University of Southern California Dissertations and Theses

Social movements and access to credit

(USC Thesis Other)

Social movements and access to credit

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content SOCIAL MOVEMENTS AND ACCESS TO CREDIT
by
Joon Sang Yoon
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BUSINESS ADMINISTRATION)
May 2023
Copyright 2023 Joon Sang Yoon
Dedication
To my wife Eun Jung, and my parents, Tae Hwa Yoon and Eun Sil Jung. I could not have done it without
you all.
ii
Acknowledgements
I am grateful to my advisor, Regina Wittenberg-Moerman, for her invaluable guidance and support. I also
thank my committee members, Richard Sloan, Mark Soliman, and Jerry Hoberg for their insightful feed-
back. I also thank Tom Chang, AJ Chen, Jaewon Cheong, Jonathan Craske, Patricia Dechow, Mark DeFond,
Jesse Gardner, Shane Heitzman, Taylor James, Eun Jung Jung, Jung Koo Kang, Anya Kleymenova, Clive
Lennox, Shelley Li, Tracie Majors, Carmen Mann, Maria Ogneva, Vivek Pandey, Lorien Stice-Lawrence,
David Tsui, Forester Wong, TJ Wong, and fellow PhD students at USC for helpful feedback. I thank seminar
participants at the University of Southern California, the Chinese University of Hong Kong-Shenzhen and
AAA Miami Rookie Camp. I gratefully acknowledge the financial support from the University of Southern
California.
iii
TableofContents
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
Chapter 1: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Chapter 2: Background and Hypothesis Development . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1 Peer-to-Peer Lending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Hypothesis Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Chapter 3: Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1 Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Sample Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Chapter 4: Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1 Primary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2 Robustness Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.2.1 Lead and Lagged Protests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.2.2 Falsification Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3 Alternative Explanations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3.1 Increased Demand for Credit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3.2 Foot Traffic to Bank Branches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.3.3 Loan Purpose Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3.4 Small Business Administration (SBA) Loans . . . . . . . . . . . . . . . . . . . . . . 31
4.4 Cross-Sectional Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.4.1 Political Affiliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.4.2 Bank Competition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.4.3 Economic Disadvantage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4.4 Single Location Banks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.4.5 Minority Bank Boards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
iv
4.5 Other Alternative Explanations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.5.1 Willingness to Lend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.5.2 Non-BLM Protests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.5.3 P2P Credit Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Chapter 5: Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
v
ListofTables
3.1 DistributionofBorrowers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 LoanDescriptives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.1 PrimaryResults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3 AlternativeExplanation: IncreaseinDemandforCredit . . . . . . . . . . . . . . . . 47
4.4 PoliticalAffiliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.5 BankCompetition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.6 EconomicDisadvantage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.7 SingleLocationBanks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.8 MinorityBankBoards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.9 LenderBehavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.10 Non-BLMProtests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.11 P2PCreditModel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
vi
ListofFigures
2.1 LendingClubScreens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1 BorrowerLocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
vii
Abstract
I investigate whether social movements, such as #BlackLivesMatter (herein “BLM”), affect local loan of-
ficers’ mood and consequently, credit approvals at local banks. Motivated by research in psychology, I
hypothesize that protests will lead to increased stress levels, inducing negative moods in loan officers and
ultimately result in a decrease in banks’ credit approvals. Because both loan officers’ sentiment and their
credit decisions are unobservable, to test this prediction I rely on the substitution between P2P lending
and traditional banking and expect that borrowers who are unable to secure credit from banks migrate to
P2P lending. Utilizing the staggered occurrences of BLM protests, I show that P2P lending significantly
increases in areas that experience BLM protests. Additionally, consistent with P2P lending serving bor-
rowers rejected by banks, I find that the increase in P2P lending is driven by low quality borrowers. These
effects are stronger for areas that are Republican leaning, have less banking competition, are economically
disadvantaged, have more single bank branches, and have a greater proportion of minority board mem-
bers. Moreover, I fail to find that the increase in P2P lending is driven by potential alternative explanations
such as an increased demand for credit. Overall, my results suggest that social movements and the related
protests may have an adverse effect on the ability of weaker borrowers to access credit.
viii
Chapter1
Introduction
Research in social sciences have examined various socio-economic implications of social movements such
as the #BlackLivesMatter (herein “BLM”).
∗
The aforementioned research has highlighted the effect of the
BLM protests on the public sector, including reduction in racial prejudice following protests (e.g., Sawyer
and Gampa 2018; Mazumder 2019), increased support of the movement by non-Black ethnic minorities
(e.g., Vani et al. 2022), and reduction in police lethal use-of-force (e.g., T. Campbell 2021; Hehman, Flake,
and Calanchini 2018; Skoy 2021). Studies have also examined the implications of the movement on different
financial market outcomes, including analysts’ recommendations (e.g., Rupar, S. Wang, and Yoon 2021),
firm disclosure (e.g., Chen, Dechow, and Tan 2021) and consumer responses (e.g., Y. Wang et al. 2022).
Despite this growing body of research, there is no evidence of how protests affect the decision-making
processes of loan officers and thus credit market outcomes. In this study, I fill this gap by investigating
how social movements affect commercial bank loan officers’ protest-induced moods (i.e., sentiment) and
consequently the ability of low quality borrowers to access credit.
†
The empirical challenge in estimating this research question is that changes in loan officers’ sentiment
are not directly observable. Moreover, the lack of data on consumer lending at the loan officer level further
hinders the researcher’s ability to measure the effect of these changes. To address these challenges, I
∗
I use the term social movement and protest interchangeably.
†
I use the term low quality and weak interchangeably.
1
exploit the fact that commercial banking and peer-to-peer (herein “P2P”) lending are substitutes, where
P2P lending serves infra-marginal borrowers who are denied access to bank credit (e.g., Tang 2019; Roure,
Pelizzon, and Thakor 2022). More specifically, by relying on the changes in P2P lending following BLM
protests, I examine how protests influence loan officers’ sentiment and the related changes in commercial
banks’ lending. I also acknowledge that my study is not without limitations. A caveat in my study is
that due to data limitations, I am unable to distinguish between the effects of racism and sentiment. More
specifically, as P2P platforms such as Lendingclub and Prosper no longer collect individual characteristics
such as ethnicity that may be used for discriminatory purposes, I am unable to identify whether a specific
borrower is related to a specific ethnicity.
Motivated by prior research in psychology and epidemiology which finds that protests lead to increased
stress levels for the general population near the protests (e.g., Ni, Li, et al. 2016; Hou et al. 2015), I hypoth-
esize that local loan officers (i.e., loan officers in the same three-digit zip code as the protests) experience
negative moods following protests (e.g., Bolger et al. 1989), leading to a decrease in credit approvals (e.g.,
Cortés, Duchin, and Sosyura 2016; Agarwal, Duchin, and Sosyura 2012). Moreover, if protest-induced neg-
ative moods affect the decision making processes of local loan officers, the effect should be stronger when
the decisions require more discretion (e.g., Klein and Barnes 1994; Wemm and Wulfert 2017; Beilock and
DeCaro 2007; Cortés, Duchin, and Sosyura 2016; Agarwal, Duchin, and Sosyura 2012). This suggests that
the effect should be concentrated in low quality borrowers. Consequently, the low quality borrowers who
are unable to secure loans at commercial banks will migrate to P2P lending (e.g., Tang 2019), resulting in
an increase in this lending to borrowers in the vicinity of the protests (i.e., the three-digit zip codes areas
of the protests).
My hypothesis, however, is not without tension. BLM protests may positively affect the sentiment
of local loan officers due to the following reasons. Prior research in various disciplines in social sciences
document positive externalities of BLM, such as shifts in public opinion and a more widespread support
2
of diverse ethnic groups (e.g., Mazumder 2019; Sawyer and Gampa 2018; T. Campbell 2021; Hehman,
Flake, and Calanchini 2018; Skoy 2021). (e.g., Vani et al. 2022; Arora and Stout 2019). In line with these
arguments, local loan officers may be positively inspired by the protests to feeling that collectively, they
can make a difference (e.g., Landmann and Rohmann 2020; Sabherwal et al. 2021), leading to a positive
mood. Consequently, the resulting positive mood is likely to lead to an increase in credit approvals at
local banks for low quality borrowers following protests. In this case, we should observe a decrease in P2P
lending in the areas in the vicinity of the protests.
To test my hypothesis, I collect the location and dates of all BLM protests from 2014 to 2018.
‡
I obtain
the P2P lending data from Lendingclub.com (herein “LC”), the largest P2P lender in the U.S.
§
. The P2P
lending data allows me to identify the number of loan originations and the amount of loans in each three-
digit zip code for each year-month. I then combine the P2P lending data with the staggered occurrences
of BLM protests across time and location to implement a differences-in-differences framework with zip
code and year-month fixed effects M. Bertrand, Duflo, and Mullainathan 2004. My design compares the
changes in P2P lending in areas that experience a BLM protest in a particular year-month (“protest areas”)
to the changes in P2P lending in areas that do not experience a BLM protest in a particular year-month
(“control areas”). To alleviate concerns regarding omitted variables that might simultaneously affect the
P2P lending in the area and the probability the area experiences a BLM protest, I match protest areas with
areas that never experience a protest based on a propensity score calculated every year.
Consistent with the hypothesis that BLM protests negatively affect the local loan officer sentiment, I
find that the number of P2P loans originated in the areas that experience a BLM protest and the total loan
amount significantly increase in the month following a protest compared to areas that do not experience
a BLM protest. Economically, there is a 1.6% increase in the P2P loan origination amount and a 2.1%
increase in the number of P2P loans in the month following a BLM protest. In addition, consistent with
‡
the sample period ends in 2018 due to the availability of the P2P lending data .
§
Refer to section?? for more details
3
P2P lending and commercial banking being substitutes with P2P lending serving infra-marginal borrowers
who are unable to secure loans at commercial banks (e.g., Tang 2019) and the fact that the effect of mood
is larger for borrowers that require more subjective judgment (e.g., Cortés, Duchin, and Sosyura 2016;
Agarwal, Duchin, and Sosyura 2012), I find that the increase in P2P lending to be driven by low quality
borrowers. My primary results are robust to a test of parallel trends and a falsification test performed on
three-digit zip codes that never experience BLM protests.
While the increase in P2P lending following BLM protests is consistent with my hypothesis that loan
officers’ sentiment is negatively affected by the protests, the increase may be also consistent with the
alternative explanation that the demand for credit increased following protests. For example, the demand
for credit may increase because rational minority entrepreneurs predict an increased support for minority-
owned businesses after protests, leading to an increase in the demand for business loans. In order to
examine whether my primary findings are demand driven, I first test the relationship between protests and
the outstanding balances of personal and small business lending at banks. If the increase in P2P lending
following protests is driven by a greater demand for credit, I expect a greater demand for bank lending
as well, which should lead to an increase in the balance of outstanding loans at banks.
¶
Using Schedule
RC-C of Call Reports, I find a negative association between protests and outstanding loan balances during
the quarter of the protests. The decrease in outstanding loans suggests that the increase in P2P lending
following protests is unlikely to be driven by the increased demand for credit and is rather consistent with
an adverse effect of protests on loan officer’s mood and loan approval decisions.
Second, I examine the level of foot traffic to bank branches. To the extent that foot traffic to banks
is positively correlated with the demand for loans, I should observe a significant increase in foot traffic
to banks following protests if my primary results are driven by demand related reasons. Using the foot
traffic data from SafeGraph, I find that the foot traffic to banks during the month of the protest is not
¶
Since P2P lending and banking are substitutes, it is unlikely that the demand for one source of credit increases while the
other stays constant.
4
significantly different from zero. This evidence further suggests that the increase in P2P lending is unlikely
to be attributed to an increased demand for credit. Third, I specifically address a concern that the increase
in P2P lending may be driven by entrepreneurs expecting an increased support for minorities following
protests, which incentivizes them to seek loans for business purposes. If this is the case, the increase in
P2P lending should be concentrated in business purpose loans. However, I find that the increase in P2P
lending is concentrated in non-business loans, such as debt consolidation, improving housing, and medical
and vacation expenses. Lastly, I show further support for the idea that the demand for business loans does
not change before and after protests by examining loans from the Small Business Administration (SBA).
Consistent with the idea that the demand for business loans does not change following BLM protests, I
find that both the SBA loan amount and the number of SBA loans do not change following BLM protests.
Fourth, to provide stronger support to the inference that changes in mood are the primary driver of
the increase in P2P lending and to further alleviate concerns that the increase in this lending is driven by
demand-related reasons, I perform a number of cross-sectional analyses. If the increase in P2P lending
following BLM protests is driven by loan officers’ moods being adversely affected by protests, I expect the
results to be stronger (weaker) for cross-sections where loan officers’ loan approval decisions are more
(less) likely to be affected by their mood. At the same time, there are no reasons to expect that the demand
for credit should be affected by loan officers’ mood.
As the first cross-sectional test, I examine the area’s political affiliation using the election results for
the 2016 presidential election. As Parker, Horowitz, and Anderson 2020 finds, there exists a partisan divide
in participation and support for the BLM protest, which suggests that the political affiliation of the area
is associated with the support of the movement. Motivated by the psychology literature which finds that
identifying with intentions of collective actions induces positive moods by invoking thoughts that indi-
viduals can make a difference (e.g., Landmann and Rohmann 2020; Sabherwal et al. 2021), I hypothesize
that local loan officers in Democrat-leaning areas are less likely to be affected by protest-induced negative
5
moods. Consistent with my prediction, I find that the increase in P2P lending in Democrat-leaning ar-
eas following a BLM protest is significantly lower than the increase in P2P lending in Republican-leaning
areas.
The second cross-sectional test is motivated by the literature which documents that the intensity of
local bank competition affects the dynamic between P2P lending and commercial lending (e.g., Havrylchyk
et al. 2016; Jagtiani and Lemieux 2018; Beck, Behr, and Madestam 2018). In this test, I explore whether the
intensity of the banking competition in the region is a factor which moderates the shift in the loan officers’
sentiment after BLM protests. I hypothesize that if the area’s banking market is more competitive then the
loan officers in the area are likely to be less affected by sentiment due to the more competitive pressure.
That is, with more competition, there is less room for mood to have an effect on decision making (e.g.,
Hong and Kacperczyk 2010). Consistent with this notion, I find evidence that the increase in P2P lending
in areas with less banking competition after a BLM protest is significantly greater than the increase in P2P
lending in areas with more banking competition.
∥
In the third cross-sectional test, I examine if there is a greater increase in P2P lending in economically
disadvantaged areas.
∗∗
Motivated by prior literature which finds that the effect of mood is greater when
decisions require more subjective judgment (e.g., Agarwal, Duchin, and Sosyura 2012; Cortés, Duchin, and
Sosyura 2016), I hypothesize that if loan officers are negatively affected by the protests and use more sub-
jective judgment in examining applications of low quality borrowers, the decrease (increase) in bank lend-
