move fast, 
break things'' mantra that was adopted by a number of 
technology firms, and now as fintech firms are entering into 
spaces and operating, as Mr. Girouard mentioned, without clear 
regulatory guidance, there will be a temptation to use 
information or data, alternative data, to facilitate what may 
be faster, more efficient, lower-cost credit evaluation 
processes.
    We do have some knowledge in advance of the types of data 
that tends to lead to bias or discrimination, based on a long 
history of legislation and court decisions and agency actions 
in this space.
    I think one of the things we can do is really identify red 
flags and target areas. Some of the data Ms. Wu mentioned 
earlier and has been talked about over the course of this 
hearing, that it is useful and be thoughtful about would be 
rental payment history, but there are any number of reasons 
why--and Ms. Wu's organization and others have thought about--
that information may disadvantage or utility bill payment may 
disadvantage certain--
    Mr. Scott. Ms. Johnson, the time is running out.
    Ms. Johnson. Thank you very much.
    Ms. Wexton. I yield back.
    Mr. Scott. Thank you very much.
    The gentleman from Missouri, Mr. Luetkemeyer, is now 
recognized for 5 minutes.
    Mr. Luetkemeyer. Thank you, Mr. Chairman.
    Mr. Girouard, since all banks are required to follow the 
ECOA and you partner with a lot of banks, what due diligence 
and ongoing monitoring does your company provide your bank 
partners to ensure that 100 percent certainty for those banks 
of no fair-lending violations?
    Mr. Girouard. Sure. That is a very good question. For sure, 
providing this technology to banks is not for the faint of 
heart. There is what I would say is a process of probably more 
than a year of them getting to understand and do diligence on 
our processes, fair lending being just one of many, to make 
sure that loans originated using this type of system are within 
the law. And also, of course, that the creditworthiness is 
real, the efficacy of the model is real. So, there is real, 
significant work before anything happens, before any 
relationship is signed.
    After the fact, there is a constant reporting and auditing 
like function. The same report that we provide for CFPB for all 
loans, we can do for an individual bank. And that gives the 
bank comfort that we are actually monitoring on a regular basis 
to make sure the loans originated in their name, under their 
charter, are within the bounds of fair lending regulation.
    Mr. Luetkemeyer. I would like to follow up on the previous 
colleague's questions here with regards to the no-action 
letter.
    I am assuming that because you have a no-action letter, it 
is very helpful when you go approach other banks to become 
partners with them. Because it would sort of seem like you 
are--it is a get-out-of-jail free card from the standpoint that 
you have already been sort of preapproved by CFPB, that the 
modeling you are doing is something that falls within the 
guidelines of everything.
    How important is that no-action letter whenever you start 
negotiating with the other entities?
    Mr. Girouard. It is certainly very important. And the 
reason we were willing to invest information and be as 
transparent as we were for several years with CFPB, I think, it 
is important because it demonstrates to banks that we are not a 
``move fast and break things'' company. That may be the name--
or sort of a label you want to paint Silicon Valley startups 
with. But we are not in that class. We are a company that takes 
regulation and working transparently with regulators seriously.
    However, as I said earlier, it is absolutely not a panacea. 
They care about the FDIC, they care about the OCC, they care 
about State regulators, all of whom could decide to accept the 
CFPB's no-action letter and its conclusions or could choose not 
to. And that is why I think ultimately it is important to 
clarify regulation.
    Mr. Luetkemeyer. Why do you think more entities like you 
have not gone the no-action letter route? There are not very 
many, if any, that have done this, is that correct?
    Mr. Girouard. There is none other to date, as far as I am 
aware.
    Mr. Luetkemeyer. Why do you think that you are the only one 
that has done this? It would seem to me to give you a marketing 
advantage from the standpoint--if I am a bank and you come to 
me and you say, look, I have already had my modeling fall 
within the guidelines of the CFPB and all of the other entities 
out here that are regulating this, and I will continue to put 
these processes in place to protect the integrity of our data, 
it looks to me like you have to a win/win there. Why is there 
nobody else doing that?
    Mr. Girouard. My only conclusion I can draw from that is 
one of a few things. Number one, they are not actually using 
alternative data in a meaningful way.
    Number two, they are using it, but they have found another 
way, another path to creating comfort that they are within the 
bounds of fair lending laws.
    Or, three, they are using it, but they are not using it 
responsibly. And I don't necessarily know which of those is the 
answer.
    Mr. Luetkemeyer. Very good.
    Mr. Evans, your testimony points out that CFPB has 
developed fair lending examinations related to credit models 
and the Federal banking regulators have issued guidance to the 
depositor institutions on third-party or vendor management, 
including fintechs. However, despite this regulatory framework, 
there seems to be a disconnect between lenders and fintechs and 
the regulators that provide uncertainty in the fintechs' place.
    Can you explain this?
