information asymmetry in relationship vs. …
TRANSCRIPT
INFORMATION ASYMMETRY IN RELATIONSHIP VS. TRANSACTIONAL DEBT MARKETS:
EVIDENCE FROM PEER-TO-PEER LENDING
A Dissertation Proposal
Submitted to the Faculty
of
Purdue University
by
Craig R. Everett
This version: September 7, 2010
ABSTRACT
This paper uses a unique new data source, online social lending (a.k.a. peer-to-peer lending),
to help answer the question of what impact borrower-lender information asymmetries have on
adverse selection and moral hazard. This data source has unique characteristics that allow
analysis of the public versus private debt choice without some of the endogeneity issues that
are present when using other data sources. Each loan contains detailed bidding information
from both public and private investors. Thus, a clean distinction can be drawn between
public and private debt without the potential problem of unobserved borrower risk
characteristics. The results support the idea that the hold-up problem is more severe with
private lenders than public lenders, and that personal relationships mitigate the moral hazard
problem.
1
Introduction
There is growing body of evidence in the banking literature that relationship lenders use the
private soft information collected from their borrowers over time (Black 1975; Fama, 1985) to extract
monopoly economic rents in the form of higher interest rates (Rajan, 1992). Borrowers experiencing
this hold-up problem cannot easily switch lenders, since the private information proving that the
borrower is not a lemon is kept confidential by the existing relationship bank. Financial
intermediaries play an important role in mitigating these information asymmetries between borrowers
and investors. By improving selection of borrowers and providing delegated monitoring services,
financial intermediaries address the adverse-selection and moral hazard issues that characterize the
lending relationship. By addressing these issues, lending costs are reduced.
Measurement of the magnitude of the reduction in lending costs resulting from private banking
relationships, as well as measurement of the increases in interest rates due to the hold-up problem, has
proven problematic due to endogeneity problems. For example, loan terms and conditions certainly
affect the loan interest rate, but they are also jointly determined with the interest rate.
This paper uses data from Prosper.com, the largest peer-to-peer lending organization (a.k.a.
online social lending) in the United States. It is essentially an online Dutch auction for loans,
removing the need for an intermediary between borrower and investor. Peer-to-peer lending data
have unique characteristics that shed new light on adverse selection issues such as the hold-up
problem while avoiding some of the common endogeneity problems. This data source also provides
new insight into the resolution of moral hazard through monitoring.
Peer monitoring is a concept adopted from microfinance, which is most prevalent in the
developing world. In most microfinance programs, borrowers are members of a lending group. The
members of the group monitor each other and thus repayment rates remain high, even without explicit
monitoring by a financial intermediary. Groups also help improve selection through private
2
information and also mitigate moral hazard via informal enforcement mechanisms such as shame and
ostracism.
Group membership is available as an optional feature in some of the online peer-to-peer lending
communities. This provides the opportunity to test whether or not the risk mitigating impacts of
membership in a lending group in the developing world have corresponding impacts in online peer-
to-peer lending. If group membership has similar effects in both microfinance and the online world,
then we should find a negative relationship between online group membership and default rate. I find
that this is indeed the case for certain types of groups. Being a member of a group with personal
relationships or a group with a high likelihood for having private information is associated with
substantially reduced default rates. In turn, lower default rates are, by definition, associated with
lower loan costs.
In this paper, I make the assumption that deals between the borrower and fellow members of the
same group are analogous to relationship banking or private debt. Both are characterized by the
potential presence of private soft information. Likewise, deals between the borrower and unaffiliated
investors in the auction are analogous to public debt auctions, where the information asymmetries are
more pronounced.
The choice between public and private debt is fundamentally an endogenous one for firms, and
thus a robust econometric analysis must somehow exogenize this choice. Just as the risk
characteristics of the borrower are a key determinant of the ultimate rate spread on the debt, the debt
terms are not exogenous and can simultaneously impact the risk profile of the borrower. The precise
measurement of the hold-up problem is thus difficult to accomplish. Previous research has either
assumed that the nature of the banking relationship is exogenous (Houston and James, 1996) or has
addressed the endogeneity issue econometrically. The latter has been done with instrumental
variables (Santos and Winton, 2008) or by exogenizing the public/private debt choice via samples that
3
are matched using propensity scoring (Hale and Santos, 2009). Unfortunately, between the samples,
there still may be unobserved risk characteristics that drive the public vs. private choice.
In peer-to-peer lending, on the other hand, both insiders (private) investors, and outsiders (public)
investors actively bid on the loans, so the exercise of finding matching sets of borrowers in order to
exogenize the public vs. private debt choice is unnecessary. Each loan contains detailed bidding
records from both public and private investors. Thus, a clean distinction can be drawn between
public and private debt choice without the potential problem of the presence (between loans) of
unobserved risk characteristics that drive the decision.
This paper contributes to the literature by providing new evidence on the impact of certain
borrower characteristics on two outcomes – default rate and interest rate spread. The borrower
characteristic that I focus on is group membership. I hypothesize that group membership is related to
adverse selection and moral hazard mitigation capabilities and thus its impact on default rate and
spread can be empirically tested. The ability to distinguish between informed and uninformed
pricing within a single loan appears to be unique to this paper, allowing for relatively unambiguous
examination of the effects of private soft information. It thus provides a potentially cleaner
measurement and pricing of information monopolies. The data support the idea that group
membership is negatively and significantly associated with default rate when the group has a potential
for either personal relationships or private information. I also find evidence that the hold-up problem
prevents these cost reductions from being efficiently passed down to the borrower.
The paper proceeds as follows: Section 1 explores the relevant literature and background
information on peer-to-peer lending and the hold-up problem. In Section 2, I provide more detail
about the impact of information asymmetry on loan default and interest rate spread. Section 3
describes the data from Prosper.com. Section 4 describes the methodology for testing the various
hypotheses. Section 5 presents the empirical results, and Section 6 concludes.
4
1. Background and Related Literature
A. Peer-to-peer Lending – A Primer
Since online social lending is still a relatively new phenomenon, it may be helpful to provide a
brief summary of how the online borrowing and lending processes actually function. For the
purposes of this paper, I will describe only the lending processes of Prosper.com, which is my source
of loan data. There are many social lending internet sites, each operated in a slightly different way.
Since Prosper is the largest U.S. peer-to-peer lending site, it has attracted the most media interest and
thus its description provides a relevant introduction to the basics of social lending.
The process begins when a borrower creates a listing that includes the loan amount requested, the
maximum interest rate they are willing to pay, personal data such as income and occupation, and a
detailed description of the purpose of the loan. The borrower is always an individual, even for
business loans. Loan amounts are limited to $25,000 (US), which imposes a type of credit rationing
that can lower overall default rates (Jaffee and Russell, 1976). Simply the threat of future credit
rationing can mitigate moral hazard problems (Stiglitz and Weiss, 1981).
Once the listing is submitted, Prosper.com pulls a credit report for the borrower and appends the
salient credit report information onto the loan listing. The credit report information includes credit
score, delinquency history, number of credit lines, credit line utilization and other pertinent
information. The borrower’s bank account information is also verified. When the listing is approved,
it appears on the website. Lenders, who are simply investors who have signed up on the site and have
also had their bank accounts verified, are then permitted to bid on loans. Lenders can bid as little as
$50 (US) toward any listing and can bid any interest rate less than or equal to the borrower’s
maximum rate.
The process then works like a Dutch auction. The loan stays in “unfunded” status until enough
bids are submitted to add up to the full amount of the loan requested. At that point, additional bids at
lower interest rates can continue to be submitted, crowding out the bids at higher rates. The winning
5
lenders are the collection of lenders who have bid the lowest interest rates and whose cumulative bid
amounts add up to at least the loan request amount. The ultimate interest rate to the borrower is
0.05% less than the lowest rate that was bid among the losing bidders.
Borrowers have the option of joining a “group.” Group leaders can impose additional
information requirements as a condition for joining. Employment verification, for example, is a
common condition imposed by some groups, but is not required by Prosper.com itself. This is the
source of private information that can exist for group lenders. Borrowers often join a group after
they have been unsuccessful in previous individual peer-to-peer loan attempts. Thus, these borrowers
may have particularly severe information asymmetry problems.
For borrowers who have opted to become members of a “group,” the winning list of bidders will
typically consist of both members of their own group (a.k.a. insider, relationship or private investors)
as well as non-members (a.k.a. outsider, arm’s-length, transactional or public investors). This fact
becomes important in drawing the parallels between the social lending market and the traditional
public and private debt markets.
Each loan auction stays open for seven to ten days. Bidders are notified by email whenever they
have been outbid and therefore no longer one of the potential winners. Lenders then have the option
of updating their bid to a lower interest rate. At the end of the ten day period, if the loan receives
enough bids to be funded, the loan is initiated and the funds are withdrawn from the lenders’ bank
accounts, and the loan proceeds are deposited to the borrower’s account. The website charges an
initiation fee to the borrower equal to one percent of the loan principal, as well as a small monthly
servicing fee to the lenders. Prosper.com then services the three-year loan on an ongoing basis, doing
automatic withdrawals from the borrower’s account and proportional deposits to lenders’ accounts
each month.
After the loan is originated, monthly email messages are sent to the borrower a few days before
the loan payment debit is made from his/her bank account. The standard enforcement mechanisms
6
for repayment are the imposition of additional fees in the case of insufficient funds, adverse credit
report postings for past due accounts, and key derogatory credit report information and referral to a
collection agency in the case of default.
If the borrower is a member of a group, then there may be additional informal enforcement
mechanisms. Lending groups are the foundation of microfinance, and the moral hazard mitigation
mechanisms available to a traditional lending group are shared risk, peer monitoring, and peer
enforcement (Brau and Woller, 2004). In online social lending, on the other hand, late payments and
defaults will affect the group’s “star” rating, which can impact the ability of groups to attract new
members, as well as the ability of existing group members to get new loans. This is the only shared
risk, since members are not responsible for the repayment of fellow members’ loans. Additionally,
group members can monitor the status of loans borrowed by other members of their group. If a late
payment is reported, they can exert pressure on the delinquent borrower via email messages. This
constitutes a peer monitoring and enforcement mechanism, albeit a potentially weak one without any
explicit sanctions.
Table 1 lists the pioneers of online social lending by their dates of inception. It is evident that this
is a rapidly developing phenomenon, with the number of new players accelerating over time to satisfy
the growing worldwide demand for unsecured business and personal loans. Returns to lenders appear
to be consistent with conventional lending. Summary statistics (Table 3) imply that return for
investors is between 9% (for loans to “AA” ranked borrowers, which is the highest credit ranking)
and 15.1% (for “D” ranked borrowers) after adjusting for default rates that range between 0.6% and
5.2%, respectively. The overall default rate, when also including “E” rated borrowers and unrated
borrowers (no credit history), is approximately 7.5%. The cumulative default rate of 7.5% is similar
to the default rates of other types of debt. For example, the average default rate on corporate bonds
between 1982 and 2001 is 4.19% (Altman et al., 2005). The default rate for small business loans
(source: SBA) during a similar time frame is approximately 11%. Market interest rates during the
7
sample period (May 31, 2006 to Nov 6, 2007) were reasonably stable until the Summer of 2007,
when mortgage-backed securities began to be downgraded. Figure 1 shows a comparison of interest
rates for Prosper.com (borrowers with credit score of 720 and above), Baa rated corporate bonds, Aaa
rated corporate bonds, federal funds overnight rate, and the 3-month T-bill rate.
Figure 1 – Interest Rate Comparisons over sample period 1
B. A Brief History of Social Lending
The advent of modern social lending is attributed to the English “friendly societies” of the 18th
and 19th century that arose spontaneously during the industrial revolution as clubs that helped their
members pool resources and risk. The friendly societies allowed members to make deposits and
receive loans, and also assisted member families in the case of negative shocks such as illness. With
the Friendly Societies Act of 1793, the British Parliament formally recognized and regulated the
1 Sources of interest rate data are Prosper.com and federalreserve.gov
8
burgeoning industry. Although the records of the era are informal and incomplete, it is thought that
by 1815 as many as 33% of the families in England were members of a friendly society (Gorsky,
1998). An even earlier implementation of a microfinance-style lending paradigm was instituted in
Ireland in the early 18th century, when Jonathan Swift (author of Gulliver’s Travels), with his own
resources, instituted a fund for lending to poor tradesmen. To address the adverse selection problem,
he required each borrower to obtain a personal guarantee from two neighbors. Any late payment
would result in a notice being sent to the borrower and both guarantors (to help ensure future
punctuality). There were apparently no defaults on the Swift loans (Hollis & Sweetman, 2001).
The concept caught on, and Irish microloan funds flourished for more than 100 years until they
eventually fell into decline in the mid 19th century as a result of dual death blows from the potato
famine and from punitive legislation introduced through the lobbying efforts of mainstream banks.
During their heyday, the Irish loan funds are said to have provided services for as much as 20% of the
population of Ireland. The Irish loan funds of the 19th century were quite similar in structure to those
instituted by Jonathan Swift a hundred years earlier, except that Swift did not charge interest. Like
modern microfinance efforts in lesser developed countries (LDCs), the Irish loans were intended for
enterprises that would generate enough cash flow to allow the borrower to make the required
payments and hopefully help lift him out of poverty (Hollis & Sweetman, 2001).
The concept of organized social lending was transplanted from England to the United States in
1831 with the Oxford Provident Building Association in Frankfort, PA (near Philadelphia).
Individuals could purchase up to five shares for five hundred dollars apiece. Whenever the
association accumulated at least five hundred dollars, a loan would be made to the member offering
the highest bid (Potter, 1954). This auction approach is remarkably similar to the structure of online
social lending today.
9
C. Parallels to Microfinance
There is often confusion surrounding some of the common terms in microfinance. The words
microcredit, microfinance and microenterprise loans are sometimes used somewhat interchangeably,
but typically differ from each other in terms of scale. Microcredit generally refers to loans less than
$500 (US), microfinance refers to loans less than $10,000, and microenterprise loans generally refer
to a larger capital expenditure loan of less than $100,000 for a business (such as bakery or small
factory). Since microfinance has traditionally been a social welfare endeavor (often not-for-profit) in
the developing world and has mainly consisted of small loan amounts, it suffers from a lack of
perceived relevance and has therefore been largely ignored by the finance academic community (Brau
& Woller, 2004). Now that the principles of group banking have migrated to the industrialized world
and to cyberspace, they are beginning to encroach upon for-profit financial markets and therefore are
becoming increasingly of interest. In fact, the three largest social lending sites (Prosper, Zopa, and
LendingClub) report that approximately 20% of their loans are for small businesses (Farrell, 2008).
Although online social lending does not exactly fit the definition of microfinance, it shares some
key characteristics. For example, the primary challenge in microfinance is overcoming the
information asymmetry problems inherent within the non-bankable community. Microfinance has
potentially serious adverse selection and moral hazard issues (Hollis and Sweetman, 2001; Williams,
2004). Primarily, the issues are related to avoiding lending to borrowers with unacceptably high
default risk (adverse selection) and then preventing borrowers from using the loan irresponsibly after
receiving the funds (moral hazard). There is an impressive body of literature surrounding
microfinance in the developing world that can be drawn upon to make predictions about other forms
of credit (such as peer-to-peer lending). Many studies have concluded that the stronger the social
connections within group, the higher the repayment rate (Woolcock, 2001; Karlan, 2007). Karlan
also finds that higher geographic proximity of the borrowers in the group reduces default risk. This is
10
consistent with Hung (2006), who demonstrates that group monitoring is more active with greater
geographic proximity.
