Online Peer-to-Peer Lending: A Lenders Perspective

Transcription

1 Online Peer-to-Peer Lending: A Lenders Perspective Michael Klafft Fraunhofer ISST, Berlin, Germany Abstract Online Peer-to-Peer lending platforms claim to be beneficial for both borrowers and lenders by eliminating expensive intermediaries and reducing transaction costs. However, are the often inexperienced lenders who operate in a pseudonymous online environment with potentially significant information asymmetries really able to obtain an attractive return on their investment? This paper discusses the question by presenting profitability data from the US platform Prosper.com. Although the overall investment performance has not been satisfactory for most rating categories, it is shown that following some simple investment rules improves profitability of a portfolio and leads to acceptable returns for all credit rating categories with exception of the high-risk one. Keywords: Peer-to-peer-lending, Return on Investment, Decision Rules. 1 Introduction Peer-to-peer (P2P) lending platforms are online platforms where borrowers place requests for loans online and private lenders bid to fund these in an auction-like process. Such platforms became available in 2005 and have increasingly been used ever since. Nowa, they are available in a wide range of countries, such as the United Kingdom (ZOPA), Germany (SMAVA) or the United States (PROSPER).The motivation behind building such platforms was to circumvent banks as intermediaries, which may have the following advantages: - an expensive middleman is replaced by a more costeffective online platform, thus reducing transaction costs [1] - borrowers are given the chance to present their loan case in much detail, providing information to lenders that banks with their standardized decision processes usually do not take into consideration [2] - The loan generation process is transparent and creates a feeling of fairness (all bids visible and traceable online, [3]). - Loans on peer-to-peer lending platforms are said to generate higher returns for investors (compared to traditional bank savings) and to be cheaper for borrowers [3]. However, the question needs to be raised if investing on peer-to-peer lending platforms is really as beneficial from a lender s perspective as claimed. For P2P lenders, it is difficult to judge the quality of the deal offered beforehand, because lenders bear the default risk and few of them are experts in risk management [4]. Additionally, pseudonymous online environments are characterized by information asymmetries, which make the exploitation of lenders particularly easy for the borrowers (opportunistic behaviour, [5]). However, the long-term success of the platforms discussed primarily relies on the lenders willingness to place bids. Rational, risk-neutral and profit-oriented lenders will only do so if they obtain at least as good a return as in comparable alternative investments. Therefore, a thorough analysis of loan profitability from a lender s perspective is needed to assess the future business potential of online peer-to-peer lending platforms. This paper is going to address the issue of loan profitability. Additionally, it discusses if the borrower information presented on the platform sufficiently reduces information asymmetries to prohibit opportunistic behaviour on the borrowers side. 2 Return on Investments in Online Peer-to-Peer lending : the Prosper case In the paper presented, the issue of loan profitability is analyzed using data from the platform Prosper.com. This platform is based in the United States, and applies a pure credit auction format where partial bidding is allowed. Borrowers request a certain loan amount online and specify their willingness to pay (i.e. the maximum interest rate). Lenders then bid the amount of money they wish to invest (min. 50 US$) and specify the interest rate which they want to receive for their loan. Auctions usually terminate after a predefined time period, although an express funding mechanism is available (termination of the auction as soon as the loan is fully funded). Upon auction termination, the bids with the lowest rates are bundled until the total loan amount has been reached and are then combined into a single loan, with each lender receiving the interest rate which he specified in his bid. The resulting loan has to be repaid in monthly rates over 3 years (early repayment possible). All financial transactions, including debt collection (if necessary) are managed by the platform. Prosper.com does not actively manage risks by pooling loans. Instead, investors are provided with very detailed information on each potential borrower, and it is then up to Electronic copy available at:

2 the investors themselves to mitigate their risks. This allows lenders maximum flexibility, at the cost that they bear more responsibility for their investment. The information provided by the platform to facilitate investment decisions includes among others: - the borrower s credit rating, as determined by an external rating agency - the borrower s debt to income ratio - verified bank account information (if applicable) - verified home ownership (if applicable) - detailed information on income and monthly expenses (not verified by the platform) - past and present delinquencies of the borrower - negative credit-related public records - current credit lines - the current credit balance - bankcard utilization and - enquiries by lenders during last 6 months. Furthermore, the platform provides loan performance data of past and ongoing loans through an online performance assessment tool [6]. This tool analyses loans on a portfolio basis in accordance with selection criteria defined by the user (e.g. only loans with certain debt-to-income ratios etc.). The tool estimates return on investments for the selected portfolio using the following calculation approach (1.): I ROI = Where chgd + F late LOSS B total average E coll + R F serv % PD[ d] I chgd : interest charged within the period under consideration F late : late payment fees collected within the period under consideration LOSS total : gross default losses caused within the period under consideration, including uncollected interests E coll : expenses paid for the collection of debt from the portfolio within the period of consideration R: charged-off debt recovered within the period under consideration F serv : servicing fees to be paid to prosper Baverage: average outstanding loan balance within the period and portfolio under consideration PD[d]: duration of the period under consideration in Within equation (1.), the determination of credit losses by Prosper needs to be explained in more detail. Prosper uses a roll-rate based approach (see e.g. [7]) for credit loss determination. Non-performing loans are grouped into delinquency categories depending on how long they have been late with payments (31-60, 61-90, etc.), and it is determined from past payment data how likely a loan is to roll over into the next category of delinquency. The so-called roll rate specifies the fraction of these loans that is expected to roll-over into the next lateness category. Loans being late for more than 150 and thus reaching lateness category 6 will be considered as delinquent and charged off entirely (for more detailed information on the matter, see [8]). The relevant roll rate data for Prosper is displayed in Table 1. Table 1: Roll rate data for Prosper loans that are late (for loans originated June 2006 October 2007) 1. Values given indicate the percentage of loans rolling over from one lateness category to the next. Days late Cat. 2: Cat. 3: Cat. 4: Cat. 5: Table 1 shows that loans from the AA-category which are between 31 and 60 late with their payments have a likelihood of 42% to roll over to the category, those AA-loans being between 61 and 90 late have a 71% likelihood to roll over into the category and so on. As mentioned before, loans are considered to be on default and charged off after reaching the 150 threshold, but this does not necessarily mean that the whole outstanding amount is lost. In some cases, debt brokers will buy the outstanding debt for a fraction of its nominal value. On Prosper, this recovery ratio varies between 23% in the AA category and 8.1% in the high risk category (as of January 1 AA A B C D E HR 2 42% 41% 36% 38% 47% 43% 48% 71% 79% 77% 80% 76% 84% 80% 95% 86% 92% 95% 93% 91% 97% % Source for all loan performance data: Prosper.com website, as of January 14 th, 2008 unless mentioned otherwise. Roll rate for lateness category 1 unavailable. 2 HR = high risk Electronic copy available at:

3 When assessing the expected net default loss of a loan portfolio, early repayment is a factor that needs to be taken into consideration as well. On the platform under analysis, borrowers have the contractual right to repay their loan early at any time, without the need to pay any fees for it. As a result, the default risk of the loan portfolio decreases, as early repayment eliminates all risks associated with the repaid amount. On the other hand, the outstanding amount of the portfolio is also reduced during the period under consideration. Table 2 shows how many debtors from various rating categories repay early on Prosper.com. Table 2: Early repayment of loans Rating category % of loans repaid early each month AA 3.5% A 2.1% B 1.8% C 1.2% D 1.0% E 0.8% HR 0.5% Investors should take all these factors into consideration while calculating expected (gross) defaults from their investments on peer to peer lending platforms. Looking at a loan, the expected value of losses due to default over the loan s duration can be approximated using an iterative algorithm. Let be LOANAM m : current Outstanding loan amount that is not late in month m. STARTAM: initial loan amount INT(m): interest payment due in month m ANNUITY(m): payment due in month m on the current outstanding, but performing loan amount 3 RAT : rating category of the loan REPAY [RAT]: proportion of loans repaid early each month in rating category RAT l max : index of the highest lateness category where loan is written off l: lateness category, l [0,1,2,..., lmax] ROR [RAT,l]: roll-rate from lateness category l to lateness category l+1 in rating category RAT M max : LOSS m : LOSS total : duration of the loan in months (on Prosper: M max = 36) expected value of losses due to default originated within month m accumulated losses due to default Under the assumption that all payments are made at the end of the month, expected (gross) losses due to default within any month m can be calculated: lmax 1 LOSS m = LOANAMm ROR[ RAT, i] (2.) i= 0 Within this month m, a fraction ROR[RAT,0] of the current outstanding loan amount LOANAM is expected to become late. Out of this amount becoming one month late, a second fraction specified by ROR[RAT,1] will roll over to the next higher lateness category and so on. As a result, the expected gross loss due to default originating in month m can be calculated by multiplying the current outstanding loan amount with all roll rate factors for the rating category of the loan, as shown in (2.). Taking into account all payments, interests and loans becoming non-performing, the expected outstanding (and still performing) loan amount at the end of month m can be updated as follows: LOANAM 0 := STARTAM (3.) LOANAM m := LOANAM m-1 * (1-REPAY[RAT] - ROR[RAT,0]) + INT(m) - ANNUITY(m) (4.) Gross expected losses over the duration of the loan are then the sum of monthly losses calculated according to equation (2.) with the initial loan amount (3.) and updated monthly (performing) loan amounts (4.). Mmax LOSS = (5.) total LOSS m = m 1 = Mmax m= 1 LOANAM m lmax 1 i= 0 ROR[ RAT, i] 3 Loans that are late are considered as non-performing within this context.

4 The expected (gross) credit loss calculated in (5.) can then be used for ROI calculations, e.g. with equation (1.). Please note that the issue of debt recovery is separately reflected in equation (1.) through recovery parameter R, which is why gross losses were considered in (5.) and not net losses after recovery. The most interesting question from a lenders perspective is whether investors on social lending platforms like Prosper s can expect their investment to be profitable. Profitability data casts a doubt on that, as Table 3 shows. Table 3: Return on Investment depending on rating category, calculated using a roll-rate based approach (all loans generated Nov March 2007, data as of February 22 nd, 2008 Prosper 2008a) Rating category Estimated average ROI (calculated using a rollrate based approach) AA 5.87% A 4.20% B 3.10% C -0.33% D 0.57% E -8.69% HR % As of February 2008, lenders were loosing money in the C, E and HR rating categories, and returns on investment in the D category were unattractive. Investments in prime borrowers (AA, A, B), however, yielded returns above those of AAA-rated US treasuries (2.23% p.a. as of February 21st, [9], for treasuries with the same 3-year duration like standard loans on Prosper). The data in Table 2 indicate that lenders need to be very selective, particularly when investing in high risk categories. As previous studies revealed [10], lenders do already take into account borrower-specific information while placing their investments, such as information on credit ratings, verified bank account information, verified home ownership, the borrowers membership in online peer groups or his/her debt-to-income ratio. However, the unsatisfactory ROIs generated with subprime borrowers show that lenders still have not been selective enough in the past. As lenders on Prosper.com primarily are private persons without particular banking and finance expertise, clear and simple investment rules are needed to improve profitability. The question, however, is whether it is possible to improve portfolio performance with such simple rules, or if good performance can only be achieved with sophisticated models that are not manageable for private lenders. In order to answer this question, the effect of three simple decision rules on portfolio profitability was tested. Rule 1: Only invest in borrowers which do not have any delinquent accounts. Rule 2: Only invest in borrowers which do not have any delinquent accounts AND a debt to income ratio below 20%. Rule 3: Only invest in borrowers which do not have any delinquent accounts AND a debt to income ratio below 20% AND where no credit inquiries have been reported during the last 6 months. Table 4 shows that based upon current loan performance data observing these rules would have improved the expected return on investment. Table 4: Annual Return on Investment depending on rating category and investment rules (for loans generated Nov March 2007, data as of February 22 nd, 2008 Prosper 2008a) Rating category ROI (all loans) ROI (rule 1) ROI (rule 2) ROI (rule 3) AA 5.87% 6.44% 6.85% 7.60% A 4.20% 4.77% 5.35% 7.33% B 3.10% 4.55% 2.84% 7.21% C -0.33% 0.50% 4.26% 10.07% D 0.57% 3.55% 5.36% 4.26% E -8.69% 2.52% 4.29% 8.45% HR % % % Insufficient data Loan portfolio returns would have already been improved by applying the very simple investment rule No. 1 (no investment into borrowers with delinquent accounts). By doing so, positive returns would have been assured for all rating categories except the high-risk one. Adding more investment constraints (debt to income ratio below 20%, no inquiries within the last 6 months) further improves loan performance (with the exception of two outliers) and would have led to returns well above the ones obtained in alternative investments like 3 year-treasuries.

5 3 Conclusion In the past, many lenders investing into Prosper.com were unable to generate acceptable returns with their investments due to a high number of loan defaults. This led to angry discussions in several online communities including users threats to quit or boycott the platform [11][12]. All these developments raised the issue whether peer-to-peer lending is suitable at all for a pseudonymous online environment. However, it could be demonstrated within this paper that careful lenders who choose their borrowers in accordance with a number of easy-to-observe selection criteria can still expect their investments to be profitable. Thus, lending online still has a chance for long term success, if the platform actively addresses the issue of bad investments and low loan performance. Several such measures have already been taken, including offering webinars (= web-based seminars) to lenders to raise their problem awareness, or making additional verified borrower information available online. On the other side, the deepening sub-prime crisis currently overshadows all these efforts and leads to soaring losses and an increased risk-averseness of investors. This indicates that, for the time being, peer to peer lending platforms like Prosper s will primarily work well with good debtors from the AAand A-category, but appear to be not so suitable with high risk lenders, meaning that P2P-platforms will not be able to solve the financial problems of the poor. 4 References [1] Rumiany, D. Internet Bidding for Microcredit: making it work in the developed world, conceiving it for the developing world, Development Gateway, March itemid= (accessed November 15th, 2007) [6] Prosper Inc. Marketplace Performance. (accessed February 22nd, 2008) [7] Starkmann, H. C. Methods and Systems for Determining Roll Rates of Loans. United States Patent No , [8] Prosper Inc. Calculation of Estimated ROI. (accessed January 14 th, 2008) [9] U.S. Treasury Daily Treasury Yield Curve Rates. (accessed February 22nd, [10] Klafft, M. Peer-to-Peer lending: Auctioning Microcredits over the Internet. Proceedings of the 2008 International Conference on Information Systems, Technology and Management, Dubai, March [11] Rose, S. The Prosper Lender Rebellion and the US Credit / Borrowing Black Hole. P2P foundation, (accessed February 25th, 2008) [12] P2P-Kredite Rebellion der Anleger im Prosper Forum. (accessed February 25th, [2] Steelman, A. Bypassing Banks, in: Region Focus, Federal Reserve Bank, Richmond, USA, pp , Summer [3] Slavin, B. Peer-to-peer lending An Industry Insight. uploads/ 2007/06/peer-to-peer-lending.pdf (accessed February 25th, [4] Heng, S., Meyer, T., Stobbe, A. Implications of Web 2.0 for financial institutions: Be a driver, not a passenger, e-conomics vol. 63, pp. 1-10, [5] Tan, Y. H., Thoen, W. Toward a generic model of trust for electronic commerce, International Journal of Electronic Commerce vol. 5 no. 2, pp , 2000.