    Mr. Evans. Well, yes. And I think it goes back to 
ultimately the fragmented nature of the regulatory system. 
Fintechs experience uncertainty in that regard, because there 
are a number of actors in that particular space.
    CFPB's position on one thing may differ from the Federal 
Reserve or the OCC's position. And so oversight of fintech 
lending requires significant coordination. And the knowledge-
building initiatives that I talked about in my opening 
statement would allow regulators to really understand the 
fintech products and ensure that the regulatory framework is 
adaptable and flexible.
    Mr. Luetkemeyer. Okay. Very good. I see that my time has 
expired. Thank you.
    Thank you, Mr. Chairman.
    Chairman Lynch. The Chair now recognizes the gentleman from 
Texas, Mr. Green, for 5 minutes.
    Mr. Green. Thank you, Mr. Chairman. And I thank you and the 
ranking member for hosting this hearing.
    I would also like to thank Mr. McAdams for allowing me to 
proceed at this time. In fact and in truth, it would be his 
turn, and he has allowed me to have the opportunity to proceed.
    I would like to move first, if I may, and rather 
expeditiously to Mr. Girouard.
    Sir, in the model that you currently utilize, do you 
maintain the traditional credit score and then do you add these 
other, what you are calling, alternative datapoints to the 
traditional score?
    Mr. Girouard. We do. We vary--we use FICO score. We use--
    Mr. Green. That is going to be enough, because I have a lot 
to cover. I appreciate it.
    Mr. Girouard. Okay.
    Mr. Green. Thank you. I don't mean to be rude, crude, and 
unrefined.
    Mr. Girouard. Not at all.
    Mr. Green. Okay. Thank you.
    Friends, I started with Mr. Girouard for a reason. What we 
are calling alternative data, in most circumstances--there may 
be some that I am not covering--is really additional data. It 
is additional data. My bill that I have is not about 
alternative data, alternative meaning one or another. It is 
about additional data. It is about what Mr. Girouard does when 
he takes the traditional data and then he adds what we are 
calling alternative, but it really is more data that we are 
adding. We are not leaving out the traditional scores.
    My bill does not require consumers to opt in. Consumers do 
this of their own volition. They can allow their additional 
data to be scored, and it can help a good many consumers, as 
evidenced by what Mr. Girouard has called to our attention.
    The bill is a bill that has metamorphosed. I confess that 
initially we used the term, ``alternative,'' but we soon 
realized that when people heard the term, ``alternative,'' they 
assumed that we were somehow going to negate what was already 
there as a traditional score.
    Now, understanding that we are talking about additional--we 
are talking about the utilities, we are talking about the rent, 
but we simply added to what is already there, and in doing 
this, I think we will give many consumers the opportunity to 
own a home, and to make purchases that they would not 
ordinarily be able to make.
    Those that don't opt in will not be--they won't have that 
traditional score in any way encroached upon, infringed upon. 
It won't have an impact on that. Only those who opt in.
    With that said, I want to give you an opportunity to ask me 
a question. Let's turn the tables, if you don't mind, so that 
we can become as clear as possible, perhaps perspicuously, so 
that there is a better understanding of what this bill is 
about.
    I am not going to debate persons who want to have an 
alternative credit scoring model. That is perfectly acceptable 
to me. I would only suggest that if we focus on this bill, that 
we use the term, ``additional credit scoring.''
    Questions from any member of the panel, please?
    Ms. Johnson. I have a question actually.
    Mr. Green. Thank you.
    Ms. Johnson. And Mr. Girouard may answer it, but it grows 
directly out of your question. Thank you, Congressman, for 
inviting us to ask.
    In the first instance, we have described credit invisibles 
as those who do not have a traditional credit score under the 
FICO criteria.
    To the extent that inclusion is our goal, which I think is 
bipartisan motivation for the committee and our thoughtfulness 
today--if inclusion is the goal and the idea that you propose 
is that alternative data is additional data supplementing an 
already robust methodology for analyzing consumer--the 
likelihood of consumer default or predicting creditworthiness, 
I am not sure I follow how credit invisibles are actually 
captured if the data that is being used is not the primary 
source of evaluation.
    Mr. Green. If I may answer, because there are only 32 
seconds left.
    You could be a great Member of Congress, by the way, with 
your question.
    Here is how they are captured. Because they can opt in. And 
if they have nothing more, that will be there, plus the 
nothing, plus the something. I hate to be so elemental with the 
explanation. But what I want to do is make it as clear as 
possible that what we are doing is leaving the traditional, 
whatever it happens to be, and then we bring these additional 
points of data to the scoring process.
    Now, given that my time is almost up, and by some standards 
up, I see--
    Mr. Scott. Will the gentleman yield for a moment?
    Mr. Green. I will yield and beg that the Chair would not 
look at the clock, if you will, please.
