Prosper provides unsecured, 36-month, fixed-rate personal loans ranging from $1,000 to $25,000. Borrowers and lenders must be legal U.S. residents with a valid domestic address and bank account. The members true identities, addresses, and other contact information are never publicly disclosed by the site. For privacy, the borrower is prohibited from releasing that information to lenders. However, the borrower's state of residence is displayed on the listing Borrowers create a listing requesting a loan for a specified amount and a maximum interest rate they are willing to accept(borrower's max rate). They set the duration of the listing(up to 14 days), specify the category of use(Debt Consolidation, Home Improvement, Business, etc. ) write a brief description, and, optionally, include an image of themselves. Each listing displays financial information about the borrower including debt-to-income ratio, income, occupation, employment status, credit grade(40-point bands of the borrowers Experian credit score), total credit lines open, number of credit inquiries in the last six months, current delinquencies, and home ownership status. Prosper posts aggregate historical data on the default and interest rates grouped by credit grade After some institutional and legal restructuring in 2008, Prosper started to collaborate with a national bank, who became the legal originator for all Prosper loans. This arrange ment makes it possible for the site to utilize a 1978 Supreme Court decision allowing all Prosper borrowers to avoid their individual state's usury law and face a uniform fixed Lenders(bidders)search through the listings for loan requests that they want to bid n,see Figure 1. The funding mechanism is a descending price uniform share auct Lenders'bids are made up of two components: amount of bid($50 minimum) and the lowest interest rate that they are willing to accept. The bidding process is proxy bidding, similar to that on eBay. Each bid is considered independent, so a lender may bid multiple times with potentially different interest rates. These price-quantity pairs form the supply curve of available funds for this loan request. The auction is partially open; lenders always see the number of bids and the quantity of money pledged of each bid. The interest rate submitted by the lenders is only shown for losing bids; accordingly, before the loan is fully funded, lenders only see the borrower's maximum rate Successfully submitted bids cannot be rescinded. Given that this platform is a collection of individualized markets for each individual loan each market clears when the amount of pledged funds is at least as large as the requested amount, with the interest rate The market description in this section refers to market conditions and mechanisms that were in place when the data was collected. Some policy and regulation changes have since occurred, making the current market slightly different. 6
eBay of loans.11 Prosper provides unsecured, 36-month, fixed-rate personal loans ranging from $1,000 to $25,000. Borrowers and lenders must be legal U.S. residents with a valid domestic address and bank account. The members’ true identities, addresses, and other contact information are never publicly disclosed by the site. For privacy, the borrower is prohibited from releasing that information to lenders. However, the borrower’s state of residence is displayed on the listing. Borrowers create a listing requesting a loan for a specified amount and a maximum interest rate they are willing to accept (borrower’s max rate). They set the duration of the listing (up to 14 days), specify the category of use (Debt Consolidation, Home Improvement, Business, etc.), write a brief description, and, optionally, include an image of themselves. Each listing displays financial information about the borrower including debt-to-income ratio, income, occupation, employment status, credit grade (40-point bands of the borrower’s Experian credit score), total credit lines open, number of credit inquiries in the last six months, current delinquencies, and home ownership status. Prosper posts aggregate historical data on the default and interest rates grouped by credit grade. After some institutional and legal restructuring in 2008, Prosper started to collaborate with a national bank, who became the legal originator for all Prosper loans. This arrangement makes it possible for the site to utilize a 1978 Supreme Court decision allowing all Prosper borrowers to avoid their individual state’s usury law and face a uniform fixed legal maximum borrower rate of 36%. Lenders (bidders) search through the listings for loan requests that they want to bid on, see Figure 1. The funding mechanism is a descending price uniform share auction. Lenders’ bids are made up of two components: amount of bid ($50 minimum) and the lowest interest rate that they are willing to accept. The bidding process is proxy bidding, similar to that on eBay. Each bid is considered independent, so a lender may bid multiple times with potentially different interest rates. These price-quantity pairs form the supply curve of available funds for this loan request. The auction is partially open; lenders always see the number of bids and the quantity of money pledged of each bid. The interest rate submitted by the lenders is only shown for losing bids; accordingly, before the loan is fully funded, lenders only see the borrower’s maximum rate. Successfully submitted bids cannot be rescinded. Given that this platform is a collection of individualized markets for each individual loan, each market clears when the amount of pledged funds is at least as large as the requested amount, with the interest rate 11The market description in this section refers to market conditions and mechanisms that were in place when the data was collected. Some policy and regulation changes have since occurred, making the current market slightly different. 6
