1. finance
2. personal finance
3. lending
4. student loans

1
Tell me a good story and I may lend you my money:
The role of narratives in peer-to-peer lending decisions
Michal Herzenstein
Assistant professor of marketing
Lerner College of Business and Economics
University of Delaware
Newark, DE 19716
Tel: (302) 831-1775
Email: [email protected]
Scott Sonenshein
Assistant professor of management
Jones Graduate School of Business
Rice University
6100 Main Street
Houston, TX 77005
Tel: (713) 348-3182
Email: [email protected]
Utpal M. Dholakia
Professor of management
Jones Graduate School of Business
Rice University
6100 Main Street
Houston, TX 77005
Tel: (713) 348-5376
Email: [email protected]
We thank Rick Andrews, Richard Schwarz, and Greg Hancock for their statistical assistance. We
further thank Editor John Lynch, the AE and two reviewers for their insightful comments.
Financial support from the Lerner College of business and economics at the University of
Delaware and the Jones graduate school of business at Rice University is greatly appreciated.
Electronic copy available at: http://ssrn.com/abstract=1840668
2
Tell me a good story and I may lend you my money:
The role of narratives in peer-to-peer lending decisions
Abstract
We examine the role of identity claims constructed in narratives by borrowers in influencing
lender decision making regarding unsecured personal loans. We study whether the number of
identity claims and their content influence decisions of lenders and whether they predict longerterm performance of funded loans. Using data from the peer-to-peer lending website
Prosper.com, we find that unverifiable information affects lending decisions above and beyond
objective, verifiable information. Specifically, as the number of identity claims in narratives
increases, so does loan funding but loan performance suffers, because these borrowers are less
likely to pay back. In addition, identity content plays an important role. Identities about being
trustworthy or successful are associated with increased loan funding but ironically they are less
predictive of loan performance compared with other identities (moral and hardworking). Thus,
some identity claims are meant to mislead lenders while others are true representations of
borrowers.
Key Words
Identities, narratives, peer-to-peer lending, decision making under uncertainty, consumer
financial decision making.
Electronic copy available at: http://ssrn.com/abstract=1840668
3
The last decade has seen a growing number of business models that facilitate economic
exchanges between individuals with limited institutional mediation. Consumers can buy products
on eBay, lend money on peer-to-peer (P2P) loan auction sites such as Prosper.com, and provide
zero-interest ―social loans‖ to entrepreneurs via Kiva.org. In all these cases, strangers must
decide whether to engage in an economic exchange and on what terms using only the provided
information. Objective quantitative data about exchange partners is often difficult to obtain,
insufficient, or unreliable in these contexts. As a result, decision makers may turn to subjective,
unverifiable, but potentially diagnostic qualitative data (Michels 2011).
One form of qualitative data useful to decision makers in such economic exchanges is
narratives constructed by potential exchange partners. A narrative is a sequentially structured
discourse that gives meaning to events unfolding around the narrator (Riessman 1993). For
example, a narrative might explain an individual’s past experiences, current situation or future
hopes (e.g., Thompson 1996; Wong and King 2008). By providing an autobiographical sketch
explaining the vicissitudes of their life, narratives often provide a window into how individuals
conceptualize themselves (Gergen and Gergen 1997) and a portrait of how they construct their
identity. As a result of either ambiguity or the strategic use of the medium to influence others
(e.g., Schau and Gilly 2003), one characteristic of narratives is that they offer only one of several
possible interpretations of self-relevant events (Sonenshein 2010). As a result, narrators can relay
interpretations of their circumstances that convey favorable identities (Goffman 1959).
RESEARCH MOTIVATIONS AND CONTRIBUTIONS
The idea that narratives may involve the construction of a favorable identity poses two key
questions for research. First, while narratives can provide diagnostic information to a decision
maker considering an economic exchange, the veracity of the narrator’s story is difficult to
4
determine. Because a narrative offers the possibility of either describing an authentic, full or true
self, or a partial, inauthentic or misleading self, potential exchange partners are left to intuit the
truth of what is presented. Accordingly, a key question to consider is, given their potential for
diagnostic and misleading information, to what extent do narratives influence economic
exchange transactions? While previous consumer research has largely focused on narratives
regarding consumption experiences (Thompson 1996) or consumption stories (Levy 1981),
scholars have not examined the role of narratives in economic exchanges. Yet, we believe that
narratives may be a particularly powerful lens as they allow the consumer to attempt to better
control an exchange and thus can provide an edge in helping consummate an exchange.
A second key question is to what extent do narratives affect the performance and outcome of
an economic exchange? Narrative scholars claim the construction and presentation of a narrative
can shape its creator’s future behavior (Bruner 1990) but rarely empirically examine the nature
of this influence. Such an examination is critical to understand how a mixture of quantitative and
qualitative factors shapes outcome quality (e.g., Hoffman and Yates 2005).
We examine these two questions using online P2P loan auctions on Prosper.com by studying
borrower-constructed narratives (particularly the identities embedded in them), the subsequent
decisions of lenders and transaction performance two years later. We define identity claims as the
ways that individuals describe themselves for others (Pratt, Rockman, and Kaufman 2006).
Borrowers can construct an identity based on a range of items, such as about religion or success.
This becomes an identity claim when it is public discourse as opposed to a private cognition. In
this framework, we intend to make the following contributions.
First, by developing and testing theory around how narratives supplement more objective
sources of information decision makers use when considering a financial transaction, we draw
5
attention to how narrators can intentionally exploit uncertainty and favorably shape
circumstantial facts to obtain resources such as access to money in unmediated environments.
The narrative, as a supplementary, yet sometimes deal-making or deal-breaking, information
source to decision makers, is predicated on compelling stories versus objective facts. It thus
offers a means for individuals to reconstruct their pasts and present their futures in positive ways.
Second, by linking narratives to objective performance measures, we show how narratives
may predict the longer-term performance of lending decisions. Because the decision stakes are
high in this unmediated and unsecured financial arena, lenders engage in highly cognitive
processing (Petty and Wegener 1998). The strong incentive of potential financial loss leads to
cognitive processing that tends to produce accurate attributions about a person and probabilities
about future events (Osborne and Gilbert 1992). Given this motivation for accuracy, we suspect
lenders use narratives to make investment decisions.
Third, from a practice perspective, the recent financial crisis has exposed flaws in criteria
used to make lending decisions. Quantitative financial metrics, such as credit scores, have proven
unreliable in accurately predicting the ability or likelihood of consumers to repay unsecured
loans (e.g., Feldman 2009). A narrative perspective on the consummation and performance of
financial transactions offers the promise of improving our systems of assessing borrowers.
RESEARCH SETTING
Our research was conducted on Prosper.com (hereafter, Prosper), which is the largest P2P
loan auction site in the U.S. with over a million members and $228 million in personal loans
originated since its inception in March 2006 through April 2011. We selected Prosper because
borrowers and lenders never meet in person, thereby allowing us to assess the role of narratives
in overcoming uncertainty that arises during financial transactions between unacquainted actors.
6
The process of borrowing and lending money through a loan auction on Prosper is as follows.
Before posting their loan request, borrowers give Prosper permission to verify relevant personal
information (household income, home ownership, bank accounts, etc.) and access to their credit
score from Experian, one of the major credit-reporting agencies. Using this and other
information, such as pay-stubs and income tax returns, Prosper assigns each borrower a credit
grade reflecting the risk to lenders. Credit grades can range from AA, indicating the borrower is
extremely low in risk (a high probability of paying back the loan), to A, B, C, D, E, to HR,
signifying the highest risk of default. Borrowers can now post loan requests for auction. When
posting their loan auction, borrowers choose the amount to ask (up to $25,000) and the highest
interest rate they will pay. They may also use an open-text area with unlimited space to write
anything they want—the borrower’s narrative—which is voluntary (see also Michels 2011).
After the listing becomes active, lenders decide whether to bid and, if relevant, how much
money to offer and at what interest rate. A $1,000 loan can be financed by one lender who lends
$1,000 or, at the other extreme, by 40 lenders each lending $25. Most lenders bid the minimum
amount ($25) on individual loans to diversify their portfolios (Herzenstein et al., 2011). After the
auction closes, listings with bids covering the requested amount are funded. If it receives bids
covering more than this amount, the bids with the lowest interest rates win. If the auction does
not receive enough bids, the request remains unfunded. Prosper administers the loan, collects
payments, and receives fees of .5 to 3.0% from borrowers and a 1% annual fee from lenders.
