The value of financial intermediation: Evidence from online debt crowdfunding

时间：2025-07-09
作者：Fabio Braggion,Alberto Manconi,Nicola Pavanini

Abstract

Most online marketplaces are peer-to-peer. Credit ones, however, are not and they have resurrected many features of traditional financial intermediaries. To understand why, we use online credit as a laboratory to investigate the value of financial intermediation. We develop a structural model of online debt crowdfunding and estimate it on a novel database. We find that abandoning the peer-to-peer paradigm raises lender surplus, platform profits, and credit provision, but exposes investors to liquidity risk. A counterfactual where the platform resembles a bank by bearing liquidity risk can generate larger lender surplus and credit provision when liquidity is low and lenders are risk averse.

1. Introduction

Many online marketplaces, such as Uber, Airbnb, and eBay, operate on a peer-to-peer paradigm focused on matching buyers with sellers. In the early days, online credit marketplaces — commonly known as debt crowdfunding platforms — adopted a similar approach, where individual lenders would fund specific loans, holding them until maturity. This peer-to-peer setup has since evolved. Most debt crowdfunding platforms now operate under a “marketplace credit” paradigm, where they aggregate loans into portfolios that are sold to investors, eliminating the need for investors to select loans individually. Because the portfolios often have shorter maturities than the underlying loans, the maturity mismatch introduces liquidity risk, i.e., the risk associated with the need to refinance the loans when the portfolio matures. Under the marketplace credit paradigm, investors have borne liquidity risk; yet many platforms have recently begun offering “bank-like” products that resemble traditional deposits, thereby shifting the liquidity risk onto the platforms themselves.

We ask why online credit evolved differently from other online marketplaces, and what is the welfare value of its unique features. These questions are important, because online debt crowdfunding is an increasingly large investment and consumer credit channel (Rau, 2020). Moreover, it provides a clean environment to quantify the welfare value of financial intermediation in general as, in comparison to traditional intermediaries, it is exclusively focused on intermediating credit and it is less exposed to the potential confounding impact of regulation (Buchak et al., 2018).

To address our questions, we build and estimate a structural equilibrium model of online debt crowdfunding that rationalizes its development. The key forces in our model are maturity mismatch and liquidity risk. By funding longer-maturity loans while also allowing lenders to liquidate their investments in the short term, the loan portfolios offered under the marketplace and bank-like credit paradigms increase credit provision and create value for lenders, borrowers, and the platform itself. However, the extent to which marketplace credit and bank-like credit generate value varies based on liquidity risk and how much investors are willing to tolerate it.

We estimate our model on a novel, hand-collected micro database of the universe of loan applications, actual loans, and loan portfolios on a leading Chinese online debt crowdfunding platform, Renrendai. During our sample period, Renrendai both sold loan portfolios to investors bearing liquidity risk (under the marketplace paradigm) and allowed direct investment in loans (under the peer-to-peer paradigm). The data show lenders’ investment choices: whether they invest directly or buy a portfolio product, and which portfolios they choose if they opt for the latter. We can match this information with the maturity of the loans in the portfolios products, and thus compute a precise measure of maturity mismatch. Renrendai, moreover, allows lenders who do not want to roll over their portfolio investments to sell the underlying loans on its internal secondary market. Our data contain every transaction in the primary and secondary markets, and reveal how fast a loan is resold, thus quantifying its liquidity.

Our main findings are as follows. First, similar to online credit platforms in the U.S. and other countries, we observe the transition from peer-to-peer to marketplace credit. In 2010, when Renrendai was launched, 100% of lending was peer-to-peer; but by the end of our sample in early 2017, over 98% of the loans on Renrendai are funded as part of a marketplace loan portfolio. The key feature of these portfolios is maturity mismatch: whereas their most common maturities are 3, 6, and 12 months, the underlying loans typically mature in 36 months. This exposes the lenders to non-trivial liquidity risk. Moreover, lender investments have become more diversified and less exposed to defaults, especially for portfolio products purchased on the platform, consistent with a change in the platform’s clientele towards investors that are more averse to risk.

Second, the estimates of our structural model shed light on lender preferences for loan and portfolio product characteristics, as well as on the platform’s preferences for individual loan attributes when assembling portfolios. Lenders prefer higher returns, especially for peer-to-peer loans, and portfolio products with lower liquidity risk (shorter resale time on the secondary market). Moreover, lender preferences are heterogeneous: the more sophisticated, active lenders have a stronger preference for yield and a weaker disutility from liquidity risk, whereas the opposite is true for less frequent investors.