    Mr. Scott. Okay. Very quickly, I think another part of 
this--
    Chairman Lynch. The gentleman will suspend. We can't be 
doing this. You are over. If the gentleman wants to conclude 
his thought, he can, but--
    Mr. Green. I can't yield?
    Chairman Lynch. The gentleman's time has expired. I'm 
sorry.
    The Chair now recognizes the gentleman from Ohio, Mr. 
Gonzalez, for 5 minutes.
    Mr. Gonzalez of Ohio. Thank you, Mr. Chairman, and Ranking 
Member Hill for holding this hearing today, and thank you to 
our witnesses. I believe this area is an incredible opportunity 
to explore how new technologies can be deployed to allow more 
Americans to gain access to credit. That is sort of the promise 
or the hope, anyway, of the machine-learning technology.
    And I share the sentiment that Ms. Johnson just raised, 
which is the goal is to expand credit to as many Americans as 
possible.
    Mr. Girouard, I want to focus on your company specifically 
in the context of the sandbox. And so, we will go there.
    Bear with me for a second. You were founded in 2012, 
according to Crunchbase anyway, and have raised, I think it was 
$144 million in total funding.
    At what point did you start working with the CFPB directly 
in the funding stream?
    Mr. Girouard. I believe our first meeting with the CFPB was 
either in 2012 or 2013, about that time.
    Mr. Gonzalez of Ohio. Okay. So really, from the beginning, 
this was a concerted effort and a decision on your part?
    Mr. Girouard. That is correct.
    Mr. Gonzalez of Ohio. Okay. How big was the A, if you are--
I don't know if you are allowed to share that, but--
    Mr. Girouard. I'm sorry?
    Mr. Gonzalez of Ohio. How big was the series A run, 
roughly? I will tell you where I am going so you can maybe 
answer this.
    I want you to talk about the benefits of the sandbox in 
terms of allowing for more startups to enter this space. 
Because you talked about the big banks potentially being able 
to get into it. But I want to see more innovation. You guys 
have an incredible team. I was on your site, a bunch of ex-
Googlers and very smart folks. I know there are plenty of folks 
in Silicon Valley who would love to get into this space.
    How would the sandbox enable that?
    Mr. Girouard. The sandbox brings some clarity, which tends 
to make the money flow in terms of these companies, first of 
all, more entrepreneurs wanting to enter this space. When you 
have a very highly regulated area with a lot of confusion, most 
entrepreneurs will opt for something else.
    If you want more entrepreneurial effort in this area, 
bringing clarity will bring both the interest of the 
entrepreneurs and the money from the investors, and that will 
create companies that are going to make a difference over time.
    Mr. Gonzalez of Ohio. So, one of the benefits of the 
sandbox is not just that it gives Wells Fargo a chance, but 
that it gives the next group of startups a chance as well?
    Mr. Girouard. Without question.
    Mr. Gonzalez of Ohio. Great. And then I want to shift to 
some of the data privacy laws that you have kind of alluded to 
as well.
    California's privacy law is going to be coming into effect. 
And we hear a lot throughout the industry about the problems 
that is going to create.
    Can you comment on how you see it affecting your business 
specifically and AI in general?
    Mr. Girouard. Sure. I believe there are real issues related 
to privacy and large technology companies that need to be 
addressed, and I know are being addressed. And I am very 
appreciative of our home State, California, taking the lead on 
this.
    We are, of course, already preparing, reviewing, and 
planning to adapt our practices, our product, to the California 
law. What I would just generally add, of course, is a business 
like ours operates at a national level, so it would certainly 
be a step forward for us to have something of that sort, sort 
of managed at a Federal level more than at a State level. But 
having said that, we appreciate that is not the way the world 
works, and we will adapt to California's law.
    Mr. Gonzalez of Ohio. Yes. I think one of my concerns--and, 
again, that I keep hearing is when you have this patchwork of 
50 different State laws and you want to operate all over the 
country, as does everybody, you are creating--not you--but 
California has created a bit of chaos. And I know one thing 
this committee is committed to is to solving that, which I am 
excited about.
    And then I guess kind of with my last question, as we are 
thinking through what that national standard should be, what is 
it about what the California law that you like, and what is it 
that you think should be changed or different?
    If you are not comfortable answering, that is fine.
    Mr. Girouard. I am not sure I am comfortable enough to try 
to state that here. Thank you.
    Mr. Gonzalez of Ohio. Okay. Thanks. With that, I yield 
back. Thanks.
    Chairman Lynch. The gentleman yields back.
    The Chair now recognizes the gentleman from Utah, Mr. 
McAdams, for 5 minutes.
    Mr. McAdams. Thank you, Mr. Chairman.
    I want to thank the panelists for being here today. And I 
care deeply about expanding financial inclusion. But I want to 
make sure we are supporting an environment where all Americans 
can access credit.
    I do know that credit decisions can mean the difference 
between a family qualifying for a home or a loan to buy a car 
and the incredible life consequences that those decisions have 
for each and every potential borrower.