We employed three dependent variables in our study. First, loan funding is the percentage of
the loan request to receive commitment of funding from lenders. For example, if a loan request
for $1,000 receives bids worth $500, then loan funding is 50%. If it receives bids worth $2,000,
loan funding is 200%. A higher value of loan funding signifies greater lender interest. Second,
7
percent reduction in final interest rate captures the decrease in interest rate between the
borrower’s maximum specified rate and the final rate. For example, if a borrower’s maximum
interest rate is 18% and the final rate is 17%, then the percent reduction in interest rate is (1817)/18 = 5.56%. The rate is reduced only if the loan request receives full funding; greater lender
interest results in greater reduction of interest rate. Third, loan performance is the payment status
of the loan two years after its origination. It is also important to explain the types of identity
claims made by prospective borrowers. Borrowers in our sample employed six identity claims in
their narratives: trustworthy, economic hardship, hardworking, successful, moral, and religious.
Table 1A provides definitions and illustrative examples of these identity claims. Borrowers
constructed an average of 1.53 (SD = 1.14) identity claims in their narratives.
RESEARCH HYPOTHESES
Who Is Likely to Provide More Identity Claims in Their Narratives?
Narratives, when viewed as vehicles for identity work, provide opportunities for individuals
to manage the impressions others hold of them. Impression management theory posits that
individuals have a goal to create and maintain a specific identity (Leary and Kowalski 1990).
Narratives provide one avenue of impression management as, through discourse, individuals can
shape situations and construct identities designed specifically to obtain a desired outcome
(Schlenker and Weigold 1992). Some scholars argue that individuals strategically use narratives
to establish, maintain, or protect desired identities (Rosenfeld, Giacalone, and Riordan 1995).
However, the use of impression management need not automatically signal outright lying—
individuals may select from a repertoire of self-images they genuinely believe to be true (Leary
and Kowalski 1990). Nevertheless, strategic use means that, at a minimum, individuals offer
those select representations of their self-image most likely to garner support.
8
In economic exchanges involving repeated transactions, individuals receive feedback from
exchange partners that either validate or dispute the credibility of their self-constructions (Leary
and Kowalski 1990) and thus, are able to determine if an identity claim has been granted. Prior
transactions also offer useful information through feedback ratings and other mechanisms that
convey and archive reputations (Weiss, Lurie, and MacInnis 2008). However, in one-time
economic exchanges, this type of feedback is not available. Instead, individuals have a single
opportunity to present a convincing public view of the self. Similarly, receivers of this
information have only this one presentation to deem the presenter as either credible—or not.
We hypothesize that in these conditions, borrowers will be strategic in their identity claims.
Borrowers with satisfactory objective characteristics will be less likely to construct identity
claims to receive funding: they may feel their case stands firmly based on its objective merits
alone. In contrast, borrowers with unsatisfactory objective characteristics may view narratives as
an opportunity to influence the attributions lenders may make by countering past mistakes and
difficult circumstances. In this scenario, borrowers make identity claims that offset the
attributions made by lenders about the borrower being fundamentally not a creditworthy person.
These dispositional attributions are often based on visible characteristics (Gilbert and Malone
1995). The most relevant objective characteristic of borrowers is the credit grade assigned by
Prosper and derived from personal credit history (Herzenstein et al. 2011). More than one
identity claim allows borrowers to present a more complex, positive self to counteract negative
objective information from a low credit grade. Thus:
H1: The lower the borrower’s credit grade, the greater will be the number of identities
claimed by the borrower in the narrative.
The Impact of the Number of Identity Claims on Lenders’ Decision
Although economists often expect that unverifiable information should not matter (e.g.,
9
Farrell and Rabin 1996), we suggest that the number of identity claims constructed in a
borrower’s narrative will play a role in lenders’ decision making for at least two reasons. First,
borrower narratives with too few identities may fail to resolve questions about the borrower’s
disposition. If a borrower fails to provide sufficient diagnostic information for lenders to make
attributions about the borrower (Cramton 2001), lenders may suspect the borrower lacks
sufficient positive or distinctive information or is withholding or hiding germane information.
Second, the limited diagnostic information provided by fewer identity claims limits a
decision maker’s resolution of outcome uncertainties. Research on perceived risk supports this
reasoning showing that decision makers gather information as a risk-reduction strategy and tend
to be risk averse in the absence of sufficient information regarding the decision at hand (e.g., Cox
and Rich 1964). In the P2P lending arena, the loan request and evaluation process unfold online
without any physical interaction between the parties. Furthermore, on Prosper, borrowers are
anonymous to lenders (real names and addresses are never revealed). This lack of seemingly
relevant information is especially salient because many decision makers view unmediated online
environments as ripe for deception (e.g., Caspi and Gorsky 2006). To the extent that the identity
claims presented in a narrative reduce uncertainty regarding a borrower, lenders should be more
likely to view the listing favorably, increase loan funding and decrease the final interest rate paid.
Therefore, the number of identity claims in a borrower’s narrative may serve as a heuristic for
assessing the borrower’s loan application and likely lead to greater interest in the listing.
While we suggest that the number of identities borrowers claim will result in favorable
lending decisions, we further suggest that these identity claims may persuade lenders erroneously
and thus, lenders may fund loans with a lower likelihood of repayment. Borrowers can use
elaborate multiple-identity narratives to craft ―not-quite-true‖ stories and make promises they
10
might find difficult to keep later on. More generally, a greater number of identity claims may
suggest that borrowers are being more strategic and positioning themselves in a manner they
believe is likely to resonate with lenders as opposed to presenting a true self. Therefore, we posit
that, whether consciously or not, borrowers that construct several identities may have more
difficulty fulfilling their obligations, and be more likely to fall behind or stop loan repayments
altogether. Thus, despite the high stakes of the decision, we believe lenders swayed by multiple
identities are more likely to fall prey to borrowers that underperform or fail (Goffman 1959).
H2: Controlling for objective, verifiable information, the more identities borrowers
claim in their loan requests, the more likely lenders will be to: (a) fund the loan, and
(b) reduce its interest rate, but (c) the poorer will be the likelihood of repayment.
The Role of the Content of Identity Claims on Lender Decision Making and Loan Performance
We also examine the extent to which select identities affect lenders’ decision making and the
longer-term performance of loans. Due to a limited theoretical basis for determining the types of
identities most likely to influence lenders’ decision making, we pose this part as exploratory.
Research on trust offers a promising starting point (Mayer, Davis, and Schoorman 1995), as
it suggests that identities may reduce dispositional uncertainty and favorably influence lenders.
Trust is a crucial element for consummation of an economic exchange (Arrow 1974). Scholars
theorize trust as involving three components: integrity (borrowers adhere to principles lenders
accepts), ability (borrowers possess skills necessary to meet obligations) and benevolence
(borrowers have some attachment to lenders and are inclined to do good) (Mayer et al.1995).
We theorize that trustworthy, religious, and moral identities will increase the perception of
integrity by leading lenders to believe that borrowers ascribe to the likely lender-endorsed
principle of fulfilling obligations, either directly (trustworthy) or indirectly by adhering to a
philosophy (religious or moral). More specifically, the moral identity tells potential lenders that
an individual has ―a self-conception organized around a set of moral traits‖ (Aquino et al. 2009,
11
p. 1424), thereby increasing perceptions of integrity. The religious identity signals a set of roleexpectations a person is likely to adhere to, and while religions vary in the content of these role
expectations (Weaver and Agle 2002), many include principles around not lying or stealing, and
generally honoring a contractual agreement. We further suggest that the hardworking identity
will increase the perception of integrity, because hardworking individuals are determined and
dependable, and as such they are often problem solvers (Witt et al. 2002), meaning that they will
do their best to meet their obligations, a disposition likely to resonate with lenders.
Additionally, we reason that the religious and moral identities will invoke in lenders a sense
of benevolence, a foundational principle of many religions and moral philosophies. Similarly, the
economic hardship identity may also invoke benevolence as the borrower exhibits forthrightness
about past mistakes, which may suggest to lenders that the borrower is trying to create a
meaningful relationship based on transparency.
We theorize that an identity claim of success can increase the perception of ability and the
belief that the narrator is capable of fulfilling promises (Butler 1991). Lenders are more likely to
lend money to a borrower if they perceive the individual as capable of on-time repayment
(Newall and Swan 2000). While the successful identity is likely to describe the past or present, it
can also serve as an indication of a probable future (i.e., that the borrower will continue to be
successful), thereby helping ―fill in the blanks‖ about the borrower in a positive way. On the
other hand, economic hardship is likely to construct the borrower as someone who has had a
setback and ultimately undermine perceived ability, thus negatively affecting lenders’ decisions.