ing (P2P lending) should be more pronounced in economically disadvantaged areas after a BLM protest.
Consistent with this prediction, I find that the increase in P2P lending is stronger in economically disad-
vantaged areas.
††
∥
In untabulated analyses, I find that my primary results are also stronger for regions dominated by single bank branches
compared to regions dominated by big bank branches. The idea is that single bank branches are more likely to rely on the local
loan officer to make the lending decision.
∗∗
I categorize a three-digit zip code as economically disadvantaged if the median income level of the area is lower than the
U.S median income.
††
The results are robust to using an alternative measure of borrower quality using both income and education levels.
6
The fourth cross-sectional test continues to examine whether the increase in P2P lending is driven by
changes to loan officers’ sentiment. Motivated by prior literature which finds that loan decisions at single
bank branches are less likely to be automated (e.g., Cortés, Duchin, and Sosyura 2016; Agarwal, Duchin,
and Sosyura 2012), I hypothesize that loan officers that work in areas that are dominated by banks which
operate in large geographical regions are less likely to be affected.
‡‡
Consistent with this, I find that the
increase in P2P loan origination amount and the number of P2P loans are significantly less for areas that
are dominated by banks with large geographical regions.
In my final cross-sectional test, I examine whether areas that are dominated by banks with a lower
percentage of minority representation on their boards are associated with a greater increase in P2P lending.
Prior banking literature provides theoretical and empirical evidence that increased minority representation
on bank boards leads to a reduced likelihood that a similarly qualified minority applicant is rejected than
their counterparts (Goenner 2023). Consistent with this, I find that P2P lending increases significantly in
areas that are dominated by banks with low minority representation on their boards. In addition, further
supporting my findings that the effect of protests is stronger when loan officer decisions require more
subjective judgment, I find that for all of the previous cross-sectional analyses, the increase in P2P lending
is driven by low quality borrowers.
Collectively, the above set of analyses mitigates the demand-driven explanation to my findings. How-
ever, I acknowledge two additional alternative explanations to the increase in P2P lending. First, P2P
investors may be more willing to lend to weaker borrowers after protests.
§§
In order to empirically test
this alternative, I follow Tang 2019 and measure whether a loan request has a similar probability of being
fully funded before and after protests.
¶¶
The idea behind this measure is that if P2P investors became more
willing to lend to low quality borrowers after protests, then the probability that a loan is fully funded will
‡‡
A three-digit zip code is classified as a many bank area if there are more banks that operate in other areas than the cross-
sectional median number of banks that operate in other areas.
§§
Institutional features of P2P lending that reduce the likelihood of P2P investors being affected by mood changes are discussed
in more detail in section??.
¶¶
A loan may be partially funded (i.e., not fully funded) if it fails to attract enough investors.
7
increase after protests.I find that the probability that a loan is fully funded does not change after a BLM
protest.
I continue examining whether investors’ willingness to lend is affected by the protests, I focus on non-
BLM protests.
∗∗∗
Non-BLM protests provide a good laboratory for examining the change in the willingness
of P2P lenders to invest in weaker borrowers due to the fact that non-BLM protests do not involve the
underprivileged. Therefore, investors’ willingness to fund weaker borrowers should not change before and
after non-BLM protests. Consistent with this idea, I continue to find that the probability that a loan is fully
funded does not change following non-BLM protests. The combined evidence from examining possible
changes to investors’ lending behavior suggests that my results are not driven by local P2P lenders being
more sympathetic to the underprivileged.
Finally, I examine whether changes to LC’s proprietary credit model are able to explain the increase
in P2P lending following protests. The credit model, which LC continuously updates to reflect macroe-
conomic and socioeconomic conditions, determines the universe of loans that investors see on the P2P
lending platform. According to LC’s annual reports, LC uses the algorithm, which leverages behavioral
data, transactional data, and employment data, to assess borrowers’ risk profiles and automatically accept
or reject borrowers.
†††
While conversations with LC representatives suggest that the platform has not
made significant changes to its credit model after BLM protests, I formally test this question by examining
whether the probability of a loan being accepted by the platform changed following protests. I expect that
if LC revised the algorithm after BLM protests to include more weak borrowers, it will lead to a higher
probability that a loan is accepted. Consistent with the idea that LC’s proprietary credit model did not
change, there is no evidence that the likelihood of a loan being accepted increases after BLM protests.
∗∗∗
Non-BLM protests are obtained from the Crowd Counting Consortium. For definitions, please refer to Appendix ??.
†††
In my setting anaccepted loan means that the loan is approved by the proprietary algorithm to be considered for funding on
the platform. Loans can also be automatically rejected by the algorithm. While LC does not disclose rejection criteria calculated
based on the model, its annual reports state that certain minimum credit requirements are a FICO score of at least 660, satisfactory
debt-to-income ratios, 36 months of credit history, and a limited number of credit inquiries in the previous six months.
8
My study makes several contributions. First, I contribute to the growing literature on the real effects
of social movements. While existing studies have largely focused on the positive externalities of the BLM
movement (e.g., Chen, Dechow, and Tan 2021; T. Campbell 2021), I contribute to this literature by docu-
menting that it can also have an adverse effect on credit market outcomes, where low quality borrowers
are led to seek out alternative sources of financing by migrating to P2P lending platforms.
I also add to the large literature on the effect of sentiment on access to credit. Extant literature has
highlighted that external factors such as sunshine, hunger, attention, and robberies can affect credit deci-
sions (e.g., Cortés, Duchin, and Sosyura 2016; Agarwal, Duchin, and Sosyura 2012; Demiroglu et al. 2021;
Morales-Acevedo and Ongena 2020; D. Campbell, Loumioti, and Wittenberg-Moerman 2019). I add to this
literature by providing evidence that social movements can also have a material effect on the local loan
officers’ mood and thus, their credit granting decisions.
Lastly, I add to the emerging literature on the relationship between commercial lending and P2P lend-
ing. Existing studies have primarily focused on regulatory changes which govern traditional sources of
financing as exogenous negative shocks to the banking credit supply to study the dynamic between com-
mercial lending and P2P lending (e.g., Tang 2019; Roure, Pelizzon, and Thakor 2022). I add to this literature
by documenting that non-banking shocks such as social movements can also have affect the substitution
dynamics of between commercial and P2P lending.
9
Chapter2
BackgroundandHypothesisDevelopment
2.1 Peer-to-PeerLending
Peer-to-Peer loans have risen to prominence in the last decade with the emergence of ‘fin-tech’ lending.
In the U.S., two of the oldest and most prominent platforms are Prosper and LendingClub (herein “LC”).
LC’s 10-K filings report that it has taken over as the largest P2P platform by facilitating more than $16.0
billion in loans since its inception in 2007.
∗
P2P lending cuts out traditional financial institutions such as
banks or credit unions and instead creates a marketplace where borrowers in need of funds can meet with
lenders who are willing to lend money (e.g., Thakor 2020). P2P lending has maintained its competitive
edge by providing benefits to both borrowers and lenders. The growing literature on P2P lending doc-
ument that the extensive use of alternative sources of data such as utility or rent payments, credit card
transactions, internet footprints and others has allowed some borrowers to gain access to cheaper credit
(e.g., Jagtiani and Lemieux 2019). On the one hand, some studies find that fintech lenders offer lower rates
than traditional lenders (e.g., Jagtiani, L. Lambie-Hanson, and T. Lambie-Hanson 2021). However, others
argue that fintech borrowers value fast and convenient services and fintech lenders charge a premium
for providing this service (e.g., Buchak et al. 2018). For lenders, P2P platforms offer the opportunity for
small and medium sized investors who are willing to take on more risk than savings notes or corporate
∗
Taken from LC’s 2015 annual report.
10
bond funds (e.g., Morse 2015; Jagtiani and Lemieux 2019). Moreover, lenders have been able to generate
abnormal returns from investing in a portfolio of loans on the P2P platform (e.g., Kräussl et al. 2018). P2P
lending has also been a popular investment choice for institutional investors such as hedge funds. For
LC, the proportion of institutional investors have grown from 63% in 2014 to 78% in 2018. The growth
of institutional investors presence on P2P platforms may be explained by the ability of P2P platforms to
offer short-term realizations that may be coupled with more traditional fund structures (e.g., Morse 2015).
Moreover, P2P platforms have offered institutional investors with an opportunity to obtain leverage via
securitization of P2P loans (e.g., Bavoso 2020).
The loan process for LC works as follows. First, the borrower submits a loan application which contains
the annual income, address, the amount requested, and the purpose of the funds. The range of funds
requested can vary from $1,000 to $35,000.
†
Pursuant to the Equal Credit Opportunity Act (ECOA), LC does
not ask for characteristics such as race or sex which may potentially be used for discriminatory practices.
LC uses its proprietary technology to perform a credit check on the borrower and then automatically
accepts or rejects borrowers.
‡
Once the application has been accepted by the algorithm, it is listed on the
platform to attract investor commitments.
Lenders who wish to lend money to borrowers can do so using the platform. Figure 2.1a shows the list
of filters that lenders on LC can sort on.
§
Some examples of the filters are characteristics that are associated
with the borrower’s credit history such as the borrower’s credit score, number of delinquencies in the last
2 years, and minimum length of employment. Lenders can choose to click on any of the filters to create
a custom list of loans to invest in. After applying the initial filter, lenders can browse the loans to select
specific loans that they would like to purchase as in Figure 2.1c. Lenders can also automate this process
by specifying a set of portfolio weights as in Figure 2.1b. The five letters (“A” through “E”) represent loan
†
LC increased the maximum borrowable amount to from $35,000 to $40,000 in 2016. As for the maturity, borrowers have the
option to choose between a three-year or a five-year maturity.
‡
According to LC’s annual reports, some of the selection criteria are a FICO score of at least 660, 36 months of credit history,
and a limited number of credit inquiries in the previous six months.
§
Accessed via the WaybackMachine as LC retired its notes platform in 2020.
11
grades that are algorithmically determined by the LC, with “A” grades being less risky than “E” grades.
The automated investment feature then purchases loans according to the lender’s pre-specified portfolio
weight.
2.2 RelatedLiterature
Researchers in accounting and finance have shown that various exogenous factors induce fluctuations in
mood and emotional state, broadly referred to as sentiment. These changes in sentiment play important
roles in economic decisions. For example, weather conditions such as sunshine or rain impact individual’s
mood and thus have financial market implications. Cortés, Duchin, and Sosyura 2016 use fluctuations in
sunshine as an instrument for sentiment and show that positive sentiment (i.e., sunshine) is associated with
higher credit approvals. Also, they find that negative sentiment (i.e., cloud cover) has an opposite effect
of a larger magnitude. Weather induced sentiment has also been shown to affect other financial market
participants such as analysts (e.g., Dong et al. 2021), both sophisticated and unsophisticated investors (e.g.,
Hirshleifer and Shumway 2003; Goetzmann et al. 2015) and consumers (e.g., Busse et al. 2015; Hu and Lee
2020). Prior literature has also highlighted factors that affect sentiment other than weather. For instance,
exogenous factors such as local sports teams winning or losing (e.g., Agarwal, Duchin, and Sosyura 2012;
Edmans, García, and Norli 2007; Drake, Gee, and Thornock 2016), traumatic events (e.g., Morales-Acevedo
and Ongena 2020), and physiological factors such as fatigue, hunger, and attention (e.g., Demiroglu et
al. 2021; D. Campbell, Loumioti, and Wittenberg-Moerman 2019) affect decision makers’ sentiment and
consequently, economic decisions.
While the above studies have broadly examined several exogenous factors that affect sentiment, a
relatively under-studied area is whether social movements influence the decision maker’s sentiment. Re-
searchers in diverse disciplines of social sciences have examined numerous socio-economic implications of
social movements. Prior studies have examined the real effects of social movements such as shifts in public
12
opinion following protests (e.g., Sawyer and Gampa 2018), reduction in fatal interactions with police (e.g.,
T. Campbell 2021), analysts’ reactions (e.g., Rupar, S. Wang, and Yoon 2021), firm disclosure (e.g., Chen,
Dechow, and Tan 2021), and consumer responses (e.g., Y. Wang et al. 2022). However, missing from the cur-
rent literature is whether social movements affect the sentiment of decision makers such as loan officers.
Therefore, I examine how BLM protests affect the sentiment of the loan officers in the same three-digit zip
code as the protests (i.e., local loan officers) and consequently credit approvals at local banks.
The empirical challenge in estimating this research question is that changes in loan officer sentiment
is not directly observable. The absence of publicly available data on consumer lending at the loan officer
level further hinders my ability to measure the effect of these changes. To address this challenge, I adopt an
indirect approach where I exploit the fact that commercial lending and P2P lending are substitutes, where
P2P lending serves borrowers who are denied access to credit at commercial banks (e.g., Tang 2019).
¶
More
specifically, Tang 2019 develops a conceptual framework in which P2P platforms may operate as substitutes
or complements to banks. The identifying assumption in her framework is that when banks experience a
negative shock to their credit supply which leads them to tighten the lending criteria, borrowers who are
unable to secure loans migrate to P2P platforms. Using FAS 166/167 as a negative shock to the banks’ credit
supply, Tang 2019 shows that P2P lending is a substitute for commercial banks as P2P lending increases
after the negative credit supply shock with the increase driven by lower quality borrowers.
2.3 HypothesisDevelopment
In this paper, I investigate how social movements such as BLM protests affect the local loan officers’ senti-
ment and consequently credit approvals at local banks by measuring changes in the P2P lending associated
with the protest areas before and after each protest. I predict that BLM protests negatively affect the local
¶
Other studies such as Jiang et al. 2021 find that P2P platforms focus on under-served borrowers in China. However, Cor-
naggia, Wolfe, and Yoo 2018 find that for the U.S. consumer credit market, P2P lending and banking behave similar to substitutes.
Additionally, Jagtiani and Lemieux 2018 compare FICO scores of borrowers on LendingClub and the FRBNY Equifax CCP and
find that both groups have similar FICO score ranges.
13
loan officers’ sentiment due to the following reasons. First, protests are logistically disruptive. The fact
that protestors occupy streets and bridges, causing traffic stops and congestion has been widely covered
by popular press. The academic literature has also documented that such unexpected events increase neg-
ative affect because interruptions create perceptions of time pressure and low task accomplishment (e.g.,
Sonnentag, Reinecke, et al. 2018). Additionally, these disruptions may be frustrating and depleting for
the agents involved (e.g., Zohar, Epstein, and Tzischinski 2003; Lanaj, Johnson, and M. Wang 2016) and
the effects are greater when the interruptions are deemed as inappropriate or necessary (e.g., Sonnentag
and Lischetzke 2018). Protests are also mentally disruptive in that they exact a substantial, pervasive, and
persistent toll on mental health even in the absence of large-scale violence such as fatalities, arson, or loot-
ing (e.g., Ni, Y. Kim, et al. 2020; Hou et al. 2015). Prior literature in epidemiology argue that non-violent
protests generate stress via interpersonal conflicts (e.g., Ni, Li, et al. 2016) and some studies have gone so
far as to suggest that stressful events are associated with greater risk of arrhythmia (e.g., Rosman et al.
2021). Overall, this discussion suggests that protests may lead to greater stress and consequently bad mood
(e.g., Bolger et al. 1989). The bad mood experienced by the local loan officers are likely to lead to lower
credit approvals at banks (e.g., Cortés, Duchin, and Sosyura 2016; Agarwal, Duchin, and Sosyura 2012). In
terms of examining the specific group of borrowers that drive the effect, the psychology literature finds
that stress-induced negative moods will lead to lower performance for tasks that require more complex
decisions (e.g., Klein and Barnes 1994; Beilock and DeCaro 2007; Wemm and Wulfert 2017). Therefore,
the effect will likely be concentrated in low quality borrowers as these borrowers require greater use of
subjective judgment by the loan officers (e.g., Agarwal, Duchin, and Sosyura 2012; Cortés, Duchin, and
Sosyura 2016). Finally, the lower credit approvals will lead to borrowers who are rejected (i.e., lower qual-
ity borrowers) to migrate to alternate sources of financing such as P2P platforms (e.g., Tang 2019). This
leads to my hypothesis, stated in an alternative form:
14
Hypothesis: TheareaswhichexperiencedBLMprotestsexperienceanincreaseinP2Plending,compared
to areas which did not experience BLM protests over the same time period, with the increase driven by low
quality borrowers.