    We need to have appropriate consumer protections, and 
consumer protections shouldn't be ignored while we get the dial 
right to maximize the benefits while minimizing any potential 
negative impacts.
    But I want to zero in on that balance, the potential 
benefits and the potential harm or questions that are raised 
from the use of alternative data.
    First, speaking towards the benefits, we have heard 
testimony today that these alternative data factors are giving 
lawmakers more confidence in who they can responsibly lend to, 
meaning more consumers have access to credit, ideally at 
competitive rates.
    My first question is to you, Mr. Girouard. What percentage 
of your loan portfolio would you estimate that your company can 
make loans to because of the inclusion of alternative data 
sources?
    Or stated another way, if you were only allowed to use 
traditional data sources, what percentage of your customers 
would you not be able to lend to because you couldn't assess 
their creditworthiness?
    Mr. Girouard. Thank you, Congressman. That is a great 
question.
    That is exactly the data that I presented in my up-front 
statements. What the CFPB asked us to do is to look at our 
model, if we removed all what you might call alternative data 
and used only traditional data.
    The difference is, among the general population, we 
described as--and this is among people who have applied for 
loans at Upstart--is about 27 percent. More people are approved 
because of the alternative data. But importantly, when you look 
at the near-prime segment, which is people with somewhat lower 
FICO scores, it is a 95 percent increase in approvals.
    So, it is a very significant difference in improvement on 
who we can approve due to the alternative data that we include.
    Mr. McAdams. Thank you. And further on that point, some 
alternative data factors are now being used to include--or 
maybe not furthering the point--but in a different direction. 
Some of those alternative data factors are now being used to 
include online behavioral data such as online shopping habits, 
and social network connections.
    My next question is for Mr. Rieke. I believe you made this 
distinction in your testimony between the types of datapoints 
and the conventional alternative data.
    Out of curiosity, how much of these alternative data 
sources are moving the needle on a credit score? And I am not 
referring to alternative data such as bill payments, or online 
utility payments, of which I think most Americans would 
intuitively understand why that could be included in this 
credit score. But the online shopping habits, social network 
usage, how much does that affect an individual's data score?
    Are we talking 5 points of credit? One point? Fifty points? 
And for someone who has a thin credit file, how much are these 
factors weighted compared to traditional data factors?
    Mr. Rieke. The short answer is, we don't know. Most of the 
fringe alternative data that has to do with social media and 
shopping habits and web behavior, there is a lot more hype than 
reality, in terms of what I have been able to ascertain in our 
research. There are a lot of start-up companies making some 
pretty strong claims to the media, and then maybe once they 
hire a lawyer kind of backing off of those or starting to 
practice overseas.
    And so there is some academic research studies that showed 
web signals like what website you come from, whether you are 
using an iPhone or Android phone, can really help kind of 
narrow in on what kind of person you are, mostly because those 
are proxies for wealth.
    But in terms of the real science and research around the 
predictiveness of fringe alternative data, it is really hard to 
say, because companies hold that data close, and I think there 
is a lot less of that really happening in the United States 
today because of issues with the ECOA.
    Mr. McAdams. Do consumers understand what information on 
them is being collected and used in their credit decisions, and 
are there industry standard practices on disclosure?
    Mr. Rieke. I am not aware of any kind of formal industry 
standard best practices. There are some private businesses, 
like Credit Karma, that do, in my view, a pretty good job of 
showing the basic FICO score factors and helping people make 
sense of that. I have seen nothing resembling that for more 
complex or fringe datasets.
    Mr. McAdams. Ms. Wu?
    Ms. Wu. If I may address that, Congressman, the Fair Credit 
Reporting Act and the Equal Credit Opportunity Act do require 
that if someone is turned down or priced higher for credit, a 
notice goes out explaining what the reasons were. That is 
really, really important because of the impact these decisions 
have on people's lives.
    One of my concerns is with machine-learning and AI, where 
the machine itself is determining what factors to use. How do 
you make sure consumers have adequate information about what is 
going on inside the black box?
    The other thing that I wanted to quickly mention is another 
type of nonfinancial data that is being used, and Upstart is 
using, which is education. And I worry about the impact of 
using education as a form of alternative data. Because we know 
of the great inequality and racial disparities in terms of what 
kind of degrees people get.
    Mr. McAdams. Thank you, Ms. Wu.
    My hope would be that we can use this data to not only 
expand access to credit for more individuals but it can also be 
used as a form of improving financial literacy, if individuals 
know what data is being used and what things they might do as 
individuals to move the needle as well. And hopefully, that 
doesn't include unfriending their friends on Facebook.
    Thank you. And I yield back.
    Chairman Lynch. I thank the gentleman.
    The gentleman from Florida, Mr. Lawson, is now recognized 
for 5 minutes.