While we have offered some preliminary theory in support of specific relationships between
identity content and loan funding/interest rate reductions, we acknowledge that this is
exploratory and therefore pose these relationships as an exploratory research question (ERQ):
12
ERQ1: Which types of identity claims influence lending decisions as indicated through
(a) an increase in loan funding, and (b) a decrease in the final interest rate?
We also examine in an exploratory way the impact of the content of identity claims on loan
performance. We envision two potential scenarios. Under the first scenario, we assume identities
are diagnostic of the borrower or serve as self-fulfilling prophecies. Examining the ability aspect
of trustworthiness, we anticipate a negative relation between the economic hardship identity and
loan performance (as individuals further validate their claim of setbacks), and a positive relation
between the successful identity and loan performance (as individuals prove their claim of past
success). Moreover, we expect the four identities used to construct the borrower as having
integrity—trustworthy, hardworking, moral, and religious—to be related to increased loan
performance. After presenting oneself as having integrity, the borrower will probably have a
strong psychological desire for consistency between what s/he says and what s/he does (Cialdini
and Trost 1998). That is, borrowers may be making an active, voluntary, and public commitment
that past literature has shown to psychologically bind narrators to a particular set of beliefs and
subsequent behaviors (Berger and Heath 2007). As these four identities speak to fundamental
self-beliefs versus predicted outcomes (e.g., success or hardship), they may greatly motivate
individuals to live up to their claims. This may explain why these identities, regardless of their
accuracy, can nevertheless become true and predict the performance of the lending decision.
However, under another scenario, identities can improve the lender’s impression of the
borrower, thereby allowing borrowers to exert control over the impressions given (Goffman
1959). Borrowers (or narrators more generally) construct positive impressions of themselves, and
therefore may misrepresent themselves and send off signals that may not be objectively
warranted. As such, despite believing that self-constructing identities are helpful for a lending
decision, they may have no impact or may even be harmful to lenders.
13
Given the mixed possibilities, we pose the following exploratory research question:
ERQ2: How are the content of identity claims and loan performance related?
STUDY
Data
Our dataset consists of 1,493 loan listings posted by borrowers on Prosper in June 2006 and
June 2007. We extracted the dataset using a stratified random sampling strategy. First, using a
web crawler we extracted all loan listings posted in June 2006 and June 2007 (approximately
5,400 and 12,500 listings, respectively). A significant percentage of borrowers on Prosper have
very poor credit histories and most loan requests listed on the site do not receive funding. To
avoid overweighting high-risk borrowers and unfunded loans, we sampled an equal number of
loan requests from each credit grade. To do so, we first separated funded loan requests from
unfunded ones. We then divided each group by the seven credit grades assigned by Prosper.
Finally, we eliminated all loan requests without text for three reasons. First, including loan
requests without a narrative could confound the borrower’s choice to write something in the open
text box but not utilize narratives with the choice to write nothing at all. Second, the vast
majority of listings lacking a narrative are not funded. Third, loan requests without text
represented only 9% of all loans posted in June 2006, and 4% of the loans posted in June 2007.
We nevertheless used the ―no text‖ loans in a robustness check for our results.
Next, we randomly sampled from the 14 sub-groups (2 funding status × 7 credit grades). In
2006, we sampled 40 listings from each sub-group (until data was exhausted) to obtain 513
listings and in 2007, we sampled 70 listings from each sub-group to obtain 980 listings—a total
of 1,493 listings. Each listing in our dataset includes the borrower’s credit grade, requested loan
amount, the maximum interest rate, loan funding, the final interest rate of funded loans, the
14
payback status of funded loans after two years, and open-ended text data. Before combining data
from 2006 and 2006 we tested for an effect of ―year‖ on how our independent variables affect the
dependent variables and found none, thus the data were combined.
Dependent Variables
The first dependent variable, loan funding, ranges from zero to 905% in our dataset, but the
equal inclusion of all credit ratings skews these statistics. The mean percent funded in our dataset
(including all listings) is 105.74% (SD = 129.2) and for funded listings is 205.45% (SD = 119.6).
Because it was skewed, we log-transformed loan funding as follows, Ln(percent funded + 1).
The second dependent variable, percent reduction in the final interest rate, ranges from zero to
56% in our dataset. The mean percent reduction in interest rate in our dataset (including all
listings) is 6.4% (SD = 10.7) and for funded listings is 11.88% (SD = 12.75). Since the
distribution is skewed, it was log-transformed (see distribution figures in web appendix).
The third dependent variable is loan performance measured two years after loan funding. For
each funded loan in our dataset, we obtained data about whether the loan was paid ahead of
schedule and in full (31.1% of the funded listings in our dataset were paid ahead), was current
(i.e., paid as scheduled; 40.5%), late (i.e., payments were between one to four months late;
7.1%), or had defaulted (21.3%). While it may seem that this dependent variable is ordered,
likelihood ratio tests reveal that a multinomial logit model fares better than an ordered logit
model when analyzing these data (for both independent variables, the number and content of
identities). Thus, in the analysis reported below we employ a multinomial logit model.
Independent Variables
We read approximately one third of all narratives and developed an inductively derived list of
six identity claims found in the data (Miles and Huberman 1994): trustworthy, economic
15
hardship, hardworking, successful, moral and religious. Next, two research assistants examined
the same data and determined these six identities were exhaustive. Table 1A provides definitions
of the identity claims and examples from our data. Then we employed five other pairs of research
assistants (10 total) to code the entire dataset. Each identity was coded as a dichotomous variable
that receives the value ―1‖ if the identity claim was present in a borrower’s narrative and ―0‖
otherwise. Each listing in our dataset was read by a pair of research assistants who first read the
listings independently and then discussed them to provide a unifying code for each listing.
Further, all research assistants coded 20 randomly sampled listings from our dataset to measure
their agreement. The Fleiss Kappas range from .66 to .96 (see Table 1B) indicating a substantial
agreement rate according to the Kappa interpretation guide offered by Landis and Koch (1977).
Table 1B also presents the correlation between each pair of identities and the correlation of the
presence of each identity with the total number of identities claimed.
**** INSERT TABLES 1A AND 1B ABOUT HERE****
To test H1 and H2, the independent variable is the number of identities constructed by the
borrower. That is, we counted how many different identities the loan request narrative included,
with 0 representing no constructed identities and 6 representing the maximum number of
identities. Most of the listings in our dataset (81.4%) included at least one identity. Since only 17
listings include five or six identities, we grouped them with listings that created four identities to
smooth this variable and create more even groups (we note that this grouping affects neither our
theory nor the substantive results obtained). Consequently, the variable representing the number
of identities has the following distribution: zero identities = 18.6%; one identity = 35.6%; two
identities = 26.7%; three identities = 13.7%; four or more identities = 5.4%.
Control Variables
16
Prior research shows that borrowers on Prosper with better credit grades, those requesting
smaller loan amounts, and offering higher initial interest rates are substantially more likely to
receive funding (Ryan, Reuk, and Wang 2007). Therefore, we control for these in our analyses.
We treat the credit grade as categorical because, although this variable is ordinal, we do not
believe it is interval (i.e., the difference between HR and E may not be the same as the difference
between A and AA). We created seven dummy variables, one for each credit grade. In our
analyses, we use dummy variables for credit grades AA-E and leave HR as the base comparison.
Additionally, we controlled for borrowers’ demographics. Our ten research assistants coded
the demographic characteristics of borrowers from their narratives and pictures (specifically their
gender, race, marital status and whether their family includes children). Two research assistants
coded each listing for demographics and the results of the two coders were virtually identical.
Disclosure of demographic information is voluntary on Prosper, and in our dataset, 66% of
borrowers included at least some demographic information in their loan requests, either in their
narrative or through pictures. We coded these demographic variables as categorical and note that,
because disclosure is voluntary, the interpretation of these variables is, for example, not ―female‖
but ―disclosing that the borrower is a female‖. For this reason, our models include dummies for
both female and male and the reference category is ―the borrower did not disclose gender‖.
Results and Discussion
H1 posited that a lower credit grade would result in a higher number of identities claimed in a
borrower’s narrative. We employed a trend analysis to test this hypothesis because, as Dawes and
Corrigan (1974) suggest, a linear function can account for almost all of the variance in any
conditionally monotonic relationship. For this purpose, we coded the credit grade information as
one variable that receives the following values: 1 if the credit grade is HR, 2=E, 3=D, 4=C, 5=B,
17
6=A, and 7=AA (we employed an ANCOVA that treats this variable as categorical). Controlling
for demographic and loan characteristic variables, the trend analysis showed, as expected, a
significant linear relation between borrowers’ credit grade and the number of identities borrowers
constructed in their narratives (t(1472) = 2.63, p < .01). Higher order polynomial trends did not
account for a significant proportion of the variance in the number of identities (pquadratic = .17,
pcubic = .81, porder4 = .16). Figure 1 provides the estimated marginal means of the number of
identities as a function of credit grade, controlling for the demographics and loan characteristics.
