Monetary Policy and Individual Investors Risk-Taking Behavior: Evidence from Peer-to-Peer Lending

Transcription

1 Monetary Policy and Individual Investors Risk-Taking Behavior: Evidence from Peer-to-Peer Lending Yongqiang Chu 1 and Xiaoying Deng 2 Current Version: April 2018 Abstract This paper examines whether and how monetary policy affects risk-taking and reachingfor-yield behavior of individual investors in the peer-to-peer lending market. Using data from Prosper.com from 2007 to 2013, we show that easy monetary policy, as measured by lower effective federal funds rates or the quantitative easing programs, induces individual investors to fund riskier loans and to fund loans with higher expected returns. We also find that loans originated during easy monetary policy regimes experience higher ex post default rates. Keywords: Peer-to-Peer Lending, Monetary Policy, Reaching-for-Yield, Risk Taking, QE JEL Code: E52, E58, G21, G28, G41 1 Department of Finance, Belk College of Business, University of North Carolina at Charlotte University City Blvd, Charlotte, NC yongqiang.chu@uncc.edu 2 School of Public Economics and Management, Shanghai University of Finance and Economics,777 Guoding Road, Shanghai,200433, P.R. China, deng.xiaoying@mail.shufe.edu.cn

2 1. Introduction The financial crisis and the subsequent unconventional monetary policy responses have triggered the renewed debate on whether and how monetary policy may induce risk-taking behavior, the so-called risk-taking channel of monetary policy transmission. The existing literature has devoted much effort to detecting and to identifying the underlying mechanism of the risk-taking channel. Most recent papers find that easy monetary policy encourages risk-taking by various financial institutions (e.g., Maddaloni and Peydró 2011; Jiménez at al. 2014; Ioannidou, Ongena, and Peydró 2015; Chodorow-Reich 2014; Choi and Kronlund 2017; Dell Ariccia, Laeven, and Suarez 2017; Di Maggio and Kacperczyk 2017). However, little attention has been paid to understanding the mechanism of the risk-taking channel and to understanding whether easy monetary policy also affects individual investors risk-taking behavior, with the only exception of Lian, Ma, and Wang (2018), who examines individuals risk-taking behavior in an experimental setting. Two factors may be responsible for the lack of empirical evidence on the effect of monetary policy on individual risk-taking. First, the theory on the risk-taking channel of monetary policy almost exclusively relies on mechanisms only relevant to financial institutions to explain the risktaking channel. Some theoretical work suggests that the agency conflict between money managers and their investors can induce the financial institutions to take excess risk when monetary policy is loose (Allen and Gale 2000 and 200; Dimond and Rajan 2012; Morris and Shin 2014). Others argue that easy monetary policy may encourage higher leverage, and hence higher risk in financial institutions (Dell Ariccia, Laeven, and Marquez 2014; Drechsler, Savov, and Schnabl 2015). These theories rely on frictions only matter to institutions and hence have no implication for individual investors. However, as pointed by Lian, Ma, and Wang (2018), individual behavioral biases can also cause individual investors to exhibit similar risk-taking and reaching-for-yield behavior. Second, data on individual investment and risk-taking behavior are limited, making it difficult to examine the effect of monetary policy on individual risk-taking. In this paper, we use the peer-to-peer (P2P) lending market as the laboratory to examine the effect of monetary policy on individual risk-taking behavior. The P2P market provides a perfect setting to study individual risk-taking behavior because the large majority of the lenders are 1

3 individuals in this market. 3 Furthermore, the popular press has long argued that the fast growth of the P2P lending market is fueled by the easy monetary policy after the financial crisis (Light, 2012), the finance academics, however, lags behind in providing robust evidence to show that easy monetary policy does play an important role in the development of the market. More specifically, we focus on the P2P lending market for the following reasons. First, the peer-to-peer lending platform enables the examination of individual behavioral and hence to disentangle different mechanisms that can cause the risk-taking and reaching-for-yield behavior. Financial institutions often face different incentives, ownership structure, and constraints, all of which can affect how they respond to monetary policy shocks. For example, banks face regulatory capital requirements and agency conflict, which can affect banks' risk-taking behavior (Jimenez et al and Dell Ariccia, Laeven, and Suarez 2016). The long-term liability of life insurance companies is often significantly affected by interest rate changes, which in turn affects how insurance companies invest in the low-interest rate environment (Becker and Ivashina 2015). Agency problems faced by mutual fund managers can also trigger the reaching-for-yield behavior (Di Maggio and Kacperczyk 2017). Individual investors have simpler incentives and ownership structure than financial institutions, which allows us to better distinguish between different channels through which monetary policy affects investment policy. Second, the setting and data of the peer-to-peer lending market enable a clean identification of the risk-taking channel of monetary policy. A critical identification challenge in studying the relationship between monetary policy and lending is the potential correlation between monetary policy and demand for credit. Monetary policy can affect the quantity and quality of loan demand either through the interest rate channel or the borrower balance sheet channel. Hence, a correlation between monetary policy and the riskiness of loans may not tell us about lenders' risk-taking or reaching for yield behavior at all. The peer-to-peer lending market allows us to better tackle this problem. First, we have access to both approved and rejected loan requests, which allows us to control for loan demand to a large extent. Second, and more importantly, we have access to all information lenders have, that is, we can control all potential demand-side factors and hence effectively eliminate the possibility that demand-side factors drive the results. Furthermore, the 3 This is true during the sample period we examine in this paper. However, the market becomes dominated by institutional lenders in later periods. 2