An excellent example of the applicable theoretical issues in the developing world is illustrated by
Grameen bank in Bangladesh, as well as the concept of Rotating Savings and Credit Associations
(ROSCAs) elsewhere in the world. The Grameen Bank of Bangladesh started as an experiment in
1976 and due, to its success, became an official financial institution in 1983. It is founded on the
proposition that small loans to fund microenterprises will help lift the borrower out of poverty, as
long as the return on assets exceeds the rate of interest on the loan. Since the borrowers lack assets to
pledge as collateral, Grameen implemented a soft system of group liability. Each Grameen group
consists of five members. Unlike ROSCAs, Grameen allows multiple group members to receive
loans simultaneously. Group members are not directly responsible for the repayment of each other’s
loans, but failure to stay current on a loan prevents any other group member from obtaining future
loans (Wahid, 1994). According to information provided by Grameen, as of January 2008, it has had
a total of 7.44 million borrowers to date. There are nearly 2500 Grameen branches, covering 96% of
Bangladesh. Grameen boasts an overall default rate of approximately 2% and has been self-sufficient
(no donors) since 1998. Grameen is one of the most extreme instances of group lending, since their
enforcement model takes legal collection action completely off the table. Grameen purposefully does
not require its borrowers to sign any sort of promissory note for the loan, thereby relying entirely
upon group dynamics for loan enforcement. Similarly, Hung (2006) shows that U.S. microlenders
can reduce staff and transaction costs by transferring the screening, monitoring, and enforcement
activities from the bank to the group members themselves.
Opinion is not universally positive concerning microcredit. While most of the literature appears
to hold it in high esteem as a pathway out of poverty, others see a dark side where unsophisticated
and uneducated people are snared into a cycle of debt (Williams, 2004). The other major concern is
one of efficiency. The reasoning behind this concern is that if the principal purpose of microfinance
11
is to help the poorest and most underprivileged people get a new start, then commercial microcredit
must inevitably fail since fixed transaction costs will drive lenders toward higher loan amounts to
improve profitability. Consequently, since the poorest of the poor will not be suitable risks for those
higher loan amounts, they will no longer qualify as borrowers in for-profit programs (Williams,
2004). Jaffee and Russell (1976) refer to this last effect as a type of credit rationing, which can be
viewed as the lender’s simplest way of mitigating the adverse selection problem.
Most microcredit in the world takes the more informal structure of ROSCAs. These are found on
nearly every continent and are in fact quite similar to the aforementioned Oxford Provident Building
Association of 1831 Philadelphia. In ROSCAs, all the members of the group contribute to a general
fund on a regular basis, and then the proceeds are distributed as loans, one at a time, to each member.
ROSCAs are known by many names in various parts of the world, including “kye” in Korea, “esusu”
in West Africa, and “partner” in Jamaica (Williams, 2004).
In microfinance literature, default risk is mitigated by two principal factors: joint financial
liability and personal relationships. Besley and Coate (1995) model the dynamics of group lending in
a strategic game with joint liability. They demonstrate an equilibrium in which the game value in a
joint liability situation exceeds the sum of the expected values in the corresponding individual
liability scenario. I do not address joint liability in this paper since it does not exist in any of the
major online social lending platforms. The lack of joint liability is actually helpful in this study
because it allows for easier isolation and measurement of the softer social consequences of default. A
primary result of this paper is that, even without joint liability, the screening, monitoring, and
enforcement effects performed by peers in microfinance lending groups can be shown to exist in
online social lending. In order to achieve similar group benefits as we see in microfinance, though,
online groups must be subject to the same possibility of real-life social sanctions.
12
D. Information Monopolies and the Hold-Up Problem
In an ideal world that is fully transparent and without information asymmetries and transaction
costs, borrowers and investors would deal directly with each other. There would be no need for
financial intermediaries such as banks. In the presence of adverse selection and moral hazard
problems, however, it is difficult for an individual investor to effectively determine the true risk
characteristics of the borrower. Likewise, it is unrealistic to expect that individual investors will all
engage in the monitoring activities that would be required to mitigate moral hazard dangers. Even in
public debt markets, in which information asymmetries are reduced because of public disclosure
requirements, these problems still persist. Since public debt is widely held, its relatively small
investors would be inefficiently duplicating effort (thereby multiplying transaction costs) if all of
them monitored the issuing firm. Diamond (1984) proposed that delegated monitoring to a bank may
be an efficient mitigation of moral hazard. In a related paper, Diamond (1991) shows that a higher
interest rate leads to riskier project choice, but only for unmonitored borrowers, since the presence of
monitoring prevents the borrower from selecting the riskier project in the higher interest rate
environment. The downside to the borrower is that during the course of the private banking
relationship, the bank will accumulate proprietary positive information about the borrower (Fama,
1985), particularly if the borrower is also a depositor at the bank (Black, 1975). The fact that other
lenders do not have this information makes it difficult for the borrower to switch banks. This, of
course, is not a problem for the borrower if the lender passes along the cost savings (due to resolution
of adverse selection) to the borrower. The bank can instead decide to keep the proprietary positive
information private and charge a higher interest rate. Thus, the borrower is “held-up” by the bank,
finding it difficult to escape the relationship. Since the bank’s information monopoly can
theoretically either raise or lower the cost to the borrower, it is ultimately an empirical issue as to
whether or not the hold-up problem exists. This is one of the questions addressed by this paper.
13
Sharpe (1990) formally models the informational hold-up problem with bank loans. In his
model, banks are unable to directly observe a firm’s quality, and thus only the bank that is currently
lending to the firm freely observes project success. Banks do not divulge this information since that
would just assist competing banks in bidding away their clients. Banks use the private information to
hold up the borrower, which in turn may distort borrower behavior. For example, the borrower may
attempt to mitigate the hold-up problem by developing multiple banking relationships, which is less
efficient and increases total lending costs (Petersen and Rajan, 1994). For the lender, the greater the
information advantage is, the greater are the rents that can be extracted (Rajan, 1992; Houston and
James, 1996).
E. The Choice between Public and Private Debt
Due to the unique nature of the online social lending data, the analyses undertaken in this paper
shed new insight into relationship versus transactional (a.k.a. arm’s-length) banking. The last twenty
years have seen tremendous growth in online banking. In 1997, less than ten percent of national banks
in the U.S. were capable of online transactions. By the end of Q3 1999, approximately twenty
percent of national banks performed web-based banking transactions, including loan applications.
(Furst, Lang and Nolle, 2000). Today, of course, that number is effectively one-hundred percent.
This phenomenon, along with widespread banking industry consolidation, has contributed to the
decline of local banks, which has been documented in a sizable body of literature regarding
relationship vs. transactional banking.
It has been long held that one of the main strengths of local savings and loan associations is the
aspect of face-to-face relationships. These relationships may reduce interest rates by reducing default
risk and also by providing more leniency in the case of temporary negative shocks to the borrower
(Potter, 1954). Agarwal and Hauswald (2008) find that the probability of obtaining a loan is
negatively and significantly related to the physical distance of the borrower from the bank branch,
14
which is consistent with their proposition that physical proximity is a good indicator of information
quality. This finding supports their hypothesis that personal relationships between borrower and
lender help overcome the information asymmetries inherent in the loan transaction. They also find
that credit delinquency is lower for relationship loans (2.7%) versus transactional loans (3.2%).
Petersen and Rajan (2002) take an opposing view. Their results indicate that the impact of
geographic distance on small business access to credit declined approximately substantially from
1973 to 1993, meaning that a long physical distance between a firm and its bank is less of a barrier to
credit in 1993 as it was in 1973. They interpret this result as confirmation that improved information
and communication technology have made both geographic distance and face-to-face relationships
less important.
Degryse and Ongena (2005) find that transportation costs, not information asymmetry, are the
cause of the decrease in loan rates with geographic distance. They also show that interest rates
decrease with proximity to competing banks, which is consistent with other studies (Petersen and
Rajan (1994, 1995); Boot, 2000; Boot and Thakor, 2000). Karlan (2007) finds that geographic
concentration in groups of borrowers leads to improved repayment rates, consistent with Agarwal and
Hauswald (2008).
The long-horizon personal interaction in relationship banking provides the bank with an insider’s
view of the borrowers’ true financial condition. Therefore, any observable actions or decisions of the
insider with regard to the borrowers’ accounts (such as renewal of a line of credit) can be interpreted
by outsiders as a meaningful signal (Lummer and McConnell, 1989). Leland and Pyle (1977) and
Berger and Udell (1995), in similar lines of reasoning, show that the willingness of an entrepreneur
(the ultimate insider) to invest in his or her own company, is also a signal of true quality. With online
social lending, the only insiders that can be observed are the fellow group members of the borrower.
Since group leaders are permitted to impose additional disclosure and other reporting requirements
upon potential members before joining, it follows that group members may have private information
15
that is unavailable to other investors. Therefore, the willingness of group members to lend money to
fellow group members can be viewed as a signal that reveals the nature of their private information.
Favorable private information should lower the interest rate, as long as lender competition is present
to overcome the hold-up and credit rationing problems. When competition is present, banks cannot
reasonably expect to share in the borrower’s surplus, regardless of the length and strength of the
relationship (Petersen and Rajan (1994, 1995); Berger and Udell, 1995; Boot, 2000; Boot and Thakor,
2000).
With private bank lending, the debt is more concentrated, which gives the lender greater incentive
to actively monitor (Diamond, 1984). Less worthy borrowers thus have greater access to credit, since
moral hazard associated with inferior borrowers can be mitigated via monitoring. Such borrowers
cannot easily switch lenders, since they will be identified as lemons if they do so (Rajan, 1992).
Rajan also predicts that hold-up problems should increase with firm’s risk of failure.
Santos and Winton (2008) test the proposition that hold-up power increases with firm risk. Since
firm risk is higher (for nearly all firms) in a recession, relationship banks should be able to extract
higher rents during economic downturns. They find that loan spreads do indeed rise during
recessions, but less so for firms with public debt access, consistent with hold-up theory.
It must also be considered that bank loans can, in some cases, be cheaper than public debt
because banks have the ability to mitigate adverse selection and moral hazard problems through
private information and monitoring (Houston and James, 1996). Banks, unlike public debt investors,
also have the flexibility to recontract. Since public debt is held diffusely, there is little incentive for
investors to produce private information or engage in monitoring activities. Houston and James find
that as growth opportunities increase, the share of bank debt actually decreases, which is inconsistent
with the theory that large information asymmetries (higher risk) are associated with more reliance on
bank loans (Rajan, 1992). When the firm has multiple bank relationships, however, the percentage of
bank debt is positively related to growth opportunities.
16
A mitigating factor for the hold-up problem is the level of competition between banks. When
there is high competition among banks, the relationship bank cannot extract rents that are as high,
since the outside banks can observe the existence of the relationship and thus, over time, infer costless
private signals about the firm’s credit worthiness (Petersen and Rajan, 1994; Petersen and Rajan,
1995; Berger and Udell, 1995; Boot, 2000; Boot and Thakor, 2000; Dinc, 2000). This drives the
credit market toward equilibrium. Therefore, the informational rents decrease as competition
increases.
So how large are these informational rents? Hale and Santos (2009) attempt to price the hold-up
problem by comparing bank loan pricing before and after borrowers gain access to the public debt
market (bond IPO). The bond IPO reveals new information about the firm’s creditworthiness, thus
reducing the information advantage of the banks. The bond IPO is typically associated with the firm
acquiring its first credit rating, which is a significant information event, since the credit rating
agencies are believed to have access to confidential information not available to the public. Since
there is likely to be endogeneity in the public/private debt choice, they exogenize this choice by
building a matching set (using propensity scoring) of firms that do a bond IPO and firms that do not.
They find that bank loans after the bond IPO have lower spreads (loan interest rate minus LIBOR). A
high credit rating before the IPO is associated with a smaller spread (reduction of 35-50 basis points).
A lower rating, however, is associated with less of a reduction. Among safe firms, not having a
previous credit rating is associated with a larger decline in rates. This supports the hypothesis that the
revelation of firm safety to the market increases outside investors’ willingness to bid, thus driving
down the spread.
Schenone (2010), like Hale and Santos (2009), seeks evidence of hold-up problem by pricing
private loans before and after a large public information event. She does this by examining loan
pricing before and after the firm’s equity IPO. The idea is that much private information becomes
public at the IPO, thus reducing the relationship bank’s information monopoly. The central
17
hypothesis is that rent extraction by the bank will be concentrated before IPO. The author develops
an intensity measure which is the frequency of new loans with the same lender. Before IPO the
relationship between interest rate and intensity is U-shaped. After IPO, the rate consistently
decreases with intensity. In general, she finds that interest rates are higher before firm IPO, as
expected.
The studies cited above reflect alternative approaches to discerning the impact of private
information monopolies on loan spread. The inherent limitation of these approaches is that they
either require an assumption that the choice between private and public debt is exogenous, or must
employ analytical techniques such as instrumental variables or propensity scoring to deal with the
endogeneity. peer-to-peer lending data, on the other hand, naturally exogenizes the private/public
choice, since the borrower never a actually makes the choice – the initiated loan is in fact simply a
collection of smaller loans (both public and private). All loans for group members have bids from
both group members (with private information) and bids from non-members (with only public
information). This allows the study of information effects on loan spread while mitigating some of
the endogeneity problems inherent in previous research.
2. Impact of Information Asymmetry on Loan Default and Interest Rate Spread
A. Adverse Selection
One of the purposes of financial intermediation is to mitigate adverse selection. Lenders are able
to improve loan selection by collecting private information about their borrower customers over time
(Fama, 1985). Using this private information to augment publicly available data, the lender is then
able to select a portfolio of loan customers with a lower average default rate than would be possible
Table 1 around here
18
using public data alone. Since loan defaults are costly to the lender, a lower average default rate for
the portfolio reduces overall lending costs.
In peer-to-peer lending, every group has a leader, typically the person who started the group. The
group leader is the final arbiter of which borrowers will be allowed to join. The group leader has the
option of requiring potential members to submit additional documentation above and beyond that
which is required by Prosper.com. This ability to collect private information in order to improve
selection is tightly analogous to the role played by traditional financial intermediation in mitigating
adverse selection.
In peer-to-peer lending, however, groups are not monolithic. There are wide ranges of purposes,
sizes and membership composition. Some group leaders may be more interested than others in
expending the additional effort required to collect private data. For example, a group named
“Quilters of Central Indiana” may be mostly interested in social networking, whereas a group named
“Debt Consolidators” is more likely to take borrower screening more seriously. Therefore, in order to
accurately capture the effects of group membership on outcomes in this lending market, it will be
necessary to categorize groups in ways that reveal their propensity to engage in screening and
monitoring efforts.
In the presence of competition between relationship lenders (group members) and arm’s-length
lenders (public bidders), the arm’s-length bidders are able to free-ride on the screening activities of
the relationship lenders. Interest rates should fall due to the arm’s-length lenders facing a reduced
threat of adverse selection (Hauswald and Marquez, 2006). An online auction is, by its very nature, a
highly competitive environment. Where competition is present, investors cannot reasonably expect to
share in the borrower’s surplus, regardless of the length and strength of the relationship (Rajan,1992;
Petersen and Rajan (1994, 1995); Berger and Udell, 1995; Boot, 2000; Boot and Thakor, 2000).