    Mr. Lawson. Thank you, Mr. Chairman.
    And I welcome the witnesses to the committee.
    You might have already responded on this particular issue, 
but it is important to me. As most of you know, credit reports 
do not tell the full story of one's economic status. As a 
matter of fact, they could have a false narrative, depending on 
the circumstances.
    It is estimated that the use of alternative data such as 
utility payments, rent payments, cellphone payments, and other 
forms could expand access to credit to over 40 million 
consumers here in the United States.
    Can everyday payments such as rent payments or cellphone 
payments paint a more accurate picture of someone's ability to 
pay? Are we headed in the right direction by saying that this 
would be a true picture of the individual's ability to get 
credit?
    And everyone--all of you, if you care to respond to that, 
it would be great. I would just like to know--and you might 
have already talked about it. But this is talked about all over 
the place, especially in Florida, where we have a high 
concentration of students in my district, about 80,000 or 
90,000 of them.
    So, I am anxious to know what your statement is going to 
be. And will I tell you, the reason being is when I was coming 
out of college, I was given all of these credit cards, Exxon, 
all of them, and so I started using them. And because the 
invoices, I guess, were going to the dormitory where I used to 
live, nobody forwarded them to me.
    When I got ready to try to get a loan or do some other 
thing, it came up. And it had been over 1 year or 2 years or 
so. And I just thought maybe, because I had graduated from 
school, they just gave me free credit. I didn't know. And that 
is one of the things that affects a lot of students, because 
they move around to different places.
    That is the reason I wanted to bring that question up and 
have all of you respond to it.
    Ms. Wu. Congressman, that is a great point. And you are 
absolutely right. Traditional credit scores and credit reports 
often don't reflect the true financial behavior of a consumer, 
precisely because of things like your experience or the fact 
that there are a lot of negative marks for things like medical 
debt, where people got sick and debts were sent to debt 
collectors.
    And we know that even among people with a subprime score, 
most of them, if you give them credit, will pay it back. 
Something like 80 percent of consumers who score a 600 will pay 
it back.
    So, alternative data could be useful, especially things 
like bank account data or rent, and if people choose, if 
consumers want to supply their utility and cellphone payments. 
Again, the devil is in the details; how you do it is important. 
Second-chance scores are better than putting this information 
in the traditional credit reports.
    We are concerned about factors that lead to more inequality 
or reflect inequality. As Mr. Rieke said, using geographic 
neighborhood or using what kind of degree you have, because we 
know that over 36 percent of non-Hispanic whites have a college 
degree, but less than 16 percent of Hispanics and 23 percent of 
African Americans do. So, if you use whether or not a consumer 
has a college degree, it is going to have some stark racial 
disparities.
    Mr. Rieke. Congressman, I want to just say I think the 
question of ability to repay is a really good target for this. 
We are talking about expanding access to credit, but we are not 
doing anyone any favors by giving them predatory products or 
too many credit products. That can destroy lives.
    So, I think ability to repay is a really important nexus 
between this question of alternative data and what are we 
trying to find out, but also a pretty strong consumer 
protection standard.
    Ms. Johnson. And I would just echo the earlier reflections. 
Thank you, Congressman, for this very important question about 
a really important demographic: students.
    We know from the New York Federal Reserve that households 
face $13 trillion in debt as of the end of the year, fourth 
quarter 2018, and $1.5 trillion in student debt.
    Student debt for a particular population, and most recently 
graduated generations of students, is staggering and crippling. 
And unlike past generations, these students are moving out of 
their parents' houses later, and they are having extended job 
searches.
    So, the predatory credit card tactics, the idea of drawing 
them into spaces where their credit histories will be marred, 
or they won't have credit histories at all because of how long 
it is taking them to dig themselves out of educational debt, 
really does prompt a need, a very significant need for 
alternative mechanisms, pathways for them to gain access to 
credit.
    I think we are all just thoughtful about how to do that in 
a way that is effective for consumers, protects their privacy, 
and is thoughtful about discrimination.
    Mr. Lawson. Mr. Chairman, I know I am out of time, so I 
yield back.
    Chairman Lynch. I thank the gentleman.
    We have agreed to just do one more brief round of 
questioning, so I yield myself 5 minutes for questioning.
    In our discussions with Facebook, in an effort to try to 
get some accountability on the protection of personal financial 
data, the issue of assigning fiduciary responsibility for the 
handling of information was suggested. And I would say that the 
response from Facebook was evasive, to be generous.
    What about that concept that there would be liability for 
mishandling the financial data that we surrender to fintech 
companies? Is that something that is workable, do you think, 
Ms. Wu?
    Ms. Wu. I thank you for the question, Chairman Lynch. I 
think that whether you call it a fiduciary duty, or you have 
legal duties or legal accountability for losing someone's data, 
there should be a regulatory scheme in place that holds Big 
Data companies, whether they be credit bureaus or Facebook, to 
accountability for losing sensitive personal information and 
data.