4 interest rates on the loans are set by Prosper, which enables us to examine lender behavior given the interest rates on the loan. Specifically, we use the data from Prosper.com (Prosper hereafter), the largest P2P lending platform in the US to conduct our analysis. We use the data from Prosper, instead of other platforms such as Lending Club, because it not only provides data on approved loans but also data on rejected loans, which enables us to better isolate demand-side factors. The data we have access to are the same data Prosper provides to all potential lenders, which include a wide array of loan characteristics, such as loan size, loan term, loan interest rate, loan risk measure, whether the loan is approved, the percent funded, and an even larger set of borrower characteristics, such as income, home ownership, employment status, all existing debt, past credit history, location, and more. Potential lenders rely on the same set of information as we have to make lending decisions. This turns out to be critically important for us to identify the risk-taking channel. In the context of bank lending, lenders often have access to much information unobservable to researchers, and hence it is difficult for researchers to isolate the effect of demand-side factors. In our case, we have exactly the same set of information as lenders do, and hence are able to control all demand-side factors that may affect lending decisions. Empirically, we follow the literature and use the effective federal funds rate to measure conventional monetary policy (e.g. Bernanke and Blinder 1992; Kashyap and Stein 2000; Dell Ariccia, Laeven, and Suarez 2017; and Di Maggio and Kacperczyk 2017) and examine the effect of monetary policy on P2P lending on the Prosper.com (Prosper, hereafter). We first examine the effect of monetary policy on loan approval and find that lower federal funds rates lead to higher approval rates of riskier loans. A one-percentage-point reduction in the effective federal funds rate leads to an increase of the approval rate of risky loans, relative to safe loans, by more than eight percentage points. Consistent with the argument that investors take excess risk to reach for yield, we also find that lower federal funds rates lead to higher approval rates of loans with higher expected returns. These results suggest that lower federal funds rates encourage individual risk-taking and reaching-for-yield. A large part of the sample period is during and after the financial crisis, during which the effective federal funds rate is close to zero and the Federal Reserve implemented the 3

5 Large Scale Asset Purchase (LSAP) or the Quantitative Easing (QE) Programs to conduct monetary policy. A big concern for these unconventional monetary policy programs is precisely that they may encourage excess risk-taking (Chodorow-Reich 2014; Woodford, 2016; Di Maggio and Kacperczyk 2017; Kandrac and Schlusche 2017). We hence also examine the effect of these QE programs on individual risk-taking in the P2P market. To this end, we find that riskier loans originated during the QE programs, relative to safe loans, are more likely to be approved, suggesting that the QE programs also encourage individual risk-taking. We also find that loans with higher expected returns are also more likely to be approved during QE programs, suggesting that the QE programs lead to more individual reaching-for-yield behavior. In addition to examining the effect of monetary policy on loan origination, we also examine ex post loan performance. We find that loans originated when the federal funds rate is low experience higher ex post default rates. The result holds even conditional on risk and other loan and borrower characteristics. Similarly, loans originated during the QE programs also experience higher ex post default rates. This effect compounds upon and magnifies the effect of monetary policy on loan origination, and leads to a much larger effect of monetary policy on risk. To alleviate the concern that monetary policy, or the federal funds rate, may be correlated with other unobservable macroeconomic factors that may affect loan demand, we follow the literature to use the Taylor rule to extract the exogenous component of the federal funds rate (Altunbasa, Gambacortab, and Marques-Ibanezc 2014; Delis, Hasan, and, Mylonids 2017; Dell Ariccia, Laeven, and Suarez 2017). We find that the Taylor rule residual has similar effects on individual risk-taking behavior in the P2P market, suggesting that the results are unlikely to be driven by unobservable macroeconomic factors that also affect individual risk-taking. Our study contributes to serval strands of literature. First, it contributes to the literature on the risk-taking channel of monetary policy transmission. The existing literature overwhelmingly focuses on risk-taking behavior by various financial institutions. To the best of our knowledge, we are the first to empirically examine the effect of monetary policy on individual risk-taking. The existing literature often only relies on agency costs or other mechanisms only relevant to institutions to explain the behavior. Individual investors, on the other hand, are not subject to these agency costs and other financial frictions financial institutions face, and hence their risk-taking 4