This hypothesis is not, however, without tension. BLM protests may have a positive effect on local loan
officers’ sentiment for several reasons. Numerous positive externalities of BLM have been documented by
prior research. For example, studies have shown that public opinions change after BLM (e.g., Mazumder
2019; Sawyer and Gampa 2018), fatal interactions with police are reduced (e.g., T. Campbell 2021; Hehman,
Flake, and Calanchini 2018; Skoy 2021), and analysts’ pessimism is reduced (e.g., Rupar, S. Wang, and Yoon
2021). Furthermore, Cole 2017 suggest that the implications of the BLM movement may extend to non-
black minorities as well. In a similar vein, various ethnic groups have shown support for the movement as
they benefit from the resulting social change (e.g., Vani et al. 2022). Finally, prior literature in psychology
argue that when individuals identify with or are more familiar with intentions of certain collective action,
they feel positively inspired by the idea that they can make a difference going forward, leading to positive
moods (e.g., Landmann and Rohmann 2020; Sabherwal et al. 2021). Building on these findings, loan officers
may identify with the movement, which may positively affect their moods, and consequently lead to an
increase in credit approvals at local banks. The increase in credit approvals may be driven by low quality
borrowers since these types of borrowers require a greater use of subjective judgment by the loan officers
(e.g., Klein and Barnes 1994; Beilock and DeCaro 2007; Wemm and Wulfert 2017; Cortés, Duchin, and
Sosyura 2016; Agarwal, Duchin, and Sosyura 2012). As a result, P2P lending will decrease in the three-
digit zip codes of the protests. Taken together, whether BLM protests negatively affect the sentiment of
local loan officers is an empirical question.
15
Figure 2.1: LendingClubScreens
(a)PanelA:LoanFilters
Figure 2.1a represents a screen of filters that investors can choose from on LendingClub.
(b)PanelB:AutomaticInvestingPortfolioWeight
Figure 2.1b represents a screen where investors can specify each weight when they choose to ‘automatically’ invest. The five
letters (“A” through “E”) represent loan grades algorithmically assigned by LendingClub. Investors can input numbers into the
box or adjust percentages using the up/down button.
16
(c)PanelC:BrowsingLoans
Figure 2.1c represents a browsing screen that investors see to browse the filtered loans after selecting the filters in Panel A. If
investors choose ‘automatic investing,’ then the algorithm automatically selects loans that correspond to the investor’s portfolio
weight as assigned in Panel B. If investors choose to ‘manually invest,’ then investors must select the loan that they wish to invest
in, specify the dollar amount ranging from a minimum of $25 up to $2,500 in $25 increments, and click on the ‘Add to Order’
button to purchase the loans. Rate presents the loan grade assigned by LC, with the letter representing the base risk and the
number representing the sub-risk. Loan grade of a “D1” is less risky than a loan grade of “D5.” Term is the loan maturity with
the options being a three-year or a five-year maturity. FICO is the borrower’s FICO score range. Amount is the amount of funds
requested by the borrower. Purpose is the loan purpose specified by the borrower. % Funded is the percentage of the loan that is
funded. Amount/Time Left is the amount and the duration left until the loan is issued.
17
Chapter3
Data
3.1 Sources
I begin with all loan applications from LendingClub (“LC”) between 2014 and 2018. The P2P loan data
contains 8,609,413 loan applications of which 1,412,362 are accepted. All of the loans contain information
such as FICO score, DTI ratio, employment length, and the borrower’s three-digit zip code.
∗
The accepted
loans (i.e., loans that passed the platform’s automated screening process) contain additional information
related to the borrower’s credit history such as number of inquiries, number of delinquencies, loan grade
assigned by LC, number of credit lines, maturity of the loan, home ownership status, and other character-
istics associated with the borrower’s credit history. Next, I scrape the locations and dates of BLM protests
from Alisa Robinson’s website from 2014 to 2018. The BLM protest data from the website has been widely
used by researchers in social sciences trying to understand more about the various effects of the protest
(e.g. Mazumder 2019; T. Campbell 2021). I then link each city with a BLM protest with its respective three-
digit zip code using the zip code data from the Census. As the final step, I merge the loan data from LC
with the BLM protest data based on the zip code. I label a three-digit zip code as a “protest area” if there is
at least one protest in a city that belongs to that three-digit zip code in a given year-month. If there is no
protest in a given year-month, the three-digit zip code is a “control area.”
∗
LC only discloses the borrower’s three-digit zip code. However, the borrower’s three-digit zip code is unavailable to investors
in real time.
18
3.2 SampleSelection
To reduce concerns that possible correlated omitted variables that affect the P2P lending in an area also
affect the probability that the area experiences a BLM protest, I first perform a nearest-neighbor matching
procedure between the three-digit zip codes that experienced a protest and the three-digit zip codes that
never experience a protest based on the area’s population, median unemployment rate, and median income.
As the first step in the matching process, I calculate a propensity score for the two areas every year. For
each protest area in a year, I match five never-protest areas with the lowest difference between the two
scores. This results in one protest area being matched to five never-protest areas. I exclude the unmatched
areas from the analysis.
†
The matching process results in the final sample of 38,393 year-month-area
observations from January 2014 to December 2018. Within that period, there are 789 unique three-digit
zip codes, with BLM protests occurring in 281 areas and never having occurred in 508 areas.
3.3 DescriptiveStatistics
In Figure 3.1, I plot the natural log of the number of borrowers in each three-digit zip code. Red areas
represent three-digit zip codes with more borrowers and blue areas represent areas with less borrowers.
In Table 3.1, I report the distribution of borrowers across each state. Among the 1,412,362 loans that are
issued, California has the greatest number of borrowers with 200,403, corresponding to 14.19% of all loans
from 2014 to 2018. The smallest number of borrowers are from Wyoming, with 593 borrowers.
In Table 3.2, I report the descriptive statistics for my sample. In Panel A, I report baseline charac-
teristics for all loan applications. The borrower characteristics that are available for all applications are
the natural log of the loan amount requested, borrower’s FICO score, debt-to-income ratio (DTI), and the
length of employment. The average loan amount is 9.056 which correspond to around $8,500. The av-
erage FICO score for all applicants is 644. The average applicant also has a DTI of 29.674% and a work
†
The results are robust to using an unmatched sample.
19
experience of approximately one and a half years. There are approximately 8.6 million loan applications
from 2014 to 2018. Panel B reports the accepted loans sample. Of the 8.6 million loan applications, ap-
proximately 1.4 million applications are accepted by the platform. Recall from Section?? that this process
is automated using proprietary technology of LC. The average borrower who is accepted by the platform
asks for approximately $12,800, which corresponds to 9.458 when log transformed. The average borrower
has a FICO score of approximately 700, a DTI of 18.587%, and an employment length of approximately
6 years. In Panel C, I report descriptive statistics for the primary sample that is used in this study. For
the dependent variables of interest that characterize P2P lending, I follow prior literature and measure the
natural log of the number of loan originations and amount of loan originations per each three-digit zip
code for a particular year-month. The variables are calculated one month after each BLM protest to allow
for the migration of borrowers between banking and P2P lending.
‡
For control variables that account for
area-level demographics and economic factors that might affect P2P lending, I include the natural log of
the total population, the median unemployment rate and the median income of the respective three-digit
zip code. For control variables that might affect the banking dynamics of the three-digit zip code, I in-
clude the Herfindahl-Hirschman Index (HHI), the geographic diversification of banks in the area, and the
total deposits of banks in the area standardized by the population. All of the variables are described in
detail in Appendix?? including their sources. The protest year-month-areas are 2.6% of the sample, which
correspond to 1,001 year-month-area observations.
‡
I spoke with loan officers in the Los Angeles area which suggested that most loan decisions are finalized within one month.
20
Figure 3.1: BorrowerLocation
Figure 3.1 plots the natural log of the number of borrowers in each three-digit zip code over the entire sample period. Red
represents areas with more borrowers and blue represents areas with less borrowers.
21
Table 3.1: DistributionofBorrowers
In Table 3.1, I present the distribution of borrowers in each state. N represents the number of borrowers that are from each state.
State N Percent
California 200,403 14.19
Texas 117,354 8.31
New York 115,747 8.2
Florida 107,198 7.59
Illinois 58,817 4.16
New Jersey 53,415 3.78
Georgia 48,541 3.44
Ohio 47,886 3.39
Pennsylvania 44,068 3.12
North Carolina 41,211 2.92
Michigan 38,169 2.7
Virginia 37,422 2.65
Maryland 35,601 2.52
Arizona 34,980 2.48
Massachusetts 33,235 2.35
Washington 30,079 2.13
Colorado 28,307 2
Tennessee 24,035 1.7
Indiana 23,959 1.7
Connecticut 23,899 1.69
Minnesota 23,037 1.63
Missouri 21,417 1.52
Nevada 20,686 1.46
South Carolina 18,633 1.32
Wisconsin 18,194 1.29
Oregon 16,499 1.17
Alabama 16,291 1.15
Louisiana 16,139 1.14
Kentucky 12,380 0.88
Oklahoma 11,928 0.84
Arkansas 9,763 0.69
Utah 9,276 0.66
Mississippi 9,214 0.65
Kansas 9,186 0.65
Hawaii 6,721 0.48
Rhode Island 6,705 0.47
New Mexico 6,669 0.47
New Hampshire 5,339 0.38
Delaware 4,320 0.31
Nebraska 3,935 0.28
Washington DC 3,340 0.24
Idaho 3,066 0.22
Alaska 2,996 0.21
Maine 2,491 0.18
West Virginia 2,465 0.17
Montana 2,066 0.15
South Dakota 1,758 0.12
Vermont 1,608 0.11
North Dakota 1,321 0.09
Wyoming 593 0.04
Total 1,412,362 100
22
Table 3.2: LoanDescriptives
In Table 3.2, I present the descriptive statistics. Panel A reports the statistics for all loan applications, Panel B reports the statistics for only the
accepted loans, and Panel C presents the statistics for the primary sample used in the study. All variables are as defined in the Appendix ??. All
continuous variables are winsorized at the 1 and 99 percentiles.
PanelA:AllLoanApplications
N P25 Median P75 Mean SD
Loan Amount 8,609,413 8.517 9.210 9.904 9.056 1.016
FICO Score 8,609,413 602.000 649.000 687.000 644.383 65.079
DTI 8,609,413 9.990 19.970 33.660 29.674 44.708
Employment 8,609,413 0.000 0.000 0.000 1.397 3.004
PanelB:AcceptedLoans
N P25 Median P75 Mean SD
Loan Amt 1,412,362 9.036 9.547 9.952 9.458 0.686
Loan Grade 1,412,362 6.000 10.000 14.000 10.724 6.095
FICO Score 1,412,362 677.000 697.000 722.000 702.402 33.471
DTI 1,412,362 11.900 17.840 24.590 18.587 9.011
Employment 1,412,362 3.000 7.000 9.000 6.035 3.471
Maturity 1,412,362 0.000 0.000 1.000 0.337 0.473
Inq Last 6 mo 1,412,362 0.000 0.000 1.000 0.502 0.772
Open Acc 1,412,362 8.000 11.000 15.000 11.757 5.547
Revol Util 1,412,362 29.600 48.000 67.300 48.571 24.622
Delinquency 1,412,362 0.000 0.000 0.000 0.285 0.713
Revol Bal 1,412,362 8.709 9.355 9.941 9.249 1.079
Total Acc 1,412,362 15.000 22.000 31.000 24.121 11.708
Homeowner 1,412,362 0.000 0.000 0.000 0.113 0.317
Mortgage 1,412,362 0.000 0.000 1.000 0.498 0.500
PanelC:Area-LevelStatistics
N P25 Median P75 Mean SD
N Funded 38,393 2.303 3.045 3.829 3.071 1.087
Amt Funded 38,393 5.325 6.061 6.846 6.056 1.126
Protest 38,393 0.000 0.000 0.000 0.026 0.159
Population 38,393 9.608 10.154 10.753 10.183 0.812
Unemployment 38,393 3.560 4.408 5.386 4.607 1.498
Income 38,393 45.364 52.870 63.799 57.362 18.265
HHI 38,393 11.429 14.420 18.426 16.479 9.039
Geo Diversification 38,393 6.000 17.000 41.000 31.420 42.097
Deposits 38,393 10.647 14.070 20.765 19.969 25.475
23
Chapter4
Results
4.1 PrimaryResults
I organize my empirical analysis in the following manner. I first start by examining the effect of BLM
protests on P2P lending volume and amount. I then test the parallel trends assumption and conduct fal-
sification tests using randomly generated pseudo-protest months. Next, I examine whether my primary
results are driven by alternative explanations based on an increased demand for credit. Next, to help sub-
stantiate that my primary results are driven by changes in loan officers’ moods, I examine whether the
results are stronger for cross-sections of the data designed to capture differential changes of loan officer
sentiment. Lastly, I examine the remaining potential alternative explanations based on possible changes
in lender behavior and the P2P platform’s credit model.
In this section, I start with an examination of the effect of BLM protests on P2P lending volume and
lending amount. Specifically, I implement the following differences-in-differences design following M.
Bertrand, Duflo, and Mullainathan 2004:
Y
it+t
=A
it
+P
it
+cX
it
+βProtest
it
+ϵ it
. (4.1)
24
In equation 4.1 the subscriptsi andt denote the three-digit zip code and year-month, respectively. Y
it
is
the dependent variables of interest that characterize P2P lending. I use the number of P2P loan originations
and the amount of P2P loan originations to capture changes in the P2P lending. In tests of borrower quality,
I use the number of P2P loan originations in the low quality and high quality group.
∗
A
it
denotes the area
fixed effect (i.e., three-digit zip code) and P
it
denotes the period fixed effect (i.e., year-month). X
it
is a vector
of area-level control variables andϵ it
is the error term. Protest
it
is an indicator that equals to one when a
particular three-digit zip code experiences a BLM protest in a given year-month and otherwise zero. The
Protest
it
variable is allowed to shift between 0 and 1 depending on when a particular area experiences or
does not experience a protest for a given year-month (e.g., Acemoglu et al. 2019; Imai, I. S. Kim, and E. H.
Wang 2021; Bilach, Roche, and Wawro 2022). The coefficient of interest is β , which compares the change
in the dependent variable of protest areas to the change in the dependent variable of control areas over
the same time period. Control variables include the list of variables that affect the P2P lending in the area.
Area fixed effects are included to control for time-invariant characteristics associated with each area that
may affect the decisions of the borrowers in the area to apply for P2P lending. Year-month fixed effects
are included to control for any time-varying economy wide factors. The standard errors are clustered at
the area and year-month level to account for any time-series or cross-sectional correlations (e.g., Petersen
2009; Gow, Ormazabal, and Taylor 2010).
Table 4.1 presents the effect of BLM protests on P2P lending volume, amount, and borrower quality.
In Panel A, Column (1), where the dependent variable of interest is the P2P loan origination amount, I
find that one month after a BLM protest, there is a significant increase in P2P lending amount (coefficient
= 0.016, t-stat 2.927). In Column (2), where the dependent variable of interest is the number of P2P loan
originations, I find that one month after a BLM protest, there is a significant increase in the number of P2P
∗
Low quality group includes all borrowers with FICO scores 740 or below. High quality group includes all borrowers with
FICO scores above 740 (e.g., Agarwal, Duchin, and Sosyura 2012). The results are also robust to other splits such as quartiles.
25
loan originations (coefficient = 0.021, t-stat 3.650).
†
Economically, these results suggest that one month
after a BLM protest, there is a 1.6% increase in the P2P lending amount and a 2.1% increase in the number
of loan originations in the three-digit zip code that experienced a protest.
‡
As an additional test, in Panel
B, I examine the composition of borrowers that drive the increase in P2P lending. Column (1) denotes the
change in the number of loan originations in the low quality borrower bucket and Column (2) denotes
the change in the number of loan originations in the high quality borrower bucket. Consistent with Tang
2019 who finds that P2P lending and commercial lending are substitutes where P2P serves infra-marginal
borrowers who are rejected at local banks, I find that, after a BLM protest, the increase in P2P loan orig-
inations is driven by the lower quality borrowers as denoted by the positive and significant coefficients
in Column (1) (coefficient = 0.050, t-stat 4.643). Taken together, the evidence from Table 3 is consistent
with the hypothesis that there is a negative shift in the loan officers’ sentiment following a BLM protest,
which has a tightening effect on credit. That is, a negative shift in the sentiment of local loan officers
after a BLM protest induces a negative shock to commercial lending supply, resulting in an influx of lower
quality borrowers into the P2P platform who are rejected from banks.
4.2 RobustnessTests
4.2.1 LeadandLaggedProtests
In this section, I carry out two analyses designed to test the identifying assumptions in my research design.
I first estimate the coefficients on lagged and lead values of protests. Lead and lag protest variables proxy
for relative months before and after a BLM protest in a given three-digit zip code and capture anticipatory
†
In untabulated tests where the Protestit indicator is replaced with a count of BLM protests, I continue to find a similar
effect.
‡
The precise interpretation of the β coefficient in the log-dummy regression (e.g., ln(Y) = α + β * D + ϵ ) is a percentage
change inY associated with switching from 0 to 1 for D. The percentage change is calculated as 100 * (e
β ˘1 ). In this case, when
β = 0.016, the expression approximately equals 1.613.
26
and persistent effects of BLM protests (e.g., Smith, Clarke, and Pease 2002; Bilach, Roche, and Wawro 2022).
I empirically test this specification in the following way:
Y
it+1
=A
it
+P
it
+cX
it
+βProtest
it
+
2
X
k=− 2;k̸=0
θ k
Protest
it+k
+ϵ it
. (4.2)
The treatment variableProtest
it
is the same as in equation (4.1). I include two leads and lags of BLM
protests indexed by the summation notation (note thatk excludes zero as this is captured by the treatment
variableProtest
it
). The lead protest variable, for example,Protest
it− 1
, equals to one when the area ex-
periences a protest in the next month and zero otherwise. As a specific example, assume that Los Angeles
experienced a BLM protest in February of 2014. The lead protest indicator,Protest