Therefore, we can reasonably expect that the lower cost of lending caused by the decreased default
19
risk in a competitive environment will be passed along to the borrower in the form of a lower interest
rate.
B. Moral Hazard
The other main purpose of financial intermediation is the mitigation of moral hazard. Once the
loan is made, the borrower may act irresponsibly with the loan proceeds or take unnecessary risks.
One way that a traditional financial intermediator addresses this issue is through monitoring of the
borrower to ensure that contractually prohibited activities are not taking place. The borrower knows
that the monitoring will happen, and thus is discouraged from engaging in those activities that would
increase the likelihood of default. Typically, the bank has contractual remedies (such as calling the
loan due) that provide teeth to its monitoring.
Like the Grameen Bank in Bangladesh, the online group lending model has no group-level
enforcement teeth to strengthen the effects of monitoring. Instead, the Grameen lending model relies
entirely upon shame and ostracism to pressure group members into compliance (Ghatak, 1999;
Williams, 2004). Effective social punishment relies heavily upon the assumption that there are social
relationships between group members outside of the lending transaction. Whether these relationships
are established before or after the formation of the group should not matter as much as the level of ex-
post importance of these relationships to the borrower. The financial impact of the relationships is
influenced by the presence of competition (or the lack thereof). Relationships should decrease the
cost of credit, since the presence of social connections has been shown to improve repayment rates
(Woolcock, 2001).
Social sanctions are one of the enforcement mechanisms that can be imposed for borrower
default. This, of course, only works if the group members actually care about the opinions of the
others. The financial impact of social sanctions in a group lending environment is addressed by De
Aghion and Morduch (2000) as part of a larger model of loan repayment incentives. In their model,
20
a borrower will repay the loan only if the perceived cost of the shame (social sanctions) is greater
than or equal to the value of the debt obligation (principal plus interest) minus the discounted project
return. They interpret this to mean that with the additional social sanctions present, lenders are able
to charge higher interest without increasing risk of default.
An alternative interpretation of the work of De Aghion and Morduch (2000) is that the presence
of social sanctions should have a negative impact on interest rates, ceteris paribus. If default risk is
exogenously reduced, then this should result in a decrease in the interest rate that the lender requires.
However, default risk is itself endogenously affected by the interest rate. Stiglitz and Weiss (1981)
demonstrate that when facing a given interest rate, if a risk-neutral borrower is indifferent between
two projects and then the interest rate is increased, the borrower will prefer the riskier project. In a
similar line of reasoning, Diamond (1991) shows that a higher interest rate leads to riskier project
choice, but only for unmonitored borrowers, since the presence of monitoring impedes the borrower
from selecting the riskier project in the higher interest rate environment.
C. Are Lending Cost Savings Passed Along to the Borrower?
The above discussion of adverse selection and moral hazard establish the basis for the first
testable hypothesis of this paper.
Hypothesis 1: Membership in a group with private information or monitoring is
associated with lower default rates.
In peer-to-peer lending, like any form of financing, it should be apparent that interest rate spread
and risk are positively associated since investors expect to be compensated for risk. In an efficient
lending market, competition should result in interest rates that are exactly equal to the lenders’
21
marginal lending costs. So if lending costs are reduced through enhanced screening and monitoring
activities, then this reduction should be reflected in lower interest rates to the borrower.
Hypothesis 2: Lower default rates are associated with lower interest rate
spreads
Alternatively, it may be the case that the lending market is not perfectly efficient, but subject to
frictions such as the hold-up problem. If this is the case, then the cost savings from improved
selection will not be passed along to the borrower, but will instead be extracted by the lender as
surplus economic rents. If the hold-up problem exists, then spread should be positively associated
with being a member of a group with an established propensity for rent extraction, since these lenders
are more likely to seek monopoly rents. If there is no association, or if the relationship is negative,
then that would be evidence against the existence of a hold-up problem.
Hypothesis 3: Group leaders with a measurable propensity to capture rents are
more likely to hold-up their borrowers and not pass along cost savings.
If the hold-up problem is indeed based on an information monopoly, then it stands to reason that
as time passes and private information leaks to the public, then monopoly rents should decrease. Hale
and Santos (2009) found evidence supporting this when comparing bank loan interest rates before and
after the bond IPO for the borrowing firm, since information about creditworthiness is released during
the first bond issue. Schenone (2010) found similar results before and after the equity IPO.
Assuming that there is information about creditworthiness released by the insider peer-to-peer lenders
as part of the process of funding the first loan, we should see a higher difference between insider and
outsider bids for the first peer-to-peer loan than for subsequent loans.
22
Hypothesis 4: The difference between group and non-group bid spreads is lower
on second loan due to a decreasing information advantage over time.
3. Data
Data are obtained from Prosper.com, which represents the largest online social lending
community in the United States. Prosper.com also has the unique feature of allowing borrowers to
join “groups” that are designed to provide risk filtering as well as peer-pressure for borrowers to
perform on the repayment of their loans. The data include detail for 13,465 loans initiated between
May 31, 2006 and November 6, 2007, borrower information including credit score (but excluding
identity), lender information, loan performance information (through May 7, 2008), group
information, and detail regarding every winning bid made for each loan listing. It is important to
recognize that these loans were the result of successful loan listings. The majority of listings (more
than 90%) never receive enough bids to become funded, and thus never become actual loans.
Although the fact that I am only including funded loans may introduce some selection bias, there
would have been significant problems with analysis of bidding behavior on the unsuccessful loan
listings due to having far fewer bid observations with higher variability and more outliers. In
successful loans, these outlier bids are self-censored as they are crowded out by competition.
The loan principal amounts range from $1,000 to $25,000 US. Total funds borrowed during this
period were $86 million (US). The historical fed funds overnight rates, which are used to calculate
the loan spread, were downloaded from the Federal Reserve website. Estimated 2007 population data
for the geographic dispersion measure was obtained from the U.S. Census Bureau.
23
There are more than 1,300 groups in the sample, which I categorized by reading each group’s
narrative self-description and then manually coding each group into the following categories: friend,
small business, company, alumni, geographic, hobby, ethnic, occupation, religious, low-risk
borrowers, high-risk borrowers, lending purpose, growth purpose, and other. Business groups (360
observations) specialize in making loans to small businesses or self-employed entrepreneurs.
Occupation groups (513 observations) consist of people that share the same profession. Depending
on the size of the profession, members may be impacted by their reputation within that profession and
thus wish to avoid default if possible. Company groups (45 observations) consist of employees of the
same company. Geographic groups (99 observations) are based in a certain geographic area (the
same city, in most cases). Alumni groups (219 observations) consist of alumni of a certain university,
who generally must prove their status as an alumnus in order to join.
Williams (2004) asserts that shame and ostracism can be a substitute for legal enforcement in the
event of default, and these effects are enhanced by geographic proximity. Karlan (2007) also finds
that geographic concentration in groups of borrowers leads to improved repayment rates, which is
consistent with Agarwal and Hauswald (2008), who also find that geographic proximity is a good
proxy for information quality. Karlan implements a measure called “geographic dispersion” which
reveals the extent to which group members are clustered by neighborhood. In this paper, I implement
a geographic dispersion measure similar to this one to determine how tightly the group members are
clustered. Since the raw data only include member locations by state, to implement Karlan’s
geographic dispersion measure I substitute states for neighborhoods. Hence, in the formula below, ��
represents the percentage of the group from a particular state, and �� is the percentage of the
aggregate total population across those represented states, that lives in each state.
Table 2 around here
24
������ � ��������
�����
Close inspection of this formula reveals that it is actually a measure of the extent to which the
geographic distribution of the group differs from the geographic distribution of the overall population
covering the same combined geographic areas. So a high value of GDSTATE indicates that some
geographic areas covered by the group are overrepresented. The pure GDSTATE calculation fails
when all group members live in the same state, since then there is no deviation from the population
distribution. GDSTATE equals zero in this case, even though the group is in fact highly
concentrated. Therefore, when all group members live in the same state, I assume perfect
concentration and substitute a value of one (1) in place of the generated value of zero.
Since shame and ostracism (or any other social sanction) can only exist in the presence of
relationships that are important to the borrower, I develop a proxy for personal connections by
reading each group’s narrative self-description and then manually coding each group into a category.
Group categories were: friend, small business, company, alumni, geographic, hobby, ethnic,
occupation, religious, low-risk borrowers, high-risk borrowers, lending purpose, and other. Personal
relationships are proxied by combining the groups: friend, company, geographic, occupation and
alumni into a single dummy variable called GRP_RELATIONSHIPS. These are the groups in which
the borrower is most likely to have direct or indirect personal relationships. Private “friend” groups
(167 observations) are the groups with descriptions such as “Family and friends of John Doe - by
invitation only.”
One of the important characteristics of this online social lending data set is that the loans have
aspects of both relationship banking as well as capital market transactions. On the one hand, like
auction-based capital markets, Prosper.com does not accept deposits – it just directly matches
borrowers with investors. On the other hand, the group membership aspect with its information
asymmetries induces behavior much like relationship banking. Each loan to a borrower that is a
25
group member has the very useful property of having both relationship lenders (investors belonging
to the same group), and arm’s-length lenders (other investors who do not belong to that group).
Because the data consists of loans that have both relationship and transactional lenders that can be
distinguished from each other within the same loan, the effects of the relationships can be measured
very cleanly with less noise and less potential for bias or endogeneity.
4. Methodology
Before being able to answer the question of what lenders do with loan cost reductions, it is
necessary to verify that membership in a group with enhanced selection or monitoring is truly
associated with lower default rates. Toward that end, I identify proxies for enhanced private
information and enhanced monitoring. During the time frame of this data set, group leaders have the
option of establishing what are called “group leader rewards.” This is an additional percentage
(ranging from zero to one percent) that would be added to the borrower’s interest rate and paid
monthly to the group leader. Therefore, I identify the groups that have an observable profit motive
as those that have their group leader rewards set at a rate greater than zero. For these groups I set the
FORPROFIT indicator to one (1). Lenders will collect private information over time (Fama, 1985)
and make lending decisions based on that information.
FORPROFIT groups have better loan selection due to private information and therefore
should have lower default rates (hence lower costs). This leads to a simple test for the
presence of a hold-up problem. If the hold-up problem exists, then spread should be
positively associated with the FORPROFIT indicator, since these lenders are more likely to
seek monopoly rents. If there is no association, or if the relationship is negative, then that
would be evidence against the existence of a hold-up problem. In this study, the association
between FORPROFIT and spread is univariate tested using t-test mean comparisons. For the
multivariate analyses, the association of private information and other loan characteristics
26
with default is tested using probit analyses. The relationship of private information with
interest rate spread is tested using OLS and tobit analyses.
Of course, as discussed previously, it is likely that there is an endogeneity problem with the
traditional choice between group and non-group loans. Fortunately, this data set has detailed bidding
data on each loan which allows the researcher to distinguish between bids from group members and
bids from outsiders. Therefore, it is possible to measure the hold-up problem without the endogeneity
problem by comparing group bids to non-group bids for the same loan.
Agarwal and Hauswald (2008) attempt to isolate the effects of private information when
comparing online eloans (transactional) to local branch loans (relationship). They do this by
orthogonalizing the internal and external (public) credit scores in order to extract the purely private
component of the overall credit screen. For the local branch loans, the internal credit score will be
impacted by proprietary information obtained by the bank over the course of their business
relationship with the borrower. They captured the proprietary information in the residual (µi) in the
following regression:
��������� �� � �� �!�� � 1�#$�% &� � &� �!��� � '�
The residuals ('�) are stored a variable called the Private Information Residual (PIR) and used as
an independent variable in subsequent regressions. XSBI is the firm’s credit score.
I employ a similar approach to extract the purely private information available to online lending
group members. The underlying assumption is that bidding behavior is based on all available
information. Public information is available to all bidders, but additional private information is only
available to insiders. Therefore, differences in insider vs. outsider bidding behavior must be driven
by differences in information. As a proxy for the internal credit score, I use the de-meaned ratio of
insider bids to outside bids on each loan. This is a reasonable proxy, since a credit score itself is
27
simply a measure of the lender’s positive opinion of the borrower. In the same sense, the ratio of
insider bids to total bids measures the insider’s relative positive opinion of the borrower as compared
to outsider opinion. The intuition is that if the private information is adverse in nature, then the
insiders will be less likely to bid (or bid a higher rate that could be crowded out by lower outsider bid
rates) resulting in a lower ratio. If the private information is positive, then insiders will bid more
aggressively, resulting in a higher ratio. Non-group loans, of course, have a ratio of zero. Thus, I
calculated the residuals (PIR1) as follows:
()�*�!*+!,���-.�
�� � ��/��+*������� � ��/��0 1)���
� 1%$%23$45#$�% &� � &�/��+*������� � &�/��0 1)���� � '�
The dependent variable RATIOBIDBYGROUP is noisy, since bidding behavior may be impacted
by factors other than soft private information about credit worthiness. When the dependent variable
is noisy, the residuals will be biased toward zero. Therefore, a significant result for PIR as an
independent variable is all the more convincing. Of course, favorable proprietary info may lead to
higher interest rates, due to information capture (Houston and James, 1987; Von Thadden, 2004).
However, in the presence of competition between relationship lenders and arm’s-length lenders,
which is certainly the case here, interest rates should fall due to the arm’s-length lenders facing a
reduced threat of adverse selection (Hauswald and Marquez, 2006).
To measure the extent to which the credit marketplace incorporates these risk factors into their
bids, I estimate the effect of these factors on the borrower interest rate by performing a tobit analysis
using interest rate spread as the dependent variable.
28
In this model, X is a vector of independent variables representing borrower characteristics, W
represents the loan characteristics and Z represents the group/investor characteristics.
The hold-up problem is about rent extraction by relationship lenders from borrowers that cannot
easily escape said relationship. Therefore, if FORPROFIT groups are excluded from the analysis,
there should be no difference in spread for relationship versus non-relationship groups.
Monitoring will only affect borrower behavior if there is some negative consequence associated
with bad behavior. Shame and ostracism only exist in the presence of relationships that are important
to the borrower, which I proxy by combining the groups: friend, company, geographic, occupation
and alumni into a single dummy variable called GRP_RELATIONSHIPS. Bertrand et al. (2004)
contend that people that share a common educational background may have personal connections that
influence financial outcomes. This is why I include alumni group membership among the proxies for
personal relationships.
The univariate effects of monitoring on default and spread are measured using t-tests of
differences in population means. Multivariate effects on default are tested using probit analysis.
Finally, the multivariate effects of monitoring on interest rate spreads are tested using OLS and tobit
analyses.
5. Results
A. Adverse Selection
Table 3 and Table 4 contain descriptive statistics from the sample of 13,465 loans. The summary
data are divided into three groups of three columns. The first three columns represent data from all
loans, regardless of group membership. The second set of three columns is restricted to only those
�� 6 � � � � &7� � 89� � :�
29
loans to borrowers who are members of a group (any group). The third set of columns represents
borrowers who are not members of any group. Table 3 reports interest rate spread means and Table
4 reports default rate means. The last column is the difference in mean between borrowers who are
group members and those who are not. On average, members of groups pay higher rates than those
who are not. However, it is also apparent that group members have higher default rates. Recall that
being a member of a group corresponds to the borrower choosing a mix of public and private funding
for their debt, whereas not being a member of a group represents the choice of purely public debt.