6 behavior cannot be driven by those factors. Our results therefore suggest that individual behavior bias or preference may also drive the risk-taking channel of monetary policy transmission. Second, the paper also contributes to the literature on P2P lending. Previous researchers have focused on informational advantages and frictions of the P2P lending market (e.g. La Ferrara 2003, Udry 1994, Hoff and Stiglitz 1990, Besley and Coate 1995; Armendariz and Morduch 2010), which are confirmed in lab and field experiments (Karlan 2005; Giné and Karlan 2010; Feigenberg, Field and Pande 2010; Bryan, Karlan and Zinman 2010; Ravina 2007; Pope and Sydnor 2011, Iyer et al. 2016, Rigbi 2013, Hampshire 2008, Freedman and Jin 2010, Lin, Viswanathan and Prabhala 2013, Paravisini, Rappoport and Ravina 2011). In contrast, our paper is the first to examine the effect of macroeconomic policy on the P2P lending market. The rest of the paper is organized as follows. Section 2 describes the data and sample construction. Section 2 presents some preliminary and graphical analysis of the effect of monetary policy on risk-taking. Section 4 presents the results of ex ante risk-taking. Section 5 presents the results of ex post loan performance. Section 6 performs additional and robustness analyses. Section 7 concludes. 2. Data and Sample Construction To examine how monetary policy affects individual investors risk-taking and reachingfor-yield in the P2P lending market, we use data of Prosper.com from We end the sample in 2013 because of the need of measuring the performance of the loans, most of which have a maturity of three years. A second reason is that institutions dominated the market after 2013.As the first and by far the largest peer-to-peer lending platform in the US, Prosper.com offers fixed rate, unsecured, fully amortized loans. One thing to note is that the Prosper data contain all borrower and loan information available to the investors, which removes the possibility that investors making lending decisions based on factors unobservable to the econometrician, a common and almost unavoidable problem in the bank lending literature. 4 We do not use the lending club data because the lending club data do not have sufficient information on loan request that are not funded, which is critical in identifying risk-taking and reaching-for-yield as shown in the later discussion. 5

7 We use the following variables to measure loan outcome. The first one is Approval, which equals one if the loan request is funded, and zero otherwise. A loan request is funded if the loan request is more than 70% funded by the deadline, otherwise the loan request is rejected. We also use a continuous variable, Percent Funded, defined as the percent of loan request funded for both approved and rejected loans. We use two different measures of borrower risk. The first one is Prosper Score, which, according to Prosper, is the probability of a loan going bad, where bad is the probability of going 60+ days past due within the first twelve months from the date of loan origination. The Prosper Score ranges from 0-11, in which 0 represents loan requests with the highest risk, and 11 represents loan requests with the lowest risk. We consider this measure as a forward-looking loanlevel risk measure. We further use the borrower s ScoreX plus credit rating as an alternative measure of borrower risk. 5 Propser reports the ScoreX rating in different-sized bins. We transform these bins to numerical values from 0 to 10, with 0 assigned to the bin with the ScoreX plus rating below 600, and with 10 assigned to the bin with the ScoreX plus rating above 778. We denote the transformed ScoreX plus credit rating as ScoreX Rating. A higher ScoreX Rating means lower risk of the borrower. We consider this measure as a backward-looking borrower level risk measure. To capture investors reaching-for-yield, we use Prosper's estimated return, Return, as the main measure of investors expected returns. After reviewing the credit profile of the potential borrower, Prosper determines the interest rate, the servicing fee, and estimates the expected loss on the loan if the loan is funded, and the estimated return is equal to interest rate minus the sum of servicing fee and expected loss. The estimated return is the expected return lenders expect to earn on the loan. In our analysis of ex post loan performance, we define the indicator variable, Default, which equals zero if the status of the loan is either completed, current or final payment in progress, and equals one otherwise (loan status being Chargedoff, Defaulted, or Past Due). We also try categorizing those less than 30 days, 60 days, or 90 days as current, and still find similar results. In the baseline analysis, we follow Dell Ariccia, Laeven, and Suarez (2017) and use the effective federal funds rate, FF, to measure monetary policy. In robustness checks, we also use the 5 We do not use the FICO score because the FICO score was not introduced or reported by Prosper until

8 Taylor rule residual as a measure of monetary policy to address the potential endogeneity problems. We include a broad set of control variables. Following the existing literature, to control loan characteristics, we include Listing Amount, Listing Term, and Listing Payment. To control individual-level characteristics, we include Monthly Income, DTI, Months Employed, Homeowners, Prior Prosper Loan, Monthly Debt, Ten Year Public Record, One Year Public Record, Seven Year Credit Lines, Six Month Inquiries, Total Inquiries, Amount Delinquent, Current Credit Lines, Open Credit Lines, Bank Card Utilization, Total Open Revolving, Installment Balance, Real Estate Balance, Real Estate Payment, Revolving Balance, Revolving Available Percent, Current Delinquency, Seven Year Delinquency. The Appendix lists all the variables used in this paper s empirical analyses. The summary statistics of main variables used in this paper are presented in Table 1. Among all loan requests received by Prosper, only about 35% are eventually funded. The average requested loan amount is around $8,000 and the average loan term is 37.8 months, that is, a little over three years. The average Prosper Score is 2.41 and the average SocreX Rating is The average estimated return is 4%. [Insert Table 1 about here] 3. Graphical Evidence Before we present the formal analysis on the effect of monetary policy on risk-taking in the P2P market, it is worthwhile to show univariate results first. We first split the sample period according to the effective federal funds rate into five bins (less than 0.25%, between 0.25% and 1%, between 1% and 2%, between 2% and 3%, and higher than 3%), and then calculate the loan approval rates in each federal funds rate bins for risky (Prosper Score less than median) and safer loans (Prosper Score greater than median) separately. The results are presented in Panel A of Figure 2. It is evident from the figure that the approval rates for risky loans are very low (less than 10%) except when the effective federal funds rate is extremely low (less than 25 basis points), during which the approval rate for risky loans is more than 70%. In contrast, for safe loan requests, the approval rates are lower when the effective federal funds rate is low. We then split the sample period into non-qe and QE periods, and also calculate the approval rates for risky and safe loan requests during these different time periods. The results are presented in Panel B of Figure 2. The approval rates of risky loan requests are much higher during 7