determined by the auction. In the case of ties, bids placed earlier take precedence over later bids. When the auction closes, the bids are sorted by bid interest rate. The bids with the lowest rates are bundled until the total loan amount has been reached and are then combined into a single loan. Each winning lender receives the same interest rate, which is determined by the marginal losing or last winning bid, depending on the auction Winning bids are either fully or partially participating in the loan; a partially participating bid means that the lender is allocated a smaller share of loan than his or her quantity bid If the loan request is not fully funded by the listings close date, the listing is closed and dismissed. Successful loan requests get further review by the site to initiate the needed legal documentation for the loan to be originated. a borrower who defaults on his or her Prosper loan is barred from using the site again 3. Data This paper uses publicly released data containing all loan requests, with their accompa nying bids, using the open auction format posted by borrowers with FICO credit scores greater than or equal to 560 that were active from January to October 2008. Prior to late 2008, any legal resident of the United States could be a lender on the site and legal residents of every state, except South Dakota, could be borrowers. During the period in my sample, 42, 657 loan requests and 2,022,910 bids were posted. A total of 9, 624 listings and 1,688, 531 bids were for listings that were fully funded. The publicly available characteristics of the borrowers are: debt-to-income ratio(henceforth DIR, the variable is top coded at 10.1), credit grade, homeowner status, whether the borrower is in a Prosper group, and state of residence Klafft(2008b)confirm that the rules that apply in the traditional banking system apply to P2P lending as well; credit grade and the dir are the two most important hard financial variables in determining the financial outcomes. The site provides members the ability to join groups that are designed to develop and foster a community of lenders and borrowers akin to what occurs in relationship lending, contract enforcement via reputation, and peer effects. Agrawal et al. (2011)and Lin et al. (2013)find that social connections seem to reduce market frictions. Therefore, i collect data on these on-site social networks. In the loan-level analysis, the In group variable indicates whether a borrower is in a Prosper group; and in the bid-level analysis, the In group variable indicate whether that particular lender is in the borrower's specific In addition to the variables collected directly from Prosper, I construct two variables Total Competition and Credit grade Competition -to measure the competition that each 12This Prosper group never quite took out and are not widely used
determined by the auction. In the case of ties, bids placed earlier take precedence over later bids. When the auction closes, the bids are sorted by bid interest rate. The bids with the lowest rates are bundled until the total loan amount has been reached and are then combined into a single loan. Each winning lender receives the same interest rate, which is determined by the marginal losing or last winning bid, depending on the auction. Winning bids are either fully or partially participating in the loan; a partially participating bid means that the lender is allocated a smaller share of loan than his or her quantity bid. If the loan request is not fully funded by the listing’s close date, the listing is closed and dismissed. Successful loan requests get further review by the site to initiate the needed legal documentation for the loan to be originated. A borrower who defaults on his or her Prosper loan is barred from using the site again. 3. Data This paper uses publicly released data containing all loan requests, with their accompanying bids, using the open auction format posted by borrowers with FICO credit scores greater than or equal to 560 that were active from January to October 2008. Prior to late 2008, any legal resident of the United States could be a lender on the site and legal residents of every state, except South Dakota, could be borrowers. During the period in my sample, 42,657 loan requests and 2,022,910 bids were posted. A total of 9,624 listings and 1,688,531 bids were for listings that were fully funded. The publicly available characteristics of the borrowers are: debt-to-income ratio (henceforth DIR, the variable is top coded at 10.1), credit grade, homeowner status, whether the borrower is in a Prosper group, and state of residence. Klafft (2008b) confirm that the rules that apply in the traditional banking system apply to P2P lending as well; credit grade and the DIR are the two most important hard financial variables in determining the financial outcomes. The site provides members the ability to join groups that are designed to develop and foster a community of lenders and borrowers akin to what occurs in relationship lending, contract enforcement via reputation, and peer effects.12 Agrawal et al. (2011) and Lin et al. (2013) find that social connections seem to reduce market frictions. Therefore, I collect data on these on-site social networks. In the loan-level analysis, the In Group variable indicates whether a borrower is in a Prosper group; and in the bid-level analysis, the In Group variable indicate whether that particular lender is in the borrower’s specific Prosper group. In addition to the variables collected directly from Prosper, I construct two variables –Total Competition and Credit Grade Competition –to measure the competition that each 12This Prosper group never quite took out and are not widely used. 7