Although it may seem counterintuitive at first that group members pay higher rates – it begs the
question of why someone would join a group if the result is merely a more expensive loan. It is
important to note that only 7.9% of loan listings during the sample period received enough bids to
become funded. Figure 2 shows failed listings versus successful listings, broken out by credit score.
Figure 2 – Listing Success by Credit Score Category
Many borrowers must make multiple attempts before successfully initiating a loan. Successful
borrowers have an average of 2.7 unsuccessful listing attempts in their history. Joining a group may
be viewed by borrowers as a way to increase the chances of funding, even if the interest rate is not as
favorable. Figure 3 illustrates this concept by showing that group members are more than twice as
likely to have their loan funded as non-group members.
30
Figure 3 – Listing Success by Group Membership
This explanation is consistent with Petersen and Rajan (1994), who find that deeper lending
relationships do not necessarily lower the interest rate, but nevertheless have a positive impact on the
availability of credit. The fact that the inclusion of private investors seems to result in higher rates is
consistent with the hold-up problem.
Examining Table 4 in more detail, when the default rates are broken out by credit risk, it is
apparent that the higher overall default rates for group members are being driven by the extremely
risky borrowers, who may be joining groups simply to gain greater access to credit (Rajan, 1992). If
the highest risk borrowers are excluded, then there is generally no statistically significant difference
in default mean between members of groups and non-members.
Table 3 around here
31
The results in Table 5 are similar to Table 4, in that the differences between the default rates of
relationship versus non-relationship groups are negative and highly significant on average for the
entire sample. However, when broken out by credit risk, the differences in mean are generally
insignificant or of low significance, except for very high risk borrowers.
The descriptive statistics in Table 6 add to the evidence of the existence of a hold-up problem in
the peer-to-peer lending market. Like Table 3, Table 6 distinguishes between groups with and
without profit motive, but only in terms of default rates. Borrowers in FORPROFIT groups default
an average of 8.1% of the time, whereas borrowers not in FORPROFIT groups have a mean default
rate 12.6%. This difference in mean is significant at the 1% level, so the hypothesis that loans in
FORPROFIT groups default less than non-profit groups cannot be rejected. So even though the loans
are lower risk, the FORPROFIT groups do not pass those savings along to the borrower, as evidence
by the results in Table 3. In fact, those lower risk loans are charged a premium. This is consistent
with the more severe hold-up problem described by Von Thadden (2004).
Table 7 reports statistics that address the question of whether groups without enhanced selection
or monitoring have higher default rates. The lack of propensity to monitor is proxied by a dummy
Table 6 around here
Table 5 around here
Table 4 around here
32
variable that is set to one (1) if the group is neither FORPROFIT nor a relationship group. Non-
monitored groups have an average default rate of 13.8%, compared to a default rate of 8.1% for the
monitored groups. The difference in mean is significant at the 1% level and is consistent with the
idea that monitoring reduces default.
Table 8 reports descriptive statistics for interest rate spread for only those loans made to
borrowers who are group members. This time, however, the SPREAD means and medians are broken
out by whether each bid came from an group (private) versus non-group (public). The means and
medians reported represent the weighted average interest rate spreads for the loan. The weighting is
done by bid dollar amount. Pairwise t-tests of differences are performed to test the significance of the
difference between insider and outsider bids. Although the insider bids are at a fairly consistently
higher spread than the outsider bids, the difference is only about ten basis points, which is not very
economically significant. This may be due to the heterogeneity of the raison d’etre of each group.
The lowest positive difference in spread is for the very lowest risk borrowers, which is consistent with
Rajan (1992) and Hale and Santos (2009), but not very meaningful due to low economic significance.
As explained above, the groups have a very wide range of purposes. Some are largely non-
commercial, while others are clearly commercial in their focus. If there is a hold-up problem and
some groups are extracting monopoly rents, we would expect that to occur within the groups with a
commercial focus rather than a hobby group, alumni group, or a friends & family group, for example.
In this data set, group leaders have the option of setting up group leader rewards (GLREWARDS) of
Table 8 around here
Table 7 around here
33
between zero and one percent. This means that the group leader will receive that portion of the
borrowers’ interest. It is reasonable to assume that groups with a non-commercial or social
networking focus will be more likely to have a group leader reward of zero. Therefore, as a proxy
indicator of a for-profit group, I code a value of one (1) for the FORPROFIT indicator for any group
that has a non-zero group leader reward.
Table 9 reports descriptive statistics for only loans to borrowers who are group members, broken
down into for-profit and non-profit groups. If there is a hold-up problem, then the non-profit groups
should exhibit a lower spread than the for-profit groups. It can be seen in the far right column that the
groups with an explicit profit motive do indeed have higher mean interest rates, consistent with the
hold-up problem.
Table 10 takes this a step further, focusing only on groups that have a profit motive and
comparing the weighted average interest rate bids between insiders and outsiders. If there is a hold-
up problem, then insiders should be bidding higher rates on average. This is indeed the case, as
shown in the column to the far right. The difference between insiders and outsiders seems to be
greatest for high loan amounts and for borrowers with good, but not great, credit scores, consistent
with Rajan (1992).
Table 11 provides descriptive bidding statistics for groups that are excluded from Table 10
because they have no explicit profit motive. We would expect to see a lower difference in this table if
in fact the profit motive groups were extracting monopoly rents. Overall, there is virtually no
Table 10 around here
Table 9 around here
34
difference (only 8 basis points). However, if we look specifically at the risk group that had the
greatest drop between insider and outsider bid rates in Table 10 (credit score ranging from 640-679)
and hence the largest potential hold-up problem, we find in Table 11 that the reduction is fourteen
basis points less. This is consistent with the idea that insiders are extracting monopoly rents from
higher risk borrowers (Hale and Santos, 2009).
Table 12 reports the first set of multivariate probit tests performed in order to test the degree to
which the explanatory risk characteristics discussed in the univariate section are actually associated
with loan defaults. This is relevant, since rational loan pricing should, in theory, depend upon the risk
profile of the loan and the borrower. The dependent variable is DEFAULT, which is a binary
variable set to one if the loan is in default status or more than 3 months past due. The coefficients
reported are marginal effects. The relevant group characteristics are FORPROFIT, a dummy
variable that indicates that the group leader has an explicit profit motive, LNGRPLOANS, a
continuous variable that is a proxy for group activity level, GDSTATE, an index between zero and
one that represents the geographic concentration of group members, and RATIOBIDBYGROUP, a
proxy for insider confidence which is the percentage of total successful bids that came from fellow
group members.
In column (7), the coefficient on the variable RATIOBIDBYGROUP is negative, very large, and
very significant. RATIOBIDBYGROUP represents the proportion of insider bids to total bids on the
loan. FORPROFIT and GDSTATE have positive coefficients, but are either not significant or barely
so. It is reasonable to expect that attributes of the group should not have an impact on borrower
performance. RATIOBIDBYGROUP, on the other hand, is negative, economically significant and
Table 11 around here
35
statistically significant at the 1% level. If insiders, who are members of the same group as the
borrower, have positive soft information about the borrower, then it is reasonable to expect that they
will bid on the loan in higher amounts. If the insiders are correct about this positive information, then
these loans should exhibit lower default rates. The result in column (7), for example, supports this
conclusion.
Table 13 reports OLS regressions with loan spread as the dependent variable and the same
explanatory variables as Table 12. Since the relationship between risk and return is expected to be
positive, it is reasonable to expect the coefficients in Table 12 and Table 13 to generally have the
same direction and significance. Therefore, it follows that the most meaningful results are the
instances where this is not the case. LNBORROWERAGE, for example remained significant but
switched signs. Although older borrowers appear to be less risky, they actually receive higher rates
when controlling for loan, group and borrower characteristics. This age effect appears to be
inconsistent with extant literature on borrower risk characteristics. FORPROFIT kept the same sign
(+) but became highly significant (1% level) in Table 13. So although the profit motive of the group
leader was not a very significant determinant of loan default, it has a significant and positive effect on
interest rates, which is consistent with the existence of a hold-up problem (Hypothesis 3).
If the hold-up problem exists and insider lenders are indeed extracting monopoly rents, then
another empirical effects should be apparent. Rajan (1992) predicts that firms with a higher risk of
default should be more susceptible to the hold-up problem. This is because for firms with a high
probability of success (low risk), the relationship bank’s information advantage is small. This
prediction has been confirmed empirically by Hale and Santos (2009). The predicted loan default
Table 12 around here
36
probability (DEFPROB), which is generated using the significant independent variables of Table 12,
is positive and highly significant in all appearances in Table 13. This is consistent with Hypothesis 2
– the fundamental idea that risk and return are positively related.
Another principal finding from Table 13 is that the presence of relationship lenders vs. arm’s-
length lenders has a significant impact on the interest rate. In column (5), the coefficient on the
variable RATIOBIDBYGROUP is negative, very large, and very significant. RATIOBIDBYGROUP
represents the proportion of insider versus outsider bids on the loan. If insiders, who are members of
the same group as the borrower, have positive soft information about the borrower, then it is
reasonable to expect that they will bid on the loan in higher numbers and with more favorable rates
than if they were less interested, which must necessarily drive down the ultimate rate charged to the
borrower. The result in column (5) supports this conclusion. Table 13 also supports the presence of a
hold-up problem by showing that after controlling for borrower characteristics, there is not a
significant relationship between group membership and interest rate spread, even though group loans
should have better selection and monitoring.
Table 14 reports results that address a key research question of this paper. SPREAD_DIFF is the
dependent variable, which is the weighted average of insider (private) interest rate bids minus the
weighted average of outsider (public) bids on each individual loan record. Since public and private
data reside on the same loan record, all borrower, group, loan and time characteristics are fixed.
There is no need to exogenize the public/private debt choice by doing any econometric exercises such
as instrumental variables, propensity score matching, or even fixed effects. In the presence of a hold-
up problem, group characteristics should be significantly associated with SPREAD_DIFF. Indeed,
Table 13 around here
37
this is the case. In fact, the only explanatory variables with significant coefficients are group-related.
FORPROFIT is positively associated with higher insider rates at the 5% level. GDSTATE, which
measures the geographic concentration of the insiders, is positively associated with insider rates and
is significant at the 1% level. Column (7) is restricted to only non-profit groups in order to test the
idea that relationships should not be associated with differences in spread for these groups.
Consistent with this, the coefficient on GRP_RELATIONSHIPS is neither statistically nor
economically significant.
Table 15 summarizes the results of the analyses testing the effects of the soft private group
information on interest rates. PIR1 is the variable that represents the orthogonalized residuals of the
regression of outside credit score onto the ratio of insider bids to total bids, which is the proxy for
inside credit score. Due to the noisiness of the dependent variable, the residuals will be biased toward
zero. Therefore a significant result when using PIR1 as an exogenous variable is particularly
believable. It is clearly demonstrated in every analysis in Table 15 that private soft information, as
captured by PIR1, is strongly significant and has a negative relationship with interest rates, consistent
with Hauswald and Marquez (2006), and not consistent with Von Thadden’s (2004) finding that
favorable info will lead to higher rates.
It could be argued that the significant results for the PIR1 residual are due to omitted variables in
the orthogonalization regression. To address this, I calculated a new residual variable called PIR2,
which, in addition to credit score, also includes all other variables that were previously shown to have
significant effects on interest rate spreads (in Table 13), as well as their interactions. The results of
the inclusion of PIR2 as an independent variable are shown in column 6 of Panel B of Table 15.
Private positive information continues to be consistent with lower interest rates.
Table 14 around here
38
B. Moral Hazard
If the lender is extracting rents via information monopoly, then a public lending event such as a
bond IPO in the public debt markets or, in the case of peer-to-peer lending, a successful loan auction,
should result in some degree of private information dissemination. Therefore, a decline in the
difference between insider and outsider interest rate bids should be observed for subsequent loans as
the information monopoly is abated. Table 16 provides descriptive bidding statistics for insiders and
outsiders for their first loan, then subsequent loans. A narrowing of the spread is indeed observed.
Overall, it is only 7 basis points, but for the previously identified borrowers with good yet sub-
optimal credit scores, the rates narrow by 56 basis points, consistent with Hale and Santos (2009).
When a borrower is a member of a group that has personal relationships, loan default rate is
significantly reduced. This result is consistent with Abbink, Irlenbusch and Renner (2006), who find
a weakly significant result that private social contact (but not professional contact) has a positive
impact on the decision to repay. Cassar, Crowley and Wydick (2007), however, performed a series of
experimental microlending games in South Africa and Armenia and found that mere social contact
was not enough, but that reduction of default risk occurred only when those relationships included
personal trust. Both of these results are consistent with the results of this paper.
Column (2) of Table 12 tests Hypothesis 1 by adding the GRP_RELATIONSHIPS dummy to the
probit specification. The coefficient is negative and significant at the 1% level, consistent with the
Table 16 around here
Table 15 around here
39
idea that relationships decrease default risk due to increased monitoring (Diamond, 1991; De Aghion
and Murdock, 2000; Williams, 2004).
Another factor that has been shown in previous studies to have a significant effect on loan
repayment rates is geographic proximity. Karlan (2007) finds that geographic concentration in
groups of borrowers leads to improved repayment rates, which is consistent with Agarwal and
Hauswald (2008). In columns (5-7) of Table 12, the specification is modified in order to test whether
geographic proximity is negatively related to default rate. The measure for this is GDSTATE, which
is a proxy for the geographic concentration of the group. The coefficient for GDSTATE is negative,
consistent with Karlan (2007) and Agarwal and Hauswald (2008), but is not statistically significant.
GDSTATE, however, is significantly associated with a higher spreads between insider and outsider
bids in Table 13. The negative relationships between geographic concentration and both default and
interest rate is consistent with economic theory and previous empirical work. Notwithstanding, the
fact that I replaced “neighborhood” with “state” in the construction of the GDSTATE measure is
conceptually troubling, since a state would seem to be too geographically large to represent a good
proxy for borrower concentration. For example, there are many borrowers in this data set from
California, a state larger than most countries in the world. The GDSTATE measure treats all
Californians as if they are within the same community, which is clearly not the case.
Column (7) of Table 12 tests the hypothesis that group leader rewards incentivize the group
leader to monitor the loans more carefully, and thus should be associated with lower default rates.
The coefficient on GLREWARDS is indeed negative, consistent with the hypothesis, but statistically
insignificant, allowing the hypothesis to be rejected.
In Table 13, I report the results from tobit analyses using the same independent variables as Table
12, but with the borrower interest rate as the dependent variable, censured at zero. If lenders are
informed and rational, then the results should closely mirror the results in Table 12, since the lenders
should be considering the default risk effects from each of these variables when setting their
40
minimum interest rate bid. However, as previously discussed, this data set contains loans that have
both insiders and outsiders. Not surprisingly, consistent with Table 12, MEMBEROFGROUP has a
positive effect on interest rate, whereas GRP_RELATIONSHIPS has a negative impact, both at the
1% significance level. Competition is proxied by LNBIDCOUNT, which also is negative and
significant. These results support the idea that in the presence of competition, relationships will lower
the interest rate to the borrower.
Table 17 reports the results of the OLS regressions with the dependent variable defined as the
difference in insider versus outsider spreads between the first peer-to-peer loan and subsequent loans.
There is no evidence that information released during the first auction or loan monitoring process has
any significant impact on the hold-up problem. This result is inconsistent with both Hale and Santos
(2009) and Schenone (2010), putting Hypothesis 4 in question. It is difficult, however, to draw a firm
conclusion from Table 17, however, since there were so few second loans in the data set. The results,
therefore, may be driven by data limitations.