9 QE periods and the difference in approval rates for safe loans is much smaller. The difference of the approval rates of risky loans in different monetary policy regimes can also be driven by demand-side factors. If loan requests during loose monetary policy regimes are riskier, and the lenders funding decisions are random, more risk loans will get funded during loose monetary policy regimes. In this case, the effect of monetary policy on the appear-to-be risk-taking is driven by changes in demand, instead of by lenders incentives for reaching-for-yield. To assess to what extent this might be a problem, we examine the risk characteristics of all loan requests in different monetary policy regimes. We again split the sample period according to the effective federal funds rate into five bins and then calculate the means of Prosper Score, ScoreX Rating, Debt-to-Income Ratio, and the Number of Delinquencies of all loan requests (both approved and rejected). The results are presented in Panel A of Figure 2. The average Prosper Score and ScoreX Rating are higher and the Debt-to-Income Ratio and the Number of Delinquencies are lower of loan requests made when the federal funds rate is low, suggesting that the loan requests made during easy monetary policy regimes are in fact safer. We then also calculate the means of these measures for different QE periods, and the results are presented in Panel B of Figure 2. Again, it suggests that the loan requests made during the QE programs are likely to be safer than those made during normal times. Overall, these figures suggest that the higher approval rates of risky loans during easy monetary policy regimes are unlikely to be the result of a shift in demand. 4. Ex Ante Risk-Taking and Reaching-for-Yield 3.1 Monetary policy and risk-taking To identify how monetary policy affects risk-taking, we follow the literature (Dell Ariccia, Laeven, and Suarez, 2017 and Jiménez et al. 2014) and estimate the following model: YY iiii = αα tt + ββffff tt RRRRRRRR iiii + δδrrrrrrrr iiii + γγzz iiii + Fixed Effects + εε iiii (1) where YY iiii is a measure of loan application outcome, which can be either Approval or the natural logarithm of Percent Funded, αα tt is time fixed effects (year-month), FFFF tt is the effective federal funds rate, RRRRRRRR iiii is either the Prosper Score or ScoreX Rating, ZZ iiii is a set of other loan and borrower characteristics. We also include city, city year-month, and borrower fixed effects 8

10 in different specifications. Under Equation (1), ββ captures the effect of monetary policy on risk taking. If easy monetary policy encourages risk-taking by individual investors in the P2P lending market, ββ should be positive. We cluster the standard errors by year-month. A common challenge in identifying the supply-side determinants of credit supply is to separate the effect of demand-side factors because potentially unobservable demand-side factors can be correlated with the supply side factors. In this regard, the Prosper data enable us to mitigate this concern. First, we observe both approved and rejected loan requests instead of just approved loans, which allows us to make analysis conditional on observable loan demand. Second, using the Prosper data, we have access to the same set of borrower and loan characteristics as those available to potential lenders, that is, there are no demand-side factors that are observable to the lenders but not observable to the econometrician. We are therefore able to better control demandside factors using all these borrower and loan characteristics. Third, many borrowers on the Prosper platform borrow multiple times, allowing us to control for borrower heterogeneity using borrower fixed effects. We first show the results of estimating Equation (1) using Prosper Score as the risk measure in Panel A of Table 2. The first four columns present results with Approval as the dependent variable. We only include year-month fixed effects in Column (1), and then further include city fixed effect to control for location-specific time-invariant demand factors. To control for time-varying location-specific demand-side factors, we include city year-month fixed effects in Column (3). Finally, we include borrower fixed effects to control borrower-specific factors. In fact, unless the lenders consider factors outside those provided by Prosper, this is not really necessary because we observe all borrower characteristics the lenders observe. In all columns, the coefficients on FF Prosper Score are positive and statistically significant, suggesting that riskier loan requests, that is, requests with lower Prosper Score, are more likely to be approved when the federal funds rate is low. The result is therefore consistent with the hypothesis that easy monetary policy encourages risk-taking by individual lenders in the P2P lending market. The coefficient estimates in Columns (1) - (4) are all greater than 0.02 or 2%. To gauge the economic magnitude, we consider the effect of reducing the effective federal funds rate by one percentage point on two loans with Prosper Score of four (the 25 th Percentile) and eight (the 75 th 9

11 Percentile). The probability of approving a riskier loan (Prosper Score of four) when the effective federal funds rate is lower is eight percent (2% 4) higher than the probability of approving a safer loan (Prosper Score of four) when the effective federal funds rate is higher. Given that only about 35% of the loan requests are approved, the economic magnitude is large. In Columns (5) - (8), we present the results with Log Percent Funded as the dependent variable. Consistent with the risk-taking hypothesis, the coefficients on FF Prosper Score are again positive and statistically significant. It is worthwhile to point out that the adjusted R-squares of these regressions are all very high (more than 70%), and the unreported unadjusted R-squares are even higher (more than 90%). This is consistent with the fact that we are able to observe and hence control almost all factors lenders can observe, 6 and hence leaves very little room for omitted variables bias. To ensure that the results are robust to alternative risk measures, we then use the ScoreX Rating as the risk measure. The results are reported in Panel B of Table 2. Similar to the results presented in Panel A, the coefficient estimates on FF ScoreX Rating are all positive and statistically significant, suggesting the loan requests from riskier borrowers are likely to receive more funding. Overall, the results in Table 2 suggest that easy monetary policy does induce individual investors in the P2P lending market to take more risk, consistent with findings on risk-taking by financial institutions. However, while the risk-taking channel of monetary policy found in financial institutions may be driven by agency problems or other financial frictions institutions face, the effect we document here is more likely to be only from individual behavioral bias. [Insert Table 2 about here] 3.2 Monetary Policy and Reaching-for-Yield Next, we proceed to examine whether individual investors risk-taking behavior is driven by their incentives for reaching-for-yield. To this end, we first estimate the following: YY iiii = αα tt + ββffff tt RRRRRRRRRRRR iiii + δδδδδδδδδδδδδδ iiii + γγzz iiii + Fixed Effects + εε iiii (2), 6 We are not getting 100% R-squares because we are only controlling linear terms, and higherorder terms and/or interaction terms are left unexplained. 10