listing faces. The competition measures are the number of current listings that are active for at least one full day at the same time that the particular listing is also active. Tota Competition is the number of total listings, regardless of credit grade, while Credit Grade ompetition is the number of listings from the same credit grade. At the bid level, I recreate the auction to determine the current standing interest rate, money pledged, and bid count that exist in the auction at the exact moment that each lender bids on a particular listing. This process allows me to observe the current state of the listing as each lender sees it before he or she bids Table 1 shows the summary statistics of the borrower's characteristics for all listings The maximum listing request amount is $25,000, but most listings request significantly less(around $6,000-9,000). Intuitively, the mean request amount increases as the credit grade improves as more credit-worthy borrowers have the ability to support larger loans Although borrowers can select the duration of their listing(3-14 days), around 80% of listings are active for one week. The borrower's max rate is the reservation price for the auction and, as one would expect, it tends to increase as the credit grade worsens However, across all credit grades, over 22% of borrowers select max rates greater than or equal to 35%. Table 2 displays that breakdown of the listings and completed loans by credit grade; over half of all of the listing requests are in the bottom credit grades. The simple funding rate decreases strictly monotonically as the credit grade worsens I also collect data on the ex-post loan outcomes, paid back or defaulted for the listing that are actually originated. However, I do not observe when in the cycle a borrower defaulted or how much of the loans principle was paid back, but only the discrete outcomes of any kind of default or not. The loan outcome data comes from a different Prosper data release which contains only outcomes of the loans that were settled by early September 2011. Out of the 9, 624 completed listings in my sample, I am able to match 9, 099 of them to their final outcomes in 2011. The last column of Table 2 displays the simple default rate by credit grade for this sample. As one would expect, the default rate has a clear positive trend as the credit grade worsens. While these default rates seem ather, it is important to recall that this is a measure of any kind of default. However, these magnitudes are well in line with national residential mortgage delinquency rates from the same time period. 13 The top half of Table 3 presents the frequency distribution of the different categorie e by credit grade. Regardless of credit grade, the three most commonly chosen categories are debt consolidation, business, and personal. The bottom half of Table 3 displays the ompletion rate of loan requests by credit grade and category of use. while the Other https://www.richmondfed.org_/-/media/richmondfedorg/banking/markets-trends_-and statistics/trends/pdf /delinquency_and- foreclosure_rates. pdf
listing faces. The competition measures are the number of current listings that are active for at least one full day at the same time that the particular listing is also active. Total Competition is the number of total listings, regardless of credit grade, while Credit Grade Competition is the number of listings from the same credit grade. At the bid level, I recreate the auction to determine the current standing interest rate, money pledged, and bid count that exist in the auction at the exact moment that each lender bids on a particular listing. This process allows me to observe the current state of the listing as each lender sees it before he or she bids. Table 1 shows the summary statistics of the borrower’s characteristics for all listings. The maximum listing request amount is $25,000, but most listings request significantly less (around $6,000–9,000). Intuitively, the mean request amount increases as the credit grade improves as more credit-worthy borrowers have the ability to support larger loans. Although borrowers can select the duration of their listing (3–14 days), around 80% of listings are active for one week. The borrower’s max rate is the reservation price for the auction and, as one would expect, it tends to increase as the credit grade worsens. However, across all credit grades, over 22% of borrowers select max rates greater than or equal to 35%. Table 2 displays that breakdown of the listings and completed loans by credit grade; over half of all of the listing requests are in the bottom credit grades. The simple funding rate decreases strictly monotonically as the credit grade worsens. I also collect data on the ex-post loan outcomes, paid back or defaulted, for the listings that are actually originated. However, I do not observe when in the cycle a borrower defaulted or how much of the loan’s principle was paid back, but only the discrete outcomes of any kind of default or not. The loan outcome data comes from a different Prosper data release, which contains only outcomes of the loans that were settled by early September 2011. Out of the 9,624 completed listings in my sample, I am able to match 9,099 of them to their final outcomes in 2011. The last column of Table 2 displays the simple default rate by credit grade for this sample. As one would expect, the default rate has a clear positive trend as the credit grade worsens. While these default rates seem rather, it is important to recall that this is a measure of any kind of default. However, these magnitudes are well in line with national residential mortgage delinquency rates from the same time period.13 The top half of Table 3 presents the frequency distribution of the different categories of use by credit grade. Regardless of credit grade, the three most commonly chosen categories are debt consolidation, business, and personal. The bottom half of Table 3 displays the completion rate of loan requests by credit grade and category of use. While the Other 13Source: https://www.richmondfed.org/-/media/richmondfedorg/banking/markets_trends_and_ statistics/trends/pdf/delinquency_and_foreclosure_rates.pdf 8