Third, we combine our estimates of the lender demand model with a platform profit function to simulate counterfactuals. We compare the baseline marketplace credit with two counterfactual scenarios: peer-to-peer credit, where only direct lending is allowed, and bank-like credit, where the platform sells portfolio products but bears liquidity risk. In the marketplace and bank-like scenarios, the platform maximizes profits by choosing portfolio product target return, the mismatch between portfolio duration and the maturity of the underlying loans, and the resale time of portfolio loans on the secondary market. Marketplace credit appears welfare-improving relative to the peer-to-peer paradigm: the counterfactual allowing only direct lending generates a 73% drop in credit provision and a 45% decline in lender surplus. We find that this is driven by the fact that under the marketplace paradigm, maturity mismatch allows the platform to offer a broader assortment of portfolios, more closely tailored to the lenders’ maturity preferences; on the other hand, the platform’s ability to search, screen, and monitor borrowers plays a lesser role. We also find that, when the platform’s cost of generating liquidity is low—i.e., it can easily repurchase loans for its portfolio products without bearing significant costs, bank-like credit yields similar outcomes to marketplace credit, with only a small drop in platform profits (0.2%).

That comparison is different, however, when we raise the platform’s cost of generating liquid loans. Under higher liquidity costs, relative to bank-like credit, the marketplace paradigm exhibits lower credit provision and lender surplus, but higher platform profits. In other words, when liquidity is low marketplace credit is preferable from the platform’s point of view, but worse for lenders and borrowers. Finally, in counterfactuals where the lenders have weaker utility from yields and stronger disutility from liquidity risk, the bank-like paradigm is a Pareto improvement, raising platform profits too.

These results are consistent with a narrative in which, in the early days of online debt crowdfunding, the platform mainly attracts risk-tolerant lenders, who seek higher returns and have higher welfare under the peer-to-peer and marketplace paradigms. As the platform’s clientele grows, it comes to encompass more risk-averse lenders, who are more sensitive to liquidity risk and have higher welfare under bank-like credit. Our findings are in line with anecdotal evidence about the most mature platforms such as LendingClub, Funding Circle, RateSetter, or Zopa, which have shut down peer-to-peer credit, offering instead securitized (marketplace) loan portfolios to a more risk-tolerant institutional investor clientele as well as, in recent years, traditional banking products to more risk-averse retail investors.

Our paper makes three main contributions. First, it contributes to the literature on the value of financial intermediation. Since the seminal work of Diamond and Dybvig (1983), the theory of financial intermediation points to maturity transformation as a central tool to facilitate the provision of credit for longer-term investment. Empirical work in this literature has used bank-level data to develop liquidity risk indexes (Berger and Bouwman, 2009, Brunnermeier et al., 2012, Bai et al., 2018, Ma et al., 2020) and has estimated the costs and benefits of maturity transformation (Fuster et al., 2017, Segura and Suarez, 2017, Drechsler et al., 2021), focusing on the relation between liquidity risk and financial stability. We also measure liquidity risk; but our focus is different, as we study how it affects credit provision and welfare. With our detailed data, we can construct a precise measure of liquidity risk both at the individual loan and portfolio product level and estimate lenders’ preferences. We are also able to simulate a rich set of counterfactual scenarios, illustrating potential conflicts of interest of the platform vis-à-vis lenders and borrowers. In addition, online debt crowdfunding constitutes a comparatively clean and tractable setting, as its business model is entirely focused on intermediating loans, and, during our sample period, it was less exposed to the potential confounding impact of regulation (Buchak et al., 2018).

Second, our paper provides new results on the design of online debt crowdfunding platforms. Much of the literature has focused on the information aspects of platform design: information provision to investors (Vallée and Zeng, 2019), efficiency of pricing mechanisms (Franks et al., 2021), and the welfare losses associated with asymmetric information (Kawai et al., 2022, DeFusco et al., 2022). We take a different, complementary angle. Building on the evidence that online credit platforms increasingly offer a combination of marketplace loan portfolios and traditional bank-like products, we focus on maturity mismatch and liquidity risk, and their impact on welfare. In that respect we also relate to the literature comparing online and offline credit intermediaries (Buchak et al., 2018, de Roure et al., 2022), as well as to the industrial organization literature on online marketplaces reviewed by Einav et al. (2016). Our results help rationalize the evolution of the design of online debt crowdfunding platforms from peer-to-peer to a combination of marketplace and bank-like credit.

Third, our paper contributes to the literature on structural estimation in financial intermediation (Egan et al., 2017, Crawford et al., 2018, Wang et al., 2022), online credit (Kawai et al., 2022, Xin, 2020, Tang, 2020, DeFusco et al., 2022), and online marketplaces in general (Dinerstein et al., 2018, Einav et al., 2018, Fréchette et al., 2019, Farronato and Fradkin, 2022). Work in this literature has so far focused on buyers and sellers or lenders and borrowers, placing less emphasis on an active role for platforms. In contrast, our approach directly models the design of portfolio products by the platform.