**** INSERT FIGURE 1 ABOUT HERE****
In H2a, we posited that an increased number of identities claimed in a narrative would have a
positive effect on loan funding. To test this hypothesis, we regressed the number of identities as
the main predictor and credit grade, requested amount, gender, marital status, race, and having
children as covariates on the natural log of loan funding. We present results in Table 2 and show
that, controlling for borrowers’ demographics and loan characteristics, the number of identities
borrowers claim in their narratives positively affects loan funding (β = .11, SE = .04, t = 2.81, p <
.001). The model including the number of identities as the main predictor fits the data better than
a model including only the controls (R2 = .46 versus .44, F(1, 1472) = 7.91, p < .01).
Furthermore, and in support of H2a, the relation between the number of identities and loan
funding is monotonic and increasing (the linear trend is significant (t(1472) = 2.71, p < .01), but
not the quadratic: pquadratic = .99; and the cubic is marginal: pcubic = .08). Figure 2 presents the
average loan funding as a function of the number of identities.
H2b stated that the percent reduction in the final interest rate would be positively affected by
the number of identity claims of the borrower. We employed the same regression analysis as in
H2a but with the natural log of percent reduction in the final interest rate as the dependent
18
variable. Results are presented in Table 2 and show that the number of identities borrowers claim
positively affects the reduction of interest rate (β = .12, SE = .03, t = 4.04, p < .001). The model
including the number of identities as the main predictor fits the data better than a model
including only the controls (R2 = .17 versus .15, F(1, 1472) = 16.33, p < .001). As before, and in
support of H2b, the relation between the number of identities and the reduction in interest rate is
monotonic and increasing (the linear trend is significant (t(1472) = 3.98, p < .001), but not the
quadratic, pquadratic = .52 or cubic, pcubic = .15). Figure 2 presents more detail about this analysis.
Additionally, to test the robustness of our specifications, we examined the effect of the
presence of text on loan funding. Our dataset for this analysis included all loans posted on
Prosper on June 2006 (n = 4,959 with no repeat listings). We created a dichotomous variable that
takes the value 1 if the borrower included text in the loan request and 0 if not. Controlling for the
financial characteristics, results show that loan funding increases when the listing includes text (β
= .32, t = 5.64, p < .001). Indeed, only 3.5% of the listings that have no text were fully funded,
compared with 8.0% of the loans with text during that month (χ2(1)= 12.63, p < .001).
**** INSERT TABLE 2 AND FIGURE 2 ABOUT HERE****
To test H2c, which posited a negative effect of number of identities on loan performance, we
employed a multinomial logistic model. The main predictor was the number of identities and we
controlled for borrower demographic and financial characteristics. Results reveal that as the
number of identity claims provided in the loan request increase, borrowers are more likely to
default than to pay late (β = .29, χ2(1) = 2.98, p < .1), and are more likely to pay late than to pay
on time (β = -.31, χ2(1) = 3.73, p = .053) or ahead of time (β = -.32, χ2(1) = 3.67, p = .054). The
number of identities does not affect inclusion in any other categories of loan performance (see
Table WA-1 in the web appendix). Figure 2 presents the percentage of borrowers that repay
19
(current and paid combined) by the number of identities constructed in their narrative. Table 3
presents the distribution of loan performance as a function of the number of identities borrowers
claim in their narratives. From this table it is clear that as the number of identities increases, the
default rate increases and the paid rate decreases.
**** INSERT TABLES 3 AND 4 ABOUT HERE****
ERQ1 concerned the roles played by the content of the identities claimed by borrowers on
lender decision making. We examined this exploratory research question using regression
analyses with the six identities as main predictors. As before, we controlled for demographic
variables and other financial characteristics to account for the data available to lenders when they
make decisions. We present our results in Table 4.
After controlling for borrowers’ demographics and loan characteristics, lenders are affected
by the trustworthy identity (β = .33, SE = .09, t = 3.79, p < .001) and the successful identity (β =
.23, SE = .11, t = 2.11, p < .05) such that loan funding increases. Lenders are also marginally
affected by the religious identity (β = -.33, SE = .18, t = -1.82, p = .07), but this identity reduces
loan funding. Borrowers claiming the trustworthy or successful identities received significantly
higher loan funding than those who did not claim these identities (120.74% versus 81.51%,
t(1491) = 5.76, p < .001; 124.89% versus 101.06%, t(1491) = 2.84, p < .01, respectively).
Are the above effects of the identities due to an omitted variable in the regression (the total
number of identities) rather than the content of these identities? That is, do the successful and
trustworthy identities always appear in narratives that have many identities and the religious
identity always appear in narratives that have fewer identities? Examining the correlations of the
appearance of the six identities with the total number of identities (see Table 1B) shows this is
not the case. For example, the low correlation of the successful identity with the overall number
20
of identities shows that this identity’s positive effect on lenders’ decisions is because of its
content rather than the number of identities in narratives that include successful claims.
Claiming any other identity neither helped nor hurt borrowers attempting to secure funding.
The model including the six identities predicts loan funding significantly better than a model
including only the controls (R2 = .46 versus .44, F(6, 1467) = 3.84, p < .001). We found similar
results for the final interest rate with the trustworthy (β = .24, SE = .07, t = 3.55, p < .001) and
successful identities (β = .27, SE = .08, t = 3.26, p < .001) helping to reduce final interest rates.
Borrowers claiming trustworthy or successful identities enjoy greater reduction in interest rate
compared with those who did not claim these identities (7.60% versus 4.54%, t(1491) = 5.41, p <
.001; 8.86% versus 5.83%, t(1491) = 4.38, p < .001, respectively). None of the other identities
had an effect on reduction in interest rate. Also, the model including the six identities predicts the
percent reduction in interest rate better than the model including only the controls (R2 = .17
versus .15, F(6, 1467) = 4.75, p < .001).
One interesting question is what affects lenders more, the number of identities or their
content. We compared the unrestricted model (the one with the six different identities) with the
restricted model (the one in which we count the identities and ignore their content) and using the
F-test for nested models we find that the unrestricted model fits the data better (F(5, 1472) =
3.02, p = .01). Next, we tested whether there is a partial effect of the number of identities after
controlling for their content. To the regressions presented in Table 4 we added a dichotomous
variable for listings by borrowers who claim many identities (4 or more). This variable has a
marginal positive effect on loan funding above and beyond the content of the identities (β = .35,
SE = .21, t = 1.67, p < .1). Moreover, the trustworthy identity remains significant in this analysis
(β = .32, SE = .09, t = 3.63, p < .001), the successful identity becomes less significant (β = .19,
21
SE = .11, t = 1.73, p < .1), and the religious identity becomes more significant (β = -.43, SE =
.19, t = -2.23, p < .05). Taken together, these analyses show that the effect of the number of
identities is weaker than the effect of the specific identities.
To examine ERQ2, we subjected the loan performance measure to a multinomial logit with
the identities as the main predictors along with other borrower and loan characteristics as
controls (see results in WA-2 in the web appendix). We find that while the successful identity
helps borrowers fund their loans and reduce their final interest rate, this identity is unrelated to
loan performance. The trustworthy identity predicts loan performance such that borrowers who
claim this identity are more likely to pay ahead of time than to pay on time (β = -.51, χ2(1) =
5.51, p < .05). More interestingly, two other identities were found to be related to loan
performance, moral (positive) and economic hardship (negative). Specifically, borrowers
claiming a moral identity are more likely to pay on time than pay late (β = .81, χ2(1) = 3.00, p <
.1) or default (β = .73, χ2(1) = 5.77, p < .05). Additionally, borrowers claiming the economic
hardship identity are more likely to default (β = .90, χ2(1) = 9.15, p < .001), pay late (β = .69,
χ2(1) = 3.10, p < .1), or on time (β = .66, χ2(1) = 6.54, p < .01) than pay their loans ahead of time.
The model including the six identities predicts loan performance better than the model including
the controls only (-2LL = 1547 versus 1575, χ2(6) = 27.88, p < .001).
In sum, our findings suggest that while the trustworthy and successful identity claims are
more likely to result in funded loans, lenders should only consider the trustworthy identity.