12 where Return is the estimated return provided by Prosper, which is the difference between the stated interest rates on the loan and the sum of servicing fees and expected credit loss. We consider Return as our measure of lenders expected return from the loan. The specification is made possible by the fact that the interest rate and servicing fees are set by the platform, instead of the lenders. Lenders in the P2P lending market takes the expected return as given. The results of estimating Equation (2) are presented in Table 3. The format of the table is exactly the same as those of Table 2. [Insert Table 3 about here] In all columns, the coefficient estimates on FF Return are all negative and statistically significant, suggesting that loans with higher estimated returns are more likely to be approved when the effective federal funds rate is lower. The coefficient estimates are around 0.2. A onepercentage point decrease of the effective federal funds rate will cause the approval rate on a higher return loan (estimated return of 12%, the 75 th Percentile) to increase by 1.2 percentage more than the increase in the approval rate of a lower return loan (estimated return of 6%, the 25 th Percentile). The results suggest that individual investors reach for yield in the P2P market especially when the monetary policy is loose. To further ascertain whether the risk-taking behavior is entirely driven by the reaching-foryield incentives, we then combine Equations (1) and (2) to estimate the following: YY iiii = αα tt + ββ 1 FFFF tt RRRRRRRRRRRR iiii + δδ 1 RRRRRRRRRRRR iiii + ββ 2 FFFF tt RRRRRRRR iiii + δδ 2 RRRRRRRR iiii + γγzz iiii + Fixed Effects + εε iiii (3). The results are presented in Table 4, with risk measured by Prosper Score in Panel A and by Scorex Rating in Panel B. In both panels, while the coefficient estimates on FF Return remain negative and statistically significant, the coefficients on FF Risk are either positive or statistically insignificant, suggesting that the risk-taking incentives triggered by easy monetary policy are entirely driven by individuals reaching-for-yield incentives. [Insert Table 4 about here] 3.3 The Effects of the Quantitative Easing Programs on Risk Taking One potential problem with the above results is that much of the sample period is during or after the financial crisis, during which the federal fund rates are low and have little variation. In fact, the federal funds rate is close to zero for most of the post-crisis period, and 11

13 monetary policy is conducted through the Large Scale Asset Purchase (LSAP) programs, that is, the quantitative easing (QE) programs. To the extent that monetary policy affects individual risktaking, the QE programs should also affect individual risk-taking. As such, we examine how the federal reserve's quantitative easing programs affect P2P investors' risk-taking and reaching-for-yield behavior. Figure 3 provides a timeline of the various Fed LSAP programs. QE1 lasted from late November 2008 until March 2010, and QE2 was first announced in mid-august 2010 and ran from November 2010 to June QE3 was announced in September In between QE2 and QE3, the federal reserve also implemented the maturity extension program (also called operational twist). [Insert Figure 3 about here] We follow Di Maggio, Kermani, and Palmer (2016) to first create dummy variables, QE1, QE2, MEP, and QE3, which equal one if the time period is during those programs, and zero otherwise. We then examine the effect of these QE programs on risk-taking using the following specification: YY iiii = αα tt + ββ 1 QQQQ1 RRRRRRRR iiii + ββ 2 QQQQ2 RRRRRRRR iiii + ββ MMMMMM MMMMMM RRRRRRRR iiii + ββ 3 QQQQ3 RRRRRRRR iiii + δδrrrrrrrr iiii + γγzz iiii + Fixed Effects + εε iiii (4) Under this specification, the ββ s capture the effect of the QE programs on risk taking. If easy monetary policy does encourage risk-taking, we expect the ββ s to be negative. The results of estimating Equation (4) are presented in Table 5, with Panel A using the Prosper Score as the risk measure, and Panel B using the ScoreX Rating as the risk measure. Consistent with the conjecture that easy monetary policy encourages risk-taking, the coefficient estimates on the interaction terms between the QE programs and the risk measures are all negative and statistically significant, riskier loan requests, relative to safer one, are more likely to be approved when the QE programs are in place. The economic magnitudes of these estimates are also large. QE1 can increase the approval rates of a loan with the prosper score of four (25 th percentile), relative to a loan with the prosper score of eight (75 th percentile), by more than 12 percentage points with the most conservative estimates (column (4) of Panel A). The effects of other programs are even greater. The results are consistent with the larger effects shown in Figure 12