it− 1
, will be 1 for Los
Angeles in January of 2014. It will be zero in February of 2014. The lag protest variable, for example,
Protest
it+1
, equals to one when the area experienced a protest in the month prior and is zero otherwise.
If Los Angeles experienced a protest in February 2014, Protest
it+1
equals to one in March of 2014. The
coefficient on the lead protest variables ( Protest
it− 1
;Protest
it− 2
) captures an anticipatory effect and the
coefficients on the lag protest variables ( Protest
it+1
; Protest
it+2
) captures persistent effects. If parallel
assumptions hold, I expect to find that the coefficients on the leading protest variables to be statistically in-
distinguishable from zero.
§
Area and year-month fixed effects are included and standard errors are double
clustered on area and year-month. Table 4.2, Panel A presents the coefficients from estimating equation
4.2. In both Columns (1) and (2), I find that the leading protest coefficient is statistically insignificant, in-
dicative of no anticipatory effects of the BLM protests on P2P lending. The effect of the protests on loan
officers’ moods seem to be short lived, as shown by the statistically insignificant coefficients on the lagged
protest variables.
§
In untabulated tests, I examine trends before the first BLM protest of the region and find that inferences do not change.
27
4.2.2 FalsificationTest
Next, to provide additional support for the mechanism that BLM protests affect the loan officer’s sentiment
which ultimately results in an influx of borrowers into the P2P platform, I conduct a falsification test using
pseudo-BLM protest dates. If BLM protests are instrumental in explaining the increase in P2P lending,
I should not find the same result as my primary analysis in a set of areas that never experience a BLM
protest. Therefore, in this analysis, I restrict the sample to the set of areas that never experience a BLM
protest. I then randomly assign a fraction of area-year-month observations to become pseudo-protest
dates. The random assignment is constructed such that the fraction of pseudo-protest dates matches the
fraction of actual protest dates. I then estimate equation 4.1 using the same three dependent variables –
P2P loan origination amount, number of P2P loan originations, and borrower quality buckets. The results
are reported in Table 4.2, Panels B and C. In both Columns (1) and (2) of Table 4 Panel A, I find that the
coefficient estimates on the Protest variable is statistically indistinguishable from zero. Consistent with this
in Table 4 Panel B, I show that the coefficients on the two splits of borrower quality are also statistically
insignificant. Overall, the evidence from Table 4 supports the mechanism that BLM protests negatively
affect the local loan officers’ sentiment, which leads to an increase in P2P lending in the subsequent month.
4.3 AlternativeExplanations
4.3.1 IncreasedDemandforCredit
While the increase in P2P lending is consistent with protest-induced moods negatively affecting loan of-
ficers, the increase in P2P lending may be driven by an increased demand for credit after protests. In this
section, I specifically investigate whether this is true. If the increase in P2P lending is driven by demand-
related reasons, I should also expect to find that bank lending also increases after protests. Because P2P
lending and banking are substitutes (e.g., Tang 2019), if demand for a particular source of credit increases
28
(i.e., P2P lending), then it is likely that demand for bank credit also increase. In order to do so, I collect the
outstanding balances of consumer and small business lending from Schedule RC-C of Call Reports. As the
Call Reports are announced quarterly, I aggregate my observations to year-quarters.
I report the coefficients from estimating equation 4.1 at the quarter level in Table 4.3, Panel A. The
independent variable is whether a particular zip code experienced a protest in a given quarter. The de-
pendent variable is the natural log of the outstanding balance of consumer and small business lending for
each zip code-quarter.
¶
In Table 4.3, Panel A, I find that the coefficient on QtrProtest
i,t
is significantly
negative. This result lends some support that the increase in P2P lending following protests is not entirely
driven by demand related reasons. In addition, the negative and significant coefficient suggests that loans
made by banks to consumers and small businesses during each quarter of protest decreases. The decrease
is more consistent with the substitution between bank lending and P2P lending. That is, certain borrowers
who are unable to secure credit at banks migrate to P2P platforms.
4.3.2 FootTraffictoBankBranches
Next, I further investigate whether there is an increased demand for credit at local banks by empirically
examining the association between protests and foot traffic to local bank branches. The foot traffic data has
been widely used by recent studies in accounting to examine consumer responses to corporate tax news
(e.g., Asay et al. 2022), earnings persistence (e.g., Jin, Stubben, and Ton 2022), and earnings announcements
(e.g., Noh, So, and Zhu 2022). Using foot traffic data from SafeGraph, I measure the aggregated number of
foot traffic to bank branches in a given year-month.
∥
As the foot traffic data is available starting in 2018,
I limit my sample to only 2018. I re-estimate equation 4.1 using the natural log of the foot traffic for all
¶
Call Reports are filed at the end of each quarter and thus I do not introduce lags in the specification. This results in the
subscripts to be concurrent (i.e., t instead of t+1).
∥
A caveat in this measure is that I am unable to capture banking activities that are conducted online. However, Noh, So, and
Zhu 2022, who document that foot traffic to stores significantly increase after firms’ earnings announcements, find similar results
in a panel of online transactions. To the extent that foot traffic to banks is positively correlated with the aggregate demand for
credit, the foot traffic data noisily proxies for the overall demand for credit.
29
banks in a three-digit zip code in a given year month as the dependent variable. The results are reported
in Table 4.3, Panel B. I find that the coefficient on Protest
i,t
is not statistically significant zero, consistent
with a no change in the demand for credit following protests.
4.3.3 LoanPurposeAnalysis
Complementary to examining the balance of loans outstanding at banks, I take a closer look at the purpose
of the loans that drive the increase in P2P lending. A potential reason for the increase in P2P lending may
be that rational borrowers foresee an increased support for minorities following the protests (e.g., Vani et
al. 2022) and request loans to open up businesses. If this is the case, I should expect to observe an increase
in loans that are requested for business purposes. I categorize loans into two categories – loans related
to business purpose and loans unrelated to business purpose. In the sample, business related loans make
up approximately 13% of the accepted loan sample and are composed of loans that are applied for “Small
Business” and “Major Purchase” purposes. On the other hand, non-business loans include loans that are
related to day-to-day expenses such as purchasing cars, consolidating debt, improving housing, medical
fees, and vacation. I use the number and amount of loan originations that pertain to the two categories as
new dependent variables and estimate equation 4.1.
I report the coefficient estimates in Table 4.3, Panel B. In Columns (1) and (2), I report the results when
the loans are in the non-business category and in Columns (3) and (4), I report the coefficient estimates
when the loans are in the business category. I find that my primary results are driven by non-business
loans and that the coefficients on Protest
it
for business loans are statistically indistinguishable from zero.
The results from examining the purpose of the loans suggest that the primary results are driven by the
loans requested for non-business purposes.
30
4.3.4 SmallBusinessAdministration(SBA)Loans
Lastly, I continue to examine whether the demand for business loans increased following BLM protests by
examining loans from the Small Business Administration (SBA). There are two types of loans offered by
the SBA: 7(a) and 504 loans. For my purpose, I focus on 7(a) loans as these loans are more flexible than 504
loans in that 1) they can be used for a variety of purposes, while 504 loans are specifically for land and real
estate; 2) 504 loans are suited for larger businesses.
∗∗
SBA loans have been used to proxy for demand for
credit for smaller entrepreneurs (e.g., Huang 2019). The SBA dataset includes the address of the business,
location of the bank which granted the loan, loan grant date, loan mount, etc.
I report the coefficient estimates from estimating equation 4.1 when the dependent variables are the
amount and the number of SBA loan originations in Table 5, Panel D. Consistent with the demand for
credit not changing following protests, I find a statistically insignificant coefficient in both columns (1)
and (2) of Table 5, Panel D. This set of results further supports the idea that the demand for business loans
did not increase following BLM protests.
4.4 Cross-SectionalAnalyses
In this section, I further substantiate that my primary results (i.e., increase in P2P lending following
protests) are driven by the substitution between bank and P2P lending by conducting three cross-sectional
tests that capture differential changes of loan officer moods. That is, if changes in loan officers’ moods
leads borrowers to migrate to P2P lending, I expect to find stronger results for cross-sections that are as-
sociated with greater changes of moods. Moreover, to the extent that the increase in the demand for credit
should not co-vary with cross-sections that capture differential changes in loan officers’ moods, this set of
analyses should also help reduce concerns that the primary results are demand driven.
∗∗
Obtained from the SBA.
31
4.4.1 PoliticalAffiliation
I first examine the three-digit zip code’s political affiliation. The political science literature finds there exists
a partisan divide in participation and support for the BLM movement. Parker, Horowitz, and Anderson
2020 find evidence that Democrat or Democrat-leaning adults are more likely to support BLM compared
to Republican or Republican-leaning adults. Prior literature in psychology finds that identifying with the
intentions of protests induces individuals to feel that they are able to make a difference in the future and
consequently, leads to positive moods (e.g., Landmann and Rohmann 2020; Sabherwal et al. 2021). Based
on this argument, I hypothesize that individuals in regions that are associated with greater support for
the movement will be less affected by the stress generated from interpersonal conflicts of opinion (e.g., Ni,
Li, et al. 2016). Therefore, to the extent that the political affiliation of the area may be generalized to the
average individual’s political affiliation, I expect my primary results to be, on average, weaker for more
Democrat leaning areas compared to more Republican leaning areas. A caveat in this research design
is that I am unable to directly measure the loan officers’ political affiliation and that I am assuming an
individual’s political affiliation aligns with the average political affiliation of the three-digit zip code.
A three-digit zip code is defined as more Democrat leaning if the percentage share of the democratic
votes in the three-digit zip code is larger than the national median percentage of votes for democrats, oth-
erwise, a three-digit zip code is classified as more Republican leaning.
††
I estimate equation 4.1 separately
for the Democrat leaning and Republican leaning subsamples. This research design allows all of the co-
efficient estimates on the right-hand-side variables to vary by each subsample. I then test the difference
between the coefficients of interest Protest
it
on the Republican sample and the Democrat sample by fol-
lowing the bootstrap procedure in Shroff, Verdi, and Yu 2014 and Barth et al. 2018. Table 4.4 reports the
coefficient estimates from estimating equation 4.1 in the Republican and the Democrat subsample. The
††
I measure the area’s political affiliation by using the 2016 presidential election data at the precinct level from the MIT data
lab. I then convert the vote counts aggregated by each county to three-digit zip codes using the county-zip code crosswalk from
HUD Crosswalk for 2016Q4.
32
difference between the coefficient estimates and its associated p-value is reported in the “Difference” row.
Columns (1) and (2) present results when the dependent variable of interest is the P2P loan origination
amount and Columns (3) and (4) present results when the dependent variable of interest is the number of
P2P loan originations. In the Republican sample in Column (1), I find that the coefficient on Protest
it
is
statistically significant (coefficient = 0.015, t-stat 2.471). However, in the Democrat sample in Column (2),
I find that the coefficient on Protest
it
is statistically insignificant (coefficient = -0.006, t-stat -0.198). The
difference between the two is 0.021 and is statistically significant at the 10% level. While the difference
between the coefficients on Protest
it
for Columns (3) and (4) is not significant, the p-value is 0.106. In
Panel B, where I examine the composition of borrowers that drive the increase in P2P lending using high
and low FICO scores, I find that the increase P2P lending is driven by low quality borrowers in Republican
leaning areas. The difference between Columns (1) and (2) in Panel B is 0.058 and is statistically significant
at the 5% level. In contrast, the difference between Columns (3) and (4) in Panel B is statistically indis-
tinguishable from zero. Overall, these findings lend support to the mechanism that BLM protests affect
the mood of local loan officers, with a weaker change in sentiment associated with loan officers in more
Democratic leaning areas.
4.4.2 BankCompetition
Next, I examine the competitiveness of the banking market for each three-digit zip code. Prior literature
finds that a factor which affects the relationship between commercial lending and P2P lending is the level
of banking competition (e.g., Havrylchyk et al. 2016; Jagtiani and Lemieux 2018; Beck, Behr, and Madestam
2018). Therefore, I explore whether the intensity of banking competition is a factor which moderates the
shift in loan officers’ sentiment after BLM protests. I hypothesize that if the area’s banking market is more
(less) competitive then the loan officers in the area are likely to be less (more) affected by sentiment due
to the more (less) competitive pressure (e.g., Hong and Kacperczyk 2010). I measure the three-digit zip
33
code’s banking competition using the Herfindahl-Hirschman index based on deposit shares. A three-digit
zip code is classified as high (low) competition area if the HHI score is lower (greater) than the median. I
then separately estimate equation 4.1 for the high and low competition subsamples.
I report the coefficient estimates in Table 4.5. In Panel A, I report the coefficient estimates for the P2P
loan origination amount and the number of P2P loan originations. Columns (1) and (2) indicate when the
dependent variable is P2P loan origination amount and Columns (3) and (4) indicate when the dependent
variable is the number of P2P loan originations. For the high competition sample. I find that the coefficient
onProtest
it
is statistically not different from zero in Column (1). However, for the low competition sample
in Column (2), I find that the coefficient on Protest
it
is positive and significantly different from zero.
The difference between the coefficients is -0.023 and is significant at the 5% level. I find similar results
when the dependent variable is the number of P2P loan originations. In Panel B, I conduct regressions
of borrower quality and find that the increase in P2P lending in low competition areas are driven by low
quality borrowers. Overall, the evidence from Table 4.5 presents evidence consistent with my hypothesis,
the effect of BLM protests on loan officers’ sentiment is lower in areas that are highly competitive and is
greater in areas that are less competitive.
4.4.3 EconomicDisadvantage
Third, to provide further evidence that the increase in the P2P lending is driven by weak borrowers who
require greater use of subjective judgment by the loan officers, I categorize each three-digit zip code as
either an economically disadvantaged or not disadvantaged area based on the area’s income level. If the
area either has an income level that is lower than the cross-sectional median income then the area is
classified as a economically disadvantaged area. Following prior literature in psychology which finds that
stress leads to lower performance for tasks that require complex decision making (e.g., Klein and Barnes
1994; Wemm and Wulfert 2017; Beilock and DeCaro 2007), I predict that the increase in P2P lending will
34
be greater in areas that are economically disadvantaged compared to the increase in areas that are not
disadvantaged.
In Table 4.6, I report the coefficient estimates. In Panel A, I report the coefficient estimates for the
P2P loan origination amount and the natural log of the number of P2P loan originations. Columns (1)
and (2) report the coefficients when the dependent variable is the loan amount and Columns (3) and (4)
report the coefficients when the dependent variable is the number of loans. I find that the difference in
the coefficients on Protest
it
between Columns (1) and (2) is 0.034 and is statistically significant. Similarly,
the difference in the coefficients between Columns (3) and (4) 0.036 and is also statistically significant.
These results support the prediction that the increase in P2P lending in economically disadvantaged areas
is greater than the increase in P2P lending in non-disadvantaged areas.
‡‡
In Panel B, I report the tests of
borrower quality and also find the increase in P2P lending in economically disadvantaged areas are driven
by low quality borrowers.
4.4.4 SingleLocationBanks
I continue to examine whether the increase in P2P lending may be driven by changes to loan officers’
sentiment by classifying three-digit zip codes into areas that are dominated by single location banks. This
cross-section is motivated by the banking literature which shows that loan officers at banks with single
branches (i.e., banks that operate in small geographic regions) are more likely to have control over their
decision making processes (e.g., Cortés, Duchin, and Sosyura 2016; Agarwal, Duchin, and Sosyura 2012).
For example, Cortés, Duchin, and Sosyura 2016 state that loan approval decisions at local community
banks are less likely to be automated. Moreover, managers at community banks are located in the same
area, which increases the chance that they are affected by similar exogenous factors such as protests. In line
with this, I hypothesize that loan officers who work in areas that are dominated by single bank branches