Lenders should also consider the moral identity and put negative weight on the economic
hardship identity as it is negatively related to loan performance.
GENERAL DISCUSSION
In this research, we examined the roles narratives play in influencing decision making in
22
economic exchanges between previously unknown transaction partners—borrowers who serve as
narrators and lenders who serve as decision makers. We reasoned and found support for the
hypothesis that narratives influence decision makers beyond objective information. Narrators
appear to strategically provide identities and favor multiple identities when their credit grades are
poor. From the lender’s perspective, claiming more identities predicts an increased likelihood of
loan funding and a reduction of the final interest rate. Yet, more identities can have a negative
impact on loan performance. Thus, talk (i.e., claiming identities) is ―cheap‖, but given that
lenders respond to it, talk affects important outcomes. Further, claiming trustworthy and
successful identities influence lenders to engage in economic exchanges, but ironically the
successful identity is not predictive of a positive outcome, and the trustworthy identity explains
the difference between borrowers who pay on time and those who pay ahead of time, both
positive outcomes. Instead, more predictive of long-term performance are the moral and
economic hardship identities (positively and negatively respectively). To the extent that lenders’
and borrowers’ interests are sufficiently aligned (e.g., negative loan performance affects the
ability to borrow again on Prosper), a misrepresentation that increases loan funding but does not
positively affect loan performance may be detrimental to borrowers and lenders.
Theoretical Implications
The narrative approach to identity that we apply to decision making builds on recent
scholarly interest around the use of narratives in facilitating economic transactions. For example,
Martens, Jennings, and Jennings (2007) found narratives help entrepreneurs secure resources for
their endeavors by providing a compelling story about the organization’s identity. Chen, Yao, and
Kotha (2009) examined business plans of entrepreneurs that, as a type of narrative, serve as a
vehicle providing vital information during financial exchanges involving uncertainty. We extend
23
the study of narratives to financial transactions involving unsecured personal loans between
individuals. We explain our findings using the idea that information embedded in identity claims
reduces dispositional uncertainty and provides contextual information that influences decision
makers. By extending our findings to include longer-term performance, we show that narratives
not only influence decision makers but also can predict the outcome of their decisions. This
demonstration expands the scope of narrative research from an exercise in persuasion to a key
factor in predicting outcomes, a point narrative scholars have frequently made (e.g., Bruner
1990), but one that has gone largely undeveloped and thus far, unproven.
In explaining our exploratory findings around the content of identity claims and lending
decision making, we found the trustworthy identity may invoke attributions of integrity and the
perception that the borrower adheres to a set of principles the lender finds acceptable. The
successful identity may invoke attributions of ability and the perception that the borrower is
competent and has the skills necessary to meet obligations. Future research may seek to explain
why the moral and religious identities did not increase attributions of integrity. It could be that
lenders discount or ignore these particular identities based on their persuasion knowledge
regarding such identity claims (Friestad and Wright 1994) or because of attributions based on
these identities regarding the borrower (Gilbert and Malone 1995).
We also posed as exploratory research the relationship between the content of identity claims
and loan performance, proposing two scenarios. In the first, we posited that constructed identities
are transparent and true indicators of the self and, as theorized, an economic hardship identity
negatively affects loan performance. However, the presentation of a successful identity did not
relate to loan performance. While future research is needed to unpack this mixed finding, one
explanation may be that negative constructs (such as economic hardship) tend to be more
24
powerful than positive ones (Baumeister et al. 2001). We also suggested that the four identities
around integrity would psychologically bind individuals to engage in behaviors to uphold their
sense of integrity. In fact, the moral and trustworthy identities exhibited this effect. One
explanation for the strong effect of the moral identity may be that this identity serves as a selfregulatory mechanism (Aquino et al., 2009). On the other hand, a second scenario suggests that
there is either no or even a negative relationship between the content of the identity claims and
loan performance. Indeed, the successful, hardworking, and religious identities are not related to
loan performance, and economic hardship is negatively related to loan performance.
More generally, our findings largely support the interpretation that individuals use narratives
strategically to manage favorable impressions, something which contributes to research on
impression management and its role in decision making. A wide body of research indicates
individuals are adept in crafting favorable self-views for themselves and others (Leary and
Kowalski 1990). What is less clear is the relationship between these impressions and reality. Our
study addresses this question by assessing whether impressions individuals construct bear out in
the future. We find that, even for a process as complex and prone to unforeseen environmental
contingencies (job loss, medical emergency, etc.) as a loan (Warren and Warren Tyagi 2003),
when individuals claim a moral identity, they strive to fulfill that presentation. However, the
economic hardship identity is negatively related to loan performance suggesting that, despite
being authentic about their difficulties, it is not always reasonable to expect these borrowers to
create a new reality. Additionally, the pattern of results suggests that individuals use an economic
hardship identity strategically to gain empathy, a tactic lenders should view as a warning sign as,
ultimately, these borrowers lack either the ability or willingness to fulfill their loan obligations.
Practical Implications
25
Our findings suggest how both borrowers (narrators) and lenders (decision makers) can be
more effective. For borrowers, our findings show the power of constructing a viable narrative
when attempting to influence a decision maker. An effective narrative should contain several
identity claims and emphasize a trustworthy or successful identity. For lenders, our findings
show the importance of analyzing narratives when making economic exchange decisions under
uncertain conditions. The analysis of narratives can provide a competitive advantage to lenders
that goes beyond traditional credit-based risk scoring models. While credit scores and other
objective economic information will continue to play an important role in lending decisions, our
findings suggest the benefits of supplementing this approach with an informed assessment of
narrative data. Specifically, lenders should favor borrowers who claim to be moral or trustworthy
and avoid borrowers claiming economic hardship. In this sense, our research offers an important
method to assess financial transactions beyond traditional credit-based models.
Limitations and Future Research Directions
Our research has several limitations and, as a result, suggests promising directions for future
research. First, while our data includes important information about borrowers, lenders’
decisions, and outcomes, we were unable to get any additional information about lenders. Based
on theory, we inferred how lenders make decisions. As such, there is an opportunity for future
research to capture the mental maps of decision makers when evaluating narratives. For example,
are lenders cognizant of the number and content of identities, or are their responses heuristically
based? Do key factors, such as lending experience, explain why some decision makers are more
adept at consummating transactions with favorable outcomes? Is it possible that narratives that
are familiar to and resonate with particular decision makers have an increased likelihood of
funding? Knowing the background and personal characteristics of lenders would be useful in this
26
regard. Second, given the novelty of this research area, we centered our hypotheses around single
identities. To build on our findings, future research should develop and test hypotheses about
interactions between identities. Third, we focused on one aspect of narratives—identities, both
their number and content. Yet, as rich mediums of communication, narratives contain more
complexity than we explored. For example, narratives convey emotions (e.g., sympathy, anger)
or attributions (internal or external) that may influence decision makers, and temporal structures
that reveal how individuals reconstruct the unfolding of events (Gergen and Gergen 1997).
Studying these aspects and their effects on consummation and performance of economic
exchanges can add further depth to our approach. Additionally, we acknowledge that the size of
the effects of narratives on loan funding is generally small (around 2%), even if it is in line with
effect sizes reported by other narrative researchers (for example, Martens et al. (2007) show that
identity claims affect underpricing in IPOs above and beyond objective information by 3%).
Our findings present both opportunities and challenges for research. Economic exchanges at
arm’s length provide little diagnostic information for exchange partners. What is especially
attractive about our approach is that the alternative source of information provided in a narrative
can reduce uncertainty and lead to more frequent successful exchanges. This information can
also enhance predictions of performance. As evidenced by the recent financial crisis, lenders
often make poor decisions that lead to sub-optimal outcomes (Feldman 2009). Our findings
suggest that decision makers might be relying on the wrong type of data or not adequately
supplementing their analyses with alternative data. A narrative, as a rich source of qualitative
data about who an individual is, offers the promise to expand current decision making models of
lending (and other economic transactions), reduce uncertainty transaction partners usually face
and limit future challenges similar to those recently experienced in the financial markets.
27
REFERENCES
Aquino, Karl F., Dan Freeman, Americus Reed II, Vivien K. G. Lim and Will Felps (2009),
―Testing a Social-Cognitive Model of Moral Behavior: The Interactive Influence of
Situations and Moral Identity Centrality,‖ Journal of Personality and Social Psychology,
97(1), 123-141.
Arrow, Kenneth (1974), The Limits of Organization. New York: Norton, York.