6. Conclusion
Relationship banking uses soft information gathered via proprietary means to overcome
information asymmetries relating to borrower credit-worthiness. Loans granted by relationship
lenders with accurate soft information have a lower risk of default and a lower cost. In the presence
of open lender competition, these savings should be passed along to the borrower in the form of lower
interest rates. But if lenders are able to maintain that soft information as proprietary, then they have
the ability to extract monopoly rents from the borrower. This is the hold-up problem. The choice
Table 17 around here
41
between public and private debt trades off the high transaction costs of public debt versus the high
hold-up costs of private debt.
This paper uses peer-to-peer lending data to illustrate the effects of group composition on adverse
selection and moral hazards problems in lending. This particular data set is effective for doing so
because it uniquely contains both public and private investor bidding data on the same loan record.
Thus, the typical econometric approaches for attempting to exogenize the public/private debt choice
are rendered unnecessary.
Through a series of univariate and multivariate analyses, I find that membership in a group with
private information or enhanced monitoring is indeed associated with lower default rates. In general,
lower default rates imply lower lending costs, and these are generally associated with lower interest
rate spreads in the data. The degree to which these cost reductions are passed down to the borrower
in the form of lower interest rates depends entirely upon the nature of the group to which the
borrower belongs. I find that private profit-centered investors do, in fact, “hold-up” their borrowers,
extracting higher economic rents. This is especially the case for borrowers with imperfect credit
ratings, which is consistent with extant literature. Investors that have social relationships with
borrowers, on the other hand, do not hold-up their borrowers but instead pass along the cost savings
associated with better monitoring and private information.
In recent years, the internet social lending community has grown from a small innovation to a
growing industry that now encompasses not only small personal loans, but has also begun to make
inroads into the small business loan markets. peer-to-peer lending represents a disruptive technology
to traditional financial intermediation by putting borrowers and investors together directly while
potentially mitigating moral hazard through informal peer monitoring, which presumably has lower
costs. As such, it is a phenomenon that merits continued attention.
42
REFERENCES
Abbink, Klaus, Bernd Irlenbusch, and Elke Renner, 2006, Group size and social ties in microfinance
institutions, Economic Inquiry 44, 614-628.
Agarwal, Sumit, and Robert Hauswald, 2008, The choice between arm's-length and relationship debt:
Evidence from eloans. Unpublished working paper.
Altman, Edward. I., Brooks Brady, Andrea Resti, and Andrea Sironi, 2005, The link between default and
recovery rates: Theory, empirical evidence, and implications, The Journal of Business 78, 2203-
2228.
Berger, Allen N., and Gregory F. Udell, 1995, Relationship lending and lines of credit in small firm
finance, The Journal of Business 68, 351.
Bertrand, Marianne, Francis Kramarz, Antoinette Schoar, and David Thesmar, 2004, Politically
connected ceos and corporate outcomes: Evidence from France, (University of Chicago).
Unpublished working paper.
Besley, T., and S. Coate, 1995, Group lending, repayment incentives and social collateral, Journal of
Development Economics 46, 1-18.
Black, Fischer, 1975, Bank funds management in an efficient market, Journal of Financial Economics 2,
323-339.
Boot, Arnoud W. A., 2000, Relationship banking: What do we know?, Journal of Financial
Intermediation 9, 7-25.
Boot, Arnoud W. A., and Anjan V. Thakor, 2000, Can relationship banking survive competition?, The
Journal of Finance 55, 679.
Brau, J., and G Woller, 2004, Microfinance: A comprehensive review of the existing literature, The
Journal of Entrepreneurial Finance and Business Ventures 9, 1-26.
Cassar, Allesandra, Luke Crowley, and Bruce Wydick, 2007, The effect of social capital on group loan
43
repayment: Evidence from field experiments, The Economic Journal 117, F85-F106.
Diamond, Douglas W., 1984, Financial intermediation and delegated monitoring, Review of Economic
Studies LI, 393-412.
Diamond, Douglas W., 1991, Monitoring and reputation: The choice between bank loans and directly
placed debt, The Journal of Political Economy 99.
De Aghion, Beatriz, and Jonathan Morduch, 2000, Microfinance beyond group lending, Economics of
Transition 8, 401-420.
Degryse, Hans, and Steven Ongena, 2005, Distance, lending relationships, and competition, The Journal
of Finance 60, 231.
Dinc, I. Serdar, 2000, Bank commitment, and the effects of competition in credit markets, Review of
Financial Studies 13, 781-812.
Fama, Eugene F., 1985, What’s different about banks?, Journal of Monetary Economics 15, 29-39.
Farrell, Maureen, 2008, Banking 2.0: New capital connections for entrepreneurs, Forbes.com (Forbes).
Furst, Karen, William W. Lang and Daniel E. Nolle, 2000, Who offers internet banking?, Quarterly
Journal, Vol. 19, No. 2, Office of the Comptroller of the Currency.
Ghatak, Maitreesh, and Timothy W. Guinnane, 1999, The economics of lending with joint liability:
Theory and practice, Journal of Development Economics 60, 195-228.
Gorsky, Martin, 1998, The growth and distribution of english friendly societies in the early nineteenth
century, Economic History Review LI, 489-511.
Hale, Galina and Joao A.C. Santos, 2009, Do banks price their information monopoly?, Journal of
Financial Economics 93, 185-206.
Hauswald, Robert, and Robert Marquez, 2006, Competition and Strategic Information Acquisition in
Credit Markets, Review of Financial Studies 19, 967-1000.
Hollis, Aidan, and Arthur Sweetman, 2001, The life-cycle of a microfinance institution: The irish loan
funds, Journal of Economic Behavior & Organization 46, 291-311.
44
Houston, Joel and Christopher James, 1996, Bank information monopolies and the mix of private and
public debt claims, Journal of Finance 51, 1863-1889.
Hung, Chi-kan R., 2006, Rules and actions: Determinants of peer group and staff actions in group-based
microcredit programs in the united states, Economic Development Quarterly 20, 75-96.
Jaffee, Dwight M., and Thomas Russell, 1976, Imperfect information, uncertainty and credit rationing,
Quarterly Journal of Economics 90, 651-666.
Karlan, Dean S., 2007, Social connections and group banking, The Economic Journal 117, F52-F84.
Kim, Jane J., 2008, Where either a borrower or a lender can be: Small-business owners turn to online
networks for funds as banks tighten credit, Wall Street Journal (Dow Jones, New York).
Leland, Hayne E., and David H. Pyle, 1977, Informational asymmetries, financial structures, and financial
intermediation, Journal of Finance 32, 371-387.
Lummer, Scott L., and John J. McConnell, 1989, Further evidence on the bank lending process and the
capital-market response to bank loan agreements, Journal of Financial Economics 25, 99-122.
Petersen, Mitchell A., and Raghuram G. Rajan, 1994, The benefits of lending relationships: Evidence
from small business data, Journal of Finance 49, 3-37.
Petersen, Mitchell A., and Raghuram G. Rajan, 1995, The effect of credit market competition on lending
relationships, Quarterly Journal of Economics 1995, 407-443.
Petersen, Mitchell A., and Raghuram G. Rajan, 2002, Does distance still matter? The information
revolution and small business lending, Journal of Finance 57, 2533-2570.
Potter, Helen C., 1954, Savings and loan associations and the consumer interest, American Journal of
Economics and Sociology 13, 191-203.
Rajan, Raghuram G., 1992, Insiders and outsiders: The choice between informed and arm's-length debt,
Journal of Finance 47, 1367-1400.
Santos, Joao A. C. and Andrew Winton, 2008, Bank loans, bonds and information monopolies across the
business cycle, Journal of Finance 63, 1315-1359.
45
Schenone, Carola, 2010, Lending relationships and information rents: do banks exploit their information
advantages?, Review of Financial Studies 23, 1151-1199.
Sharpe, Steven A., 1990, Asymmetric information, bank lending and implicit contracts: a stylized model
of customer relationships, Journal of Finance 45, 1069-1087.
Stiglitz, Joseph E., and Andrew Weiss, 1981, Credit rationing in markets with imperfect information,
American Economic Review 71, 393-410.
Von Thadden, Ernst-Ludwig, 2004, Asymmetric information, bank lending and implicit contracts the
winner’s curse, Finance Research Letters 1, 11-23.
Wahid, Abu N. M., 1994, The grameen bank and poverty alleviation in bangladesh, American Journal of
Economics and Sociology 53, 1-15.
Williams, Toni, 2004, Requiem for microcredit? The demise of a romantic ideal, Banking and Finance
Law Review 19.
Woolcock, Michael, 2001, Microenterprise and social capital: A framework for theory, research, and
policy, Journal of Socio-Economics 30, 193-198.
46
Table 1 The World of Online Social Lending
This table lists the principal online social lending services that have appeared since 2005, but before May 2008.
All listings are commercial for-profit enterprises unless otherwise indicated.
Lending Site Inception Description
Zopa.com (UK) Mar 2005 Peer-to-peer (P2P) lending in the UK Kiva.org Oct 2005 Non-profit. Allows individuals to make loans to entrepreneurs
in developing countries. Prosper.com Feb 2006 peer-to-peer lending in the US. Allows borrowers to list their
loan requirements and members can then bid as lenders. Prosper then services the loan for the 3-year term.
MyC4 Dec 2006 Denmark for-profit Microfinance Institution (MFI) in Africa Boober.nl Jan 2007 peer-to-peer lending in the Netherlands Virginmoneyus.com Feb 2007 Formally circelending.com. manages and services direct family
and friend lending Smava.de Mar 2007 peer-to-peer lending in Germany Fairrates.dk Apr 2007 peer-to-peer lending in Denmark LendingClub.com May 2007 peer-to-peer lending in the US, originally limited to
Facebook.com members PPdai Jun 2007 peer-to-peer lending in China Microplace.com Oct 2007 Ebay MFI investing in LDCs IGrin.com.au Oct 2007 Peer-to-peer lending in Australia Zopa.com (Italy) Nov 2007 Peer-to-peer lending in Italy Boober.it (Italy) Nov 2007 Peer-to-peer lending in Italy Loanland Dec 2007 Peer-to-peer lending in Sweden Zopa.com (US) Dec 2007 Peer-to-peer lending in the US IOUCentral.ca Feb 2008 Peer-to-peer lending in Canada Zopa.com (Japan) Mar 2008 Peer-to-peer lending in Japan CommunityLend.com Apr 2008 Peer-to-peer lending in Canada
47
Table 2 Variable Definitions
Variable Type Description
DEFAULTIND Dummy Value of 1 if the loan is default, 0 otherwise. The soonest that a loan can transition from past due to default, even if the borrower never makes a payment, is six months.
MEMBEROFGROUP Dummy Value of 1 if borrower is a member of any group, 0 otherwise. GRP_RELATIONSHIPS Dummy Value of 1 if borrower is a member of a group that has a higher likelihood
of personal relationships (group categories: private, geographic, alumni, occupation and company), value of 0 if member of group in the following categories: small business, hobby, ethnic, high-risk, low-risk or other.
GLREWARDS Continuous Percentage of loan interest in a group that is paid to the group leader. This practice has been discontinued so newer loans all have a value of zero in this field.
LISTREV Dummy Value of 1 indicates that group members in this group must have their loan listings reviewed by the group leader before being posted.
LNGROUPLOANS Continuous Natural log of the total number of loans initiated in a particular group. This represents a proxy for the effective size and experience of the group.
RATIOBIDBYGROUP Continuous Numerator is total dollar value of bids made by fellow members of the borrower’s group. Denominator is total dollar value of bids against the loan listing.
LNAMT Continuous Natural log of the original loan principal amount AGEINMONTHS Discrete Number of months from loan initiation until current date or until default. BORROWERRATE Continuous Interest rate to the borrower, expressed as a decimal. QUICKFUNDING Dummy Value of 1 if the borrower submitted loan listing with the
QUICKFUNDING option selected. This means that instead of waiting for the full bidding period to end (often resulting in the interest rate being bid down), the borrower wants the bidding to end and the loan to be originated as soon as the loan has received enough bids to be funded.
CREDITSCORE Discrete ScoreXPlus Credit Score received from credit bureau (Experian). This is an integer between 520 and 800. Prosper does not provide this numeric score directly to lenders. Instead, lenders see a credit grade corresponding to the following ranges: 7 AA 760+ 6 A 720-759 5 B 680-719 4 C 640-679 3 D 600-639 2 E 560-599 1 HR 520-559 (high risk)
LNSCORE Continuous Natural log of the 1-7 values of the CREDITSCORE HOMEOWNERIND Dummy Value of 1 if the borrower owns a home ENDORSEMENTSIND Dummy Value of 1 if the borrower has received personal endorsements from other
members of the social lending site LNBORROWERAGE Discrete Calculated as the number of years since the initiation of the borrower’s
credit report file minus 18. The assumption is that most people first receive credit in their own name at around age 18
GDSTATE Continuous Calculated value between zero and one, with one representing a very highly concentrated group membership.
SPREAD Continuous The difference between BORROWERRATE and FEDFUNDS LNDATEBIDCOUNT Continuous Natural log of the total number of bids placed on ALL loans on the
website on the last day of bidding for the loan being analyzed. This is a proxy for total website traffic.
FEDFUNDS Continuous This is the fed funds overnight rate on the last day of bidding for the loan. LNAMT * LNSCORE Continuous Interaction term FORPROFIT * LNGRPLOANS Continuous Interaction term
48
Table 3 Descriptive Statistics Concerning Loan and Borrower Characteristics Sorted by Whether
they are Members of a Lending Group
This table reports summary statistics of the full sample of loans initiated between May 31, 2006 and November 6, 2007. It also includes performance data for these same loans through May 7, 2008. Summary results are broken down into three groups: 1) all borrowers; 2) borrowers that are members of any lending group; and 3) borrowers that are not members of a group. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively for t-tests.
Total Member of Group Not a Member
Count Mean Median Count Mean Median Count Mean Median Diff in
Obs Spread Spread Obs Spread Spread Obs Spread Spread Mean
Loan Amount
1st quartile $0 - 2550 3446 0.13340 0.13715 2080 0.14192 0.14730 1366 0.12042 0.11430 0.0215 ***
2nd quartile $2551 to 4600 3253 0.14500 0.14760 1928 0.15040 0.15690 1325 0.13715 0.13250 0.0133 ***
3rd quartile $4601 to 8000 3380 0.12368 0.11755 1975 0.12820 0.12470 1405 0.11731 0.10950 0.0109 ***
4th quartile $8001 + 3386 0.11693 0.10760 1999 0.11931 0.11180 1387 0.11350 0.10190 0.0058 ***
Total 13465 0.12962 0.12480 7982 0.13491 0.13210 5483 0.12192 0.11270 0.0130 ***
Credit Score
760+ 1398 0.04449 0.03720 627 0.04469 0.03710 771 0.04433 0.03720 0.0004
720 - 759 1359 0.06592 0.05950 645 0.06471 0.05960 714 0.06701 0.05910 -0.0023
680 - 719 1752 0.09187 0.08575 901 0.09168 0.08500 851 0.09207 0.08690 -0.0004
640 - 679 2465 0.12096 0.11440 1400 0.11810 0.11025 1065 0.12472 0.11770 -0.0066 ***
600 - 639 2522 0.15141 0.14750 1497 0.14528 0.14250 1025 0.16037 0.15950 -0.0151 ***
560 - 599 1874 0.19103 0.19740 1259 0.18462 0.18750 615 0.20415 0.21780 -0.0195 ***
520 - 559 2095 0.18834 0.19760 1653 0.18708 0.19580 442 0.19306 0.23005 -0.0060 *
Total 13465 0.12962 0.12480 7982 0.13491 0.13210 5483 0.12192 0.11270 0.0130 ***
Borrower Age
1st quartile 0 - 27 2125 0.13353 0.12790 1235 0.14058 0.14080 890 0.12376 0.11575 0.0168 ***
2nd quartile 28 - 31 3062 0.13090 0.12605 1861 0.13780 0.13510 1201 0.12022 0.10760 0.0176 ***
3rd quartile 32 - 36 4163 0.13108 0.12740 2497 0.13654 0.13540 1666 0.12290 0.11425 0.0136 ***
4th quartile 37 + 4079 0.12502 0.11760 2362 0.12782 0.12225 1717 0.12117 0.11310 0.0066 ***
Total 13429 0.12959 0.12470 7955 0.13487 0.13190 5474 0.12191 0.11270 0.0130 ***
Is Borrower Homeowner?