14 1. [Insert Table 5 about here] 3.4 The Effects of the Quantitative Easing Programs on Reaching-for-Yield We then also examine how the quantitative easing programs affect individual investors reaching-for-yield incentives by replacing the risk measures in Equation (4) with the estimated return, that is, YY iiii = αα tt + ββ 1 QQQQ1 RRRRRRRRRRRR iiii + ββ 2 QQQQ2 RRRRRRRRRRRR iiii + ββ MMMMMM MMMMMM RRRRRRRRRRRR iiii + ββ 3 QQQQ3 RRRRRRRRRRRR iiii + δδrrrrrrrrrrrr iiii + γγzz iiii + Fixed Effects + εε iiii (5) Under this specification, the ββ s should be positive if the QE programs encourage reaching-foryield behavior, that is, loan requests with higher estimated returns are more likely to be approved during the QE programs. The results of estimating Equation (5) are presented in Table 6. Consistent with our conjecture, the coefficient estimates on the interaction terms between the QE programs and the estimated return are all positive and mostly statistically significant (except for two for the maturity extension program). The results, therefore, suggest that these unconventional monetary policy programs also encourage reaching-for-yield. [Insert Table 6 about here] We then examine the effect of the QE programs on risk-taking and reaching-for-yield simultaneously, that is, including both the interaction terms between QE programs and the risk measures and the interaction terms between QE programs and the estimated return in the regressions. The results are presented in Table 7, with Panel A using Prosper Score as the risk measure and Panel B using the ScoreX Rating as the risk measure. In all these results, coefficients on some of the interaction terms between QE programs and estimated returns remain positive and statistically significant. On the other hand, coefficients on the interaction terms between QE programs and the risk measures either change signs or become statistically insignificant. These results again suggest that individuals risk-taking behavior is driven by their reaching-for-yield incentives when monetary policy is loose. 13

15 [Insert Table 7 about here] 5. Ex Post Performance Next, we move to examine the effect of monetary policy on ex post loan performance. This analysis serves two purposes. First, the ex ante risk measures we use, namely, Prosper Score and ScoreX Rating, are summary measures, and may not capture all risk. Examining ex post loan performance therefore provides a complete account of the effect of monetary policy on risk-taking. Second, examining ex post loan performance allows us to assess the consequence of investors ex ante risk-taking and reaching-for-yield behavior. 4.1 Monetary Policy and Loan Default Empirically, we first examine the effect of the effective federal funds rate at the time of loan origination on ex post default. To this end, we follow Dell Ariccia, Laeven, and Suarez (2017) and Di Maggio and Kacperczyk (2017) and estimate the following specification on all approved loans, DD iiii = ββffff tt + δδδδδδδδδδ iiii + γγzz iiii + θθxx tt + Fixed Effects + εε iiii (6), where DD iiii is the indicator of loan default for loan i originated at time t, which equals one if the loan status is defaulted, chargedoff, or past due, and zero otherwise. Note that under this specification, we cannot include time fixed effects because they will subsume the federal funds rate. Instead, we include macroeconomic variables XX tt to mitigate the concern that the effect may be driven by other macroeconomic factors. However, we acknowledge that the identification in Equation (6) is not as clean as those in the previous sections because we cannot control all macroeconomic factors with time fixed effects. It is therefore possible that the results may be driven by unobservable demand-side factors. We also control for risk and all other loan and characteristics to examine the impact of monetary policy on ex post performance beyond those captured by the ex ante risk measures. We include city fixed effects or borrower fixed effects to control for additional demand side factors. [Insert Table 8 about here] The results of estimating Equation (6) are presented in Table 8, with Columns (1) and (2) using Prosper Score as the risk measure and Columns (3) and (4) using ScoreX Rating as the risk measure. The coefficient estimates of ββ are all negative and statistically significant, suggesting 14

16 that loans originated when the federal funds rate is low experience higher ex post default rates, even when after controlling for risk and all other loan and borrower characteristics constant. A one percentage point reduction of the federal funds rate can lead to a more than one percentage point increase in the default probability. The effect is at least similar, if not larger than, the effect of increasing the Prosper Score or the ScoreX Rating by one notch. 4.2 QE Programs and Loan Default Next, we also examine the effect of the QE programs on ex post loan performance by estimating the following specification, DD iiii = ββ 1 QQQQ1 + ββ 2 QQQQ2 + ββ MMMMMM MMMMMM + ββ 3 QQQQ3 + δδrrrrrrrr iiii + γγzz iiii + θθxx tt +Fixed Effects + εε iiii (7). As in Equation (6), we also cannot control for time-fixed effects, and therefore cannot rule out the possibility that some unobservable demand-side factors correlated with macroeconomic conditions drive the results. The results are presented in Table 9, with Columns (1) and (2) using Prosper Score as the risk measure and Columns (3) and (4) using ScoreX Rating as the risk measure. In all columns, the coefficient estimates for QE1 are all positive and three out of the four estimates are statistically significant. The coefficient estimates for QE2 and MEP are all positive and statistically significant. These results suggest that loans originated during QE1, QE2, and MEP programs have higher ex post default rates than loans with similar ex ante risk measures and borrower and loan characteristics but originated during non-qe times. The economic magnitudes are also large. In some cases, the effects of the QE programs are more than double the effect of improving the risk measures by one notch. [Insert Table 9 about here] Different from the coefficients on QE1, QE2, and MEP, the coefficient estimates on QE3 are all negative and statistically significant with city fixed effects (but not significant with borrower fixed effects). Several factors may be responsible for this result. First, after QE1, QE2, and MEP, the interest rates of the economy are already very low and QE3 probably had little impact on individuals investment opportunity set. Second, investors probably anticipated the exit of the QE programs after QE3. Third, it may be driven by demand side factors because economic conditions during QE3 have already significantly improved. As we discussed above, the 15