    Chairman Lynch. Yes. I guess I should just put a finer 
point on that.
    When I say ``fiduciary'', I mean in the classic financial 
services sense where a fiduciary is required to handle that 
information in the best interest of the customer, and not sell 
it or deploy it for other purposes. That is what I am getting 
at.
    I am trying to make sure what happens with personal 
financial data is not what happened with general data that is 
being vacuumed up and used and deployed without the knowledge 
or consent, meaningful consent, of individual consumers. That 
is what I am trying to get at.
    Ms. Johnson?
    Ms. Johnson. Yes. I'd just say thank you, Congressman.
    We have examples and models of how to protect financial 
transaction data that exist in current regulation. The Gramm-
Leach-Bliley Act, for example, specifically requires that 
financial institutions disseminate initial and annual privacy 
notices to customers regarding financial transactions. The 
provision of the Gramm-Leach-Bliley Act that I am describing 
enables consumers to specifically opt out in certain instances 
of other uses of financial data.
    It also requires financial institutions to anonymize data, 
essentially to the extent that they use data for other 
purposes, to aggregate the data and ensure that the data is 
anonymous and not directly reflective or you couldn't easily 
discern that it refers to a particular consumer based on the 
profile.
    Now, I will say that data scientists at Princeton and 
Stanford recently published a study illustrating that they 
could successfully decode, if you will, anonymized data and 
establish users' identities based on social networking 
profiles. The idea that this could happen is obviously 
concerning and gives us pause.
    But I do think that we have some examples in existing 
legislation and regulation that could offer a point of 
departure for having a conversation about how to create 
accountability, responsibility, and transparency for anyone 
who--or entities who are gathering, storing, and distributing 
personal consumer financial information.
    Chairman Lynch. Great. Let me just jump over to Mr. 
Girouard. I pulled up Upstart's terms of service agreement. And 
it is a lot shorter than Facebook's. Thank you very much. It is 
about 8 pages.
    But there is one section in here on limitation of 
liability. And it says the customer--``you agree that all 
access and use of the site and its contents and your use of the 
products and services is at your own risk.''
    In no event shall we or any lender be held liable for any 
damages, including direct or indirect, special, incidental or 
consequential damages, losses or expenses arising in connection 
with the site or any linked site or use thereof or inability to 
use by any party or in connection, or for failure of 
performance, error, omission, interruption, defect, delay in 
operation, transmission, computer viruses, et cetera.'' It is 
very, very broad.
    And this is one of those things where you have to click, 
``I agree.'' And either you agree to all of this or you don't 
use the site, you don't use Upstart.
    Is that fair to the consumer, do you think?
    Mr. Girouard. Chairman Lynch, I certainly wish we had a 
better option. But it is a complicated world. And certainly a 
business needs to protect its interests.
    Somebody could say the internet crashes and I was in the 
middle of getting a loan, and that just cost me my ability to 
buy a home or do something else.
    Chairman Lynch. This basically shuts off the consumer from 
any recovery at all under any circumstances. I understand cases 
like that where the technology breaks down, you could say in 
that case, we don't accept any liability.
    But in the terms of this, it is airtight where, you 
basically block off any type of accountability; you are beyond 
reach by this agreement. This is the type of thing I worry 
about.
    And I just--
    Mr. Girouard. It is a fair concern. I genuinely believe we 
have the highest consumer ratings we have ever found in our 
industry in terms of our respect and the way we treat customers 
or prospective customers.
    Chairman Lynch. I appreciate that. I am just concerned that 
no one has any recourse based on the terms of this agreement.
    With that, I yield to my friend, the gentleman from 
Arkansas, for 5 minutes.
    Mr. Hill. Thank you, Mr. Chairman.
    Just following up on that, Mr. Girouard, that particular 
thing he read, which obviously we haven't read, but I admired 
him in real-time going to your website--and it is a thing of 
beauty. And that is the difference between the House and the 
Senate, Ms. Johnson. You seem rather concerned about our 
technological capabilities here.
    That is really talking about your--the connectivity between 
the customers and you, isn't it, protecting you from liability, 
from the internet or from the website or the connection? Isn't 
that what that is mostly addressing?
    Mr. Girouard. Certainly. Any commercial agreement between a 
consumer and a business has to have reasonable protections in 
it. I am not an attorney, let me just admit that. So, for me to 
say what is an appropriate limitation of liability is not 
something I am probably equipped to speak about today.
    Mr. Hill. But we appreciate that. And that is something 
that we all deal with in any kind of commercial transaction. 
And I think it is made worse sometimes over the internet, 
because you don't have any kind of face-to-face explanation and 
it is a little bit more passive. But I think making sure 
consumers know what they are getting into is important.