category is the fourth most common request type, it has the highest simple completion rate across all credit grades. It is well established that business loan requests have noticeably more difficulty being fully funded on P2P lending sites than other types of crowdfunding, especially for lower credit grades(Lin and Viswanathan, 2014). 4 The sizeable share of listings that are business loan requests across all credit grade is a contributing factor in explaining the rather low completion rate for the bottom credit grades Table 4 presents the descriptive statistics of the bid amounts and bid interest rates, grouped by credit grade. The median bid amount, regardless of credit grade, is the minimum bid (S50). This result is in-line with previous Prosper research that most lenders tend to diversify across listings by pledging small amounts in any one particular listing The mean bid amount is non-monotonic across credit grades; lenders in the better credit grades tend to be bid larger amounts, but bidding in E listings has the largest mean However, when bid amount is viewed as a share of the loan request amount, the mean and median become significantly closer to being monotonically increasing as the credit grade worsens. This is a function of the loan request amounts generally becoming smaller as the credit grade worsens. One immediate question that might arise, given the size of an individual bid, is why a lender would invest in this unsecured market. It has been observed that most lenders commit to invest around $50-100 across dozens of loans Aggregated, a portfolio of an individual lender on a P2P lending site resembles a new asset lass, different from the traditional ones. If diversified correctly, they can offer lenders returns that do not directly follow the motions of stocks and bonds (Lieber, 2011) The nature of the auction mechanism does not allow me to observe the bid interest winning bids. Similarly to how eBay operates, the bid are displayed for losing bids, while only bid amount is shown for winning bids. The current standing interest rate is als hich is either the borrower's max rate o interest rate of the first losing bid. The final interest rate sets an upper bound on what the actually bid interest rates may be for the winning bids. Unless otherwise noted, following Bajari and Hortassu(2003)and the rest of the literature, I assume that winning bids equal the final interest rate. Not surprisingly, the mean and median bid interest rate increase as the credit grade worsens. Additionally, the amount of variation in bid interest rates appears to increase as the credit grade worsens(the correlation between credit grade and standard deviation of bid interest rate is 0. 833) a recent industry survey, it was found that 22% of all funds obtained by startups from crowdfunding sites came via debt-based platforms(Massolution. com, 2013)
category is the fourth most common request type, it has the highest simple completion rate across all credit grades. It is well established that business loan requests have noticeably more difficulty being fully funded on P2P lending sites than other types of crowdfunding, especially for lower credit grades (Lin and Viswanathan, 2014).14 The sizeable share of listings that are business loan requests across all credit grade is a contributing factor in explaining the rather low completion rate for the bottom credit grades. Table 4 presents the descriptive statistics of the bid amounts and bid interest rates, grouped by credit grade. The median bid amount, regardless of credit grade, is the minimum bid ($50). This result is in-line with previous Prosper research that most lenders tend to diversify across listings by pledging small amounts in any one particular listing. The mean bid amount is non-monotonic across credit grades; lenders in the better credit grades tend to be bid larger amounts, but bidding in E listings has the largest mean. However, when bid amount is viewed as a share of the loan request amount, the mean and median become significantly closer to being monotonically increasing as the credit grade worsens. This is a function of the loan request amounts generally becoming smaller as the credit grade worsens. One immediate question that might arise, given the size of an individual bid, is why a lender would invest in this unsecured market. It has been observed that most lenders commit to invest around $50–100 across dozens of loans. Aggregated, a portfolio of an individual lender on a P2P lending site resembles a new asset class, different from the traditional ones. If diversified correctly, they can offer lenders returns that do not directly follow the motions of stocks and bonds (Lieber, 2011). The nature of the auction mechanism does not allow me to observe the bid interest rates for winning bids. Similarly to how eBay operates, the interest rate and amount of a bid are displayed for losing bids, while only bid amount is shown for winning bids. The current standing interest rate is also shown, which is either the borrower’s max rate or interest rate of the first losing bid. The final interest rate sets an upper bound on what the actually bid interest rates may be for the winning bids. Unless otherwise noted, following Bajari and Hortaçsu (2003) and the rest of the literature, I assume that winning bids equal the final interest rate. Not surprisingly, the mean and median bid interest rate increase as the credit grade worsens. Additionally, the amount of variation in bid interest rates appears to increase as the credit grade worsens (the correlation between credit grade and standard deviation of bid interest rate is 0.833). 14In a recent industry survey, it was found that 22% of all funds obtained by startups from crowdfunding sites came via debt-based platforms (Massolution.com, 2013). 9