Yes 5682 0.11251 0.10400 3152 0.11734 0.11050 2530 0.10649 0.09690 0.0109 ***
No 7783 0.14211 0.14260 4830 0.14638 0.14740 2953 0.13513 0.12960 0.0113 ***
Total 13465 0.12962 0.12480 7982 0.13491 0.13210 5483 0.12192 0.11270 0.0130 ***
First Loan?
Yes 13175 0.13019 0.12660 7830 0.13540 0.13250 5345 0.12256 0.11430 0.0128 ***
No 290 0.10372 0.09135 152 0.10989 0.09805 138 0.09693 0.07955 0.0130 *
Total 13465 0.12962 0.12480 7982 0.13491 0.13210 5483 0.12192 0.11270 0.0130 ***
49
Table 4 Descriptive Statistics Concerning Default Rates Sorted by
Whether the Borrower is in a Group or Not
This table reports summary statistics of the full sample of loans initiated between May 31, 2006 and November 6, 2007. It also includes performance data for these same loans through May 7, 2008. Summary results are broken down into three groups: 1) all borrowers; 2) borrowers that are members of a group; and 3) borrowers that do not belong to a group. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
Count Mean Count Mean Count Mean Diff in
Obs Default Obs Default Obs Default Mean
Loan Amount
1st quartile $0 - 2550 3446 0.10795 2080 0.14375 1366 0.05344 0.0903 ***
2nd quartile $2551 to 4600 3253 0.09560 1928 0.12189 1325 0.05736 0.0645 ***
3rd quartile $4601 to 8000 3380 0.04704 1975 0.05570 1405 0.03488 0.0208 ***
4th quartile $8001 + 3386 0.04017 1999 0.04952 1387 0.02668 0.0228 ***
Total 13465 0.07263 7982 0.09308 5483 0.04286 0.0502 ***
Credit Score
760+ 1398 0.00572 627 0.00638 771 0.00519 0.0012
720 - 759 1359 0.01398 645 0.01860 714 0.00980 0.0088
680 - 719 1752 0.01941 901 0.01998 851 0.01880 0.0012
640 - 679 2465 0.03570 1400 0.04357 1065 0.02535 0.0182 **
600 - 639 2522 0.05234 1497 0.05812 1025 0.04390 0.0142
560 - 599 1874 0.12327 1259 0.13185 615 0.10569 0.0262 *
520 - 559 2095 0.22243 1653 0.23896 442 0.16063 0.0783 ***
Total 13465 0.07263 7982 0.09308 5483 0.04286 0.0502 ***
Borrower Age
1st quartile 0 - 27 2125 0.10306 1235 0.13603 890 0.05730 0.0787 ***
2nd quartile 28 - 31 3062 0.08589 1861 0.11231 1201 0.04496 0.0674 ***
3rd quartile 32 - 36 4163 0.06918 2497 0.08730 1666 0.04202 0.0453 ***
4th quartile 37 + 4079 0.04903 2362 0.05927 1717 0.03494 0.0243 ***
Total 13429 0.07223 7955 0.09239 5474 0.04293 0.0495 ***
Is Borrower Homeowner?
Yes 5682 0.04453 3152 0.05330 2530 0.03360 0.0197 ***
No 7783 0.09315 4830 0.11905 2953 0.05080 0.0683 ***
Total 13465 0.07263 7982 0.09308 5483 0.04286 0.0502 ***
First Loan?
Yes 13175 0.07385 7830 0.09464 5345 0.04341 0.0512 ***
No 290 0.01724 152 0.01316 138 0.02174 -0.0086
Total 13465 0.07263 7982 0.09308 5483 0.04286 0.0502 ***
Group Members Non-Group MembersTotal
50
Table 5 Descriptive Statistics Concerning Default Rates of Borrowers in a Lending Group Sorted by
whether the Group has the Potential for Personal Relationships
This table reports summary statistics of the full sample of loans initiated between May 31, 2006 and November 6, 2007. It also includes performance data for these same loans through May 7, 2008. Summary results are broken down into three groups: 1) all borrowers who belong to a group; 2) borrowers that are members of a group where there is a potential for personal relationships; and 3) borrowers that belong to all other groups. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
Count Mean Count Mean Count Mean Diff in
Obs Default Obs Default Obs Default Mean
Loan Amount
1st quartile $0 - 2550 2080 0.14375 223 0.04933 1857 0.15509 -0.1058 ***
2nd quartile $2551 to 4600 1928 0.12189 215 0.06512 1713 0.12901 -0.0639 ***
3rd quartile $4601 to 8000 1975 0.05570 291 0.02405 1684 0.06116 -0.0371 ***
4th quartile $8001 + 1999 0.04952 309 0.03883 1690 0.05148 -0.0127
Total 7982 0.09308 1038 0.04239 6944 0.10066 -0.0583 ***
Credit Score
760+ 627 0.00638 166 0.00000 461 0.00868 -0.0087 **
720 - 759 645 0.01860 126 0.01587 519 0.01927 -0.0034
680 - 719 901 0.01998 158 0.00633 743 0.02288 -0.0166 **
640 - 679 1400 0.04357 176 0.02273 1224 0.04657 -0.0238 *
600 - 639 1497 0.05812 166 0.03012 1331 0.06161 -0.0315 **
560 - 599 1259 0.13185 146 0.05479 1113 0.14196 -0.0872 ***
520 - 559 1653 0.23896 100 0.24000 1553 0.23889 0.0011
Total 7982 0.09308 1038 0.04239 6944 0.10066 -0.0583 ***
Borrower Age
1st quartile 0 - 27 1235 0.13603 146 0.07534 1089 0.14417 -0.0688 ***
2nd quartile 28 - 31 1861 0.11231 235 0.05532 1626 0.12054 -0.0652 ***
3rd quartile 32 - 36 2497 0.08730 319 0.04389 2178 0.09366 -0.0498 ***
4th quartile 37 + 2362 0.05927 337 0.01780 2025 0.06617 -0.0484 ***
Total 7955 0.09239 1037 0.04243 6918 0.09988 -0.0575 ***
Is Borrower Homeowner?
Yes 3152 0.05330 495 0.01818 2657 0.05984 -0.0417 ***
No 4830 0.11905 543 0.06446 4287 0.12596 -0.0615 ***
Total 7982 0.09308 1038 0.04239 6944 0.10066 -0.0583 ***
First Loan?
Yes 7830 0.09464 1009 0.04361 6821 0.10218 -0.0586 ***
No 152 0.01316 29 0.00000 123 0.01626 -0.0163
Total 7982 0.09308 1038 0.04239 6944 0.10066 -0.0583 ***
Relationship Group Non-Relationship GroupTotal
51
Table 6 Descriptive Statistics Concerning Default Rates of Borrowers in a Lending Group Sorted by
whether the Group Leader has Observable Profit Motive
This table reports summary statistics of the full sample of loans initiated between May 31, 2006 and November 6, 2007. It also includes performance data for these same loans through May 7, 2008. Summary results are broken down into three groups: 1) all borrowers who belong to a group; 2) borrowers that are members of a group where the leader takes a percentage of the loan fee (profit motive); and 3) borrowers that belong to a group without a profit motive. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
Count Mean Count Mean Count Mean Diff in
Obs Default Obs Default Obs Default Mean
Loan Amount
1st quartile $0 - 2550 2080 0.14375 1408 0.13707 672 0.15774 -0.0207
2nd quartile $2551 to 4600 1928 0.12189 1394 0.10689 534 0.16105 -0.0542 ***
3rd quartile $4601 to 8000 1975 0.05570 1485 0.04646 490 0.08367 -0.0372 ***
4th quartile $8001 + 1999 0.04952 1593 0.04269 406 0.07635 -0.0337 **
Total 7982 0.09308 5880 0.08146 2102 0.12559 -0.0441 ***
Credit Score
760+ 627 0.00638 401 0.00748 226 0.00442 0.0031
720 - 759 645 0.01860 486 0.02263 159 0.00629 0.0163 *
680 - 719 901 0.01998 701 0.01854 200 0.02500 -0.0065 *
640 - 679 1400 0.04357 1127 0.03904 273 0.06227 -0.0232
600 - 639 1497 0.05812 1162 0.05766 335 0.05970 -0.0020
560 - 599 1259 0.13185 885 0.12090 374 0.15775 -0.0369 *
520 - 559 1653 0.23896 1118 0.20930 535 0.30093 -0.0916 ***
Total 7982 0.09308 5880 0.08146 2102 0.12559 -0.0441 ***
Borrower Age
1st quartile 0 - 27 1235 0.13603 892 0.12332 343 0.16910 -0.0458 **
2nd quartile 28 - 31 1861 0.11231 1329 0.09330 532 0.15977 -0.0665 ***
3rd quartile 32 - 36 2497 0.08730 1872 0.07532 625 0.12320 -0.0479 ***
4th quartile 37 + 2362 0.05927 1765 0.05496 597 0.07203 -0.0171
Total 7955 0.09239 5858 0.08057 2097 0.12542 -0.0449 ***
Is Borrower Homeowner?
Yes 3152 0.05330 2426 0.04782 726 0.07163 -0.0238 **
No 4830 0.11905 3454 0.10510 1376 0.15407 -0.0490 **
Total 7982 0.09308 5880 0.08146 2102 0.12559 -0.0441 ***
First Loan?
Yes 7830 0.09464 5767 0.08271 2063 0.12797 -0.0453 ***
No 152 0.01316 113 0.01770 39 0.00000 0.0177
Total 7982 0.09308 5880 0.08146 2102 0.12559 -0.0441 ***
Group Profit Motive No Group Profit MotiveTotal
52
Table 7 Descriptive Statistics Concerning Default Rates of Borrowers in a Lending Group Sorted by
whether the Group has the Potential for Monitoring This table reports summary statistics of a sample of loans initiated between May 31, 2006 and November 6, 2007. It also includes performance data for these same loans through May 7, 2008. Summary results are broken down into three groups: 1) all borrowers who belong to a group; 2) borrowers that are members of a group that is neither a for-profit group nor a relationship group; and 3) borrowers that belong to all other groups. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
Count Mean Count Mean Count Mean Diff in
Obs Default Obs Default Obs Default Mean
Loan Amount
1st quartile $0 - 2550 2080 0.14375 596 0.16946 1484 0.13342 0.0360 **
2nd quartile $2551 to 4600 1928 0.12189 455 0.16703 1473 0.10794 0.0591 ***
3rd quartile $4601 to 8000 1975 0.05570 406 0.09360 1569 0.04589 0.0477 ***
4th quartile $8001 + 1999 0.04952 294 0.08844 1705 0.04282 0.0456 ***
Total 7982 0.09308 1751 0.13764 6231 0.08056 0.0571 ***
Credit Score
760+ 627 0.00638 169 0.00592 458 0.00655 -0.0006
720 - 759 645 0.01860 111 0.00901 534 0.02060 -0.0116
680 - 719 901 0.01998 143 0.02797 758 0.01847 0.0095
640 - 679 1400 0.04357 224 0.05804 1176 0.04082 0.0172
600 - 639 1497 0.05812 287 0.06620 1210 0.05620 0.0100
560 - 599 1259 0.13185 330 0.17576 929 0.11625 0.0595 **
520 - 559 1653 0.23896 487 0.29774 1166 0.21441 0.0833 ***
Total 7982 0.09308 1751 0.13764 6231 0.08056 0.0571 ***
Borrower Age
1st quartile 0 - 27 1235 0.13603 297 0.17845 938 0.12260 0.0559 **
2nd quartile 28 - 31 1861 0.11231 455 0.16923 1406 0.09388 0.0754 ***
3rd quartile 32 - 36 2497 0.08730 521 0.13244 1976 0.07540 0.0570 ***
4th quartile 37 + 2362 0.05927 474 0.08650 1888 0.05244 0.0341 **
Total 7955 0.09239 1747 0.13738 6208 0.07974 0.0576 ***
Is Borrower Homeowner?
Yes 3152 0.05330 574 0.08014 2578 0.04732 0.0328 ***
No 4830 0.11905 1177 0.16568 3653 0.10402 0.0617 ***
Total 7982 0.09308 1751 0.13764 6231 0.08056 0.0571 ***
First Loan?
Yes 7830 0.09464 1719 0.14020 6111 0.08182 0.0584 ***
No 152 0.01316 32 0.00000 120 0.01667 -0.0167
Total 7982 0.09308 1751 0.13764 6231 0.08056 0.0571 ***
Non-Prof & Non-Rel All Other GroupsTotal
53
Table 8 Descriptive Statistics Concerning Weighted Average of Interest Rate
Bids for Insiders vs. Outsiders for Loans to Group Members
This table reports summary statistics of loans to borrowers that are members of a lending group. The loans were initiated between May 31, 2006 and November 6, 2007. It also includes performance data for these same loans through May 7, 2008. Summary results are broken down into two groups: 1) bids from insiders, meaning lenders who are members of the same group as the borrower, and 2) bids from outsiders, meaning bids from everyone else. The differences between the means of these two groups are reported. Statistical significance is tested using pairwise t-tests of differences. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
Insider Bidders Outsider Bidders
Count Mean Median Count Mean Median Diff in
Obs Spread Spread Obs Spread Spread Mean
Loan Amount
1st quartile $0 - 2550 2080 0.12396 0.13219 2080 0.12283 0.13376 0.0011
2nd quartile $2551 to 4600 1928 0.13400 0.14605 1928 0.13372 0.14342 0.0003
3rd quartile $4601 to 8000 1975 0.12205 0.12030 1975 0.12119 0.12012 0.0009
4th quartile $8001 + 1999 0.11786 0.11290 1999 0.11640 0.11226 0.0015 ***
Total 7982 0.12326 0.12230 7982 0.12226 0.12142 0.0010 ***
Credit Score
760+ 627 0.05075 0.04505 627 0.05051 0.04536 0.0002
720 - 759 645 0.07473 0.07271 645 0.07323 0.07300 0.0015 **
680 - 719 901 0.09238 0.08700 901 0.09118 0.08660 0.0012 *
640 - 679 1400 0.11994 0.11709 1400 0.11808 0.11390 0.0019 ***
600 - 639 1497 0.14775 0.14720 1497 0.14686 0.14708 0.0009
560 - 599 1259 0.17773 0.17755 1259 0.17781 0.17852 -0.0001
520 - 559 1653 0.18315 0.18720 1653 0.18230 0.18626 0.0009
Total 7982 0.12326 0.12230 7982 0.12226 0.12142 0.0010 ***
Borrower Age
1st quartile 0 - 27 1235 0.12597 0.12806 1235 0.12529 0.12622 0.0007
2nd quartile 28 - 31 1861 0.12302 0.11818 1861 0.12206 0.12058 0.0010
3rd quartile 32 - 36 2497 0.12478 0.12504 2497 0.12438 0.12472 0.0004
4th quartile 37 + 2362 0.12043 0.11760 2362 0.11859 0.11715 0.0018 ***
Total 7955 0.12319 0.12229 7955 0.12217 0.12134 0.0010 ***
Is Borrower Homeowner?