    Mr. Evans, I wanted to ask you. I read your testimony--
thanks for it--about this harmonization between the regulatory 
agencies. You have urged them to adopt a harmonized approach to 
guidance under use of alternative data and also on the sandbox 
issue.
    Did they give you a timeframe when they would have a 
harmonized view on that?
    Mr. Evans. They did not. They all agreed with the 
recommendation and appreciated the spirit of it.
    Mr. Hill. Right. I think that is something we have all 
talked about here. We will be certainly pressing them for this 
more unified approach on vendor due diligence and an IT exam, 
guidance on what is an appropriate bank risk profile in this 
arena, how to do the board review of vendor due diligence. All 
of this is important.
    Mr. Evans, did you, in your work on this, see any reason to 
make statutory changes to the Fair Credit Reporting Act or the 
Equal Credit Opportunity Act?
    Mr. Evans. There are certainly some issues. The scope of 
the work didn't allow us to rigorously collect all the evidence 
for us to provide a conclusion on that, but for sure, the 
complexity of some of the algorithms could limit the type of 
information that a company is able to provide if they were to 
deny credit to an individual.
    Mr. Hill. Yes. I read the reliability of data point in your 
testimony, and we have talked about that with other witnesses 
at previous panels, with just asking the simple question, using 
an AI-based model, a machine-learning model, that uses 
additional data, just as Mr. Girouard has described, we have 
asked the question, is it auditable? And in the instances that 
we have had, the answer has been consistently yes. And the 
evidence of that is not hypothetical because the loans 
originated are subject to a fair lending exam by a commercial 
bank or portfolio buyer. And then, of course, in Mr. Girouard's 
case, they are also auditable by the CFPB's analysis of this 
data. So, as long as a commercial bank is a partner in it, from 
my point of view, that seems like the disparate impact test, 
the HMDA test if it would be a mortgage, or fair lending, or 
equal opportunity type assessments would be made. Is that 
generally your view from the work that you did? While you may 
have found six industry stakeholders who you had concerns 
about, it is doable to validate a model and have an audit trail 
as to how the determination was made, isn't it?
    Mr. Evans. In the models that we considered, I would say 
yes, but I would say our work was limited to the fintech 
companies we actually talked to. There could be classes of 
models that--
    Mr. Hill. Yes, but the obligation is on that user, that 
innovator, whether it is a bank partner or a fintech nonbank 
partner, to demonstrate that they comply with all the 
compliance obligations of the Federal Government.
    Mr. Girouard, you have said already, it is auditable, and 
the CFPB audits it, and then your bank partner audits it. That 
is correct, right, you can validate your model and backtrack 
it?
    Mr. Girouard. That is correct.
    Mr. Hill. And one question that didn't come up today--I 
didn't hear it--is back-testing. We have had the most ideal 
circumstances of the past 11 years, thanks to the unbelievable 
policies of our Federal Reserve, so that we have a very benign 
interest rate environment, we have rising real wages, we have a 
rising economy. What about back-testing your $4 billion you 
have originated? Looking back under more adverse credit 
circumstances, what have you learned?
    Mr. Girouard. Sure. That is a valid concern, and certainly 
any lender, to earn its stripes, really needs to perform 
through an economic cycle. First, we test the best we can by 
simulating higher unemployment. So, there are ways we can 
simulate higher unemployment and look at the impact we expect 
it to have on our loan portfolio.
    Second, because there actually are recessions, what I might 
call microrecessions in parts of the country, small parts of 
the country, we can actually look at loan performance in those 
particular areas. It is not a perfect proxy, but it is a way to 
understand how our loans would perform in a weaker economy.
    Mr. Hill. I thank the panel, and I yield back, Mr. 
Chairman.
    Chairman Lynch. The gentleman yields back.
    The Chair now recognizes the gentleman from Georgia, Mr. 
Scott, for 5 minutes.
    Mr. Scott. Yes, Mr. Lynch, this has indeed been a very, 
very informative hearing, and our panelists are well-prepared 
and very informative. Thank you for this. As I said in my 
opening statement, we are at a new frontier here, and it is an 
exciting frontier. But let's go back to the alternative data 
because I think that is really the fundamental foundation of 
this hearing.
    Now, there are different kinds of alternative data that I 
am hearing. So, you may say a utility bill, or you may say your 
online habits, or you may say your educational attainment. Tell 
me, how does a lender weigh these? Would they give more 
preference to your educational attainment? Does that have the 
same weight as your utility bill? How is it used by lenders 
when they are making these underwriting decisions? Let me start 
with you, Ms. Wu--or is there anybody who has an immediate 
answer to this? Or if we don't have an answer to it, don't you 
think we should?
    Mr. Girouard. I am glad to answer, Congressman, as one who 
does exactly what you are asking about. Traditional lending 
systems are what you might describe as rules-based. Okay? A 
series of, if the person's FICO is between this and that, if 
their income is between this and that, here is what we can 
offer them. That is what you might call a rules-based system. 