‡‡
Inferences do not change when the cross-section split is based on income and education following prior literature on socio-
economic disadvantage (e.g., Laws et al. 2014).
35
are more likely to be affected by protest-induced sentiment, leading to a greater increase in P2P lending in
those areas.
Consistent with this hypothesis, Table 4.7, Panel A finds that the increase in P2P loan origination
and amount is significantly greater in areas dominated by single bank branches compared to areas with
banks that operate in large geographical areas. The difference in the coefficients on Protest
it
is 0.037 for
columns (1) and (2); 0.039 for columns (3) and (4). Both coefficients are statistically significant at the 1%
level. Moreover, in Table 4.7, Panel B, I further show that the increase in P2P lending is driven by lower
quality borrowers.
4.4.5 MinorityBankBoards
Finally, my final cross-sectional test examines how minority representation on bank boards affects the
substitution dynamic between P2P and commercial lending.
§§
This cross-sectional test is motivated by the
banking literature which finds that increased minority representation on bank boards leads to a reduced
likelihood that a similarly qualified minority applicant is rejected compared to their non-minority coun-
terpart (Goenner 2023). Building on this, I predict that areas that are dominated by banks with greater
minority representation on bank boards are more likely to experience a smaller increase in P2P lending
compared to areas that are dominated by banks with smaller minority representation. That is, if there is
a greater percentage of minorities on bank boards, it will lead to smaller likelihood that a loan is rejected
following protests.
I report coefficients from estimating equation 4.1 in Table 4.8. In Panel A, I find that the coefficient
onProtest
it
is significantly greater for areas dominated by low minority banks. The difference between
columns (1) and (2) is -0.037 and is significant at the 10% level and the difference between columns (3) and
§§
I compute minority representation by using a computer algorithm rethnicity. I first compute the percentage of black indi-
viduals on the banks’ board. I then compute the median percentage for each three-digit zip code. An area is classified as a high
minority area if the median percentage for the zip code exceeds the cross-sectional median.
36
(4) is -0.017 and is also similarly significant. In Panel B, I continue to show that the increase in P2P lending
associated with low minority areas are driven by low quality borrowers.
Overall, my cross-sectional analyses provide support for the hypothesis that local loan officers’ senti-
ment is adversely affected by BLM protests, leading to a decrease in credit approvals at local banks. Con-
sequently, the borrowers who are rejected migrate to P2P lending, leading to an increase in P2P lending
in the areas where BLM protests occur.
4.5 OtherAlternativeExplanations
Finally, I examine other possible alternative explanations that may be drivers of the increase in P2P lend-
ing. More specifically, I examine the possibility that the increase is driven by 1) changes to P2P lenders’
willingness to lend and 2) changes to the P2P platform’s credit model.
4.5.1 WillingnesstoLend
First, I examine whether P2P investors willingness to lend to weak borrowers changed before and after
protests.
¶¶
The ideal setting that allows the researcher to examine the change in the behavior of P2P investors
would be to estimate the percentage of loans that are sponsored by local P2P investors (i.e., P2P investors
that are located near the protests). However, as LC does not make their lender dataset available, I adopt
an alternate approach in Tang 2019 and estimate the change in the probability that a loan is fully funded
before and after protests.
∗∗∗
If P2P investors became more willing to lend to underprivileged borrowers
¶¶
P2P investors may be less affected by sentiment related reasons compared to loan officers in the vicinity of the protests (i.e.,
local loan officers). Cortés, Duchin, and Sosyura 2016 state that loan officers typically arrive at lending decisions in less than one
day. P2P investors, however, do not face a time constraint and are able to invest on their own time. Therefore, P2P investors are
less likely to be affected by changes in mood in the short run. Also, in contrast to loan officers who may work in the vicinity of
the protests, P2P investors and borrowers are often located in different regions (e.g., Agrawal, Catalini, and Goldfarb 2011) and
thus in different regions than the related protests. Therefore, the decisions of P2P lenders will be less affected by protest-induced
moods.
∗∗∗
If a loan request passes the credit algorithm which automatically accepts or rejects the loan, it is listed on the platform to
attract investors. If for some reason, a loan fails to attract enough borrowers, the loan is considered partially funded (i.e., not fully
funded).
37
after protests, then the probability that a loan is fully funded will increase, ceteris paribus. On the other
hand, however, if P2P investors’ willingness to fund did not change, then the probability that a loan is fully
funded will not change. I estimate the following equation:
Funded
it+1
=A
it
+P
it
+cXLoan
it
+dXArea
it
+βProtest
it
+ϵ it
. (4.3)
Fund
it+1
is an indicator that equals to one if a loan is fully funded and zero otherwise. XLoan
it
are control variables constructed at each loan level that include borrower characteristics such as debt-
to-income ratio, the loan request amount, FICO score, length of employment and others. XArea
it
are
control variables constructed at each area-level which are also utilized in equation 4.1. All variables used
in the regression are described in detail in Appendix. The coefficient of interest is β . I include area and
year-month fixed effects and double cluster the standard errors at the area and year-month.
Table 4.9 reports the regression results from estimating equation 4.3. Columns (2) and (3) incrementally
build on Column (1) by including loan-level control variables, and both loan-level and area-level control
variables, respectively. In all three specifications, I find that the coefficient on Protest
it
is statistically
indistinguishable from zero, indicating that lenders’ willingness to lend did not change before and after
BLM protests.
†††
In Column (3), I observe that fundamental loan-level characteristics are significantly
associated with the probability that a loan is fully funded. For example, a loan more likely to be fully
funded if the borrower has a lower debt-to-income ratio, lower loan request amount, a higher FICO score,
and a longer employment history. Moreover, the loan is less likely to be fully funded if the borrower had a
larger number of delinquencies in the last 2 years. Taken together, the results from Table 4.9 suggests that
lenders’ behavior on the platform did not change before and after the protest.
†††
A limitation in my study is that investors are able to filter on the state where the borrower lives in. To mitigate this concern,
I conduct several robustness tests. In untabulated analyses where I replace zip code fixed effects with state fixed effects, the
inferences do not change. Moreover, I consider the fact that LC removed the location filter. Using loans issued between 2016-
2017, J. Bertrand and Weill 2021 state that investors cannot select borrowers based on geographic location. Consistent with their
sample period, I conduct a robustness test using loans issued after 2015 and find the primary results to be unchanged.
38
4.5.2 Non-BLMProtests
To provide further evidence that the willingness of P2P investors did not change before and after protests,
I replicate equations 4.1 and 4.3 for a set ofnon-BLM protests.
‡‡‡
The intuition here is that if loan officers
are similarly impacted by non-BLM protest-induced stress, then the local loan officers will likely face
negative moods. Consequently, P2P lending in the areas of non-BLM protests will increase. Thus, I expect
to find that P2P lending increases following non-BLM protests. Moreover, unlike BLM protests, non-BLM
protests do not involve the underprivileged. Thus, after non-BLM protests, P2P lenders are unlikely to be
more willing to fund the underprivileged. Therefore, I expect to find that the probability of a loan being
fully funded to not change after non-BLM protests.
I report the results from estimating equations 4.1 and 4.3 in Table 4.10. Columns (1) and (2) estimate
regressions when the dependent variable is the amount of P2P loans and the number of P2P loans originated
from the area of the non-BLM protest. Columns (3), (4), and (5) estimate regressions of lender behavior
with Columns (4) and (5) building on Column (1) by adding loan controls and both loan and area controls,
respectively. In Columns (1) and (2), I continue to find that the coefficient on NonBLMProtest
it
is
statistically significant.
§§§
In Columns (3) to (5), I continue to find that P2P investors’ willingness to lend
does not change after non-BLM protests. In summary, the results from Table 4.10 provide support that the
primary results are not driven by local P2P lenders who feel a need to support weak borrowers.
4.5.3 P2PCreditModel
Lastly, I examine whether the increase in P2P lending is driven by changes to LC’s credit model. LC uses
the proprietary algorithm to assess a borrower’s profile and automatically accepts or rejects borrowers.
LC’s annual reports state that the algorithm leverages behavioral data, transactional data and employment
‡‡‡
I scrape the location and dates of non-BLM protests from the Crowd Counting Consortium. In order to maintain compa-
rability of protests, I define non-BLM protests to be those related to relatively well-known issues such as environmental rights.
Formal definitions are provided in Appendix.
§§§
The sample size is smaller as the non-BLM protests span 2016-2018, a subset of my sample. I replicate the matching process
for the non-BLM protests to arrive at the final sample.
39
information to supplement traditional risk measures such as FICO scores. For LC, when a loan isaccepted,
it means that it has been approved by the proprietary credit model to be considered for funding on the
platform. This is different from the traditional meaning of a loan being accepted in the literature, which
corresponds to the loan being originated. LC states that it continues to revise the proprietary algorithm
to reflect changes in socioeconomic and macroeconomic conditions. Therefore, it is possible that the in-
crease in P2P lending may be driven by the platform’s increased willingness to accept lower quality loans
after protests. If LC revised the algorithm after BLM protests and became more willing to lend to weaker
borrowers, the probability that a loan request is accepted,ceterisparibus, will be higher after BLM protests.
On the other hand, if LC did not revise their algorithm, the probability that a loan request is accepted will
stay the same before and after protests.
¶¶¶
I formally examine this question by estimating the following
equation:
Accepted
it+1
=A
it
+P
it
+cXLoan
it
+dXarea
it
+βProtest
it
+ϵ it
. (4.4)
Accepted
it+1
is an indicator variable that equals one if a loan is accepted by the platform.XLoan
it
are
loan-level control variables that are available to both accepted and rejected loans. They include the loan
amount, the FICO score, debt-to-income ratio, and the length of the employment. XArea
it
are area-level
control variables that are used in prior analyses. All variables are defined in Appendix. I employ the same
two-way fixed effects and double cluster the standard errors on year-month and area as in equation 4.1.
The coefficient of interest is the β coefficient on Protest
it
.
Table 4.11 reports the regression results from estimating equation 4.4. I find that all of the coefficients
onProtest
it
are not statistically significant in Columns (1), (2), and (3). Personal characteristics such as
FICO scores, debt-to-income ratio, and employment length are highly correlated with whether the loan is
¶¶¶
I spoke with LC representatives which suggested that they do not believe their credit models were adjusted in a way that
are closely correlated with BLM protests.
40
accepted. Overall, the results from Table 4.11 suggests that the platform’s credit model did not experience
a change before and after BLM protests.
41
Table 4.1: PrimaryResults
In Table 4.1, I present the results from estimating equation 4.1: Yit+1= Ait+Pit+cXit+βProtest it+ϵ it. Ait and Pit are three-
digit zip code and year-month fixed effects. Xit are area-level control variables. Protestit is an indicator that equals to one if
a particular three-digit zip code experiences a BLM protest in a given year-month. Standard errors are double clustered on area
and year-month. Panel A presents results when the dependent variable is the natural log of the amount of P2P loan originations
(Column 1) and the natural log of the number of P2P loan originations (Column 2). Panel B presents results when the dependent
variables are the natural log of the number of loans in each category of borrower quality. All variables are as defined in Appendix.
All continuous variables are winsorized at the 1 and 99 percentiles. ***, **, and * indicate statistical significance at the 0.01, 0.05
and 0.10 levels.
PanelA:LoanOriginations
(1) (2)
Amt Fundi,t+1 N Fundi,t+1
Protest 0.016*** 0.021***
(2.927) (3.650)
Population 0.955*** 2.317***
(4.485) (9.739)
Unemployment -0.029*** -0.038***
(-4.007) (-5.005)
Income -0.004** -0.005***
(-2.288) (-3.447)
HHI 0.001 0.000
(0.517) (0.272)
Oper Area -0.000** -0.000
(-2.062) (-1.355)
Deposits 0.001 0.000
(1.553) (0.698)
Observations 38,390 38,390
MODEL OLS OLS
Year-Month FE YES YES
ZipCode FE YES YES
Cluster YES YES
AdjR
2
0.772 0.934
42
Table 3: Primary Results (Continued)
PanelB:BorrowerQuality
(1) (2)
Low Quality High Quality
Protest 0.050*** 0.022
(4.643) (0.861)
Population 3.630*** 8.573***
(7.132) (12.620)
Unemployment -0.059*** -0.082***
(-4.671) (-4.409)
Income -0.010*** 0.013**
(-3.326) (2.501)
HHI 0.001 -0.008**
(0.480) (-2.201)
Oper Area -0.000 -0.001
(-0.795) (-0.814)
Deposits 0.000 0.004***
(0.480) (3.079)
Observations 38,390 38,390
MODEL OLS OLS
Year-Month FE YES YES
ZipCode FE YES YES
Cluster YES YES
AdjR
2
0.926 0.788
43
Table 4.2: Robustness
Table 4.2 presents the results from the robustness tests. Panel A present results for the lead and lag analysis, estimated using
equation 4.2. Panels B and C present the results from the falsification tests. Panel B reports results from the loan origination
test and Panel C reports results from tests of borrower quality. The sample consists of three-digit zip codes that have never
experienced BLM protest. I randomly generate pseudo-protest year-months and ensure to keep the proportion of the pseudo-
protest year-months as the same as the proportion of actual BLM protest. All continuous variables are winsorized at the 1 and 99
percentiles. ***, **, and * indicate statistical significance at the 0.01, 0.05 and 0.10 levels.
PanelA:LeadandLagAnalysis
(1) (2)
Amt Fundi,t+1 N Fundi,t+1
Protest t-2 0.012 0.014
(0.878) (1.448)
Protest t-1 -0.001 0.005
(-0.129) (0.720)
Protest 0.014*** 0.019***
(2.761) (3.424)
Protest t+1 0.010 0.008
(1.415) (1.565)
Protest t+2 0.009 0.010
(1.098) (1.094)
Population 0.955*** 2.316***
(4.503) (9.771)
Unemployment -0.029*** -0.038***
(-3.998) (-4.998)
Income -0.004** -0.005***
(-2.256) (-3.410)
HHI 0.001 0.000
(0.544) (0.303)
Oper Area -0.000** -0.000
(-2.061) (-1.356)
Deposits 0.001 0.000
(1.509) (0.664)
Observations 38,390 38,390
MODEL OLS OLS
Year-Month FE YES YES
ZipCode FE YES YES
Cluster YES YES
AdjR
2
0.772 0.934
44
Table 4: Robustness (Continued)
PanelB:Falsification-LoanOriginations
(1) (4)
Amt Fundi,t+1 N Fundi,t+1
Protest 0.0046 0.0076
(0.3037) (0.5989)
Population 1.2120*** 2.5372***
(4.3665) (8.3787)
Unemployment -0.0237** -0.0328***
(-2.4005) (-3.3818)
Income -0.0011 -0.0015
(-0.5015) (-0.7284)
HHI 0.0000 -0.0027
(0.0053) (-1.0729)
Oper Area -0.0003 -0.0001
(-1.0561) (-0.2606)
Deposits 0.0050** 0.0017
(2.4882) (1.0222)
Observations 21,681 21,681
MODEL OLS OLS
Year-Month FE YES YES
ZipCode FE YES YES
Cluster YES YES
AdjR
2
0.741 0.913
45
Table 4: Robustness (Continued)
PanelC:Falsification-BorrowerQuality
(1) (2)
Low Quality High Quality
Protest -0.002 -0.012
(-0.139) (-0.406)
Population 4.070*** 8.290***
(6.366) (11.375)
Unemployment -0.051*** -0.064***
(-3.426) (-3.253)
Income -0.005 0.020***
(-1.404) (2.719)
HHI -0.002 -0.012**
(-0.463) (-2.246)
Oper Area 0.000 -0.000
(0.678) (-0.106)
Deposits 0.003 0.010
(1.113) (1.397)
Observations 21,681 21,681
MODEL OLS OLS
Year-Month FE YES YES
ZipCode FE YES YES
Cluster YES YES
AdjR
2
0.902 0.740
46
Table 4.3: AlternativeExplanation: IncreaseinDemandforCredit
Table 4.3 present results from examining potential alternative explanations. In this table, I specifically examine whether the in-
crease in P2P lending is driven by the increased demand for credit following protests. Panel A present results when the dependent
variable is the balance of consumer and small business loans outstanding at each zip-code. Observations are aggregated to the
year-quarter level. Panel B presents results when the dependent variable is the natural log of the total foot traffic to banks branches
in a three-digit zip code. Panel C present results from the loan purpose analysis. Business loans make up approximately 13% of
the loan application sample and are composed of loans that are applied for “Small Business” and “Major Purchase.” Non-business
loans include loans are composed of loans that are related to day-to-day expenses such as purchasing cars, consolidating debt,
improving housing, medical fees, and vacation. Panel D present results from analyzing loans from the Small Business Adminis-
tration (SBA). All variables are as defined in the Appendix. All continuous variables are winsorized at the 1 and 99 percentiles.
***, **, and * indicate statistical significance at the 0.01, 0.05 and 0.10 levels.
PanelA:LoansOutstanding
(1)
Loans Outstandingi,t
QtrProtest -0.063*
(-1.816)
Population -0.596
(-0.230)
Unemployment 0.025
(0.267)
Income 0.053*
(1.787)
HHI -0.028
(-0.758)
Oper Area -0.005
(-0.426)
Deposits -0.047*
(-1.820)
Observations 11,657
MODEL OLS
Year-Qtr FE YES
ZipCode FE YES
Cluster YES
Adjusted R-Squared 0.848
47
Table 5: Alternative Explanation: Increase in Demand for Credit (Continued)
PanelB:FootTraffictoBanks
(1)
Foot Traffic i,t+1
Protest 0.002
(0.137)
Population -2.417***
(-5.714)
Unemployment -0.015***
(-4.268)
Income -0.014***
(-8.715)
HHI 0.000
(0.512)
Oper Area 0.001**
(3.016)
Deposits 0.000
(0.532)
Observations 7,716
MODEL OLS
Year-Month FE YES
ZipCode FE YES
Cluster YES
Adjusted R-Squared 0.995
48
Table 5: Alternative Explanation: Increase in Demand for Credit (Continued)
PanelC:AnalysisofLoanPurpose
(1) (2) (3) (4)
NonBusinessLoans BusinessLoans
Amt Fundi,t+1 N Fundi,t+1 Amt Fundi,t+1 N Fundi,t+1
Protest 0.020*** 0.022*** 0.004 0.020
(3.724) (4.304) (0.361) (0.978)
Population 0.999*** 1.941*** 1.341*** 3.456***
(4.810) (9.172) (5.187) (7.136)
Unemployment -0.032*** -0.030*** -0.012** -0.037***
(-4.895) (-4.890) (-2.155) (-4.302)