2. Institutional background, data, and descriptive evidence

2.1. Development of the business model of online debt crowdfunding

China, the U.S., and the U.K. are the largest markets for online credit, accounting for about two-thirds of total lending volume (Cornelli et al., 2020). Over 2014–2019, online credit accounted for about 7.5% of total consumer credit in China.¹

Initially, online credit platforms operated solely through direct, peer-to-peer lending, where lenders selected and held loans until maturity. Over time, two key innovations emerged: platforms began offering portfolio products, often assembled by robo-advisors, and established secondary markets where loans could be traded before maturity. These features define a new “marketplace credit” paradigm of online debt crowdfunding. Marketplace credit enables maturity mismatch in portfolio products, allowing them to include longer-term loans that can be resold on the secondary market when the portfolio matures. A defining aspect of this paradigm is that investors bear liquidity risk, meaning they might have to sell at a discount or wait longer to liquidate their investments. In the U.S., LendingClub introduced a secondary market for loans in 2008 and Prosper in 2009; in the U.K., Funding circle, RateSetter, and Zopa opened a secondary market in 2010, whereas in continental Europe, Bondora launched it in 2013. Virtually every Chinese online credit platform offered portfolio products and set up secondary markets shortly after their establishment.

More recently, many online credit platforms have increased their reach to retail investors by selling bank-like savings products. In bank-like products, the investor can liquidate at any time, but the intermediary bears the liquidity risk. The platform still provides marketplace portfolio products, but they are targeted to institutional investors. For instance, in 2021 LendingClub acquired Radius Bank and started to offer deposit services to retail investors; but at the same time, institutional investors on LendingClub can still invest in portfolio products where they bear the liquidity risk.

Online debt crowdfunding in China experienced a similar evolution despite recently undergoing a restructuring driven by regulation. A number of platforms have shut down, and others have become “loan aid agencies” selling services to traditional intermediaries. However, several Chinese platforms that continue to operate offer bank-like products. For instance, in 2019 FinVolution (formerly Paipaidai) acquired a 4.99% stake in Fujian Strait Bank and formed an alliance with the bank focused on consumer lending, and 9fgroup Tech invested in Hubei Consumer Finance Company through its wholly-owned subsidiary; in 2023, Lufax Holding announced its acquisition of Ping An OneConnect Bank (Hong Kong) Limited.2 Despite the regulatory tightening, Chinese online credit companies continue to pursue bank-like activities through alliances, acquisitions, and by shifting their focus to Hong Kong.

2.2. Renrendai

We base our analysis on a novel, hand-collected database covering the universe of loan applications and credit outcomes on debt crowdfunding platform Renrendai (人人贷). During our sample period, Renrendai was the fifth largest player in the sector in China, and as of 2019 it had a 5% market share.3 Between its launch in 2010 and the end of our sample period in February 2017, Renrendai had a cumulative turnover of ¥25 bn ($3.7 bn) and registered over 1 million active users between borrower and lender accounts.

During our sample period, Renrendai is representative of a typical debt crowdfunding platform as it operates like most online credit platforms in China as well as other countries. In Renrendai, users can be borrowers or lenders. Borrowers pay a small participation fee to apply for a loan on the platform.4 When submitting a loan application, a prospective borrower specifies the amount she seeks, and proposes an interest rate and maturity. Renrendai pre-screens loan applications, assigning a credit rating to borrowers. Following this step, loan applications become visible to prospective lenders, and are available on Renrendai’s platform for one week. If an application is not fully funded within that time window, it is considered unsuccessful and it is turned down; Renrendai then removes the application from its website and the borrower does not receive the funds she requested.

Lenders pay no fees and can invest on Renrendai via two channels: direct (peer-to-peer) credit, where the lender selects the individual loans she intends to fund, and marketplace credit, where the platform sells the lender a share in a diversified portfolio of loans. Marketplace lenders can choose from a menu of portfolios known as Uplan (Uu计划). Renrendai offers every day a new set of Uplan portfolios, differentiated by target annual return (ranging between 6% and 11%), maturity (between 3 and 24 months), and minimum investment amount (¥1000 or ¥10,000). At maturity, Uplan lenders can roll their investment over or liquidate it. If they liquidate, the platform places the underlying loans on the secondary market, and does not bear the liquidity risk: the lenders do not receive a payment until all the corresponding loans have been resold. The loan is sold “at par”, i.e., at a fixed price of ¥1 for each ¥ loaned. As the price does not adjust to market conditions, the seller may not be able to find immediately a buyer and might be forced to wait before disposing of the loan. Renrendai makes a profit on Uplan based on the spread between the interest payments it receives on the underlying loans and the returns it pays to the lenders.