These newer models, what are sometimes termed machine-learning 
or AI, are far more sophisticated than that. What they really 
do is they look at the history of loans and the data that has 
come in about those loans, and it learns about how each of 
those factors actually impacted the performance of the loans. 
So, it is not a human sitting there trying to evaluate whether 
education or FICO or something else is more predictive. The 
software will learn over time what is the best combination of 
that information, that will be the most accurate model.
    The goal of a company like Upstart is to build a more 
accurate model, which tends to lead to higher approval rates, 
and we do that by relying on the software to do things that 
humans can't realistically do, which is to consider not two or 
three variables, but hundreds or maybe even a thousand 
variables, and that results in a more accurate credit model. 
And, fortunately, it also results in one that approves more 
people at lower rates.
    Mr. Scott. So, you are saying that the machine has a more 
accurate ability to give a certain alternative data more weight 
over the other? I guess what I am asking is, is there more 
benefit for one type of alternative data to be helpful to the 
unbanked or underbanked? There is a variety of things. Maybe 
also added in there, did he serve in the service? What was his 
rank? Was she a schoolteacher? What was the caliber of her 
employment structure? Do you see what I am saying? There seems 
to me that if we just put all this up to the machine, I am not 
sure it is giving it--there ought to be some weight here.
    Ms. Wu. That is a great point, Congressman Scott, and 
something I am concerned about. Because data that kind of looks 
like credit, as Mr. Rieke said, rent or bank account or utility 
bills, everybody, if they have a good history, can benefit from 
that. But if you are talking about things like education, how 
many college grads really are credit invisible? Are we really 
expanding access to credit if we say, we will give you a higher 
score if you graduate from college, especially if you graduate 
from an elite institution?
    Mr. Scott. Right.
    Ms. Wu. And then, in terms of machine-learning, one thing I 
want to add is, yes, it might be up to the lender to give that 
weight, but the lender has to be able to explain it. And if all 
this data is going into a big black box and the machine is 
deciding what is more important or not, you have to be able to 
put it on a piece of paper and explain to the consumer what was 
more important. The law requires it, because we need 
transparency in lending.
    Mr. Scott. Yes. Thank you, Mr. Chairman. Great panel.
    Mr. Evans. And so there is--I'm sorry.
    Chairman Lynch. If you can be quick, Mr. Evans.
    Mr. Evans. There is an important tradeoff to think about: 
transparency versus predictability. And that is something we 
have to grapple with and it is something about which the 
regulators can offer guidance.
    Chairman Lynch. Thank you very much. I appreciate that.
    The Chair now recognizes the gentleman from Florida, Mr. 
Lawson, for 5 minutes.
    Mr. Lawson. Thank you.
    Mr. Hill. Mr. Chairman, I need to just, I think, politely 
object. I thought we were going to do just ourselves for a 
final round of questioning, and I have no more Members here. 
And so, with all due respect to my friends, that is not really 
what we agreed to, so--
    Chairman Lynch. Okay. I understand the gentleman is short 
on time, and I totally respect him, but when I asked for a 
second round, I meant a second round for the Members.
    Mr. Hill. But you said a second round for the two of us, 
sir.
    Chairman Lynch. Sir, I was not aware that that is the way 
you understood that.
    Mr. Hill. That is the way you said it, and that is the way 
I understood it.
    Chairman Lynch. Perhaps I meant the two of us, meaning the 
two sides. I know the gentleman had no other--
    Mr. Hill. We have no other Members, so I think just in 
fairness under the rules, with the deference of Mr. Scott being 
the last questioner, that would be appreciated.
    Chairman Lynch. The gentleman, Mr. Green, has yielded and--
    Mr. Lawson. I yield back.
    Chairman Lynch. --the gentleman, Mr. Lawson, agrees as 
well?
    Mr. Lawson. Yes.
    Chairman Lynch. Okay.
    Without objection, the Chair moves to include in the record 
of this hearing a letter from the Cato Institute, Center for 
Monetary and Financial Alternatives, dated July 24, 2019; a 
letter from the Financial Data and Technology Association, 
dated July 23, 2019; and also an article from the Student 
Borrower Protection Center, entitled, ``Educational Redlining: 
The Use of Educational Data in Underwriting.'' Without 
objection, it is so ordered.
    I would like to thank our witnesses for their testimony 
today.
    The Chair notes that some Members may have additional 
questions for this panel, which they may wish to submit in 
writing. Without objection, the hearing record will remain open 
for 5 legislative days for Members to submit written questions 
to these witnesses and to place their responses in the record. 
Also, without objection, Members will have 5 legislative days 
to submit extraneous materials to the Chair for inclusion in 
the record.
    This hearing is now adjourned.
    [Whereupon, at 12:12 p.m., the hearing was adjourned.]

                            A P P E N D I X


                             July 25, 2019
                             
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]

                              [all]