Income -0.004*** -0.006*** 0.001 -0.000
(-3.226) (-4.021) (0.300) (-0.043)
HHI 0.001 0.001 -0.003* -0.003
(1.010) (1.081) (-1.883) (-1.620)
Oper Area -0.000* -0.000 -0.000 0.000
(-1.907) (-1.317) (-1.071) (0.087)
Deposits 0.000 -0.000 0.002*** 0.003***
(0.485) (-0.425) (3.222) (2.996)
Observations 38,390 38,390 38,390 38,390
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
AdjR
2
0.732 0.944 0.202 0.590
49
Table 5: Alternative Explanation: Increase in Demand for Credit (Continued)
PanelD:SmallBusinessAdministration(SBA)Loans
(1) (2)
SBA Amt Fundi,t+1 SBA N Fundi,t+1
Protest -0.004 0.007
(-0.080) (0.548)
Population 2.309** 1.924***
(2.544) (4.819)
Unemployment -0.015 -0.016*
(-0.591) (-1.727)
Income 0.005 -0.000
(1.003) (-0.142)
HHI 0.012*** 0.005
(3.034) (1.417)
Oper Area 0.001 -0.000
(0.939) (-0.002)
Deposits -0.000 -0.001
(-0.400) (-0.733)
Observations 28,588 28,588
MODEL OLS OLS
Year-Month FE YES YES
ZipCode FE YES YES
Cluster YES YES
Adjusted R-Squared 0.507 0.789
50
Table 4.4: PoliticalAffiliation
Table 4.4 presents the results from cross-sectional tests based on the area’s political affiliation. The political affiliation data is from
MIT’s datalab on the 2016 presidential election by each precinct. A three-digit zip code is considered more democratic leaning if
its democratic vote share is greater than the national median. Panel A presents results when the dependent variable is the natural
log of the amount of P2P loan originations (Columns 1 and 2) and the natural log of the number of P2P loan originations (Columns
3 and 4). Panel B presents results when the dependent variables are the natural log of the number of loans in each category of
borrower quality. Standard errors are double clustered on area and year-month. All variables are as defined in Appendix. All
continuous variables are winsorized at the 1 and 99 percentiles. ***, **, and * indicate statistical significance at the 0.01, 0.05 and
0.10 levels.
PanelA:LoanOriginations
(1) (2) (3) (4)
Republican Democrat Republican Democrat
Amt Fundi,t+1 Amt Fundi,t+1 N Fundi,t+1 N Fundi,t+1
Protest 0.015** -0.006 0.019*** 0.007
(2.471) (-0.198) (3.766) (0.315)
Difference 0.021* 0.012
P-value: 0.060 P-value: 0.106
Observations 20,641 14,946 20,641 14,946
Controls YES YES YES YES
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
AdjR
2
0.818 0.702 0.947 0.887
51
Table 6: Political Affiliation (Continued)
PanelB:BorrowerQuality
(1) (2) (3) (4)
Republican Democrat Republican Democrat
Low Quality High Quality
Protest 0.047*** -0.011 0.018 0.040
(4.718) (-0.300) (0.673) (0.609)
Difference 0.058** -0.022
P-value: 0.019 P-value: 0.337
Observations 20,641 14,946 20,641 14,946
Controls YES YES YES YES
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
AdjR
2
0.939 0.867 0.806 0.683
52
Table 4.5: BankCompetition
Table 4.5 presents the results from cross-sectional tests based on the area’s bank competition. The bank competition score is
based on the three-digit zip code’s Herfindahl-Hirschman Index. A three-digit zip code is classified as a "High Competition" area
if its HHI is below cross-sectional median. Panel A presents results when the dependent variable is the natural log of the amount
of P2P loan originations (Columns 1 and 2) and the natural log of the number of P2P loan originations (Columns 3 and 4). Panel
B presents results when the dependent variables are the natural log of the number of loans in each category of borrower quality.
Standard errors are double clustered on area and year-month. All variables are as defined in Appendix. All continuous variables
are winsorized at the 1 and 99 percentiles. ***, **, and * indicate statistical significance at the 0.01, 0.05 and 0.10 levels.
PanelA:LoanOriginations
(1) (2) (3) (4)
HighCompetition LowCompetition HighCompetition LowCompetition
Amt Fundi,t+1 Amt Fundi,t+1 N Fundi,t+1 N Fundi,t+1
Protest 0.000 0.023** 0.006 0.030***
(0.007) (2.391) (0.803) (2.828)
Difference -0.023** -0.024**
p-value: 0.042 p-value: 0.026
Observations 19,199 19,190 19,199 19,190
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
AdjR
2
0.805 0.744 0.936 0.930
53
Table 7: Bank Competition (Continued)
PanelB:BorrowerQuality
(1) (2) (3) (4)
HighCompetition LowCompetition HighCompetition LowCompetition
Low Quality High Quality
Protest 0.033** 0.062*** 0.017 0.021
(2.209) (3.888) (0.357) (0.649)
Difference -0.029* -0.004
p-value: 0.097 p-value: 0.474
Observations 19,199 19,190 19,199 19,190
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
AdjR
2
0.923 0.927 0.781 0.794
54
Table 4.6: EconomicDisadvantage
Table 4.6 presents the results from cross-sectional tests based on whether the three-digit zip code is classified as an economically
disadvantaged area or not. The area is classified as a disadvantaged area if it has an income level that is below the national median.
Panel A presents results when the dependent variable is the natural log of the amount of P2P loan originations (Columns 1 and
2) and the natural log of the number of P2P loan originations (Columns 3 and 4). Panel B presents results when the dependent
variables are the natural log of the number of loans in each category of borrower quality. Standard errors are double clustered on
area and year-month. All variables are as defined in Appendix. All continuous variables are winsorized at the 1 and 99 percentiles.
***, **, and * indicate statistical significance at the 0.01, 0.05 and 0.10 levels.
PanelA:LoanOriginations
(1) (2) (3) (4)
LowIncome HighIncome LowIncome HighIncome
Amt Fundi,t+1 Amt Fundi,t+1 N Fundi,t+1 N Fundi,t+1
Protest 0.034** -0.000 0.041*** 0.005
(2.505) (-0.014) (2.797) (0.662)
Difference 0.034*** 0.036***
P-value: 0.008 P-value: 0.003
Observations 19,195 19,194 19,195 19,194
Controls YES YES YES YES
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
AdjR
2
0.690 0.799 0.892 0.948
55
Table 8: Economic Disadvantage (Continued)
PanelB:BorrowerQuality
(1) (2) (3) (4)
LowIncome HighIncome LowIncome HighIncome
Low Quality High Quality
Protest 0.090*** 0.019 0.038 0.008
(4.723) (1.093) (1.180) (0.247)
Difference 0.071*** 0.030
P-value: 0.004 P-value: 0.112
Observations 19,195 19,194 19,195 19,194
Controls YES YES YES YES
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
AdjR
2
0.885 0.938 0.664 0.804
56
Table 4.7: SingleLocationBanks
Table 4.7 presents the results from cross-sectional tests based on whether the three-digit zip code is classified as an area with
banks at multiple locations in the continental United States. The area is classified as a many bank area if the number of banks
that operate across the continental U.S. is greater than the cross-sectional median. Panel A presents results when the dependent
variable is the natural log of the amount of P2P loan originations (Columns 1 and 2) and the natural log of the number of P2P loan
originations (Columns 3 and 4). Panel B presents results when the dependent variables are the natural log of the number of loans
in each category of borrower quality. Standard errors are double clustered on area and year-month. All variables are as defined
in Appendix. All continuous variables are winsorized at the 1 and 99 percentiles. ***, **, and * indicate statistical significance at
the 0.01, 0.05 and 0.10 levels.
PanelA:LoanOriginations
(1) (2) (3) (4)
SingleBank ManyBank SingleBank ManyBank
Amt Fundi,t+1 Amt Fundi,t+1 N Fundi,t+1 N Fundi,t+1
Protest 0.034*** -0.003 0.036*** -0.003
(3.651) (-0.492) (3.821) (-0.464)
Difference 0.037*** 0.039***
p-value:<0.001 p-value:<0.001
Observations 28,170 10,220 28,170 10,220
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
Adj R2 0.725 0.918 0.880 0.953
57
Table 9: Single Location Banks (Continued)
PanelB:BorrowerQuality
(1) (2) (3) (4)
SingleBank ManyBank SingleBank ManyBank
Low Quality High Quality
Protest 0.058*** 0.018 0.050 -0.013
(3.883) (1.192) (1.578) (-0.431)
Difference 0.039** 0.063
p-value: 0.091 p-value: 0.162
Observations 28,170 10,220 28,170 10,220
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
Adj R2 0.852 0.925 0.639 0.791
58
Table 4.8: MinorityBankBoards
Table 4.8 presents the results from cross-sectional tests based on whether the three-digit zip code is classified as a high minority
area. A three-digit zip code is classified as a high minority area in the following manner. First, I compute the percentage of
black individuals on the bank’s board for each three-digit zip code. If the median percentage for a given area is greater than the
national median percentage, then the area is given a value of 1 and 0 otherwise. Panel A presents results when the dependent
variable is the natural log of the amount of P2P loan originations (Columns 1 and 2) and the natural log of the number of P2P loan
originations (Columns 3 and 4). Panel B presents results when the dependent variables are the natural log of the number of loans
in each category of borrower quality. Standard errors are double clustered on area and year-month. All variables are as defined
in Appendix. All continuous variables are winsorized at the 1 and 99 percentiles. ***, **, and * indicate statistical significance at
the 0.01, 0.05 and 0.10 levels.
PanelA:LoanOriginations
(1) (2) (3) (4)
HighMinority LowMinority HighMinority LowMinority
Amt Fundi,t+1 Amt Fundi,t+1 N Fundi,t+1 N Fundi,t+1
Protest 0.011 0.028*** 0.012* 0.030***
(1.388) (3.830) (1.713) (3.847)
Difference -0.037* -0.017*
p-value: 0.060 p-value: 0.056
Observations 19,641 18,748 19,641 18,748
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
Adj R2 0.764 0.725 0.955 0.934
59
Table 10: Minority Bank Boards (Continued)
PanelB:BorrowerQuality
(1) (2) (3) (4)
HighMinority LowMinority HighMinority LowMinority
Low Quality High Quality
Protest 0.016* 0.044*** 0.009 0.018
(1.970) (3.040) (1.008) (1.446)
Difference -0.027* -0.009
p-value: 0.057 p-value: 0.296
Observations 19,641 18,748 19,641 18,748
MODEL OLS OLS OLS OLS
Year-Month FE YES YES YES YES
ZipCode FE YES YES YES YES
Cluster YES YES YES YES
Adj R2 0.925 0.890 0.918 0.883
60
Table 4.9: LenderBehavior
Table 4.9 presents the results from estimating shifts to P2P lenders’ behaviors before and after each protest as in equation 4.3:
Fundedit+1=Ait+Pit+cXLoanit+dXareait +βProtest it+ϵ it. The dependent variable,Fundedit+1 is an indicator variable
that equals to one if a loan is fully funded. In this test, I condition on loans that were accepted by the LendingClub. Columns
(2) and (3) incrementally build on Column (1) by adding loan-level controls and both loan-level and area-level control variables.
Standard errors are double clustered on area and year-month. All variables are as defined in Appendix. All continuous variables
are winsorized at the 1 and 99 percentiles. ***, **, and * indicate statistical significance at the 0.01, 0.05 and 0.10 levels.
(1) (2) (3)
Fundedi,t+1 Fundedi,t+1 Fundedi,t+1
Protest -0.000 -0.000 -0.000
(-1.419) (-1.324) (-1.545)
Loan Grade -0.000 -0.000
(-1.671) (-1.666)
DTI -0.000*** -0.000***
(-3.440) (-3.446)
Loan Amt -0.005*** -0.005***
(-4.194) (-4.192)
FICO Score 0.000*** 0.000***
(2.860) (2.863)
Employment 0.000*** 0.000***
(3.770) (3.779)
Maturity -0.003** -0.003**
(-2.131) (-2.135)
Inq Last 6 mo -0.002*** -0.002***
(-3.682) (-3.687)
Open Acc 0.000*** 0.000***
(2.957) (2.892)
Revol Util -0.000 -0.000
(-1.361) (-1.331)
Delinquency 0.000 0.000
(0.676) (0.675)
Revol Bal 0.000 0.000
(0.431) (0.465)
Total Acc -0.000 -0.000
(-1.517) (-1.521)
Homeowner -0.000 -0.000
(-0.546) (-0.556)
Mortgage 0.000 0.000
(0.383) (0.387)
Population 0.021**
(2.401)
Unemployment -0.000
(-0.318)
Income -0.000***
(-3.942)
HHI -0.000
(-0.086)
Oper Area 0.000*
(1.985)
Deposits -0.000
(-0.635)
Observations 1,412,358 1,412,358 1,412,358
MODEL OLS OLS OLS
Loan Controls NO YES YES
ZipCode Controls NO NO YES
Year-Month FE YES YES YES
ZipCode FE YES YES YES
Cluster YES YES YES
AdjR
2
0.0172 0.0229 0.0229
61
Table 4.10: Non-BLMProtests
Table 4.10 presents the results from estimating shifts to P2P lenders’ behaviors before and after each non-BLM protest. In this
test, I condition on loans that were accepted by the LendingClub. Columns (1) and (2) estimate equation 4.1 and Columns (3) to
(5) estimate equation 4.3 with the exception that the independent variable of interestProtestit is replaced with an indicator for
non-BLM protests. Standard errors are double clustered on area and year-month. All variables are as defined in Appendix. All
continuous variables are winsorized at the 1 and 99 percentiles. ***, **, and * indicate statistical significance at the 0.01, 0.05 and
0.10 levels.
(1) (2) (3) (4) (5)
Amt Fundi,t+1 N Fundi,t+1 Fundedi,t+1 Fundedi,t+1 Fundedi,t+1
NonBLMProtest 0.012* 0.009* -0.000 -0.000 -0.000
(1.832) (1.995) (-1.237) (-1.369) (-1.226)
Observations 29,904 29,904 1,101,927 1,101,927 1,101,927
Loan Controls NA NA NO YES YES
Area Controls YES YES NO NO YES
MODEL OLS OLS OLS OLS OLS
Year-Month FE YES YES YES YES YES
ZipCode FE YES YES YES YES YES
Cluster YES YES YES YES YES
AdjR
2
0.604 0.948 0.0270 0.0305 0.0305
62
Table 4.11: P2PCreditModel
Table 4.11 presents the results from estimating shifts to P2P platform’s credit model before and after each protest as in equation
4.4:Acceptedit+1 =Ait +Pit +cXLoanit +dXareait+βProtest it +ϵ it. The dependent variable,Acceptedit+1 is an indicator
variable that equals to one if a loan is accepted by the platform. Columns (2) and (3) incrementally build on Column (1) by adding
loan-level controls and both loan-level and area-level control variables. Standard errors are double clustered on area and year-
month. All variables are as defined in Appendix. All continuous variables are winsorized at the 1 and 99 percentiles. ***, **, and *
indicate statistical significance at the 0.01, 0.05 and 0.10 levels.
(1) (2) (3)
Acceptedi,t+1 Acceptedi,t+1 Acceptedi,t+1
Protest 0.002 0.001 0.001
(1.442) (1.349) (1.379)
Loan Amt 0.003 0.003
(1.304) (1.303)
FICO Score 0.001*** 0.001***
(8.807) (8.809)
DTI -0.000*** -0.000***
(-8.129) (-8.128)
Employment 0.066*** 0.066***
(43.445) (43.424)
Population -0.013
(-0.360)
Unemployment 0.000
(0.159)
Income 0.001*
(1.932)
HHI 0.000
(0.509)
Oper Area -0.000
(-0.891)
Deposits 0.000
(1.649)
Observations 8,609,413 8,609,413 8,609,413
MODEL OLS OLS OLS
Loan Controls NO YES YES
ZipCode Controls NO NO YES
Year-Month FE YES YES YES
ZipCode FE YES YES YES
Cluster YES YES YES
AdjR
2
0.242 0.564 0.564
63
Chapter5
Conclusions
I examine the effect of social movements, such as #BlackLivesMatter, on borrower’s ability to access credit.
While a large proportion of the literature focuses on real effects of social movements on socio-economic
outcomes, the effect of social movements on credit market outcomes has not been systematically exam-
ined. Exploiting the fact that commercial lending and peer-to-peer lending are substitutes, I show that P2P
lending volume and lending amount increase significantly in zip codes that experience BLM protests com-
pared to zip codes that do not experience BLM protests. Moreover, consistent with prior literature which
documents that P2P lending is a substitute for commercial lending, where P2P serves infra-marginal bor-
rowers who fail to secure credit at commercial banks, I find that the increase in P2P lending is driven by
lower quality borrowers. Consistent with loan officers’ moods being adversely affected by the protests,
I find that the results are stronger for areas that are more Republican-leaning, have less banking compe-
tition, are associated with lower income, are dominated by single bank branches, and have less minority
representation on bank boards. Moreover, I find that the increase in P2P lending is not driven by alterna-
tive explanations such as increased demand for credit following protests, changes in investors’ willingness
to lend, and changes to the P2P platform’s credit model.
64
I contribute to the growing literature that examines socio-economic implications of social movements
by documenting the effect of social movements on low quality borrower’s ability to access credit. My re-
sults suggest that social movements such as the #BlackLivesMatter movement adversely affect local loan
officers’ moods, thereby reducing credit approvals at local banks. Consequently, borrowers who are re-
jected by the banks migrate to P2P lending. Moreover, by documenting that social movements can also
have material effects on local loan officers’ sentiment, I add to the literature on the effect of sentiment on
economic decisions. Lastly, I contribute to the literature on the relationship between P2P lending and com-
mercial lending by highlighting that non-banking shocks such as social movements can also have similar
effects as shifts in regulations, which prior studies focus on.
65
Bibliography
Acemoglu, Daron, Suresh Naidu, Pascual Restrepo, and James A Robinson (2019). “Democracy Does
Cause Growth”. In: Journal of Political Economy 127.1.
Agarwal, Sumit, Ran Duchin, and Denis Sosyura (2012). “In the Mood for a Loan: The Causal Effect of
Sentiment on Credit Origination”. In: Working Paper.
Agrawal, Ajay K, Christian Catalini, and Avi Goldfarb (2011). “The Geography of Crowdfunding”. In:
National Bureau of Economic Research.url: www.netinst.org.
Arora, Maneesh and Christopher T. Stout (2019). “Letters for Black Lives: Co-ethnic Mobilization and
Support for the Black Lives Matter Movement”. In: Political Research Quarterly 72.2, pp. 389–402.
issn: 1938274X.doi: 10.1177/1065912918793222.
Asay, Scott H, Jeffrey L Hoopes, Jacob R Thornock, and Jaron H Wilde (2022). “Tax Boycotts”. In: Working
Paper.url: https://www.pwc.com/gx/en/tax/publications/assets/pwc-tax-strategy-and-corporate-
reputation.pdf.
Barth, Mary E, Wayne R Landsman, Mark Lang, and Christopher D Williams (2018). “Effects on
Comparability and Capital Market Benefits of Voluntary IFRS Adoption”. In: Journal of Financial
Reporting 3.1, pp. 1–22.
Bavoso, Vincenzo (2020). “The promise and perils of alternative market-based finance: the case of P2P
lending in the UK”. In: Journal of Banking Regulation 21.4, pp. 395–409.issn: 17502071.doi:
10.1057/s41261-019-00118-9.
Beck, Thorsten, Patrick Behr, and Andreas Madestam (2018). “Sex and credit: Do gender interactions
matter for credit market outcomes?” In: Journal of Banking and Finance 87, pp. 380–396.issn:
03784266.doi: 10.1016/j.jbankfin.2017.10.018.
Beilock, Sian L. and Marci S. DeCaro (2007). “From Poor Performance to Success Under Stress: Working
Memory, Strategy Selection, and Mathematical Problem Solving Under Pressure”. In: Journal of
Experimental Psychology: Learning Memory and Cognition 33.6, pp. 983–998.issn: 02787393.doi:
10.1037/0278-7393.33.6.983.
66
Bertrand, Jérémie and Laurent Weill (2021). “Do Algorithms Discriminate Against African Americans in
Lending?” In: Economic Modelling.url:
https://www.nytimes.com/2018/03/07/opinion/mortage-minority-income.html.
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan (2004). “How much should we trust
differences-in-differences estimates?” In: The Quarterly Journal of Economics 119.no. 1, pp. 249–275.
url: https://academic.oup.com/qje/article/119/1/249/1876068.
Bilach, Thomas J., Sean Patrick Roche, and Gregory J. Wawro (2022). “The effects of the Summer All Out
Foot Patrol Initiative in New York City: a difference-in-differences approach”. In: Journal of