Baumeister, Roy F., Ellen Bratslavsky, Cartin Finkenauer, and Kathleen D. Vohs (2001), ―Bad Is
Stronger Than Good,‖ Review of General Psychology, 5(4), 323-370.
Berger, Jonah and Chip Heath (2007), ―When Consumers Diverge from Others: Identity
Signaling and Product Domains,‖ Journal of Consumer Research, 34(2), 121-134.
Bruner, Jerome S. (1990), Acts of Meaning. Cambridge, MA: Harvard University Press.
Butler, John K. (1991), ―Toward Understanding and Measuring Conditions of Trust: Evolution of
a Conditions of Trust Inventory,‖ Journal of Management, 17(3), 643-663.
Caspi, Amer and Paul Gorsky (2006), ―Online Deception: Prevalence, Motivation, and
Emotion,‖ CyberPsychology and Behavior, 9(1), 54-59.
Chen, Xiao-Ping, Xin Yao, and Suresh Kotha (2009), ―Passion and Preparedness in
Entrepreneurs’ Business Plan Presentations: A Persuasion Analysis of Venture Capitalists’
Funding Decisions,‖ Academy of Management Journal, 52(1), 199-214.
Cialdini, Robert B. and Melanie R. Trost (1998), ―Social Influence: Social Norms, Conformity,
and Compliance,‖ in The Handbook of Social Psychology, 4th ed. Daniel T. Gilbert, Susan T.
Fiske and Gardner Lindzey, eds. Boston: McGraw-Hill, 151-192.
Cox, Donald F. and Stuart U. Rich (1964), ―Perceived Risk and Consumer Decision-Making:
The Case of Telephone Shopping,‖ Journal of Marketing Research, 1(4), 32-39.
Cramton, Catherine Durnell (2001), ―The Mutual Knowledge Problem and Its Consequences for
Dispersed Collaboration,‖ Organization Science, 12 (3), 346-371.
Dawes, Robyn M. and Bernard Corrigan (1974), ―Linear Models in Decision Making,‖
Psychological Bulletin, 81(2), 95-106
Duarte, Jefferson, Stephen Siegel, and Lance Young (2009), ―Trust and Credit,‖ Working paper,
Rice University.
Farrell, Joseph, and Matthew Rabin (1996), ―Cheap Talk,‖ Journal of Economic Perspectives,
10(3), 103-118.
Feldman, Amy (2009), ―Peer-to-Peer Lending for Nanks, Too?‖ BusinessWeek, April 27.
Friestad, Marian and Peter Wright (1994), ―The Persuasion Knowledge Model: How People
28
Cope with Persuasion Attempts,‖ Journal of Consumer Research, 21(1), 1-31.
Gergen, Kenneth J. and Mary K. Gergen (1997), ―Narratives of the Self,‖ in Memory, Identity,
Community: The Idea of Narrative in the Human Sciences, Lewis P. Hinchman and Sandra
Hinchman, eds. Albany, NY: State University of New York, 161-184.
Gilbert, Daniel T., Patrick S. Malone (1995), ―The Correspondence Bias,‖ Psychological
Bulletin, 117(1), 21-38.
Goffman, Erving (1959), The Presentation of Self in Everyday Life. New York: Doubleday.
Herzenstein, Michal, Utpal M. Dholakia, and Rick L. Andrews (2011), ―Strategic Herding
Behavior in Peer-to-Peer Online Loan Auctions,‖Journal of Interactive Marketing,25(1),27-36.
Hoffman, Robert R. and J. Frank Yates (2005), ―Decision(?) Making(?),‖ IEEE Intelligent
Systems, 20(4), 76-83.
Landis, J. Richard and Gary G. Koch (1977), ―The Measurement of Observer Agreement for
Categorical Data,‖ Biometrics, 33(1), 159-174.
Leary, M., R. Kowalski. 1990. Impression Management: A Literature Review and a TwoComponent Model. Psychological Bulletin 107(1) 34-47.
Levy, Sidney J. (1981), ―Interpreting Consumer Mythology: A Structural Approach to Consumer
Behavior,‖ Journal of Marketing, 45, 49-62.
Martens, Martin L., Jennifer E. Jennings, and P. Devereaux Jennings (2007), ―Do the Stories
They Tell Get Them the Money They Need? The Role of Entrepreneurial Narratives in
Resource Acquisition,‖ Academy of Management Journal, 50(5), 1107-1132.
Mayer, Roger C., James H. Davis, and F. David Schoorman (1995), ―An Integrative Model of
Organizational Trust,‖ Academy of Management Review, 20, 709-734.
Michels, Jeremy (2011), ―Do Unverifiable Disclosures Matter? Evidence from Peer-to-Peer
Lending,‖ Working Paper, University of Colorado.
Miles, Matthew B. and A. Michael Huberman (1994), Qualitative Data Analysis: An Expanded
Sourcebook, 2nd ed. Thousand Oaks, CA: Sage.
Newall, Susan and Jacky Swan (2000), ―Trust and Inter-organizational Networking,‖ Human
Relations, 53(10), 1287-1328.
Osborne, Randall E. and Daniel T. Gilbert (1992), ―The Preoccupational Hazards of Social Life,‖
Journal of Personality and Social Psychology, 62(2), 219-228.
Petty, Richard and Duane T. Wegener (1998), ―Matching Versus Mismatching Attitude
Functions: Implications for Scrutiny of Persuasive Messages,‖ Personality and Social
Psychology Bulletin, 24(3), 227-240.
29
Pratt, Michael G., Kevin W. Rockmann, and Jeffrey B. Kaufmann (2006), ―Constructing
Professional Identity: The Role of Work and Identity Learning Cycles in the Customization
of Identity among Medical Residents,‖ Academy of Management Journal, 49(2), 235-262.
Riessman, Catherine Kohler (1993), Narrative Analysis. Newbury Park, CA: Sage Publications.
Rosenfeld, Paul, Robert A. Giacalone, and Catherine A. Riordan (1995), Impression
Management in Organizations. London: Routledge.
Ryan, Joe, Katya Reuk, and Charles Wang (2007), ―To Fund or Not to Fund: Determinants of
Loan Fundability in the Prosper.com Marketplace,‖ working paper, Stanford University.
Schau, Hope Jensen and Mary C. Gilly (2003), ―We Are What We Post? Self-presentation in
Personal Web Space,‖ Journal of Consumer Research, 30(3), 385-404.
Schlenker, Barru R. and Michael F. Weigold (1992), ―Interpersonal Processes Involving
Impression Regulation and Management,‖ Annual Review of Psychology, 43, 133-168.
Shafir, Eldar, Itamar Simonson, and Amos Tversky (1993), ―Reason-based Choice,‖ Cognition,
49, 11-36.
Sonenshein, Scott (2010), ―We’re Changing or Are We? Untangling the Role of Progressive,
Regressive and Stability Narratives during Strategic Change Implementation,‖ Academy of
Management Journal, 53(3), 477-512.
Thompson, Craig J. (1996), ―Caring Consumers: Gendered Consumption, Meanings and the
Juggling Lifestyle,‖ Journal of Consumer Research, 22(4), 388-407.
Warren, Elizabeth and Amelia Warren Tyagi (2003), The Two Income Trap: Why Middle-Class
Mothers and Fathers are Going Broke. New York: Basic Books.
Weaver, Gary R. and Bradley R. Agle (2002), ―Religiosity and Ethical Behavior in
Organizations: A Symbolic Interactionist Perspective,‖ Academy of Management Review,
27(1), 77-97.
Weiss, Allen M., Nicholas H. Lurie, and Deborah J. MacInnis (2008), ―Listening to Strangers:
Where Responses are Valuable, How Valuable Are They and Why?‖ Journal of Marketing
Research, 45(4), 425-436.
Witt, L. A., K. Michele Kacmar, Dawn S. Carlson, and Suzanne Zivnuska (2002), ―Interactive
effects of personality and organizational politics on contextual performance,‖ Journal of
Organizational Behavior, 23(8), 911–926.
Wong, Nancy and Tracey King (2008), ―The Cultural Construction of Risk Understanding
through Illness Narratives,‖ Journal of Consumer Research, 34(5), 579-594.
Woolcock, Michael J. V. (1999), ―Learning From Failures in Microfinance: What Unsuccessful
Cases Tell Us About How Group-Based Programs Work,‖ American Journal of Economics
and Sociology, 58(1), 17-42.
30
Table 1A.