Yes 3152 0.11045 0.10480 3152 0.10931 0.10271 0.0011 ***
No 4830 0.13388 0.13581 4830 0.13300 0.13628 0.0009 **
Total 7982 0.12326 0.12230 7982 0.12226 0.12142 0.0010 ***
First Loan?
Yes 7830 0.12377 0.12280 7830 0.12277 0.12186 0.0010 ***
No 152 0.09740 0.09337 152 0.09621 0.09199 0.0012
Total 7982 0.12326 0.12230 7982 0.12226 0.12142 0.0010 ***
54
Table 9 Descriptive Statistics Concerning Loan and Borrower Characteristics of Borrowers in a Lending Group Sorted by whether the Group Leader has Observable Profit Motive
This table reports summary statistics of the full sample of loans initiated between May 31, 2006 and November 6, 2007. It also includes performance data for these same loans through May 7, 2008. Summary results are broken down into three groups: 1) all borrowers who belong to a group; 2) borrowers that are members of a group where the leader takes a percentage of the loan fee (profit motive); and 3) borrowers that belong to a group without a profit motive. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
Total Group Profit Motive No Group Profit Motive
Count Mean Median Count Mean Median Count Mean Median Diff in
Obs Spread Spread Obs Spread Spread Obs Spread Spread Mean
Loan Amount
1st quartile $0 - 2550 2080 0.14192 0.14730 1408 0.14333 0.14750 672 0.13896 0.14690 0.0044
2nd quartile $2551 to 4600 1928 0.15040 0.15690 1394 0.15193 0.15760 534 0.14641 0.14750 0.0055
3rd quartile $4601 to 8000 1975 0.12820 0.12470 1485 0.13037 0.12730 490 0.12163 0.11735 0.0087 ***
4th quartile $8001 + 1999 0.11931 0.11180 1593 0.12091 0.11260 406 0.11302 0.10550 0.0079 ***
Total 7982 0.13491 0.13210 5880 0.13602 0.13260 2102 0.13180 0.12710 0.0042 **
Credit Score
760+ 627 0.04469 0.03710 401 0.04859 0.04190 226 0.03778 0.03075 0.0108 ***
720 - 759 645 0.06471 0.05960 486 0.06766 0.06430 159 0.05567 0.04860 0.0120 ***
680 - 719 901 0.09168 0.08500 701 0.09410 0.08730 200 0.08321 0.07965 0.0109 ***
640 - 679 1400 0.11810 0.11025 1127 0.12037 0.11240 273 0.10873 0.10250 0.0116 ***
600 - 639 1497 0.14528 0.14250 1162 0.14690 0.14305 335 0.13967 0.13720 0.0072 ***
560 - 599 1259 0.18462 0.18750 885 0.18598 0.18750 374 0.18138 0.18760 0.0046 *
520 - 559 1653 0.18708 0.19580 1118 0.18831 0.19230 535 0.18451 0.19740 0.0038
Total 7982 0.13491 0.13210 5880 0.13602 0.13260 2102 0.13180 0.12710 0.0042 **
Borrower Age
1st quartile 0 - 27 1235 0.14058 0.14080 892 0.14445 0.14695 343 0.13051 0.12740 0.0139 ***
2nd quartile 28 - 31 1861 0.13780 0.13510 1329 0.13737 0.13430 532 0.13885 0.13695 -0.0015
3rd quartile 32 - 36 2497 0.13654 0.13540 1872 0.13652 0.13510 625 0.13658 0.13650 -0.0001
4th quartile 37 + 2362 0.12782 0.12225 1765 0.12996 0.12660 597 0.12148 0.11730 0.0085 ***
Total 7955 0.13487 0.13190 5858 0.13595 0.13260 2097 0.13186 0.12710 0.0041 **
Is Borrower Homeowner?
Yes 3152 0.11734 0.11050 2426 0.12001 0.11465 726 0.10842 0.09770 0.0116 ***
No 4830 0.14638 0.14740 3454 0.14727 0.14750 1376 0.14414 0.14690 0.0031
Total 7982 0.13491 0.13210 5880 0.13602 0.13260 2102 0.13180 0.12710 0.0042 **
First Loan?
Yes 7830 0.13540 0.13250 5767 0.13642 0.13370 2063 0.13253 0.12740 0.0039 **
No 152 0.10989 0.09805 113 0.11569 0.10200 39 0.09311 0.08230 0.0226 **
Total 7982 0.13491 0.13210 5880 0.13602 0.13260 2102 0.13180 0.12710 0.0042 **
55
Table 10 Descriptive Statistics Concerning Weighted Average of Interest Rate
Bids for Insiders vs. Outsiders for Loans to Borrowers who are Members of Groups that have Group Leader with Profit Motive
This table reports summary statistics of loans to borrowers that are members of a lending group. The loans were initiated between May 31, 2006 and November 6, 2007. It also includes performance data for these same loans through May 7, 2008. Summary results are broken down into two groups: 1) bids from insiders, meaning lenders who are members of the same group as the borrower, and 2) bids from outsiders, meaning bids from everyone else. The differences between the means of these two groups are reported, but significance cannot be tested since the bids from the two groups are on the same loan record. The profit motive refers only to the group leader, who makes the private information decisions. All investors are assumed to have a profit motive. Statistical significance is tested using pairwise t-tests of differences. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
Insider Bidders Outsider Bidders
Count Mean Median Count Mean Median Diff in
Obs Spread Spread Obs Spread Spread Mean
Loan Amount
1st quartile $0 - 2550 1408 0.12600 0.13736 1408 0.12533 0.13957 0.0007
2nd quartile $2551 to 4600 1394 0.13656 0.14825 1394 0.13583 0.14694 0.0007
3rd quartile $4601 to 8000 1485 0.12497 0.12619 1485 0.12452 0.12420 0.0004
4th quartile $8001 + 1593 0.11962 0.11507 1593 0.11849 0.11424 0.0011 **
Total 5880 0.12513 0.12515 5880 0.12433 0.12418 0.0008 **
Credit Score
760+ 401 0.05436 0.04955 401 0.05413 0.05133 0.0002
720 - 759 486 0.07833 0.07695 486 0.07684 0.07659 0.0015 *
680 - 719 701 0.09323 0.08726 701 0.09238 0.08688 0.0008
640 - 679 1127 0.12226 0.11787 1127 0.12017 0.11729 0.0021 ***
600 - 639 1162 0.14884 0.14740 1162 0.14831 0.14787 0.0005
560 - 599 885 0.17849 0.17755 885 0.17847 0.17885 0.0000
520 - 559 1118 0.18372 0.18695 1118 0.18423 0.18928 -0.0005
Total 5880 0.12513 0.12515 5880 0.12433 0.12418 0.0008 **
Borrower Age
1st quartile 0 - 27 892 0.12885 0.13145 892 0.12743 0.12630 0.0014
2nd quartile 28 - 31 1329 0.12534 0.12644 1329 0.12438 0.12466 0.0010
3rd quartile 32 - 36 1872 0.12568 0.12540 1872 0.12572 0.12638 0.0000
4th quartile 37 + 1765 0.12245 0.12043 1765 0.12109 0.11801 0.0014 **
Total 5858 0.12503 0.12496 5858 0.12422 0.12407 0.0008 **
Is Borrower Homeowner?
Yes 2426 0.11453 0.11003 2426 0.11350 0.10710 0.0010 **
No 3454 0.13450 0.13730 3454 0.13390 0.13721 0.0006
Total 5880 0.12513 0.12515 5880 0.12433 0.12418 0.0008 **
First Loan?
Yes 5767 0.12561 0.12590 5767 0.12484 0.12552 0.0008 **
No 113 0.09993 0.09337 113 0.09745 0.09199 0.0025
Total 5880 0.12513 0.12515 5880 0.12433 0.12418 0.0008 **
56
Table 11 Descriptive Statistics Concerning Weighted Average of Interest Rate
Bids for Insiders vs. Outsiders for Loans to Borrowers who are Members of Groups without a Profit Motive
This table reports summary statistics of loans to borrowers that are members of a lending group. The loans were initiated between May 31, 2006 and November 6, 2007. It also includes performance data for these same loans through May 7, 2008. Summary results are broken down into two groups: 1) bids from insiders, meaning lenders who are members of the same group as the borrower, and 2) bids from outsiders, meaning bids from everyone else. The differences between the means of these two groups are reported, but significance cannot be tested since the bids from the two groups are on the same loan record. The profit motive refers only to the group leader, who makes the private information decisions. All investors are assumed to have a profit motive. Statistical significance is tested using pairwise t-tests of differences. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
Insider Bidders Outsider Bidders
Count Mean Median Count Mean Median Diff in
Obs Spread Spread Obs Spread Spread Mean
Loan Amount
1st quartile $0 - 2550 672 0.11894 0.11765 672 0.11669 0.11750 0.0023 *
2nd quartile $2551 to 4600 534 0.12840 0.13008 534 0.12911 0.13297 -0.0007
3rd quartile $4601 to 8000 490 0.11296 0.10305 490 0.11084 0.10114 0.0021 **
4th quartile $8001 + 406 0.10922 0.10208 406 0.10611 0.10092 0.0031 ***
Total 2102 0.11729 0.11471 2102 0.11564 0.11334 0.0017 ***
Credit Score
760+ 226 0.04398 0.03892 226 0.04375 0.03864 0.0002
720 - 759 159 0.06193 0.05558 159 0.06044 0.05501 0.0015
680 - 719 200 0.08929 0.08363 200 0.08682 0.08417 0.0025 *
640 - 679 273 0.10846 0.10146 273 0.10773 0.10117 0.0007
600 - 639 335 0.14377 0.14055 335 0.14155 0.14004 0.0022
560 - 599 374 0.17550 0.17875 374 0.17587 0.17794 -0.0004
520 - 559 535 0.18189 0.18760 535 0.17797 0.18202 0.0039 ***
Total 2102 0.11729 0.11471 2102 0.11564 0.11334 0.0017 ***
Borrower Age
1st quartile 0 - 27 343 0.11798 0.10760 343 0.11935 0.12355 -0.0014
2nd quartile 28 - 31 532 0.11489 0.10365 532 0.11393 0.10296 0.0010
3rd quartile 32 - 36 625 0.12174 0.12417 625 0.11985 0.12047 0.0019 *
4th quartile 37 + 597 0.11442 0.10770 597 0.11114 0.10561 0.0033 ***
Total 2097 0.11729 0.11471 2097 0.11564 0.11334 0.0017 ***
Is Borrower Homeowner?
Yes 726 0.09521 0.09015 726 0.09363 0.08999 0.0016 *
No 1376 0.13214 0.13230 1376 0.13045 0.13118 0.0017 **
Total 2102 0.11729 0.11471 2102 0.11564 0.11334 0.0017 ***
First Loan?