4. Empirical analysis Although existing theory states that distance between investors and borrowers is important, recent empirical work has been inconclusive on this issue. Intuitively, since the internet makes it cheaper and easier to connect and share information with more people, online platforms could reduce informational frictions and improve the efficiency of the credit market. Additionally, several features of this market make the presence of geography- related frictions less plausible: loans are unsecured, lenders have little legal recourse other than the standard collection process and reporting the loan default to all credit reporting bureaus. These constraints minimize the ability of lenders to individually monitor and nforce the contract. Therefore, physical proximity should be less important online as compared to offline lending. Moreover, given this is an online market, participants have to possess at least a minimum level of computer competency. Therefore, it is highly probable that these lenders have the ability to research general local conditions like population, demographics, median household income, unemployment rate, and housing starts the above, it has yet to be determined if there exist a meaningful ful differenc in lender behavior based on geography. If the differences are driven strictly by information, hen these P2P lending sites might be able to eliminate these geographic frictions, and the behavior of local and nonlocal lenders should be observationally equivalent. However, if there are differences between behavior based upon the location of the lender relative to the borrower, it might be caused by two different mechanisms. One channel, as predicted by theory, is an informational asymmetry story where distance-related frictions still matter The other channel is a preference story: local lenders are not any better informed than nonlocal lenders, but they simply prefer local projects Following the literature on online markets(Wolf, 2000; Hillberry and Hummels, 2003; Hortacsu et al. 2009), I define localness to be when the lender and the borrower reside in the same state. Admittedly, a smaller unit measure is preferred; however, Prosper does not require individuals to publicly post their city, and they actually discourage it to prevent borrowers from personally identifying themselves. If the actual effect of information asymmetry is limited to a smaller physical proximity, then my state definition of localness is counting a significant amount of nonlocal lenders as local. Therefore, I am making the two groups of lenders more similar and weakening the potential differences that I can measure. Thus this data limitation mean that my estimates are actually a lower bound on the true effect of localness. As a robustness check, I also run my analysis using the smaller, restricted sample where local status is determined by the lender living within the same city as the borrower. The results are qualitatively the same but suffer from power issues due to the small sample size; thus for brevity, the tables can be found in the appendix
4. Empirical Analysis Although existing theory states that distance between investors and borrowers is important, recent empirical work has been inconclusive on this issue. Intuitively, since the internet makes it cheaper and easier to connect and share information with more people, online platforms could reduce informational frictions and improve the efficiency of the credit market. Additionally, several features of this market make the presence of geographyrelated frictions less plausible: loans are unsecured, lenders have little legal recourse other than the standard collection process and reporting the loan default to all credit reporting bureaus. These constraints minimize the ability of lenders to individually monitor and enforce the contract. Therefore, physical proximity should be less important online as compared to offline lending. Moreover, given this is an online market, participants have to possess at least a minimum level of computer competency. Therefore, it is highly probable that these lenders have the ability to research general local conditions like population, demographics, median household income, unemployment rate, and housing starts. Considering the above, it has yet to be determined if there exist a meaningful difference in lender behavior based on geography. If the differences are driven strictly by information, then these P2P lending sites might be able to eliminate these geographic frictions, and the behavior of local and nonlocal lenders should be observationally equivalent. However, if there are differences between behavior based upon the location of the lender relative to the borrower, it might be caused by two different mechanisms. One channel, as predicted by theory, is an informational asymmetry story where distance-related frictions still matter. The other channel is a preference story: local lenders are not any better informed than nonlocal lenders, but they simply prefer local projects. Following the literature on online markets (Wolf, 2000; Hillberry and Hummels, 2003; Hortaçsu et al., 2009), I define localness to be when the lender and the borrower reside in the same state. Admittedly, a smaller unit measure is preferred; however, Prosper does not require individuals to publicly post their city, and they actually discourage it to prevent borrowers from personally identifying themselves. If the actual effect of information asymmetry is limited to a smaller physical proximity, then my state definition of localness is counting a significant amount of nonlocal lenders as local. Therefore, I am making the two groups of lenders more similar and weakening the potential differences that I can measure. Thus this data limitation mean that my estimates are actually a lower bound on the true effect of localness. As a robustness check, I also run my analysis using the smaller, restricted sample where local status is determined by the lender living within the same city as the borrower. The results are qualitatively the same but suffer from power issues due to the small sample size; thus for brevity, the tables can be found in the appendix. 10