Definitions of identities and examples for data coding
Identity
Definition
Trustworthy
Lenders can trust the borrower to ―I am responsible at paying my bills and
pay back the money on time
lending me funds would be a good
investment.‖ (Listing #17118)
(i.e., Duarte, Siegel,
and Young 2009)
Successful
(i.e., Shafir,
Simonson, and
Tversky 1993)
Hardworking
(i.e., Woolcock 1999)
Examples
The borrower is someone with a
successful business or job/career
―I have (had) a very solid and successful
career with an Aviation company for the
last 13 years.‖ (Listing #18608)
The borrower will work very
hard to pay the loan back
―I work two jobs. I work too much
really. I work 26 days a month with both
jobs.‖ (Listing #18943)
The borrower is someone of
need because of hardship as a
(i.e., Woolcock 1999) result of difficult circumstances,
bad luck or other misfortunes
that were, or were not, in the
borrower’s control
―Unfortunately, a messy divorce and an
irresponsible ex have left me with awful
credit.‖ (Listing #20525)
The borrower is honest or moral
person
―On paper I appear to be an extremely
poor financial risk. In reality, I am an
honest, decent person.‖ (Listing #17237)
The borrower is a religious
person
―One night, the Lord awaken me and my
spouse… our business has been an
enormous success with G-d on our
side.‖ (Listing #21308)
Economic
Hardship
Moral
(i.e., Aquino et al.
2009)
Religious
(i.e., Weaver and
Agle 2002)
31
Table 1B.
Frequencies and coders’ agreement measures for identity claims, correlations of each identity claim with the total number of identities,
and co-occurrence of each two identities (out of all listings)
Frequency a
Agreement
rate b
Fleiss
Kappa
Interpretation
of Kappa c
Number of
identities
Trustworthy
Trustworthy
61.8%
83%
0.66
Substantial
.57**
1
Successful
19.6%
79%
0.75
Substantial
.38**
.07**
1
Economic Hardship
25.0%
90%
0.80
Substantial
.49**
.03
-.07**
1
Hardworking
25.7%
81%
0.71
Substantial
.56**
.09**
.07*
.14**
1
Moral
15.4%
78%
0.78
Substantial
.51**
.15**
-0.02
.14**
.16**
1
Religious
5.7%
98%
0.96
Almost perfect
.35**
.12**
-0.01
.11**
.05*
.12**
Successful
a. This frequency represents all listings of borrowers that claimed the identity, with or without claiming other identities.
b. Agreement rate between five sets of two coders for the twenty listings that everyone coded.
c. The interpretation of the Kappa size is based on the one suggested by Landis and Koch (1977).
*. ** represent significance levels at .05, and .01 respectively.
Economic
hardship
Hardworking
Moral
32
Table 2.
Loan funding (H2a) and percent reduction in interest rate (H2b) as a function of number of
identities
β
Log of Loan funding
SE
t
p
Log of % reduction in interest rate
β
SE
t
p
Constant
-6.60
0.20
-32.52
0.00
-5.64
0.16
-36.00
0.00
Number of identities
0.11
0.04
2.81
0.01
0.12
0.03
4.04
0.00
Requested loan
amount (thousands)
-0.10
0.01
-16.16
0.00
-0.03
0.01
-6.78
0.00
Initial interest rate
21.72
0.77
28.15
0.00
7.26
0.60
12.19
0.00
AA
4.23
0.18
23.16
0.00
1.59
0.14
11.29
0.00
A
3.65
0.18
20.43
0.00
1.38
0.14
9.98
0.00
B
2.85
0.17
16.85
0.00
1.08
0.13
8.28
0.00
C
1.91
0.16
11.94
0.00
0.63
0.12
5.14
0.00
D
1.30
0.16
8.27
0.00
0.46
0.12
3.81
0.00
E
0.27
0.15
1.78
0.08
0.18
0.12
1.54
0.12
Male
0.51
0.13
4.00
0.00
0.16
0.10
1.63
0.10
Female
0.38
0.14
2.83
0.00
0.13
0.10
1.21
0.23
Married
0.25
0.12
2.07
0.04
0.20
0.09
2.16
0.03
Divorced
-0.19
0.21
-0.92
0.36
0.04
0.16
0.24
0.81
Single
0.49
0.20
2.42
0.02
0.29
0.16
1.85
0.07
Engaged
-0.10
0.24
-0.43
0.67
0.21
0.19
1.13
0.26
Caucasian
-0.11
0.11
-1.04
0.30
0.12
0.08
1.44
0.15
African American
-0.37
0.18
-1.98
0.05
-0.04
0.14
-0.29
0.77
Hispanic
0.03
0.35
0.07
0.94
0.05
0.27
0.20
0.84
Children
-0.06
0.11
-0.60
0.55
-0.04
0.08
-0.53
0.59
Reference categories for the predictors: for gender it is ―no gender‖ (borrowers who did not disclose
their gender), for marital status it is ―no marital status or widow‖ (we have only two widows in our
sample), for race it is ―no race or race other than Caucasian, African American, or Hispanic‖ (there were a
few Asian and Indian in our sample), for credit grade it is HR (high risk).
33
Table 3.
Distribution of loan performance as a function of the number of identities borrowers claim
No
identities
One
identity
Two
identities
Three
identities
Four
identities
Total
Default
12.2%
18.4%
24.2%
26.4%
28.8%
21.3%
Late
8.9%
6.5%
7.4%
6.4%
7.7%
7.1%
Current
40.0%
42.1%
37.2%
41.8%
44.2%
40.5%
Paid
38.9%
33.0%
31.2%
25.5%
19.2%
31.0%
Total
100.0%
100.0%
100.0%
100.0%
100.0%
100.0%
90
261
215
110
52
728
n
34
Table 4.
Loan funding and percent reduction in interest rate as a function of the identities (ERQ1)
Log of Loan funding
Log of % reduction in interest rate
β
SE
t
p
β
SE
t
p
Constant
-6.59
0.20
-32.27
0.00
-5.62
0.16
-35.59
0.00
Trustworthy
0.33
0.09
3.79
0.00
0.24
0.07
3.55
0.00
Successful
0.23
0.11
2.11
0.03
0.27
0.08
3.26
0.00
Economic Hardship
0.00
0.10
0.01
0.99
0.01
0.08
0.14
0.89
Hardworking
0.05
0.10
0.54
0.59
0.07
0.08
0.98
0.33
Moral
0.01
0.12
0.08
0.94
-0.04
0.09
-0.46
0.65
Religious
-0.33
0.18
-1.82
0.07
0.13
0.14
0.91
0.36
Requested loan amount
(thousands)
-0.10
0.01
-16.13
0.00
-0.03
0.01
-6.84
0.00
Initial interest rate
21.48
0.77
27.80
0.00
7.08
0.60
11.85
0.00
AA
4.10
0.19
22.03
0.00
1.50
0.14
10.42
0.00
A
3.54
0.18
19.61
0.00
1.30
0.14
9.29
0.00
B
2.76
0.17
16.22
0.00
1.03
0.13
7.79
0.00
C
1.85
0.16
11.57
0.00
0.60
0.12
4.85
0.00
D
1.28
0.16
8.17
0.00
0.45
0.12
3.69
0.00
E
0.28
0.15
1.84
0.07
0.19
0.12
1.64
0.10
Male
0.48
0.13
3.83
0.00
0.16
0.10
1.63
0.10
Female
0.40
0.14
2.91
0.00
0.16
0.10
1.56
0.12
Married
0.27
0.12
2.25
0.02
0.21
0.09
2.22
0.03
Divorced
-0.18
0.21
-0.83
0.41
0.05
0.17
0.33
0.74
Single
0.53
0.20
2.60
0.01
0.32
0.16
2.02
0.04
Engaged
-0.08
0.24
-0.31
0.75
0.23
0.19
1.21
0.23
Caucasian
-0.10
0.11
-0.89
0.38
0.12
0.08
1.46
0.14
African American
-0.36
0.18
-1.94
0.05
-0.05
0.14
-0.34
0.73
Hispanic
0.01
0.35
0.02
0.98
0.03
0.27
0.10
0.92
Children
-0.05
0.11
-0.42
0.67
-0.03
0.08
-0.37
0.71
Reference categories for the predictors: for identities it is ―no identities‖ (borrowers who did not claim
any of the six identities), for gender it is ―no gender‖, for marital status it is ―no marital status or
widow‖, for race it is ―no race or race other than Caucasian, African American, or Hispanic‖, for credit
grade it is HR.
35
Figure 1.