17 specification in Equation (7) cannot rule out the possibility that the results may be driven by unobservable demand-side factors. 6. Robustness Checks 6.1 Subsample Analysis In the results presented above, we pool together time periods covering both conventional and unconventional monetary policy regimes. In this subsection, we examine the effect of conventional and unconventional monetary policy separately. First, we examine the effect of conventional monetary policy in the pre-qe period (January May 2008). The results are presented in Panel A of Table 10, with Columns (1) (3) using Prosper Score as the risk measure and with Columns (4) (6) using ScoreX Rating as the risk measure. For the results on loan approval and loan percent funded, the coefficient estimates on the interaction term between the federal funds rate and the risk measures are all positive and statistically significant, suggesting that riskier loan are more likely to be approved when the federal funds rate is low. For the results on loan default, the coefficient estimates on the federal funds rate all both negative and statistically significant, suggesting that loans originated when the federal funds rate is low are more likely to default ex post. Next, we examine the effect of the QE programs only during the QE period (June 2008 December 2013). The results are presented in Panel B of Table 10, with Columns (1) (3) using Prosper Score as the risk measure and with Columns (4) (6) using ScoreX Rating as the risk measure. For the results on loan approval and loan percent funded, the coefficient estimates on the interaction term between the QE programs and the risk measures are all negative and statistically significant, suggesting that riskier loan are more likely to be approved when the QE programs are in place. For the results on loan default, the coefficient estimates on the QE1, QE2, MEP are all positive and statistically significant, suggesting that loans originated during the QE programs are more likely to default ex post. Similar to the results presented in Table 9, the coefficient estimates on QE3 are negative and statistically significant, suggesting that loans originated during QE3 experience lower default rates. [Insert Table 10 about here] 6.2 Using the Taylor Rule Residual to Measure Monetary Policy Shocks 16

18 One concern for the analysis is that monetary policy is endogenously determined and may be correlated with past and future economic conditions that may affect the quantity and riskiness of loan demand. To this end, we follow the literature to use the Taylor rule residual to capture the exogenous component of the federal funds rate and re-examine the effect of the Taylor rule residual on risk-taking in the P2P market. Specifically, we run rolling regressions of the federal funds rate on the deviation of CPI inflation from the 2% target rate and the difference between the actual and potential GDP growth rates, and then calculate the residuals from those regressions. We then replace the effective federal funds rate with the Taylor rule residual in the regressions above. The results are reported in Table 11, with Columns (1) (3) using Prosper Score as the risk measure and with Columns (4) (6) using ScoreX Rating as the risk measure. For the results on loan approval and percent funded, consistent with the results presented above, the coefficient estimates on the interaction terms between risk measures and the Taylor rule residual are all positive and statistically significant. For the results on loan default, the coefficient estimates on Taylor rule residual are both negative and statistically significant, suggesting that loans originated when the Taylor rule residual is low experience higher default rates ex post. [Insert Table 11 about here] Overall, the results in Table 11 suggest that the baseline results are unlikely to be driven by the endogeneity of monetary policy because the Taylor rule residual is likely to capture the exogenous component of monetary policy. 7. Conclusions This paper examines whether and how monetary policy affects risk-taking and reaching for yield behavior by individual investors in the peer-to-peer lending market. The setting and data of the peer-to-peer lending market enable a cleaner identification by controlling the demand side factors. Using the data of Prosper.com from 2007 to 2013, we show that easy monetary policy induces investors in the peer-to-peer lending market to fund riskier loans and to fund loans with higher expected returns. 17

19 References Allen, F., Gale, D., Bubbles and Crises. The Economic Journal, 110(460), Allen, F.,Gale, D.,2004. Asset Price Bubbles and Monetary Policy. Global Governance and Financial Crises, edited by Desai, M., Said. Y., New York and London: Routledge, Chapter 3, Altunbasa, Y., Gambacortab, L., Marques-Ibanezc, D., Does Monetary Policy Affect Bank Risk? International Journal of Central Banking, 10(1): Armendariz, B., Morduch, J., 2010.The Economics of Microfinance. Second Edition. MIT Press. Becker, B., Ivashina, V., Reaching for Yield in the Bond Market. Journal of Finance, 70: Bernanke, B.S., Blinder, A.S., The Federal Funds Rate and the Channels of Monetary Transmission. American Economic Review, 82(4): Besley, T., Coate, S., Group Lending, Repayment Incentives and Social Collateral. Journal of Development Economics, 46(1): Bryan, G., Karlan, D., Zinman, J., Referrals: Peer Screening and Enforcement in a Consumer Credit Field Experiment. American Economic Journal: Microeconomics, 7(3): Chodorow-Reich, G., Effects of Unconventional Monetary Policy on Financial Institutions. Brookings Papers on Economic Activity, 45(1): Choi, J., Kronlund, M., Reaching for Yield in Corporate Bond Mutual Funds. Review of Financial Studies, 31(5): Delis, M.D., Hasan, I., Mylonidis, N., The Risk Taking Channel of Monetary Policy in the US: Evidence from Corporate Loan Data. Journal of Money, Credit and Banking, 49(1): DellʼAriccia, G., Laeven, L., Marquez, R., Real Interest Rates, Leverage, and Bank Risk- Taking. Journal of Economic Theory, 149 (1): Dell'Ariccia, G., Laeven, L., Suarez, G. A., Bank Leverage and Monetary Policy's Risk Taking Channel: Evidence from the United States. Journal of Finance, 72: Di Maggio, M., Kacperczyk, M., The Unintended Consequences of the Zero Lower Bound Policy. Journal of Financial Economics, 123:

20 Di Maggio, M., Kermani, A., Palmer C., How Quantitative Easing Works: Evidence on the Refinancing Channel. NBER Working Paper No Diamond, D.W., Rajan, R.G., Illiquid Banks, Financial Stability, and Interest Rate Policy. Journal of Political Economy, 120(3): Drechsler, I., Savov, A., Schnabl, P., A Model of Monetary Policy and Risk Premia. Journal of Finance, 73: Freedman, S.M., Jin, G.Z., Learning by Doing with Asymmetric Information: Evidence from Prosper.com. NBER Working Paper No Feigenberg, B., Field, E., Pande, R., Building Social Capital through Microfinance. NBER Working Paper No Ferrara, E.L., Kin Groups and Reciprocity: A Model of Credit Transactions in Ghana American Economic Review, 93(5): Hampshire, R., Group Reputation Effects in Peer-to-Peer Lending Markets: An Empirical Analysis from a Principle-Agent Perspective. mimeo. Hoff, K., Stiglitz, J.E., Introduction: Imperfect Information and Rural Credit Markets Puzzles and Policy Perspectives. The World Bank Economic Review, 4(3): Giné, X., Jakiela, P., Karlan, D., Morduch, J., Microfinance Games. American Economic Journal: Applied Economics, 2(3): Kandrac, J., Schlusche, B., Quantitative Easing and Bank Risk Taking: Evidence from Lending. Finance and Economics Discussion Series , Board of Governors of the Federal Reserve System (U.S.). Karlan, D., Using Experimental Economics to Measure Social Capital and Predict Real Financial Decisions. American Economic Review, 95(5): Kashyap, A.K., Stein, J.C., What do a Million Observations on Banks Say about the Transmission of Monetary Policy? American Economic Review, 90(3): Ioannidou, V., Ongena, S., Peydró, J.L., Monetary Policy, Risk-Taking, and Pricing: Evidence from a Quasi-natural Experiment. Review of Finance, 19(1): Iyer, R., Khwaja, A.I., Luttmer, E.F.P., Shue K., Screening Peers Softly: Inferring the Quality of Small Borrowers. Management Science, 62(6): Jimenez, G., Ongena, S., Peydró, J.-L., Saurina, J., Hazardous Times for Monetary Policy: 19

21 What do Twenty-three Million Bank Loans Say about the Effects of Monetary Policy on Credit Risk-taking? Econometrica, 82, Light, J Would You Lend Money to These People? Wall Street Journal. April 13. Lin, M., Viswanathan, S., R. Prabhala, N., Judging Borrowers by the Company They Keep: Social Networks and Adverse Selection in Online Peer-to-Peer Lending. Management Science, 59(1): Maddaloni, A., Peydró, J.L., Bank Risk-taking, Securitization, Supervision, and Low Interest Rates: Evidence from the Euro-area and the US Lending Standards. Review of Financial Studies, 24(6): Morris, S., Shin, H.S., Risk-taking Channel of Monetary Policy: A Global Game Approach. Working Paper. Paravisini, D., Rappoport, V., Ravina, E., Risk Aversion and Wealth: Evidence from Personto-Person Lending Portfolios. Management Science,63(2): Pope, D.G., Sydnor, J.R., What s in a Picture? Evidence of Discrimination from Prosper.com. Journal of Human Resources, 46(1): Ravina, E., Love & Loans: The Effect of Beauty and Personal Characteristics in Credit Markets. Working Paper. Rigbi, O., 2013.The Effects of Usury Laws: Evidence from the Online Loan Market. Review of Economics and Statistics,95(4): Udry, C., Risk and Insurance in a Rural Credit Market: An Empirical Investigation in Northern Nigeria. Review of Economic Studies, 61(3): Woodford, M., Quantitative Easing and Financial Stability. NBER Working Paper No

22 Figure 1 Monetary Policy, Loan Riskiness, and Loan Approval Rates This figure reports the approval rates of risky (prosper score less than the sample median) and safe loan requests under different monetary policy regimes. Panel A: Effective federal funds rate FF< <=FF<=1 1<FF<=2 2<FF<=3 FF>3 Risky Safe Panel B: Quantitative easing programs Normal QE1 QE2 MEP QE3 Risky Safe 21

23 Figure 2 Monetary Policy and Risk of Loan Requests This figure reports the means of Prosper Score, ScoreX Rating, Debt-to-Income Ratio, and the Number of Delinquencies of all loan requests under different monetary policy regimes. Panel A: Federal Funds Rate and Risk of Loan Requests Prosper Score ScoreX Rating DTI Deliquencies FF< <=FF<=1 1<FF<=2 2<FF<=3 FF>3 Panel B: QE Programs and Risk of Loan Requests Prosper Score ScoreX Rating DTI Deliquencies Normal QE1 QE2 MEP QE3 22

24 Figure 3 Timeline of Quantitative Easing Program 23