Experimental Criminology 18.2, pp. 209–244.issn: 15728315.doi: 10.1007/s11292-020-09445-8.
Bolger, Niall, Anita Delongis, Ronald C Kessler, and Elizabeth A Schilling (1989). “Effects of Daily Stress
on Negative Mood”. In: Journal of Personality and Social Psychology 57.5, pp. 808–818.
Buchak, Greg, Gregor Matvos, Tomasz Piskorski, and Amit Seru (2018). “Fintech, regulatory arbitrage,
and the rise of shadow banks”. In: Journal of Financial Economics 130.3, pp. 453–483.issn: 0304405X.
doi: 10.1016/j.jfineco.2018.03.011.
Busse, Meghan R., Devin G. Pope, Jaren C. Pope, and Jorge Silva-Risso (2015). “The psychological effect of
weather on car purchases”. In: Quarterly Journal of Economics 130.1, pp. 371–414.issn: 15314650.doi:
10.1093/qje/qju033.
Campbell, Dennis, Maria Loumioti, and Regina Wittenberg-Moerman (2019). “Making sense of soft
information: interpretation bias and loan quality”. In: Journal of Accounting and Economics 68.2-3.
issn: 01654101.doi: 10.1016/j.jacceco.2019.101240.
Campbell, Travis (2021). “Black Lives Matter’s Effect on Police Lethal Use-of-Force”. In: Working Paper.
url: https://ssrn.com/abstract=3767097.
Chen, A J, Patricia M Dechow, and Samuel T Tan (2021). “Beyond Shareholder Value? Why Firms
Voluntarily Disclose Support for Black Lives Matter”. In: Working Paper.
Cole, Courtney E. (2017). “Culturally sustaining pedagogy in higher education: Teaching so that Black
Lives Matter”. In: Equality, Diversity and Inclusion 36.8, pp. 736–750.issn: 17587093.doi:
10.1108/EDI-01-2017-0005.
Cornaggia, Jess, Brian Wolfe, and Woongsun Yoo (2018). “Crowding Out Banks: Credit Substitution by
Peer-to-Peer Lending”. In: Working Paper.
Cortés, Kristle, Ran Duchin, and Denis Sosyura (2016). “Clouded judgment: The role of sentiment in
credit origination”. In: Journal of Financial Economics 121.2, pp. 392–413.issn: 0304405X.doi:
10.1016/j.jfineco.2016.05.001.
Demiroglu, Cem, Oguzhan Ozbas, Rui C. Silva, and Mehmet Fatih Ulu (2021). “Do Physiological and
Spiritual Factors Affect Economic Decisions?” In: Journal of Finance 76.5, pp. 2481–2523.issn:
15406261.doi: 10.1111/jofi.13032.
67
Dong, Rui, Raymond Fisman, Yongxiang Wang, and Nianhang Xu (2021). “Air pollution, affect, and
forecasting bias: Evidence from Chinese financial analysts”. In: Journal of Financial Economics 139.3,
pp. 971–984.issn: 0304405X.doi: 10.1016/j.jfineco.2019.12.004.
Drake, Michael S., Kurt H. Gee, and Jacob R. Thornock (2016). “March Market Madness: The Impact of
Value-Irrelevant Events on the Market Pricing of Earnings News”. In: Contemporary Accounting
Research 33.1, pp. 172–203.issn: 19113846.doi: 10.1111/1911-3846.12149.
Edmans, Alex, Diego García, and Øyvind Norli (2007). “Sports sentiment and stock returns”. In:Journalof
Finance 62.4, pp. 1967–1998.issn: 00221082.doi: 10.1111/j.1540-6261.2007.01262.x.
Goenner, Cullen F. (Jan. 2023). “Majority-Minority Boards of Directors and Decision Making: The Effects
of Homophily on Lending Decisions”. In: Business and Society 62.1, pp. 54–86.issn: 15524205.doi:
10.1177/00076503211062182.
Goetzmann, William N., Dasol Kim, Alok Kumar, and Qin Wang (2015). “Weather-induced mood,
institutional investors, and stock returns”. In: Review of Financial Studies 28.1, pp. 73–111.issn:
14657368.doi: 10.1093/rfs/hhu063.
Gow, Ian D., Gaizka Ormazabal, and Daniel J. Taylor (2010). “Correcting for cross-sectional and
time-series dependence in accounting research”. In: Accounting Review 85.2, pp. 483–512.issn:
00014826.doi: 10.2308/accr.2010.85.2.483.
Havrylchyk, Olena, Carlotta Mariotto, Talal Rahim, and Marianne Verdier (2016). “What drives the
expansion of the peer-to-peer lending?” In: Working Paper.url:
http://ssrn.com/abstract=2841316Electroniccopyavailableat:http://ssrn.com/abstract=2841316.
Hehman, Eric, Jessica K. Flake, and Jimmy Calanchini (2018). “Disproportionate Use of Lethal Force in
Policing Is Associated With Regional Racial Biases of Residents”. In: Social Psychological and
Personality Science 9.4, pp. 393–401.issn: 19485514.doi: 10.1177/1948550617711229.
Hirshleifer, David and Tyler Shumway (2003). “Good Day Sunshine: Stock Returns and the Weather”. In:
Journal of Finance 58.3, pp. 1009–1032.issn: 00221082.doi: 10.1111/1540-6261.00556.
Hong, Harrison and Marcin Kacperczyk (2010). “Competition and Bias”. In: The Quarterly Journal of
Economics 124.4, pp. 1683–1725.url: https://academic.oup.com/qje/article/125/4/1683/1916282.
Hou, Wai Kai, Brian J. Hall, Daphna Canetti, Kam Man Lau, Sin Man Ng, and Stevan E. Hobfoll (2015).
“Threat to democracy: Physical and mental health impact of democracy movement in Hong Kong”.
In: Journal of Affective Disorders 186, pp. 74–82.issn: 15732517.doi: 10.1016/j.jad.2015.07.005.
Hu, Maggie Rong and Adrian D. Lee (2020). “Outshine to outbid: Weather-induced sentiment and the
housing market”. In: Management Science 66.3, pp. 1440–1472.issn: 15265501.doi:
10.1287/mnsc.2019.3517.
Huang, Ruidi (2019). “The Financial Consequences of Customer Satisfaction: Evidence from Yelp Ratings
and SBA Loans”.url: https://www.sba.gov/about-sba/sba-.
68
Imai, Kosuke, In Song Kim, and Erik H Wang (2021). “Matching Methods for Causal Inference with
Time-Series Cross-Sectional Data”. In: American Journal of Political Science, pp. 1–19.doi:
10.7910/DVN/ZTDHVE.
Jagtiani, Julapa, Lauren Lambie-Hanson, and Timothy Lambie-Hanson (2021). “Fintech Lending and
Mortgage Credit Access”. In: The Journal of FinTech 1.01.issn: 1962-5361.doi:
10.21799/frbp.wp.2019.47.
Jagtiani, Julapa and Catharine Lemieux (2018). “Do Fintech Lenders Penetrate Areas That Are
Underserved by Traditional Banks?” In: Journal of Economics and Business 100, pp. 43–54.issn:
01486195.doi: 10.1016/j.jeconbus.2018.03.001.
— (2019). “The Roles of Alternative Data and Machine Learning in Fintech Lending: Evidence from the
LendingClub Consumer Platform”. In: Financial Management 48.4, pp. 1009–1029.issn: 1755053X.
doi: 10.1111/fima.12295.
Jiang, Jinglin, Li Liao, Zhengwei Wang, and Xiaoyan Zhang (2021). “Government Affiliation and
Peer-To-Peer Lending Platforms in China”. In: Journal of Empirical Finance 62, pp. 87–106.issn:
09275398.doi: 10.1016/j.jempfin.2021.02.004.
Jin, Hengda, Stephen R Stubben, and Karen Ton (2022). “Customer Shopping Behavior and the
Persistence of Revenues and Earnings”. In: Working Paper.url:
https://ssrn.com/abstract=3744417Electroniccopyavailableat:https://ssrn.com/abstract=3744417.
Klein, Kitty and Devon Barnes (1994). “The Relationship of Life Stress to Problem Solving: Task
Complexity and Individual Differences”. In: Social Cognition 12.3, pp. 187–204.
Kräussl, Roman ; Zsofia ; Kräussl, Joshua ; Pollet, and Kalle Rinne (2018). “The performance of
marketplace lenders: Evidence from lending club payment data”. In: Working Paper.url:
https://nbn-resolving.de/urn:nbn:de:hebis:30:3-469928.
Lanaj, Klodiana, Russell E. Johnson, and Mo Wang (2016). “When lending a hand depletes the will: The
daily costs and benefits of helping”. In: Journal of Applied Psychology 101.8, pp. 1097–1110.issn:
00219010.doi: 10.1037/apl0000118.
Landmann, Helen and Anette Rohmann (2020). “Being moved by protest: Collective efficacy beliefs and
injustice appraisals enhance collective action intentions for forest protection via positive and
negative emotions”. In: Journal of Environmental Psychology 71.issn: 15229610.doi:
10.1016/j.jenvp.2020.101491.
Laws, Rachel, Karen J. Campbell, Paige Van Der Pligt, Georgina Russell, Kylie Ball, John Lynch,
David Crawford, Rachael Taylor, Deborah Askew, and Elizabeth Denney-Wilson (2014). “The impact
of interventions to prevent obesity or improve obesity related behaviours in children (0-5 years) from
socioeconomically disadvantaged and/or indigenous families: A systematic review”. In: BMC Public
Health 14.1.issn: 14712458.doi: 10.1186/1471-2458-14-779.
Mazumder, Soumyajit (2019). “Black Lives Matter for Whites’ Racial Prejudice: Assessing the Role of
Social Movements in Shaping Racial Attitudes in the United States”. In: Working Paper.url:
http://smazumder.me.
69
Morales-Acevedo, Paola and Steven Ongena (2020). “Fear, Anger, and Credit. on Bank Robberies and Loan
Conditions”. In: Economic Inquiry 58.2, pp. 921–952.issn: 14657295.doi: 10.1111/ecin.12826.
Morse, Adair (2015). “Peer-to-Peer Crowdfunding: Information and the Potential for Disruption in
Consumer Lending”. In: Annual Review of Financial Economics 7, pp. 463–482.issn: 19411375.doi:
10.1146/annurev-financial-111914-041939.
Ni, Michael Y., Yoona Kim, Ian McDowell, Suki Wong, Hong Qiu, Irene O.L. Wong, Sandro Galea, and
Gabriel M. Leung (2020). “Mental health during and after protests, riots and revolutions: A systematic
review”. In: Australian and New Zealand Journal of Psychiatry 54.3, pp. 232–243.issn: 14401614.doi:
10.1177/0004867419899165.
Ni, Michael Y., Tom K. Li, Herbert Pang, Brandford H.Y. Chan, Betty Y. Yuan, Ichiro Kawachi,
C. Mary Schooling, and Gabriel M. Leung (2016). “Direct Participation in and Indirect Exposure to the
Occupy Central Movement and Depressive Symptoms: A Longitudinal Study of Hong Kong Adults”.
In: American Journal of Epidemiology 184.9, pp. 636–643.issn: 14766256.doi: 10.1093/aje/kww103.
Noh, Suzie, Eric C So, and Christina Zhu (2022). “Financial Reporting and Consumer Behavior”. In:
Working Paper.
Parker, Kim, Juliana Menasce Horowitz, and Monica Anderson (2020). “Deep Partisan Divides over
Factors Underlying George Floyd Demonstrations”. In: Pew Research Center, pp. 1–19.url:
https://www.pewresearch.org/social-trends/2020/06/12/amid-protests-majorities-across-racial-
and-ethnic-groups-express-support-for-the-black-lives-matter-movement/.
Petersen, Mitchell A. (2009). “Estimating Standard Errors in Finance Panel Data Sets: Comparing
Approaches”. In: Review of Financial Studies 22.1, pp. 435–480.issn: 08939454.doi:
10.1093/rfs/hhn053.
Rosman, Lindsey, Elena Salmoirago-Blotcher, Rafat Mahmood, Hannan Yang, Quefeng Li,
Anthony J. Mazzella, Jeffrey Lawrence Klein, Joseph Bumgarner, and Anil Gehi (2021). “Arrhythmia
risk during the 2016 us presidential election: The cost of stressful politics”. In:JournaloftheAmerican
Heart Association 10.11.issn: 20479980.doi: 10.1161/JAHA.120.020559.
Roure, Calebe de, Loriana Pelizzon, and Anjan V. Thakor (2022). “P2P Lenders versus Banks: Cream
Skimming or Bottom Fishing?” In: The Review of Corporate Finance Studies 11.no. 2, pp. 213–262.doi:
10.2139/ssrn.3174632.
Rupar, Kathy, Sean Wang, and Hayoung Yoon (2021). “Does Wall Street Discriminate by Race? Evidence
from Analyst Target Prices”. In: Working Paper.url:
https://www.eeoc.gov/statistics/employment/jobpatterns/eeo1..
Sabherwal, Anandita, Matthew T. Ballew, Sander van der Linden, Abel Gustafson, Matthew H. Goldberg,
Edward W. Maibach, John E. Kotcher, Janet K. Swim, Seth A. Rosenthal, and Anthony Leiserowitz
(2021). “The Greta Thunberg Effect: Familiarity with Greta Thunberg predicts intentions to engage in
climate activism in the United States”. In: Journal of Applied Social Psychology 51.4, pp. 321–333.issn:
15591816.doi: 10.1111/jasp.12737.
70
Sawyer, Jeremy and Anup Gampa (2018). “Implicit and Explicit Racial Attitudes Changed During Black
Lives Matter”. In: Personality and Social Psychology Bulletin 44.7, pp. 1039–1059.issn: 15527433.doi:
10.1177/0146167218757454.
Shroff, Nemit, Rodrigo S. Verdi, and Gwen Yu (2014). “Information environment and the investment
decisions of multinational corporations”. In:AccountingReview 89.2, pp. 759–790.issn: 00014826.doi:
10.2308/accr-50643.
Skoy, Evelyn (2021). “Black Lives Matter Protests, Fatal Police Interactions, and Crime.” In: Contemporary
Economic Policy 39.2, pp. 280–291.issn: 14657287.doi: 10.1111/coep.12508.
Smith, Martha J, Ronald V Clarke, and Ken Pease (2002). “Anticipatory Benefits in Crime”. In: Crime
Prevention Studies 13, pp. 71–88.
Sonnentag, Sabine and Tanja Lischetzke (2018). “Illegitimate tasks reach into afterwork hours: A
multilevel study”. In: Journal of Occupational Health Psychology 23.2, pp. 248–261.issn: 19391307.
doi: 10.1037/ocp0000077.
Sonnentag, Sabine, Leonard Reinecke, Jutta Mata, and Peter Vorderer (2018). “Feeling interrupted—Being
responsive: How online messages relate to affect at work”. In: Journal of Organizational Behavior
39.3, pp. 369–383.issn: 10991379.doi: 10.1002/job.2239.
Tang, Huan (2019). “Peer-to-Peer Lenders Versus Banks: Substitutes or Complements?” In: Review of
Financial Studies 32.5, pp. 1900–1938.issn: 14657368.doi: 10.1093/rfs/hhy137.
Thakor, Anjan V. (2020). “Fintech and Banking: What Do We Know?”” In: Journal of Financial
Intermediation 41.issn: 10960473.doi: 10.1016/j.jfi.2019.100833.
Vani, Preeti, Shilaan Alzahawi, Jennifer E. Dannals, and Nir Halevy (2022). “Strategic Mindsets and
Support for Social Change: Impact Mindset Explains Support for Black Lives Matter Across Racial
Groups”. In: Personality and Social Psychology Bulletin.issn: 15527433.doi:
10.1177/01461672221099710.
Wang, Yang, Marco Qin, Xueming Luo, and Yu Kou (2022). “How Support for Black Lives Matter Impacts
Consumer Responses on Social Media”. In: Marketing Science, Forthcoming.
Wemm, Stephanie E. and Edelgard Wulfert (2017). “Effects of Acute Stress on Decision Making”. In:
Applied Psychophysiology Biofeedback 42.1, pp. 1–12.issn: 10900586.doi: 10.1007/s10484-016-9347-8.
Zohar, Dov, R. Epstein, and O. Tzischinski (2003). “Effects of Energy Availability on Immediate and
Delayed Emotional Reactions to Work Events”. In: Journal of Applied Psychology 88.6, pp. 1082–1093.
issn: 00219010.doi: 10.1037/0021-9010.88.6.1082.
71
AppendixA
72
Appendix
Variable Definition Source
DependentVariables
N Fund The natural log of the number of loan originations in a three-
digit zip code measured at each year-month.
LendingClub
Amt Fund The natural log of the total amount of loan originations in a
three-digit zip code measured at each year-month (in thou-
sands of dollars).
LendingClub
Low Quality The natural log of the number of loan originations for bor-
rowers with FICO scores 740 or below.
LendingClub
High Quality The natural log of the number of loan originations for bor-
rowers with FICO scores greater than 740.
LendingClub
Loans Outstanding The natural log of the number of consumer and small busi-
ness loans outstanding, measured for each year-qtr.
Call Reports
Foot Traffic The natural log of the number of foot traffic to bank branches.
The addresses of bank branches are obtained from the FDIC
SOD.
SafeGraph
Business Loans Loans applied for small business and major purchase. LendingClub
Non-Business Loans Loans applied for debt consolidation, car purchase, house im-
provement, medical fees, and vacation.
LendingClub
SBA Amount and number of loans from the Small Business Ad-
ministration 7(a) program.
SBA FOIA
Funded Indicator that equals to one if the loan is fully funded by in-
vestors.
LendingClub
Accepted Indicator that equals to one if the loan has been accepted by
LC’s proprietary algorithm. Accepted loans are listed on the
platform to solicit investor funds.
LendingClub
TreatmentVariable
Protest Indicator which equals one if a three-digit zip code experi-
ences a BLM protest in a particular year-month.
Elephrame
NonBLM Protest Indicator which equals one if a three-digit zip code experi-
ences a Non-BLM protest in a particular year-month. Non-
BLM protests are protests related to environmental rights,
women’s rights, reproductive rights, LGBTQ rights, and re-
ligion
Crowd Count-
ing Consor-
tium
73
Variable Definition Source
LoanLevelControls
Loan Grade Loan grade assigned by LendingClub using proprietary
credit scoring model. The variable can take on a value be-
tween 1 to 35 with lower numbers being loans of higher qual-
ity.
LendingClub
DTI Debt-to-income ratio calculated using the borrower’s total
monthly debt payments on the total debt obligations, exclud-
ing mortgage and the requested LC loan, divided by the bor-
rower’s self-reported monthly income.
LendingClub
Loan Amt Natural log of one plus the loan request amount LendingClub
FICO Score FICO score provided by LendingClub. LendingClub gives
FICO score as a range between two numbers. I take the av-
erage of the two.
LendingClub
Employment The borrower’s employment history. It takes on a value be-
tween 0 and 10, with 0 being no prior employment and 10
being more than 10 years of employment.
LendingClub
Maturity An indicator variable that equals to one if the maturity of the
loan is 60 months. The variable equals to zero if the maturity
is 36 months.
LendingClub
Inq Last 6 mo The number of inquiries in past 6 months (excluding auto
and mortgage inquiries).
LendingClub
Open Acc The number of open credit lines in the borrower’s credit file. LendingClub
Revol Util Revolving line utilization rate, defined as the amount of
credit the borrower is using relative to all available revolving
credit.
LendingClub
Delinquency The number of 30+ days past-due incidences of delinquency
in the borrower’s credit file for the past 2 years
LendingClub
Revol Bal Natural log of one plus the total credit revolving balance LendingClub
Total Acc Total number of credit lines currently in the borrower’s
credit file.
LendingClub
Homeowner Indicator variable that equals to one if the borrower owns the
home without mortgage.
LendingClub
Mortgage Indicator variable that equals to one if the borrower’s home
is mortgaged.
LendingClub
74
Variable Definition Source
AreaLevelControls
Population Natural log of one plus the total monthly population per
three-digit zip code. The monthly population is calculated
by dividing the annual population by 12.
Census
Unemployment Median unemployment rate by each three-digit zip code Census
Income Median income by each three-digit zip code scaled by 1,000 Census
HHI Herfindahl-Hirschman Index of the banking market of the
three-digit zip code computed using deposit shares. The HHI
score is calculating by summing up each bank’s squared mar-
ket share. Each bank’s market share is computed as the ratio
of the bank’s deposits in the three-digit zip code and the total
deposits of all banks in the three-digit zip code.
FDIC SOD
Oper Area Geographic diversification score of the three-digit zip code.
The score is computed by taking the median of the number
of three-digit zip codes a bank operates in.
FDIC SOD
Deposits The total bank deposits in the three-digit zip code scaled by
the population of the three-digit zip code.
FDIC SOD
75
Variable Definition Source
Cross-SectionalSplits
Democrat Indicator that equals to one if the three-digit zip code’s demo-
cratic vote share exceeded the national median democratic
vote share.
MIT Datalab
High Competition Indicator that equals to one if the HHI for the three-digit zip
code is lower than the cross-sectional median.
FDIC SOD
High Income Indicator that equals to one if the area has income below the
cross-sectional median.
Census
Single Bank Indicator that equals to one if the number of single bank
branches are greater than the cross-sectional median.
FDIC SOD
High Minority Minority board percentage is computed by using the reth-
nicity package in R for first and last names obtained from
BoardEx. I then compute the cross-sectional median percent-
age and the median percentage for each three-digit zip code.
The High Minority indicator equals to one if the area’s me-
dian minority board % is greater than the cross-sectional me-
dian.
BoardEX
76