Yes 2063 0.11787 0.11513 2063 0.11613 0.11413 0.0017 ***
No 39 0.09005 0.08144 39 0.09264 0.08306 -0.0026
Total 2102 0.11729 0.11471 2102 0.11564 0.11334 0.0017 ***
57
Table 12 Determinants of Loan Default
This table reports marginal effects from probit analyses with DEFAULTIND as the dependent variable indicating the loan is in default status as of May 7, 2008. The loans were initiated between May 31, 2006 and November 6, 2007 and the data includes performance information for these same loans through May 7, 2008. z-scores from the marginal effects analyses are in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
(1) (2) (3) (4) (5) (6) (7)
QUICKFUNDING 0.021*** 0.034*** 0.033*** 0.029*** 0.031***
(5.576) (5.560) (5.357) (4.813) (5.108)
LNDATEBIDCOUNT -0.040*** -0.042*** -0.041*** -0.040*** -0.043***
(-13.362) (-9.456) (-8.973) (-8.855) (-9.499)
LNSCORE -0.066*** -0.088*** -0.089*** -0.103*** -0.099*** -0.086*** -0.086***
(-23.308) (-18.144) (-18.443) (-20.745) (-20.572) (-18.155) (-18.177)
LNBORROWERAGE -0.025*** -0.045*** -0.046*** -0.043*** -0.043***
(-2.869) (-3.394) (-3.413) (-3.325) (-3.298)
ENDORSEMENTSIND -0.029*** -0.045*** -0.043*** -0.034*** -0.033***
(-8.719) (-8.340) (-7.893) (-6.150) (-6.130)
HOMEOWNERIND -0.000 -0.010* -0.010* -0.008 -0.008
(-0.059) (-1.830) (-1.836) (-1.560) (-1.592)
LNAMT 0.022*** 0.022*** 0.011*** 0.016*** 0.022*** 0.023***
(5.969) (5.817) (2.714) (4.053) (6.167) (6.286)
FORPROFIT 0.035*** 0.063*** 0.049*** 0.024*
(2.695) (4.856) (4.184) (1.952)
LNGRPLOANS 0.004** 0.011*** 0.025*** 0.017*** 0.007** 0.001
(2.242) (3.839) (7.587) (4.936) (2.473) (0.303)
FORPROFIT * LNGRPLOANS -0.009** -0.022*** -0.017*** -0.005*
(-2.507) (-5.645) (-4.542) (-1.645)
GDSTATE 0.001 0.005 0.016
(0.024) (0.122) (0.376)
LISTREV -0.044*** -0.013** -0.013**
(-7.319) (-2.303) (-2.311)
RATIOBIDBYGROUP -0.139** -0.130**
(-2.340)
GLREWARDS -0.022
(-0.848)
GLREWARDS * LNGROUPLOANS 0.008
(1.456)
GRP_RELATIONSHIPS -0.028***
(-4.494)
N 13429 7955 7955 7982 7705 7680 7680
Pseudo R^2 0.196 0.203 0.201 0.143 0.152 0.203 0.203
58
Table 13 Determinants of Loan Interest Rate Spread
This table reports ordinary least squares analyses with SPREAD as the dependent variable, indicating the difference between the interest rate on the loan and the Fed Funds rate. The loans were initiated between May 31, 2006 and November 6, 2007 and the data includes performance information for these same loans through May 7, 2008. Column (1) is restricted to non-group loans only. t-statistics from the regressions are in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
(1) (2) (3) (4) (5) (6) (7)
QUICKFUNDING 0.041*** 0.036*** 0.030*** 0.030***
(33.727) (43.983) (29.164) (29.149)
LNDATEBIDCOUNT -0.003** 0.001 0.000 0.000
(-2.207) (0.984) (0.493) (0.508)
LNSCORE -0.080*** -0.070*** -0.083*** -0.080*** -0.080***
(-68.195) (-107.304) (-97.765) (-95.716) (-95.709)
LNBORROWERAGE 0.030*** 0.027*** 0.020*** 0.020***
(10.359) (13.487) (8.507) (8.509)
ENDORSEMENTSIND 0.001 0.001 -0.001 -0.001
(0.994) (1.369) (-1.380) (-1.341)
HOMEOWNERIND -0.001 -0.002*** -0.007*** -0.007***
(-1.009) (-3.079) (-6.803) (-6.807)
FORPROFIT 0.028*** 0.027*** 0.018*** 0.018***
(7.956) (9.795) (6.840) (6.828)
LNGRPLOANS 0.003*** 0.004*** 0.003*** 0.003***
(4.368) (7.726) (6.441) (6.444)
FORPROFIT * LNGRPLOANS -0.003*** -0.005*** -0.003*** -0.003***
(-3.378) (-8.129) (-5.001) (-5.003)
DEFPROB 0.334*** 0.372***
(56.308) (75.387)
GDSTATE 0.002 -0.007 -0.007
(0.361) (-1.170) (-1.181)
LISTREV -0.009*** -0.005*** -0.005***
(-9.305) (-4.895) (-4.894)
LNAMT 0.022*** 0.023*** 0.023***
(33.430) (36.533) (36.322)
RATIOBIDBYGROUP -0.103*** -0.107***
(-12.338) (-6.031)
PIR2 0.005
(0.256)
Intercept 0.125*** 0.099*** 0.083*** 0.103*** 0.016** -0.061*** -0.061***
(8.549) (10.689) (28.675) (173.975) (2.065) (-4.777) (-4.766)
N 5474 13429 7955 13429 7705 7680 7680
R^2 0.626 0.575 0.305 0.297 0.574 0.632 0.632
59
Table 14 Impact of Loan and Borrower Risk Characteristics on the Difference
Between Private and Public Bids
This table reports ordinary least squares analyses with DIFF_SPREAD as the dependent variable, indicating the difference between the weighted average of insider (private) bid interest rates and the weighted average of outsider (public) bid interest rates. The loans were initiated between May 31, 2006 and November 6, 2007 and the data includes performance information for these same loans through May 7, 2008. t-statistics from the regressions are in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
(1) (2) (3) (4) (5) (6) (7)
QUICKFUNDING -0.000 -0.001 -0.002 -0.002 -0.002
(-0.059) (-0.223) (-0.295) (-0.307) (-0.216)
LNDATEBIDCOUNT 0.000 -0.000 0.000 0.000 0.002**
(0.325) (-0.029) (0.439) (0.101) (1.976)
LNSCORE 0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.002**
(0.333) (-0.796) (-0.618) (-0.466) (-0.473) (-0.541) (-1.984)
LNBORROWERAGE 0.002 0.002 0.002 0.002 0.007**
(1.245) (1.188) (1.250) (1.260) (2.408)
ENDORSEMENTSIND 0.001 -0.000 -0.000 -0.000 -0.001
(1.587) (-0.278) (-0.316) (-0.479) (-0.826)
HOMEOWNERIND -0.000 -0.000 -0.000 -0.000 -0.001
(-0.045) (-0.112) (-0.069) (-0.135) (-0.428)
LNAMT 0.001 0.001 0.001 0.001 0.001 0.001
(1.495) (1.318) (1.462) (1.421) (1.470) (1.220)
FORPROFIT 0.000 0.004* 0.004* 0.004* 0.005**
(0.266) (1.860) (1.716) (1.811) (2.050)
LNGRPLOANS -0.001*** -0.000 -0.000 -0.001* -0.001
(-4.568) (-0.865) (-1.048) (-1.842) (-1.635)
FORPROFIT * LNGRPLOANS -0.001* -0.001* -0.001* -0.001*
(-1.872) (-1.748) (-1.722) (-1.849)
GDSTATE 0.020*** 0.020*** 0.002
(3.309) (3.344) (0.170)
LISTREV 0.000 0.000 -0.000
(0.010) (0.066) (-0.103)
RATIOBIDBYGROUP 0.007 -0.004
(1.308) (-0.688)
GRP_RELATIONSHIPS -0.000
(-0.200)
Constant -0.009 -0.005 -0.011 -0.001 -0.018*** -0.027*** -0.049***
(-1.134) (-0.633) (-1.226) (-0.385) (-2.878) (-2.722) (-2.909)
N 2012 2012 2012 2016 2016 2012 480
R^2 0.003 0.015 0.016 0.015 0.021 0.023 0.033
60
Table 15 Effects of Insider Soft Information on Interest Rate Spread
This table reports tobit regressions for 13,429 peer-to-peer loans initiated between May 31, 2006 and November 6, 2007. Performance data for these same loans is included through May 7, 2008. Panel A represents the orthogonization utilized to produce the PIR residuals used in Panel B. In Panel A, the dependent variable is RATIOBIDBYGROUP, the ratio of internal bids to external bids for each loan. For PIR1, the independent variables are external credit score, arms-length dummy, and the interaction between arms-length dummy and the external credit score. PIR2 includes all other variables that may have an effect on the internal opinion of group members on the relative value of the loan application. In Panel B, rate spread (the difference between borrower interest rate and the fed funds rate) is the dependent variable, and it is censured at zero. The independent variable PIR represents the private information residual from the orthogonalization of public credit score against a proxy for internal credit score. p-values are in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
PIR1 PIR2
SCORE -0.001*** -0.000
(-4.537) (-0.959)
LNAMT -0.004***
(-5.221)
INCOME 0.000
(1.319)
QUICKFUNDING -0.003**
(-2.471)
LNDATEBIDCOUNT 0.003**
(1.965)
GLREWARDS -0.026***
(-12.168)
GRP_RELATIONSHIPS 0.007***
(4.156)
LISTREV -0.000
(-0.324)
LNGRPLOANS -0.001
(-1.484)
HOMEOWNERIND -0.002
(-1.547)
ENDORSEMENTSIND 0.009***
(7.136)
ARMSLENGTH -0.003** -0.025
(-2.402) (-0.542)
A/L * SCORE 0.001*** 0.000
(3.044) (0.160)
A/L * AMOUNT 0.004
(1.492)
A/L * INCOME -0.001
(-0.451)
A/L * QUICKFUNDING 0.003
(0.698)
A/L * LNDATEBIDCOUNT -0.002
(-0.450)
A/L * GLREWARDS 0.026***
(4.248)
A/L * LISTREV 0.001
(0.275)
A/L * LNGROUPLOANS 0.001
(0.146)
A/L * HOMEOWNER 0.003
(0.652)
A/L * ENDORSEMENTS -0.009*
(-1.861)
Intercept 0.003*** 0.020
(3.976) (1.588)
Number of Observations 13465 7982.000
R^2 0.002 0.037
PANEL A: Orthogonalization
61
Table 15 – continued
(1) (2) (3) (4) (5) (6)
LNAMT 0.020*** 0.020*** 0.019*** 0.019*** 0.021*** 0.021***
(48.954) (48.733) (47.526) (47.231) (37.968) (38.240)
SCORE -0.028*** -0.028*** -0.028*** -0.028*** -0.028*** -0.028***
(-151.121) (-151.955) (-152.149) (-152.987) (-108.180) (-108.907)
QUICKFUNDING 0.034*** 0.034*** 0.033*** 0.033*** 0.029*** 0.029***
(50.530) (50.466) (49.338) (49.335) (30.858) (31.068)
LNBORROWERAGE 0.001*** 0.001***
(12.341) (12.202)
INCOME 0.001*** 0.001*** 0.001*** 0.001***
(3.386) (3.637) (3.432) (3.447)
LNDATEBIDCOUNT -0.004*** -0.004*** -0.004*** -0.004***
(-5.112) (-4.969) (-3.891) (-3.923)
GLREWARDS 0.024*** 0.024***
(15.431) (15.545)
LISTREV -0.005*** -0.005***
(-5.940) (-5.989)
LNGRPLOANS 0.000 0.000
(0.447) (0.460)
HOMEOWNERIND -0.002** -0.002**
(-2.240) (-2.264)
ENDORSEMENTSIND -0.005*** -0.005***
(-4.970) (-4.989)
GRP_RELATIONSHIPS -0.009*** -0.009***
(-7.341) (-7.378)
PIR1 -0.114*** -0.112***
(-13.138) (-13.072)
PIR2 -0.090***
(-10.355)
Constant 0.055*** 0.056*** 0.072*** 0.073*** 0.073*** 0.073***
(16.770) (17.384) (10.011) (10.211) (7.939) (7.984)
N 13465 13465 13429 13429 7982 7982
R^2 -0.449 -0.454 -0.456 -0.461 -0.424 -0.429
PANEL B: Effect of PIR on Spread
62
Table 16
Descriptive Statistics Concerning the Difference between First and Subsequent Loans for the Weighted Average of
Interest Rate Bids of Insiders vs. Outsiders for Loans to Borrowers who are Members of Groups
This table reports summary statistics of loans to borrowers that are members of a lending group. The loans were initiated between May 31, 2006 and November 6, 2007. It also includes
performance data for these same loans through May 7, 2008. Summary results are broken down into two groups: 1) bids from insiders, meaning lenders who are members of the same
group as the borrower, and 2) bids from outsiders, meaning bids from everyone else. Statistical significance is tested using pairwise t-tests of differences. *, **, and *** indicate
significance at the 10%, 5%, and 1% levels respectively. For subsequent peer-to-peer loans, none of the differences are significant due to the low number of observations.
First Peer-to-Peer Loan
Subsequent Peer-to-Peer Loans
Insider Bidders
Outsider Bidders
Insider Bidders
Outsider Bidders
Count
Mean
Median
Mean
Median
Diff in
CountMean
Median
Mean
Median
Diff in
Diff in
Obs
Spread
Spread
Spread
Spread
Mean
Obs
Spread
Spread
Spread
Spread
Mean
Diff
Loan Amount
1st quartile $0 - 2550
2057
0.12400
0.13167
0.12230
0.13320
0.00170
23
0.09887
0.10267
0.09246
0.09610
0.00641
-0.00471
2nd quartile $2551 to 4600
1907
0.13400
0.14530
0.13281
0.14275
0.00119
21
0.05855
0.05855
0.06015
0.06015
-0.00160
0.00279
3rd quartile $4601 to 8000
1936
0.12234
0.12030
0.12105
0.12116
0.00129
39
0.08255
0.08029
0.08307
0.08629
-0.00052
0.00181
4th quartile $8001 +
1930
0.11791
0.11327
0.11608
0.11218
0.00183***
69
0.10379
0.10490
0.10329
0.10283
0.00050
0.00133
Total
7830
0.12343
0.12245
0.12190
0.12142
0.00153***
152
0.09661
0.09213
0.09582
0.09171
0.00079
0.00074
Credit Score
760+
603
0.05051
0.04355
0.04999
0.04504
0.00052
24
0.05036
0.05058
0.05063
0.04772
-0.00027
0.00079
720 - 759
634
0.07431
0.07180
0.07284
0.07328
0.00147**
11
0.06531
0.04885
0.06122
0.05137
0.00409
-0.00262
680 - 719
877
0.09276
0.08750
0.09093
0.08708
0.00183*
24
0.06754
0.06880
0.07136
0.06528
-0.00382
0.00565
640 - 679
1358
0.11950
0.11750
0.11775
0.11468
0.00175**
42
0.11453
0.10733
0.11201
0.10741
0.00252
-0.00077
600 - 639
1473
0.14748
0.14715
0.14575
0.14579
0.00173
24
0.14663
0.13840
0.14924
0.14258
-0.00261
0.00434
560 - 599
1245
0.17736
0.17760
0.17688
0.17779
0.00048
14
0.15750
0.15750
0.14610
0.14610
0.01140
-0.01092
520 - 559
1640
0.18299
0.18730
0.18065
0.18445
0.00234
13
0.15730
0.15730
0.14484
0.14484
0.01246
-0.01012
Total
7830
0.12343
0.12245
0.12190
0.12142
0.00153***
152
0.09661
0.09213
0.09582
0.09171
0.00079
0.00074
Borrower Age
1st quartile 0 - 27
1207
0.12669
0.12761
0.12506
0.12468
0.00163
28
0.09073
0.09069
0.09141
0.09112
-0.00068
0.00231
2nd quartile 28 - 31
1824
0.12366
0.11842
0.12197
0.12029
0.00169*
37
0.09301
0.08320
0.09519
0.08981
-0.00218
0.00387
3rd quartile 32 - 36
2454
0.12478
0.12535
0.12372
0.12463
0.00106
43
0.11309
0.12263
0.11007
0.11232
0.00302
-0.00196
4th quartile 37 +
2318
0.12017
0.11760
0.11829
0.11750
0.00188***
44
0.08781
0.08632
0.08392
0.08615
0.00389
-0.00201
Total
7803
0.12335
0.12230
0.12181
0.12122
0.00154***
152
0.09661
0.09213
0.09582
0.09171
0.00079
0.00075
Is Borrower Homeowner?
Yes
3071
0.11035
0.10490
0.10901
0.10265
0.00134***
81
0.09461
0.07750
0.09262
0.08642
0.00199
-0.00065
No
4759
0.13416
0.13670
0.13248
0.13621
0.00168**
71
0.09948
0.09789
0.10042
0.09725
-0.00094
0.00262
Total
7830
0.12343
0.12245
0.12190
0.12142
0.00153***
152
0.09661
0.09213
0.09582
0.09171
0.00079
0.00074
63
Table 17 Impact of Loan and Borrower Risk Characteristics
on the Difference between First and Subsequent Loans for the Weighted Average of Interest Rate Bids of Insiders vs. Outsiders for Loans
to Borrowers who are Members of Groups
This table reports ordinary least squares analyses with the dependent variable defined as the difference between DIFF_SPREAD for the first loan and DIFF_SPREAD for the second loan. The loans were initiated between May 31, 2006 and November 6, 2007 and the data includes performance information for these same loans through May 7, 2008. t-statistics from the regressions are in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels respectively.
(1) (2) (3) (4) (5)
QUICKFUNDING -0.006 -0.003 -0.004
(-0.849) (-0.317) (-0.398)
LNDATEBIDCOUNT 0.005 0.002 0.003
(0.521) (0.176) (0.202)
LNSCORE 0.006 0.014* 0.014* 0.012 0.013
(1.057) (1.676) (1.737) (1.552) (1.571)
LNBORROWERAGE -0.004 -0.018 -0.019
(-0.246) (-0.896) (-0.915)
ENDORSEMENTSIND -0.002 -0.000 -0.001
(-0.344) (-0.004) (-0.097)
HOMEOWNERIND -0.004 -0.004 -0.004
(-0.707) (-0.481) (-0.503)
LNAMT -0.011** -0.011** -0.011** -0.010*
(-1.982) (-2.024) (-2.116) (-1.905)
FORPROFIT 0.003 0.018 0.016 0.015
(0.287) (0.830) (0.740) (0.706)
LNGRPLOANS -0.002 0.001 0.001 -0.000
(-0.688) (0.296) (0.151) (-0.003)
FORPROFIT * LNGRPLOANS -0.004 -0.003 -0.003
(-0.786) (-0.689) (-0.677)
GDSTATE 0.045
(0.961)
LISTREV 0.005
(0.594)
Constant -0.033 0.127 0.118 0.077* 0.027
(-0.326) (0.938) (0.866) (1.720) (0.413)
N 278 147 147 147 146
R^2 0.011 0.049 0.054 .043 0.051