Estimated marginal means of the number of identities constructed by borrowers as a function of
borrowers’ credit grade, controlling for demographics and loan financial characteristics
1.75
Estimated marginal means of the number of Identities
1.71
1.70
1.68
1.65
1.60
1.55
1.50
1.48
1.45
1.46
1.44
1.45
1.40
1.40
HR
E
D
C
B
A
AA
Credit Grade
Covariates appearing in the model are evaluated at the following values: requested loan amount =
8310.24, initial interest rate = .169081, male = .32, female = .24, married = .23, divorced = .04, single =
.05, engaged = .03, Caucasian = .35, African American = .07, Hispanic = .01, children = .27.
36
Figure 2.
Number of identities predicts loan funding, reduction in interest rate, and loan performance*
Loan funding
Loan performance
Reduction in interest rate
1.4
0.1
1.3
0.08
1.1
1
0.07
0.9
0.06
0.8
0.7
0.05
Percent reduction in interest rate
Loan funding / Loan performance
0.09
1.2
0.6
0.04
0.5
0.4
0.03
No identities
One identity
Two identities
Three identities
Four or more
identities
Number of identities
* For the purposes of plotting this figure, we grouped the four loan performance levels into two
levels (pay/no pay) and present the percentage of listings by borrowers who pay.
37
Web appendix
Table WA-1
Loan performance as a function of the number of identities (H2c). Late is the reference category
Default
Paid
Current
β
0.60
SE
1.33
χ2(1)
0.20
p
0.65
β
3.86
SE
1.23
χ2(1)
9.81
p
0.00
β
3.83
SE
1.29
χ2(1)
8.85
p
0.00
0.29
0.17
2.98
0.08
-0.31
0.16
3.73
0.053
-0.32
0.17
3.67
0.054
0.05
0.04
1.61
0.20
-0.04
0.04
1.43
0.23
-0.08
0.04
5.23
0.02
2.52
4.77
0.28
0.60
-9.50
4.43
4.59
0.03
-12.63
4.66
7.35
0.01
0.68
1.38
0.24
0.62
1.77
1.28
1.90
0.17
3.02
1.31
5.32
0.02
A
-0.60
0.97
0.39
0.53
0.31
0.87
0.13
0.72
0.81
0.91
0.79
0.37
B
-0.20
0.79
0.06
0.80
0.17
0.73
0.05
0.82
0.54
0.77
0.48
0.49
C
-0.36
0.63
0.33
0.57
0.24
0.60
0.16
0.69
0.62
0.64
0.92
0.34
D
0.37
0.61
0.37
0.54
0.70
0.59
1.41
0.24
1.20
0.63
3.59
0.06
E
0.22
0.51
0.18
0.67
0.69
0.50
1.90
0.17
0.66
0.57
1.34
0.25
Male
-0.98
0.57
2.92
0.09
-0.72
0.54
1.74
0.19
-0.38
0.56
0.45
0.50
Female
-1.69
0.57
8.77
0.00
-0.96
0.54
3.17
0.08
-1.38
0.57
5.88
0.02
Married
0.22
0.47
0.22
0.64
-0.37
0.43
0.72
0.40
-0.35
0.46
0.58
0.44
Divorced
-0.27
0.61
0.20
0.66
-0.95
0.59
2.61
0.11
-1.53
0.71
4.64
0.03
Single
0.95
0.72
1.74
0.19
0.34
0.71
0.24
0.63
0.47
0.74
0.40
0.53
Engaged
0.24
0.89
0.07
0.79
0.14
0.82
0.03
0.86
-1.08
0.93
1.33
0.25
Caucasian
-0.03
0.41
0.01
0.94
0.27
0.38
0.50
0.48
0.56
0.40
1.94
0.16
-0.07
0.57
0.02
0.90
-0.94
0.60
2.46
0.12
0.30
0.62
0.23
0.63
-1.98
1.22
2.66
0.10
-1.12
0.87
1.67
0.20
-1.73
1.09
2.54
0.11
0.07
0.38
0.03
0.86
-0.03
0.36
0.01
0.93
-0.35
0.39
0.79
0.37
Intercept
Number of
identities
Requested
loan amount
(thousands)
Initial interest
rate
AA
African
American
Hispanic
Children
Reference categories for the predictors: for gender it is ―no gender‖, for marital status it is ―no marital
status or widow‖, for race it is ―no race or race other than Caucasian, African American, or Hispanic‖, for
credit grade it is HR.
38
Table WA-2. Loan performance as a function of identities (ERQ2). Paid is the reference category
2
Current
Late
Default
2
β
SE
χ (1)
p
β
SE
χ (1)
P
β
SE
χ2(1)
p
Intercept
-3.29
0.91
12.92
0.00
-3.93
1.30
9.15
0.00
0.04
0.68
0.00
0.96
Trustworthy
-0.32
0.28
1.35
0.24
-0.55
0.37
2.16
0.14
-0.50
0.21
5.51
0.02
Successful
-0.43
0.31
1.92
0.17
-0.10
0.43
0.06
0.81
-0.38
0.25
2.41
0.12
Economic
Hardship
0.90
0.30
9.15
0.00
0.69
0.39
3.10
0.08
0.66
0.26
6.54
0.01
Hardworking
0.13
0.28
0.21
0.64
-0.45
0.41
1.16
0.28
0.18
0.23
0.60
0.44
Moral
-0.53
0.35
2.37
0.12
-0.62
0.50
1.54
0.21
0.19
0.27
0.53
0.47
Religious
0.49
0.52
0.88
0.35
0.01
0.76
0.00
0.99
0.43
0.47
0.86
0.35
Requested
loan amount
(thousands)
0.14
0.03
26.75
0.00
0.08
0.04
4.65
0.03
0.05
0.02
4.86
0.03
Initial
interest rate
15.77
3.34
22.34
0.00
12.79
4.70
7.40
0.01
3.83
2.56
2.24
0.13
AA
-2.17
0.77
7.96
0.00
-2.89
1.32
4.79
0.03
-1.00
0.52
3.62
0.06
A
-1.17
0.70
2.80
0.09
-0.60
0.92
0.43
0.51
-0.22
0.50
0.20
0.66
B
-0.65
0.58
1.26
0.26
-0.45
0.78
0.33
0.56
-0.26
0.46
0.31
0.58
C
-0.88
0.49
3.18
0.07
-0.54
0.65
0.70
0.40
-0.26
0.42
0.38
0.54
D
-0.87
0.45
3.82
0.05
-1.23
0.64
3.74
0.05
-0.53
0.40
1.72
0.19
E
-0.53
0.45
1.37
0.24
-0.73
0.58
1.59
0.21
-0.04
0.43
0.01
0.92
Male
-0.68
0.37
3.40
0.07
0.31
0.56
0.30
0.59
-0.39
0.28
1.89
0.17
Female
-0.48
0.41
1.42
0.23
1.29
0.58
4.93
0.03
0.23
0.32
0.54
0.46
Married
0.51
0.35
2.11
0.15
0.29
0.46
0.40
0.53
-0.05
0.27
0.03
0.86
Divorced
0.88
0.65
1.86
0.17
1.16
0.72
2.57
0.11
0.38
0.58
0.42
0.52
Engaged
0.38
0.48
0.61
0.43
-0.53
0.75
0.49
0.49
-0.23
0.43
0.28
0.60
Single
1.43
0.72
3.91
0.05
1.25
0.94
1.76
0.18
1.26
0.60
4.43
0.04
Caucasian
-0.54
0.30
3.10
0.08
-0.51
0.41
1.57
0.21
-0.25
0.24
1.11
0.29
African
American
-0.31
0.51
0.37
0.54
-0.14
0.62
0.05
0.83
-1.28
0.52
6.13
0.01
Hispanic
-0.06
1.37
0.00
0.96
1.97
1.10
3.22
0.07
0.84
0.88
0.91
0.34
Children
0.34
0.30
1.24
0.26
0.30
0.40
0.59
0.44
0.23
0.25
0.81
0.37
Reference categories for the predictors: for identities it is ―no identities‖, for gender it is ―no gender‖,
for marital status it is ―no marital status or widow‖, for race it is ―no race or race other than Caucasian,
African America, or Hispanic‖, for credit grade it is HR.
Percent funded
over 500%
450-499%
400-449%
350-399%
300-349%
250-299%
200-249%
150-199%
100-149%
50-99%
0-49%
0
Number of listings
39
Figure WA-1
Distribution of loan funding percent
500
450
400
350
300
250
200
150
100
50
0
40
Figure WA-2.
Distribution of percent reduction in interest rate (funded listings only)
300
250
Number of listings
200
150
100
50
0-5%
5-10%
10-15%
15-20%
20-25%
25-30%
30-35%
Percent reduction in the final interest rate
35-40%
over 40%