Abstract (if available)

Abstract I investigate whether social movements, such as #BlackLivesMatter (herein “BLM”), affect local loan officers’ mood and consequently, credit approvals at local banks. Motivated by research in psychology, I hypothesize that protests will lead to increased stress levels, inducing negative moods in loan officers and ultimately result in a decrease in banks’ credit approvals. Because both loan officers’ sentiment and their credit decisions are unobservable, to test this prediction I rely on the substitution between P2P lending and traditional banking and expect that borrowers who are unable to secure credit from banks migrate to P2P lending. Utilizing the staggered occurrences of BLM protests, I show that P2P lending significantly increases in areas that experience BLM protests. Additionally, consistent with P2P lending serving borrowers rejected by banks, I find that the increase in P2P lending is driven by low quality borrowers. These effects are stronger for areas that are Republican leaning, have less banking competition, are economically disadvantaged, have more single bank branches, and have a greater proportion of minority board members. Moreover, I fail to find that the increase in P2P lending is driven by potential alternative explanations such as an increased demand for credit. Overall, my results suggest that social movements and the related protests may have an adverse effect on the ability of weaker borrowers to access credit.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Gone with the big data: institutional lender demand for private information

PDF

Comparative analysis of peer-to-peer lending in China and the United Kingdom: an assessment of the Lending Plaza’s market entry prospects

PDF

Expectation dynamics and stock returns

PDF

The changing policy environment and banks' financial decisions

PDF

Understanding the disclosure practices of firms affected by a natural disaster: the case of hurricanes

PDF

Do shared auditors improve audit quality? Evidence from banking relationships

PDF

Private disclosure contracts within supply chains and managerial learning from stock prices

PDF

Essays on information and financial economics

PDF

The structure of strategic communication: theory, measurement, and effects

PDF

Shareholder litigation as a disciplining device: evidence from firms' financial policies

PDF

Insider editing on Wikipedia

PDF

Disclosure distance and earnings announcement returns

PDF

Does fair value accounting exacerbate the pro-cyclicality of bank lending?

PDF

Implied equity duration: international evidence

PDF

Essays on the firm and stakeholders relationships: evidence from mergers & acquisitions and labor negotiations

PDF

Investment behavior of mutual fund managers

PDF

Political connections and access to bond capital: Reputation or collusion?

PDF

Three essays on macro and labor finance

PDF

Effectiveness of the SEC’s comment letters in initial public offerings

PDF

Risks, returns, and regulations in real estate markets

Asset Metadata

Creator Yoon, Joon Sang (author)

Core Title Social movements and access to credit

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Business Administration

Degree Conferral Date 2023-05

Publication Date 04/06/2023

Defense Date 03/31/2023

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag access to credit,banking,mood,OAI-PMH Harvest,peer-to-peer lending,social movements

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Wittenberg-Moerman, Regina (committee chair), Hoberg, Gerard (committee member), Sloan, Richard (committee member), Soliman, Mark (committee member)

Creator Email joonsang.yoon.phd@marshall.usc.edu,jun.yoon77@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC112952500

Unique identifier UC112952500

Identifier etd-YoonJoonSa-11576.pdf (filename)

Legacy Identifier etd-YoonJoonSa-11576

Document Type Dissertation

Format theses (aat)

Rights Yoon, Joon Sang

Internet Media Type application/pdf

Type texts

Source 20230406-usctheses-batch-1017 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu