Role of Verification in Peer-to-Peer Lending

Transcription

1 Role of Verification in Peer-to-Peer Lending Oleksandr Talavera School of Management Swansea University Haofeng Xu School of Management Swansea University Abstract Using data from a leading Chinese Peer-to-Peer (P2P) lending platform from 2012 to 2015, we investigate the role of verification in the P2P lending market. We find that borrowers with thorough and complete verification are more likely to obtain funding and also less likely to default on loans. We also find that borrowers that have incomplete verification are more likely to upwardly misrepresent their income. This leads to higher default rates for this group when compared to the default rates of more thoroughly verified borrowers. The further analysis documents that returning borrowers are more likely to maintain a good credit record. We discuss the implications of our findings for the role of verification in the growing P2P lending sector and the design of a stable financial system. JEL classification: G21, G23 Keywords: Information asymmetry, verification, P2P, income exaggeration Oleksandr Talavera thanks ESRC-NSFC (Grant ES/P004741/1, Peer-to-peer lending market in China: pricing, risk management and regulation) for financial support. Corresponding author. Address: School of Management, Swansea University, Bay Campus, Fabian Way, Swansea SA1 8EN, UK. Tel.: +44 (01792) address: oleksandr.talavera@gmail.com (O. Talavera). address: @swansea.ac.uk (H. Xu). 1

2 1 Introduction Peer-to-peer (P2P) lending is a rapidly growing branch of Fintech that has attracted significant debate from both practitioners and academics. The emergence of P2P lending strengthens the consumer lending sector regarding allocating financial resources efficiently. P2P proponents claim that P2P lending provides loans to those borrowers who have difficulties accessing the traditional banking system (Milne and Parboteeah, 2016), and lowers the cost of debt financing. P2P lenders not only achieve higher returns on their invested capital but also enhance consumer access to financing, which contributes to financial stability and economic growth. However, information asymmetry between borrowers and lenders still exists, resulting in the misallocation of financial resources such as the over-financing for risky borrowers (Demyanyk et al., 2017). An explanation is that the authenticity of information about a borrower 1 is fully/partially unknown (Freedman and Jin, 2008). Recent studies document that obtaining more borrower information and market context (e.g., soft information such as friendship) can mitigate the effect of asymmetric information in the P2P market (Miller, 2015; Lin et al., 2013), but little is known about the properties of verification on loan outcomes. The objective of this paper is to investigate the role of verification in the P2P market by analyzing data from a leading Chinese platform (Renrendai.com) between July 2012 and October To this end, we first evaluate the research question: how the number and types of verification for a borrower affect the borrower s funding success rate, cost of finance, and ex post delinquency rate? We further examine whether borrowers with fewer verification types (low levels of documentation) have more incentives to overstate their income. We then analyze the effects of this income exaggeration on loan outcomes. By comparison existing P2P studies are solely based on application related information (e.g., US platform: Prosper.com and Chinese platform: Ppdai.com). The Renrendai marketplace provides an excellent source of verification status data for borrowers. Borrowers initiate loan requests on the platform and a crowd of peer lenders can bid the loan (fully or partially) at a certain interest rate. 1 Prosper.com (one of largest P2P platform in US) reports verified employment and/or income on approximately 59% of the borrower loans originated through the marketplace on a unit basis (227,419 out of 388,617) and approximately 73% of such loans on a dollar basis ($3.531 billion out of $4.856 billion) between July 2009 and September 2015 (based on start time of the applicable bidding period). 2

3 Lenders can view loan information and the borrower s characteristics. In addition, the verification status is displayed by a table with various verification levels (Figure 1). Figure 2 shows that the average number of verification for all listings and funded loans over the month. Borrowers raise the loan requests with average 1.6 types of verification and 3.8 for funded loans 2 (12 in total). To explore income misrepresentation, we match a representative sample with similar characteristics but more reliable verification methods (high cost) to calculate the extent of income exaggeration. Beginning with the introduction of the first P2P platform in 2007, the Chinese P2P lending has been growing rapidly reaching 600 billion RMB ($91 billion) in total outstanding loans in July 2016 (Figure 3). Despite this, in China, there is no fully developed system of credit referencing an individual borrower (Milne and Parboteeah, 2016). Accordingly, the potential for loans fraud and adverse selection has always been present, resulting in a severe hazard of information asymmetry. The verification system offers an alternative way to fill this gap by validating borrower s documentation. The optimal number and types of verification can facilitate debt financing and reduce ex-post default rates. We evaluate the hypotheses for the role of verification stemming from two folders. First, we examine whether thorough and complete verification of borrowers can facilitate funding success rates and lower default rates. Although the P2P platform discourses various personal information regarding borrowers, lenders are unable to distinguish quality borrowers from bad ones if the information is not feasible. Akerlof (1970) points out that lack of credible information about borrowers and using that information to screen applicants could result in financial underdevelopment. To demonstrate their creditworthiness, borrowers can provide proof of documentation (e.g., income) to show the capability of repaying the debt. Increasing the number of verification means that borrowers attempt to indicate better qualities than those borrowers who are relatively less verified. Thus complete verification of borrowers represents an effective signal that leads to better loan performance. The informational value is different for each type of verification. Lenders cannot directly perceive trustworthiness itself but inferred from signals (Bacharach and Gambetta, 2001). Borrowers show mixed signals for P2P lenders to proof their quality and reduce the adverse selection. We then 2 Renrendai.com change the verification policies in October 2015 for the regulatory requirement. Borrowers must verify their income, employment, identity and providing credit report issued by the People s Bank of China. Other types of verification are not needed. 3

4 assess which verification can more intuitively signal the borrower s creditworthiness in the P2P lending market. A useful verification item can help the platform to improve its credit scoring system and benefit to lenders screening device. In other words, the highly qualified signal means that the inherent risk of moral hazard could be identified in the ex ante process. Second, to better understand the consequences of incomplete verification, we further examine the implication of income exaggeration by considering the effects of verification. In consumer lending market, lenders heavily rely on borrowers income to achieve the expected return, but income misrepresentation of borrowers is a potential risk for lenders. Jiang et al. (2014) find that borrowers who provide a lower level of documentation 3 resulting in borrower information misrepresentation and elevated delinquency rates in the mortgage market. Ambrose et al. (2016) confirm the findings and further evaluate the effects of the borrower heterogeneity with respect to employment status. The consequences of misrepresentation drive the expansion of mortgage market between 2002 and 2006 and inflate house price (Mian and Sufi, 2009, 2017), and as one of the possible causes of the Great Recession (Ambrose et al., 2016). Our empirical analysis uncovers the substantial problem in P2P lending. P2P loans are based on Internet and borrowers can raise multiple loan requests without cost. If there is no verification process, P2P borrowers may have incentives to access financing through misrepresentation of their income. The borrowers have higher chance to obtain loans as well gain lower interest rates compared with other similar characteristic borrowers. The cost of default would be relatively lower. Because the P2P lenders are unable to commence legal action for enforcement of the defaulted loans. In particular, the P2P platform in China cannot access to the credit system of People s Bank of China, this leads to defaulted borrowers will not be recorded in the formal banking system. In other words, the fraudulent P2P borrowers still can obtain financing from other financial institutions, which threatens financial stability and other negative social effects 4. Our results show that loans with increased verification are more likely to obtain funding. Lenders may use this complete verification as a signal because it is a strong indicator regarding endorsement which distinguishes it from low information loans. Specifically, social media verification and edu- 3 A low doc (or low documentation) loan is a type of mortgage that can be approved without the normal income verification requirements. See 4 For instance, Ezubo scam: Popular online peer-to-peer lender leaves 900,000 Chinese investors $7.69 billion out of pocket 4

5 cation credentials of borrowers are associated with lower interest rates and ex post default rates. We examine the relation between the verification and income exaggeration; the results show that borrowers who provide less verified documents have a greater extent of income exaggeration. Additionally, we provide novel evidence that borrowers with incomplete verification are more likely to have increased probability of default, but if these borrowers that are concerning about future credit availability through the platform can mitigate the effects of adverse selection and reduce the incidence of delinquency. This suggests that returning borrowers are willing to maintain their good record, which can compensate for the effects of lacking verification on the P2P platform. Our study is the first to discuss the impact of verification on loan outcomes and income exaggeration in P2P lending. This paper contributes to the literature on the growing P2P lending sector. Recent studies include Pope and Sydnor (2011); Zhang and Liu (2012); Duarte et al. (2012); Lin et al. (2013); Burtch et al. (2015); Miller (2015); Iyer et al. (2015); Wei and Lin (2016); Tao et al. (2017). Furthermore, our study contributes to the established literature on information asymmetry, credit rationing and signalling theory (Akerlof, 1970; Stiglitz and Weiss, 1981; Spence, 1973, 2002). This paper also contributes literature about misrepresentation and income overstatement. Ambrose et al. (2016); Garmaise (2015); Mian and Sufi (2017) illustrate how misreporting and falsification are associated with adverse loan outcomes in the mortgage market. Our study provides novel evidence to confirm income overstatement does occur in the P2P lending market and thorough verification can mitigate the extent of exaggeration. The rest of this paper proceeds as follows. Section 2 reviews the literature and summarizes our research. Section 3 describes the institutional background in China and the operation of the P2P platform. Section 4 describes our data set. Section 5 and section 6 contain the empirical strategy and discuss the results of the study. Section 7 concludes this paper. 2 Research Context 2.1 Information Asymmetry and P2P Lending Akerlof (1970) proposed the information asymmetry framework that sellers of used cars know more about car quality than buyers. The sellers are unable to present quality resulting in high-quality sellers withdraw from the market, leading to a market failure. Spence (1973) argues that the 5

6 adverse selection problem can be mitigated if high-quality types use signals to communicate quality. Low-quality types cannot deliver same information due to the costs of acquiring signals. In the credit market, asymmetric information between borrowers and lenders leads to a variety of consequence, e.g. adverse selection and moral hazard. In the framework, lenders cannot identify ex-ante which type of borrower is good, lenders are unable to distinguish quality borrowers from bad ones if the information is not feasible even borrowers offered to pay higher interest rates (Stiglitz and Weiss, 1981). Borrowers are characterized by their profiles, which are assumed to have the same expected returns but differ from one another in their risk. Traditionally, lenders overcome the adverse selection and moral hazard problems by requiring collateral from borrowers that signal their high-quality (Bester, 1985). However, The use of risk-based pricing in consumer loans, including credit card loans and mortgages, has become widespread, reflecting the increased ability of lenders to distinguish between borrowers with different risk profiles (Edelberg, 2003; Chomsisengphet and Pennington-Cross, 2006). Blackwell et al. (1998) provide evidence that borrowers voluntarily submit verified financial statements with their application leads to reduce interest rate in private business lending. The P2P lending is mainly serving on unsecured consumer loans for those who have difficulties accessing the traditional banking system (Milne and Parboteeah, 2016). P2P platforms claim that they have lower screening costs and better quality in underwriting borrowers. Demyanyk et al. (2017) argue that P2P lenders provide loans to the riskiest clients who often continue accessing other financing channels, resulting in leveraged growth of borrowers after P2P loan origination. The P2P market is new, and P2P lenders face severe information problems relative to offline credit markets due to the authenticity of information is unknown (Freedman and Jin, 2011). Lenders are difficult to know what type of borrower is risky or not. To increase the probability of funding, P2P borrowers provide proof of documentation (e.g., income) to demonstrate the capability of repaying the debt. In recent years, many of empirical studies have been made focusing on P2P lending, and most of them have used data from the USA, e.g. Prosper and Lending Club. Puro et al. (2010) found that smaller loan amounts can help increase the success rate and decrease the interest rate. Also, Lin et al. (2009) showed that it is difficult for loans over a longer period to be funded because sufficient liquidity for the lenders would not be provided. Freedman and Jin (2008) found credit rating to be 6

7 the most crucial factor affecting the interest rates. Debt-to-income ratio takes second place, and if a borrower has bank accounts, then his/her loan requests are more likely to be successful. Research on soft information shows evidence of the impact of friends, borrower narratives, and photos. Lin et al. (2013) distinguish different types of friendship and found that friendships of borrowers increase the probability of successful funding and reducing ex-post default rates. Pope and Sydnor (2011) showed that discrimination exists due to borrowers photos in the P2P market. Duarte et al. (2012) find that borrowers with trustworthy appearance are more likely obtain loans and less likely to default. These hard-to-quantify factors influence users experiences with the platforms and serve as the indirect evidence of a users trustworthiness (Collier and Hampshire, 2010). For the regulation change, Wei and Lin (2016) tested two market mechanism (Auction price and Post price) and found that loans are funded with higher probability under post price, but interest rates are higher than auction model. Miller (2015) found that exposing more information of borrowers can significantly improve the screening performed by existing lenders and attract new lenders who were better at screening loan applicants and earned higher returns. Zhang and Liu (2012) show that psychological factor affects behaviours of borrowers and lenders. Rational herding behaviour does exist in P2P lending, well-funded loans are more likely to attract more funding. But little is to study the role of verification in P2P lending. Kumar (2007) find that completing bank account verification was related to a lower probability of loan default, while loan size was positively correlated with the default rates. With respect to Chinese P2P market, Chen and Han (2012) compared online P2P lending between the USA and China. Feng et al. (2015) carried out an empirical study of lenders and borrowers strategies in which a larger loan amount increased the probability of funding and attracted more lenders by using a small amount of data (1057 listings) from Ppdai.com. Using data from Renrendai.com, Tao et al. (2017) find that borrowers with higher income or own a car are more likely to obtain a loan, pay lower interest rates, and default less often. They also indicate the role of offline verification in the lending process. Our study aims to investigate the role of verification in the P2P lending market. We evaluate the number and each type of verification and their outcomes. 7

8 2.2 Income Exaggeration The overstatement of income on traditional loan application has been well documented in various commission reports and the existing literature (Garmaise, 2015; Ambrose et al., 2016; Mian and Sufi, 2017). The role of borrower income misrepresentation leading up to the ex-post loan default is a source of considerable debate. For example, Jiang et al. (2014) show that borrowers who provide a lower level of documentation resulting in borrower information misrepresentation and elevated delinquency rates in the mortgage market, and particularly for those borrowers from brokers. Supporting this argument, Ambrose et al. (2016) provide new evidence to indicate that borrowers with low-doc mortgages originated were more likely to exaggerate their income and cause a higher risk of default. However, to my best knowledge, few papers studied the income misreporting in the P2P lending market. Eid et al. (2016) found that rounding of income by a borrower is more likely to default and less likely to repay than borrowers with more accurate income reporting. In contrast to the study, our analysis measures income exaggeration by comparing audited income through an offline source of a higher level of verification. Specifically, we examine the relation between the number and each type of verification and income exaggeration. Whether a borrower inflates his/her income has a higher probability to obtain a loan and resulting in elevated defaults. We then evaluate how the behaviour of repeating borrowing and misreporting affect loan performance. 3 Institutional Background 3.1 P2P Lending Market in China P2P online lending was first established in the UK and rapidly developed in other countries such as the USA, German, and China. Beginning with the introduction of the first P2P platform in 2007, Chinese P2P lending has been growing rapidly. The growth of online P2P lending platforms in China has been considered to challenge the market share of the bank in consumer lending. This because the current bank system could not locate the financial resources efficiently for the private sector- the main driver of economic growth (Allen et al., 2005). However, Chinese P2P platforms operated in an unregulated environment until July As a result, many platforms 8

9 have operational difficulties, and investors would be facing potential loss. Despite the negative effect, Chinese P2P platforms realised total outstanding balances of $91 billion in July But expected annual return has been declined from 20% to 10% over the 2014 to 2016 time period (Figure 3). The nascent industry is changing very quickly. Up to now, the P2P lending in China has been experiencing three evolution stages. Inspired by Zopa.com in the UK and Prosper.com in the US, Ppdai.com went live online in June 2007 as the first P2P lending website in China. It was followed by the appearance of some major online lending platforms of a similar type, containing Renrendai.com and Lufax.com. At this stage, most of the transactions came from personal loans that occurred between lenders and borrowers directly. The platforms only play a role of exchanging information. Because of lacking official credit information on the borrower, the practitioners have to find ways reducing information asymmetry and providing integrated services to increase the stickiness of clients. The early strategy of P2P platforms particularly intended to attract borrowers who can not access to traditional financial institutions. The risk control for borrowers is of lower quality compared to traditional banks. P2P platforms somehow provide multiple resources regarding the partnerships with external guarantee institutions. For instance, Renrendai.com can provide the guarantee for investors, which means that the platform repays the investment to lenders in advance if a loan defaulted. However, many platforms fail to control risk for both internal operation and overseeing borrowers. The P2P lenders are unable to commence legal action for enforcement of the defaulted loans. In 2012, the average bad loan over $1.5 million for each platform. From 2012 onward, P2P industry in China has been growing dramatically including the number of investors, borrowers, platforms and business turnover. Some P2P platforms experiment additional business model for market share expansion (e.g., offline channels). CreditEase and Renrendai are representative examples of this marketing model. Also, P2P platforms usually provide wealth management products offering categorised investment modules based on aggregated P2P loans as a certain interest rate with a fixed term. P2P companies sometimes collect money from investors building a funding pool if there are no enough loans to invest, which means that these institutions can invest money for any purpose without regulation. Some platforms also develop the secondary market which is designed for those investors who want to release liquidity. The government issued a guideline including ten regulatory authorities in July 2015 requiring 9

10 that online P2P lending platforms must separate their own money from investors money by opening a third-party depository account in a bank which is subject to the supervision of the People s Bank of China and other financial regulatory authorities. This regulatory framework completely changes the industry and results in a lot of consequences. According to wdzj.com, there are 3701 problematic platforms out of total 5890 platforms by the end of April 2017, which account for 63% overall historical platforms. After the policy intervention, a large number of P2P platforms were closed or changed their major business type. The regulated platforms must register themselves as information agency firms with authorities. In the present, a significant number of platforms switch to sell wealth products, which is less transparency for the information about borrowers. As a result, just a few online platforms remain the pure P2P function. Many others intend to diversify their business model to subject to regulation. 3.2 The Operation of Renrendai.com Renrendai.com is a leading P2P platform in China which was established in October By the end of March 2017, it had confirmed 385,696 loans with a total lending amount of over $3.95 billion, and 2.66 million registered users for lending and 384,288 lenders invest in loans. Figure 4 shows that the total and average principal amount are increasing over the year, and dramatically boosting for both newly registered borrowers and lenders. There are over 1 million registered users for borrowing and 367,413 borrowers successfully obtain loans up to now. The aggregated statistics shows that the platform exists $1.94 billion outstanding amount in total, and the reported default rate is 1.73% over the time period Loan Application. On the platform, lenders easily search for information about prospective funding opportunities provided by each borrower, then invest, and complete transactions with capital gain (although sometimes, loss). The operation of Chinese P2P platforms has evolved from original P2P type into several derivatives. The operation of Renrendai.com is illustrated in Figure 5. Potential borrowers post loan requests on an online platform directly, or via a third-party offline institution partnered with the platform. A standard loan request contains two sections (See Appendix Figure 6). One section consists of the loan amount, maturity and maximum interest rate 10

11 they can accept. The other sections are information on borrowers demographic factor and supporting documents. Some supporting documents are compulsory such as identity card, payslips, mobile number, address, employment status, and credit report 5. Other documents that may be helpful in obtaining funding (for example, their asset information and education certificate) could be uploaded voluntarily by borrowers. The borrower must fill in an application form with loan purpose description and other information based on above two sections. However, online platforms in China are unable to access this report from People s Bank of China, so a borrower has to provide the personal credit report with his/her application. To offset the lack of a national credit scoring system, platforms usually evaluate the loan request and assign a credit grade 6 (e.g. from AA to HR-high to low) for the borrower after verifying the documents provided by borrowers Verification and Offline Audited. The credit profile can be verified by the staff of Renrendai.com. For instance, the staff will make an essential phone call to verify the employment of a borrower. Once the information is validated, the verification status in the loan listing will show pass for the validation (See Figure 1). There are many types of verification which include employment, income, credit report, identity, mobile phone number, address, homeowner, car owner, social media, video interview, marital status, and education. To mitigate the effect of lacking reliable credit system, the platform introduced the thirdparty offline system in July The offline institution provided loan requests (borrowers don t directly register in Renrendai.com) to the online platform where lists the same types of information as borrowers from online. The verification types are same as online, but field audited for the authenticity of borrowers profiles. For example, the adviser in the offline office can request an interview in the home of a potential borrower or visit his/her employer. The adviser can verify the income a potential borrower via various sources (check payslip, accompany the customer to print bank statement in a branch and require an employer reference). Note that loan requests from a third-party offline institution have to be verified by the institution before listing to Renrendai.com. 5 It has to be noticed that the personal credit report is a credit evaluation document issued by People s Bank of China. It only reports the number of defaults of loan or credit card repayments from Chinese banks or institutions (and the records of lawsuits). 6 The credit grade will be updated by any action of borrowers. We collected the sample in February The grade for borrowers is not same as the origination date. This indicator will be excluded in our analysis. Tao et al. (2017) point out that the credit grade assigned by the lending platform may not represent the creditworthiness of potential borrowers. 11

12 Since December 2012, Renrendai.com launched their own offline institution (Ucredit.com) starting from Beijing. Until now, it has been over 200 offline offices spreading into the whole country. Once loan requests are pre-approved by Ucredit.com, the requests will be posted on Renrendai.com. Then lenders can bid these loan requests like investing normal online ones Listing, Bidding, Funding, and Repayment. As with borrowers, registered lenders are also subject to verification of mobile phone number, identity card, and bank account number. The true identities of lenders (including all personal information) are not publicly revealed on the website. Lenders can invest their capital through two different channels. The first channel is a purely intermediary P2P lending type, which provides each loan listing with borrowers information and verification status (See Figure 1). Lenders examine listings and screening which listings to invest in and how much to invest. They are not required to invest in the full amount. The minimum amount of each bid is 50 RMB. The platform also provides an automatic bid instrument. Borrowers can easily use the instrument by setting their investment criteria (e.g., credit grade, interest rate) to bid automatically. Lenders can diversify their investment risk by multiple bids into different borrowers. Once the amount of loan request has been met by the aggregate amounts of lenders bids, the loan will be fully funded and unlisted from the website. Then, loan proceeds are credited to the borrowers bank account from which repayments are automatically withdrawn. The second channel is called investment plan. The investment plan is a package of loan requests created by the platform. It contains requests with specified interest rate and maturity. Lenders who invest the packages are only able to know the fixed interest rate and term of the package, but unable to access the information of the components in the package individually. 4 Data and Summary Statistics 4.1 Data By using the Python program, we scraped data from the online P2P lending website, Renrendai.com. We collected 945,089 loan listings, including each loan transaction from October 2010 to January As we introduced in the previous section, the listings contain online P2P loans and the loans from the offline institution (Ucredit.com). Our analysis focuses on online loans because it is 12

13 a costless way for facilitating financing. Although the offline institution can verify the borrowers profile more accurately and may have better risk management. The cost of debt (the actual annual interest rate) is from 15.36% to 30.96% if a borrower raises a loan request via Ucredit.com, which is much higher than online borrowing cost. But listed interest rate of offline loans on the P2P platform is much lower than the rate charged by Ucredit.com. Because the local offices are established across the country, but the huge cost finally transfers to borrowers. After dropping missing values, our sample covers loans originated from 2012 to 2015, with performance data ending in October 2015, which includes 738,411 listings for both online and offline loans. Because after October 2015 Renrendai.com changed the types of verification and business model for regulating purpose. There are 567,955 online requests and 170,456 offline requests; of these, 21,549 online requests and 169,960 offline requests had been successfully funded respectively. We use 58,866 of offline loans to be our matching sample. Because the borrowers in the sample have a higher level of income verification. The offline requests have high acceptance rate because the platform only publishes the pre-approved requests for listings. Each listing contains the loan with the specific conditions of the annual interest rate, the amount of loan, the period of repayment, the guarantee type, a credit score issued by Renrendai.com, and various pieces of personal characteristic information (age, income, location, occupation, employer s information, education, marital status, homeowner,car owner and borrowing histories in the platform). Furthermore, Renrendai.com provides verification status, and the lenders can see which profile of a borrower has been validated. The variable definitions can be seen on Table Summary Statistics Table 2 provides summary statistics for key variables of all listings and funded loans through online requests respectively. We find that only 3.8% of the borrowers can successfully obtain requested loans. The default rate is calculated by the completed loans because the loan can default at any time between origination and maturity, it would be biased for loan performance if the loan is still repaying. The default rate reaches to 17.3% in 20,191 completed loans. Comparing with funded loans, the average interest rate that borrowers are willing to pay is 13.24% with minimum rate 7% and maximum rate 24.4%, while the average contract interest rate is slightly lower as 12.39%. The average loan amount requested for listings is 60,721 Chinese RMB, while the average loan size 13

14 is as small as RMB for funded loans. The average maturity for all listings is 13 months compared with 16.2 months, which indicates that lenders prefer to fund loans with a shorter term. Lenders also can see the borrower s histories in the platform, that is, the average loan requests and borrowers have been successfully funded and pay off in the platform. The loan listing report times of borrowers arrears in the platform (up to 30 days and up to 90 days). Note that the loan is described as default if the borrower doesn t keep up repaying over 90 days. Turning to observable borrower characteristics, Table 2 shows that the average borrower is 32 years old with reported monthly income 10,649 RMB for all listings, while the funded average borrower is slightly turning to 34 years old with much higher reported monthly income 14,984 RMB. Borrowers can provide their asset information such as property and car. We can see that 30% of loan requests report property and only 1% with the mortgage, while 54.2% of funded loans claim homeowner and 22.5% with the mortgage. 17.4% of requests declare car owner and 4% existing car finance, while 37.7% of funded loans state car and 8.4% with the car loan. Borrowers can write a self-description of loan purpose to show more information regarding the requests. We count the length of words because Dorfleitner et al. (2016) find that investors react more strongly to soft information and text information can significantly affect loan outcome and performance in P2P platforms. In consistence with (Dorfleitner et al., 2016), the average length of self-disclosure in all listings is 130 words which are lower than 150 words for funded loans. Regarding verification, Renrendai.com provides the unique way for lenders to screen loan requests. Borrowers can choose a different combination of verification types to facilitate finance. Table 2 shows small proportion for all types of verification in all loan listings of online requests, while borrowers who get funded have much more percentage of credit validation. We can see that the identity verification account on 32.4% for all requests and 99.4% for funded loans, which means for those borrowers who want to obtain loans must verify their identity. The first evidence provides a snapshot that the average number of verification for a funded loan is 3.84, which is higher than all listings with average 1.61 verification. Figure 7 shows average verification percentage for funded loans over the month. The monthly percentage of income, employment and credit report verification fluctuate around 15% over the period, increasing rapidly from 2015 and reach to almost 100% in October 2015 (policy change). The identity verification is almost a horizontal line at 100%. Because the verification is compulsory for borrowers. The monthly percentage of other types of 14

15 verification maintain a significant level until 2015 and reduce to 0 in October Verification and Loan Outcomes We begin our analysis by examining the relationship between the number of verification and probability of a loan successfully funded, the interest rate, loan size and maturity of funded loans. We then conduct statistical analysis in which we examine the relation between the number of verification and ex-post default rates for completed loans. In the second part of the section, we study the effect of each verification on loan performance. 5.1 Number of Verification and Loan Outcomes This study employs regression analysis to find out the number of verification that influences the probability of loan funded, the interest rate charged, loan size, loan maturity, and ex-post default rate. Following the univariate analysis in the previous section, we estimate the following regressions of loan status and characteristics: P robability(loanstatus i = 1) = Φ(α + βnumv erification i + δx i + ɛ i ), (1) LoanInformation i = α + βnumv erification i + δx i + ɛ i, (2) Where the dependant variable LoanStatus i in equation 1 measures whether a loan i had been funded (Success i ) and an indicator of measuring default of on loan i (Default i ) and Φ is the cumulative distribution function of the standard normal. NumV erification i is the number of verification submitted by a borrower. In equation 2, LoanInformation i contain contract interest rate, loan size, and maturity. The vector X i includes set of controls, for instance, loan characteristics (interest rate, loan amount, maturity and the length of self-disclosure) and borrower characteristics (the borrower s histories in the platform, borrower age, borrower income, whether the borrower is a homeowner, and with or without mortgage, car owner, and with or without car finance, borrower location, occupation, employment sector, employment length, the firm size for a borrower employed, education and marital status). We also include the variable indicating loan origination year to 15

16 control the cohort effects. The parameter β is the primary coefficients of the differential effect on the probability of successfully funded rate and default, and interest rate, loan size, and maturity for the number of verification. Table 3 shows the relationship between the number of verification and loan performance. Borrowers with thorough and complete verification can significantly increase the probability of loan funded and mitigate the effect of ex-post default and are more likely to request the shorter period of loans. The number of verification has the negative relation with interest rate and amount but statistically significant. In consistence with credit rationing theory, our empirical results show that increasing interest rate can result in lower funding probability, which means that interest rate cannot be used to clear excess loan demand in the P2P market (Jaffee and Stiglitz, 1990). Hence, the adverse selection occurs in the P2P market. Also, loans with higher interest rate are more like to default. As expected, other loan characteristics such as loan size, maturity have a significant impact on funding probability and default rate. Borrowers who exist mortgage are more likely to obtain the loans and have a lower probability of default. It is an indicator to show potential lenders that the borrower had already accessed to finance. The results of other borrower characteristics are consistent with recent P2P lending papers (Feng et al., 2015; Freedman and Jin, 2014; Miller, 2015; Serrano-Cinca et al., 2015), particularly, Tao et al. (2017) study that borrower characteristics and loan characteristics impact on loan funding outcomes by using data from Renrendai.com. 5.2 Verification Types and Loan Outcomes We further discuss the relation between each verification type and probability of funding, default rates and interest rates. We estimate each verification type by the following specifications: P robability(loanstatus i = 1) = Φ(γ + λv erification i + δx i + ɛ i ), (3) InterestRate i = γ + λv erification i + δx i + ɛ i, (4) Where V erification i represent each verification type, which includes validation status for employment, income, credit report, identity, mobile phone number, address, homeowner, car owner, social media, video interview, marital status, and education. The dependent variable LoanStatus i, 16

17 InterestRate i and control variables X i are defined as in equation 1 and equation 2. Equation 3 estimates whether the funding probability of loan requests are correlated with each verification type, and examines the relation between the probability of default and each verification type. We also test that borrower s debt cost by verifying the different type of documentation in equation 4. Table 4 presents regression results for the probability of successfully funded across each verification type. We find that most of the verification type can increase funding probability. Specifically, borrowers who verify income, employment status and providing credit report particularly have higher chance to obtain loans. The verification of mobile number is negatively related to successfully funded of requests. That is because the mobile number was a compulsory process when borrowers had registered as members in the platform. Thus, the mobile number verification is no longer as an effective signal for screening loan requests. The results in Table 5 provides an analysis of the relationship between verification type and the default for completed loans. Borrowers who verify credit report, social media account, marriage status, car ownership and education level are less likely to default on their loans. Verifying income, employment and home ownership are not correlated with ex-post default rate, which raises the concern of income misrepresentation and effectiveness of online verification. The mobile number and address are additional verification which inflates the probability of default. Table 6 shows that borrowers who verify their social media account, education and car ownership can obtain loans with lower interest rate. To summarize, borrowers are more likely obtain loans if they choose to verify more documents whichever type of document with the application. Lenders catch up the signal for bidding. However, it cannot be used to a prediction of ex-post default. Such important verification types (income, employment) are not correlated with default. Although credit report issued by People s Bank of China only provides limited information about credit history, it still shows a signal to lenders for the creditworthiness in past banking accessing for the borrower. The lower probability of default for verifying credit report illustrates that borrowers who expose their credit histories can indicate ex-post loan performance. It is not surprising that verifying education level and car ownership can relief the effect of default. Education level and asset are significant characteristics dividing social class which can determine the borrower s risk level (Hollingshead et al., 1975). 17

18 6 Income Exaggeration and Loan Performance In the previous section, we evaluate the impact of verification on loan performance. Borrowers with complete verification facilitate the debt financing and mitigate default rates. Having considered the problem that the particular set of borrowers have the incentive to inflate or falsify their income. We now can examine the consequence of incomplete verification by exploring income exaggeration and its effect. 6.1 Measure Income Exaggeration The common method of measuring income exaggeration is to find a baseline group such as World Values Survey compared with the income of treatment group. However, the data of survey is collected from a general source of the whole population. It could be a selection bias in our data because for those borrowers who seek finance in P2P platforms is a certain group in China. Fortunately, Renrendai.com provides an ideal matching sample (offline audited loans) to examine income exaggeration. As we mentioned in section 3, Renrendai.com launched a new financing channel through offline institution from The loan requests from offline are strictly audited (higher level of verification compared with the original sample). The offline audit only includes four types of verification: income, employment, identity and credit report (verify all four types together). Other information is consistent with the online application. Table 7 shows the summary statistics for funded loans from the online and offline channel. We can see that borrowers who verify their documentation via offline channel have lower average income 7 than the online group. It shows initial evidence that borrowers could exaggerate their income if incomplete verification provision is carried out. To formally measure income exaggeration, we employed nearest-neighbour matching with bias correction to obtain the matching estimators for average treatment effects (Abadie and Imbens, 2006, 2011). It provides a consistent estimator for the large sample variance. The bias-corrected matching estimators have the advantage of correcting for a large-sample bias that exists when 7 Note that the original income reported by a range, e.g. 5,000 RMB to 10,000 RMB. We take the average value for the income (7,500 RMB for the range 5,000 RMB to 10,000 RMB). For the reported income below 500 RMB, we take the value as 500 RMB. The reported income above 50,000 RMB, we take the value as 50,000 RMB. It won t affect the results because both online and offline group use the same method of record. 18

19 matching on more than one continuous covariate. Our matching includes several continuous covariates (loan amount and age). We can get more accurate outcome using bias-corrected matching estimators. Other studies to measure income exaggeration such as Ambrose et al. (2016); Jiang et al. (2014) estimate income by employing semi-log model from borrower characteristics, loan characteristics, area characteristics and other covariates. Our matching covariates contain borrower characteristics (age, education, marital status, province dummies, employment sector, employment length, firm size, industry sector, home ownership and car ownership) and loan characteristics (loan size and loan origination year dummies). We calculate the measure of income exaggeration by subtracting the estimated logarithm income from the reported logarithm income. The measure means the percentage difference between estimated income and reported income. Table 8 presents the average treatment effect on logarithmic income compared with offline audited loans. The results show that the matching estimate of income reported by online applicants is negatively related to the offline audited group, which indicates the evidence of inflating income by online borrowers. The result is consistent with the matching estimate of propensity score matching method. 6.2 Number and Type of Verification against Income Exaggeration To further examine the extent of income misrepresentation, we employ the following specification: IncExg i = γ + λv erreleated i + δx i + ɛ i, (5) Where IncExg i is the measure of income exaggeration, V erreleated i represents NumV erification i and V erification i defined as in equation 1. X i are defined as in equation 1. The model examines the number or type of verification impact on the extent of income exaggeration. Table 9 reports the results of the measure of income exaggeration and the number of verification for funded online loans. We find that borrowers with incomplete verification are more likely to upwardly misrepresent their income. In other words, borrowers with thorough verification can mitigate the extent of income exaggeration. Table 10 presents that the relationship between income exaggeration and each type of verification. Carrying out verification on income, employment, credit reports and video interview can reduce the level of income overstatement. Other types of verification are no 19

20 statistically significance with income misrepresentation, suggesting no matter with higher cost but lower effect documents. For instance, borrowers have less incentive to falsify education certificate, homeowner certification, and car owner certificate. The income verification should be an effective way to obtain the loans. That could be one of the causes that the platform introduces offline audited loans. The results are consistent with other studies such as (Jiang et al., 2014; Ambrose et al., 2016; Mian and Sufi, 2017). We then test the effect of income exaggeration on the loans for the probability of default. The model are specified as follows: P robability(default i = 1) =Φ(γ + νincexg + λv erreleated i + ηincexg V erreleated i + δx i + ɛ i ), (6) Where Default i indicate bad debt that a loan has not been kept up repaying over 90 days. Other independent variables IncExg i and V erreleated i and X i are described as in previous sections. Table 9 illustrate that income overstatement results in higher probability of delinquency, but borrowers with complete verification can mitigate the bad effects of the default. Table 11 shows that borrowers who validate their income and employment can reduce default rates against income exaggeration. There is no evidence that other types of verification have the same effect on loan default. Also, Figure 8 shows the average marginal effects of income verification at different levels of estimated income exaggeration. Note that 2.3 and 1.86 are the 5th and 95th percentiles of income exaggeration, respectively. The marginal effects are derived from the probit model of loan default described in equation 6. The default rates slightly decrease if increasing the extent of income exaggeration with income verification, while the default rates go high when inflating the income without the verification. 6.3 Low-doc Verification, Returning Borrower and Income Exaggeration In the previous section, we examined the relationship between income exaggeration and the number and type of verification. A borrower normally provides a different combination of credential files during the verification process. We found that borrowers who verify thorough and complete verification can reduce the probability of default. To better explain the combination of verification, we 20

21 classify the verification as two types: incomplete verification (low-doc) and complete verification (high-doc). To classify these two levels of verification, we employ Latent class analysis (LCA) to find groups or subtypes of cases in the multivariate categorical verification.the LCA will attempt to detect the presence of latent classes (the disease entities), creating patterns of association in the verification. As in factor analysis, the LCA can also be used to classify case according to their maximum likelihood class for the number of verification. Followed Lanza et al. (2003), the model can be described as below p i1,i 2,...,i N T N p t p n i n,t (7) t n Where T is the number of latent classes and equals 2 (two latent classes: low-doc and high-doc). p t are the unconditional probabilities and p n i n,t are the conditional probabilities. The output includes the probability of a response to number of verification for each latent class. In other words, we can obtain the probability that members of each class had of engaging in each verification type. We then examine whether low-doc verification with overstated income is associated with ex-post loan performance, and we discuss a situation if the borrowers consider future credit availability. The models are shown as follows: P robability(default i = 1) =Φ(α + β 1 Lowdoc i + β 2 Returning i + β 3 Lowdoc i Returning i + λ 1 IncExg i + λ 2 Lowdoc i IncExg i (8) + λ 3 IncExg i Returning i + λ 4 Lowdoc i Returning i IncExg i + δx i + ɛ i ) Where Lowdoc i means borrower with less verified credential files measured by LCA. Returning i means returning borrower in the platform, i.e. borrowers successfully obtained loans at least one time and requested loans for more times. The explanation of other variables is described as in previous sections. Table 12 illustrate that inflating income is associated with higher probability to predict the loan defaulted. Low-doc borrowers are more likely to default. However, we find 21

22 that low-doc borrowers with relatively low ex-ante concerns about future credit access would raise their income are less likely to default. Figure 9 shows the average marginal effects of Low-doc verification by returning borrowers. The positive sloping of marginal effect for first time borrower means that amplifies the impact on income exaggeration and is associated with a significantly higher probability of delinquency than returning borrowers. However, the low-doc borrower has a larger impact on the probability of default, while the slope of the marginal effect for returning borrowers is negative but not significantly different from 0.2. It indicates that these borrowers are more likely to maintain their credit records. 7 Conclusion This paper studies the impact of financial innovation in the form of P2P lending on the role of verification and its consequences for income exaggeration. We present empirical evidence consistence with borrowers with complete verification can facilitate debt financing and reduce default rate. However, borrowers with incomplete verification lead to misrepresenting their income upwardly. In other words, borrowers who provide less verified documents have a more significant extent of income exaggeration. This leads to higher default rates for this group when compared to the default rates of more thoroughly verified borrowers. Our further analysis uncovers that low-doc borrowers who exaggerate their income are more likely to increase the probability of default, but if they concern about future credit availability in the platform are less likely to default. From a policy perspective, a regulation blanket mandating adjusted verification (i.e., only remain income, employment, identity and credit report for verification ) is consistence with our prediction. However, the platforms is turning to attract more offline loans subject to higher verification cost. A possible explanation is that the government unveiled a regulatory framework in The platform is accused of misappropriating operation that leads to financial instability. The regulation may be excessively restrictive and lead to credit rationing for the subset of the population that faces high information verification costs. In the long run, the misallocation of financial resources may have serious unintended consequences for the economic growth. Our analysis therefore suggests that regulators/platforms are seeking to establish a sharing verification system with significant ex ante low information costs. 22

23 References Abadie, A. and Imbens, G. W. (2006). Large sample properties of matching estimators for average treatment effects. econometrica, 74(1): Abadie, A. and Imbens, G. W. (2011). Bias-corrected matching estimators for average treatment effects. Journal of Business & Economic Statistics, 29(1):1 11. Akerlof, G. A. (1970). The market for lemons : Quality uncertainty and the market mechanism. The quarterly journal of economics, pages Allen, F., Qian, J., and Qian, M. (2005). Law, finance, and economic growth in china. Journal of financial economics, 77(1): Ambrose, B. W., Conklin, J., and Yoshida, J. (2016). Credit rationing, income exaggeration, and adverse selection in the mortgage market. The Journal of Finance, 71(6): Bacharach, M. and Gambetta, D. (2001). Trust in signs. Trust in society, 2: Bester, H. (1985). Screening vs. rationing in credit markets with imperfect information. The American Economic Review, 75(4): Blackwell, D. W., Noland, T. R., and Winters, D. B. (1998). The value of auditor assurance: Evidence from loan pricing. Journal of accounting research, 36(1): Burtch, G., Ghose, A., and Wattal, S. (2015). The hidden cost of accommodating crowdfunder privacy preferences: a randomized field experiment. Management Science, 61(5): Chen, D. and Han, C. (2012). A comparative study of online p2p lending in the usa and china. Journal of Internet Banking and Commerce, 17(2):1 15. Chomsisengphet, S. and Pennington-Cross, A. (2006). The evolution of the subprime mortgage market. Federal Reserve Bank of St. Louis Review. Collier, B. C. and Hampshire, R. (2010). Sending mixed signals: Multilevel reputation effects in peer-to-peer lending markets. In Proceedings of the 2010 ACM conference on Computer supported cooperative work, pages ACM. 23

24 Demyanyk, Y. S., Loutskina, E., and Kolliner, D. (2017). The taste of peer-to-peer loans. Available at SSRN: Dorfleitner, G., Priberny, C., Schuster, S., Stoiber, J., Weber, M., de Castro, I., and Kammler, J. (2016). Description-text related soft information in peer-to-peer lending evidence from two leading european platforms. Journal of Banking & Finance, 64: Duarte, J., Siegel, S., and Young, L. (2012). Trust and credit: The role of appearance in peer-to-peer lending. The Review of Financial Studies, 25(8): Edelberg, W. (2003). Risk-based pricing of interest rates in household loan markets. FEDS Working Paper No Available at SSRN: Eid, N., Maltby, J., and Talavera, O. (2016). Income rounding and loan performance in the peerto-peer market. Available at SSRN: Feng, Y., Fan, X., and Yoon, Y. (2015). Lenders and borrowers strategies in online peer-to-peer lending market: An empirical analysis of ppdai. com. Journal of Electronic Commerce Research, 16(3):242. Freedman, S. and Jin, G. Z. (2008). Do social networks solve information problems for peer-to-peer lending? evidence from prosper. com. NET Institute Working Paper No Available at SSRN: Freedman, S. and Jin, G. Z. (2014). The signaling value of online social networks: lessons from peer-to-peer lending. Technical report, National Bureau of Economic Research. Freedman, S. M. and Jin, G. Z. (2011). Learning by doing with asymmetric information: evidence from prosper. com. Technical report, National Bureau of Economic Research. Garmaise, M. J. (2015). Borrower misreporting and loan performance. The Journal of Finance, 70(1): Hollingshead, A. B. et al. (1975). Four factor index of social status. Iyer, R., Khwaja, A. I., Luttmer, E. F., and Shue, K. (2015). Screening peers softly: Inferring the quality of small borrowers. Management Science, 62(6):

25 Jaffee, D. and Stiglitz, J. (1990). Credit rationing. Handbook of monetary economics, 2: Jiang, W., Nelson, A. A., and Vytlacil, E. (2014). Liar s loan? effects of origination channel and information falsification on mortgage delinquency. Review of Economics and Statistics, 96(1):1 18. Kumar, S. (2007). Bank of one: Empirical analysis of peer-to-peer financial marketplaces. AMCIS 2007 Proceedings, page 305. Lanza, S. T., Flaherty, B. P., and Collins, L. M. (2003). Latent class and latent transition analysis. Handbook of psychology. Lin, M., Prabhala, N. R., and Viswanathan, S. (2009). Judging borrowers by the company they keep: Social networks and adverse selection in online peer-to-peer lending. SSRN elibrary. Lin, M., Prabhala, N. R., and Viswanathan, S. (2013). Judging borrowers by the company they keep: friendship networks and information asymmetry in online peer-to-peer lending. Management Science, 59(1): Mian, A. and Sufi, A. (2009). The consequences of mortgage credit expansion: Evidence from the us mortgage default crisis. The Quarterly Journal of Economics, 124(4): Mian, A. and Sufi, A. (2017). Fraudulent income overstatement on mortgage applications during the credit expansion of 2002 to The Review of Financial Studies, 30(6): Miller, S. (2015). Information and default in consumer credit markets: Evidence from a natural experiment. Journal of Financial Intermediation, 24(1): Milne, A. and Parboteeah, P. (2016). The business models and economics of peer-to-peer lending. Pope, D. G. and Sydnor, J. R. (2011). Whats in a picture? evidence of discrimination from prosper. com. Journal of Human Resources, 46(1): Puro, L., Teich, J. E., Wallenius, H., and Wallenius, J. (2010). Borrower decision aid for peopleto-people lending. Decision Support Systems, 49(1): Serrano-Cinca, C., Gutiérrez-Nieto, B., and López-Palacios, L. (2015). Determinants of default in p2p lending. PloS one, 10(10):e

26 Spence, M. (1973). Job market signaling. The quarterly journal of Economics, pages Spence, M. (2002). Signaling in retrospect and the informational structure of markets. American Economic Review, pages Stiglitz, J. E. and Weiss, A. (1981). Credit rationing in markets with imperfect information. The American economic review, 71(3): Tao, Q., Dong, Y., and Lin, Z. (2017). Who can get money? evidence from the chinese peer-to-peer lending platform. Information Systems Frontiers, pages Wei, Z. and Lin, M. (2016). Market mechanisms in online peer-to-peer lending. Management Science. Zhang, J. and Liu, P. (2012). Rational herding in microloan markets. Management science, 58(5):

27 A Tables Table 1. Variable Definitions Variables Definition Loan information Successfully funded 1 if loan is successful funded, else 0 Default 1 if loan is defaulted, else 0 Interest rate (%) Annual percentage rate on the loan Loan amount (RMB) loan amount on request Maturity (Months) Period of a loan Borrower s history in the platform Number of loan requests How many loan requests of borrowers have been in the platform Number of successful requests How many loan requests of borrowers have been successful in the platform Number of repaid How many times borrowers have been repaid in the platform Number of arrears How many times of borrowers arrears in the platform (up to 30 days) Number of severe arrears How many times of borrowers arrears in the platform (up to 90 days) Borrowers characteristics Monthly income (RMB) Borrower s monthly income in RMB Occupation Borrowers occupation Employment length Length of employment in years Employment sector The sector of Borrowers employment, such as government, sales Firm size Employees number of borrowers firm Location Borrowers working city Self-description Borrowers disclosure to describe loan purpose Age Borrowers age Education 0 is high school and below, 1 is college, 2 is undergraduate and 3 is postgraduate Marital status 1 is widowed, 2 is married, 3 is single and 4 is divorced Home owner 1 means homeowner, else 0 Mortgage 1 means borrowers have existing mortgage, else 0 Vehicle owner 1 means borrowers have vehicle(s), else 0 Vehicle loan 1 means borrowers have existing vehicle loan, else 0 Verification status Income verification 1 means income is validated, else 0 Employment verification 1 means employment is validated, else 0 Credit report verification 1 means credit report is validated, else 0 Identity verification 1 means identity is validated, else 0 Address verification 1 means address is validated, else 0 Marriage verification 1 means marital status is validated, else 0 Education verification 1 means education is validated, else 0 Mobile phone number verification 1 means mobile phone number is validated, else 0 Homeowner verification 1 means homeowner is validated, else 0 Car owner verification 1 means car owner is validated, else 0 Video verification 1 means video interview is validated, else 0 Social Media verification 1 means social media such as Weibo, kaixin is validated, else 0 27

28 Table 2. Descriptive statistics: All listings and funded loans. Online sample. This table reports average/min/max loan characteristics, borrowers characteristics and verification carried out by Renrendai.com for the online sample for years Funded loans means that all loans in this group are successfully granted. Full sample Funded loans mean sd min max mean sd min max Successfully funded Default Interest rate (%) Loan amount (RMB) 60, , , , , , , , Maturity (Months) Monthly income (RMB) 10, , , , , , Number of loan requests Number of successful requests Number of repaid Number of arrears Number of severe arrears Age Length of Self-description , , Home owner Mortgage Vehicle owner Vehicle loan Employment Verification Income Verification Identity Verification Credit Report Verification Homeowner Verification Car Verification Education Verification Mobile Verification Address Verification Marriage Verification Video Verification Social Media Verification Number of Verification Number of observations 567,955 21,549 28

29 Table 3. Relationship between the number of verification and loan performance This table reports the results of the following analysis: P robability(success i = 1) = Φ(α + βnumv erification i + δx i + ɛ i ), P robability(default i = 1) = Φ(α + βnumv erification i + δx i + ɛ i ), LoanInformation i = α + βnumv erification i + δx i + ɛ i, Where the dependant variable Success i measures whether a loan i had been funded and Default i is an indicator of measuring default of on loan i and Φ is the cumulative distribution function of the standard normal. NumV erification i is the number of verification submitted by a borrower. LoanInformation i contain contract interest rate, loan size, and maturity. The vector X i includes information loan characteristics (e.g. interest rate, loan amount, maturity and the length of self-disclosure) and borrower characteristics (the borrower s histories in the platform, borrower age, borrower income, whether the borrower is a homeowner, and with or without mortgage, car owner, and with or without car finance, borrower location, occupation, employment sector, employment length, the firm size for a borrower employed, education and marital status). We also include the variable indicating loan origination year to control the cohort effects. The online sample is from Renrendai.com. *** p < 0.01, ** p < 0.05, * p < 0.1 (1) (2) (3) (4) (5) Dependent Variable: Successfully funded Default Interest rate (%) Maturity (Months) Log(Amount) Number of Verification (0.003) (0.008) (0.007) (0.029) (0.003) Interest rate (%) (0.002) (0.008) (0.030) (0.003) Maturity (Months) (0.001) (0.002) (0.002) (0.001) log amount (0.006) (0.022) (0.019) (0.079) Borrower Characteristics Yes Yes Yes Yes Yes Origination Year Fixed Effects Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes R Number of observations 322,427 16,723 17,913 17,913 17,913 29

30 Table 4. Probability of funding and different type of verification This table reports the results of the following analysis: P robability(success i = 1) = Φ(γ + λv erification i + δx i + ɛ i ), Where the dependent variable Success i and control variables X i are defined as in Table 3. V erification i represent the range of verification types, which contains validation status for employment, income, credit report, identity, mobile phone number, address, homeowner, car owner, social media, video interview, marital status, and education. We estimate whether the funding probability of loan requests are correlated with each verification type. The online sample is from Renrendai.com. *** p < 0.01, ** p < 0.05, * p < 0.1 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Dependent Variable: Success Success Success Success Success Success Success Success Success Success Success Income Verification (0.017) Employment Verification (0.017) Credit Report Verification (0.011) Social Media Verification (0.018) Marriage Verification (0.015) Mobile Verification (0.015) Homeowner Verification (0.014) Car Verification (0.016) Education Verification (0.014) Address Verification (0.014) Video Verification (0.018) Borrower Characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Origination Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of observations 322, , , , , , , , , , ,427 30

31 Table 5. Probability of default and different type of verification for completed online loans This table reports the results of the following analysis: P robability(default i = 1) = Φ(γ + λv erification i + δx i + ɛ i ), Where the dependent variable Default i and control variables X i are defined as in Table 3. The variable V erification i is defined as in Table 4. We examine the relation between the probability of default and each verification type. The online sample is from Renrendai.com. *** p < 0.01, ** p < 0.05, * p < 0.1 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Dependent Variable: Default Default Default Default Default Default Default Default Default Default Default Income Verification (0.034) Employment Verification (0.034) Credit Report Verification (0.033) Social Media Verification (0.055) Marriage Verification (0.039) Mobile Verification (0.042) Homeowner Verification (0.037) Car Verification (0.045) Education Verification (0.045) Address Verification (0.037) Video Verification (0.048) Borrower Characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Origination Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of observations 16,723 16,723 16,723 16,723 16,723 16,723 16,723 16,723 16,723 16,723 16,723 31

32 Table 6. Interest rate and different type of verification for funded online loans This table reports the results of the following analysis: InterestRate i = γ + λv erification i + δx i + ɛ i, Where the dependent variable InterestRate i and control variables X i are defined as in Table 3. The variable V erification i is defined as in Table 4. We test that borrower s debt cost for verifying the different type of documentation. The online sample is from Renrendai.com. *** p < 0.01, ** p < 0.05, * p < 0.1 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Dependent Variable: Interest Interest Interest Interest Interest Interest Interest Interest Interest Interest Interest Income Verification (0.030) Employment Verification (0.030) Credit Report Verification (0.028) Social Media Verification (0.042) Marriage Verification (0.034) Mobile Verification (0.032) Homeowner Verification (0.033) Car Verification (0.039) Education Verification (0.034) Address Verification (0.034) Video Verification (0.043) Borrower Characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Origination Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes R Number of observations 17,913 17,913 17,913 17,913 17,913 17,913 17,913 17,913 17,913 17,913 17,913 32

33 Table 7. Descriptive statistics: Funded loans from online and offline. Online sample and matched offline sample. This table reports average monthly income, loan characteristics and verification carried out by Renrendai.com for the online sample and matched offline sample for years (1) (2) Online loans Offline audited mean sd mean sd Monthly income (RMB) 14, , , , Interest rate (%) Loan amount (RMB) 24, , , , Maturity (Months) Employment Verification Income Verification Identity Verification Credit Report Verification Homeowner Verification Car Verification Education Verification Mobile Verification Address Verification Marriage Verification Video Verification Social Media Verification Number of Verification Number of observations 21,549 58,408 33

34 Table 8. Average treatment effect on Log(income) by offline audited group This table presents the average treatment effect on logarithmic income compared with offline audited loans. The results show that the matching estimate of income reported by online applicants is negatively related to the offline audited group, which indicates the evidence of inflating income by online borrowers. The result is consistent with the matching estimate of propensity score matching method. *** p < 0.01, ** p < 0.05, * p < 0.1 ATE Log(income) Nearest-neighbor matching with bias correction Propensity score matching Offline Audited (0.033) (0.046) Number of observations 72,930 72,930 34

35 Table 9. Income exaggeration and the number of verification for funded online loans This table reports the results of the following analysis: IncExg i = α + βnumv erification i + δx i + ɛ i P robability(default i = 1) = Φ(α + βnumv erification i + νincexg + δx i + ɛ i ), Where the dependent variable IncExg i represents the extent of income exaggeration. Default i is defined as in Table 3, V erification i and control variables X i are defined as in Table 4. We examine the relationship between the number of verification and the extent of income exaggeration and the their impact on delinquency rates. The online sample is from Renrendai.com, income exaggeration is calculated by comparing the matched offline sample. *** p < 0.01, ** p < 0.05, * p < 0.1 (1) (2) Dependent variable: IncExg Default Number of Verification (0.003) (0.008) IncExg (0.027) IncExg Number of Verification (0.006) Number of observations

36 Table 10. Income exaggeration and different type of verification for funded online loans This table reports the results of the following analysis: IncExg i = γ + λv erification i + δx i + ɛ i Where the dependent variable IncExg i is defined as in Table 9. V erification i and control variables X i are defined as in Table 4. We examine which type of verification impact on the extent of income exaggeration. The online sample is from Renrendai.com, income exaggeration is calculated by comparing the matched offline sample. *** p < 0.01, ** p < 0.05, * p < 0.1 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Dependant variable: IncExg IncExg IncExg IncExg IncExg IncExg IncExg IncExg IncExg IncExg IncExg Income Verification (0.017) Employment Verification (0.017) Credit Report Verification (0.016) Social Media Verification (0.023) Marriage Verification (0.019) Mobile Verification (0.018) Homeowner Verification (0.018) Car Verification (0.022) Education Verification (0.019) Address Verification (0.019) Video Verification (0.024) Borrower Characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Origination Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of observations 17,913 17,913 17,913 17,913 17,913 17,913 17,913 17,913 17,913 17,913 17,913 36

37 Table 11. The impact of income exaggeration with each type verification on delinquency rates for funded online loans This table reports the results of the following analysis: P robability(default i = 1) = Φ(α + λv erification i + νincexg + δx i + ɛ i ), Where the dependent variable Default i is defined as in Table 3. IncExg i is defined as in Table 9, V erification i and control variables X i are defined as in Table 4. We examine different type of verification with income exaggeration effect on the extent of income exaggeration. The online sample is from Renrendai.com, income exaggeration is calculated by comparing the matched offline sample. *** p < 0.01, ** p < 0.05, * p < 0.1 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Dependent variable: Default Default Default Default Default Default Default Default Default Default IncExg (0.017) (0.017) (0.017) (0.016) (0.016) (0.016) (0.016) (0.016) (0.016) (0.016) Income Verification IncExg (0.027) Employment Verification IncExg (0.027) Credit Report Verification IncExg (0.025) Social Media Verification IncExg (0.045) Marriage Verification IncExg (0.030) Homeowner Verification IncExg (0.026) Car Verification IncExg (0.029) Education Verification IncExg (0.038) Address Verification IncExg (0.030) Video Verification IncExg (0.028) Borrower Characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Origination Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of observations p < 0.05, p < 0.01, p <

38 Table 12. The impact of Low doc, returning borrowers and income exaggeration on delinquency rates for funded online loans This table reports the results of the following analysis: P robability(default i = 1) =Φ(α + β 1 Lowdoc i + β 2 Returning i + β 3 Lowdoc i Returning i + λ 1 IncExg i + λ 2 Lowdoc i IncExg i + λ 3 IncExg i Returning i + λ 4 Lowdoc i Returning i IncExg i + δx i + ɛ i ) Where the dependent variable Default i is defined as in Table 3. IncExg i is defined as in Table 9, V erification i and control variables X i are defined as in Table 4. Lowdoc i means borrower with incomplete verification. Returning i means returning borrower in the platform, i.e. borrowers successfully obtained loans at least one time and requested loans for more times. We examine the impact of Low doc, returning borrowers and income exaggeration on delinquency rates. The online sample is from Renrendai.com, income exaggeration is calculated by comparing the matched offline sample. *** p < 0.01, ** p < 0.05, * p < 0.1 Dependent variable: Default IncExg (0.026) Lowdoc (0.049) Lowdoc IncExg (0.032) Returning Borrower (0.041) Returning Borrower IncExg (0.027) Lowdoc Returning Borrower (0.058) Lowdoc Returning Borrower IncExg Borrower Characteristics Yes Origination Year Fixed Effects Yes Other controls Yes Number of observations

39 B Figures Figure 1. Loan Listing and Verification Status The figure shows that a loan listing contains an annual interest rate, the amount of loan, the period of repayment, the guarantee type, a credit score issued by Renrendai.com, and various pieces of personal characteristic information (age, income, location, occupation, employer s information, education, marital status, homeowner,car owner and borrowing histories in the platform). Furthermore, the below table exhibits verification status which includes various type of verification. The validate documentation is displayed by a green tick with verification time. For example, the below figure shows that the borrower validates his/her credit report, identity card, education, employment, income, mobile phone number and social media. Source: Renrendai.com Loan information Borrower s information 39

40 Figure 2. Average number of verification for all listings and funded loans over the month The figure shows the average number of verification for all listings and funded loans over the month. Source: Renrendai.com, Observation number are 567,955 for all listings and 21,549 for funded loans. 40

41 Figure 3. Transaction volume, outstanding debt and average interest of P2P lending in China The figure shows the trend of total transaction volume (left axis) and outstanding loan volume of P2P lending in China and average interest (right axis) in the marketplace between January 2014 and July Source: Wind Information, 41

42 Figure 4. Total and average principal amount issued over year of Origination; Total new borrowers and lenders over Year of origination The figure shows the total and average principal amount issued over the year of origination and total new borrowers and lenders over Year of origination. Source: Renrendai.com Total Principal Amount Issued over Year of Origination Average Principal Amount Issued over Year of Origination Total Principal Amount Issued 0 5.0e e Source: Renrendai.com/We.com Average Principal Amount Issued 0 20,000 40,000 60,000 80, Source: Renrendai.com/We.com Total New Borrowers for Funded Loans 0 50, , ,000 Total New Borrowers for Funded Loans over Year of Origination Source: Renrendai.com/We.com Total New Lenders for Funded Loans 0 50, Total New Lenders for Funded Loans over Year of Origination Source: Renrendai.com/We.com 42

43 Figure 5. The operation of Renrendai.com The figure shows the operation of Renrendai.com. For the online channel, potential borrowers raise a requisition on the platform. After evaluating the credentials of the borrowers, the loan request is made open for the lenders to screening. Lenders then can bid all listed loans. The borrower will receive the fund if a loan request is fully funded. Borrowers can apply for loans via the offline channel (Ucredit.com), the documents provided are similar to the online application, but the verification provision is different (see section 3). Other processes are same as online operation. Offline- Ucredit.com Borrowers Renrendai.com Lenders 43

44 Figure 6. Loan Application Process The figures show the loan application form of Renrendai.com. A borrower needs to fill in the loan purpose, amount, maturity, maximum interest rate they can accept and self-description. The borrower must provide personal characteristics and related supporting documents subject to verification. Source: 44

45 Figure 7. Average verification percentage for funded loans over the month The figure shows average verification percentage for funded loans over the month. The monthly percentage of income, employment and credit report verification fluctuate around 15% over the period, increasing rapidly from 2015 and reach to almost 100% in October 2015 (policy change). The identity verification is almost a horizontal line at 100%. Because the verification is compulsory for borrowers. The monthly percentage of other types of verification maintain a significant level until 2015 and reduce to 0 in October Source: Renrendai.com 45