Fig. 1 breaks down credit at Renrendai during our sample period between direct and marketplace loans. When Renrendai was first launched, online debt crowdfunding was based on the older peer-to-peer paradigm, and 100% of loans were direct. Portfolio investment was introduced in December 2012, and since then we observe a steady rise of marketplace credit, reaching 98% of total investment at the end of our sample period in February 2017. We build on this stylized fact, and investigate the welfare effects of the marketplace credit model in comparison to alternative platform designs. In October 2020, regulatory pressure to limit online lending led to withdrawals and low liquidity in Renrendai’s secondary market. Since this happened more than three years after our sample period, it is unlikely that loans from before March 2017 are the root of these developments.⁵

Fig. 1. Direct and marketplace loans at Renrendai, 2010Q4–2017Q1.

The figure plots the outstanding volumes of loans at Renrendai, for each calendar quarter over the period 2010–2017. The dark bars denote direct, or peer-to-peer, loans, and the lighter-shaded bars loans that are part of portfolio products, i.e., marketplace loans.

2.3. Data; loan applications, funded loans, and portfolio products

Our data cover 955,405 loan applications and 376,219 funded loans, associated with 358,383 borrowers and 351,333 lenders on Renrendai. They report detailed information on loan applications, funded loans, portfolio products, borrower characteristics, and individual lender IDs. Table 1 presents descriptive statistics for loan applications and funded loans. Around 40% of loan applications ultimately obtain funding, and among those the average default rate is 1%. The median loan funded on the platform has size of about ¥62,000 ($9000) and maturity of 36 months; it pays a 10.8% annual interest rate, and is financed by 45 lenders. To contain dimensionality, we aggregate these data into categories based on loan size, maturity, interest rate, and borrower creditworthiness, defined in Appendix A.

Table 2 provides descriptive statistics for the portfolio products sold on Renrendai. The median portfolio product offers an 8.5% return, has a maturity of 6 months, a total size of ¥3 million, and a minimum investment amount of ¥1000. For each portfolio product, we also observe every investment that the platform makes on behalf of each lender and the exact time of the investment, as well as whether the lenders roll their investments over at maturity; just over 12% of portfolio investments are rolled over on average. When lenders liquidate their investment, we can measure the time until the portfolio share is sold on the secondary market, or resale time: on average, about half a day.6

Table 1. Summary statistics, loans.

The table reports summary statistics for loan applications (panel A) and funded loans (panel B) on Renrendai, over the period 2010–2017. One observation corresponds to a loan. All variables are defined in detail in Appendix A.

The resale time of portfolio shares at maturity plays an important role in our analysis, as it captures the liquidity risk that lenders face when investing in a portfolio product. On average the secondary market for loans is liquid, but the resale time distribution has a thick right tail. Out of 2810 portfolio products in our data, around 9.5% have resale time in excess of one day. For these cases, the mean resale time is 4.2 days and the maximum is 88 days. Note that all lenders investing in the same portfolio face the same resale time, as the platform waits until all non-rolled over loans are sold on the secondary market before liquidating lenders.

Table 2. Summary statistics, portfolio products.

The table reports summary statistics for portfolio products offered on Renrendai, over the period 2010–2017. One observation corresponds to a portfolio product. The number of observations is smaller for Rollover rate and amount, because portfolio products in the earlier years did not provide the rollover option, and for Resale time because around one third of portfolio products have not reached maturity by the end of our sample period, so that a resale time cannot be observed.

Fig. 2. Maturity mismatch on Renrendai’s portfolio products.

The figure plots the outstanding amounts of portfolio products sold by Renrendai and their underlying loans by maturity bins. The light bars represent the total outstanding amounts of portfolio products. The dark bars represent the total amount of outstanding loans underlying the portfolio products.

2.4. Borrowers and lenders; maturity mismatch and liquidity risk

Table 3 displays descriptive statistics for Renrendai’s borrowers and lenders. The average borrower is 34 years old, male, and has a monthly gross income of ¥12,520 ($1880). Annual income per capita in China as of the end of our sample in 2017 is ¥25,974 ($3900; ¥2,165 per month), and in Beijing, the wealthiest part of the country, ¥57,230 ($8600; ¥4,769 per month; source: National Bureau of Statistics of China).

Fig. 2 describes the distribution of the maturities of portfolio products and their underlying loans. The most popular portfolio products have maturities under 12 months, and no portfolio has maturity beyond 24 months. Their underlying loans, on the other hand, have longer maturities, with the bulk of the distribution beyond 15 months. This evidence indicates the extent of maturity mismatch and the potential exposure to liquidity risk: portfolio products with maturity 3, 6, or 12 months comprise loans with maturity almost exclusively 24 or 36 months, and the weighted-average portfolio product maturity mismatch is about 22 months. This value is close to estimates of maturity mismatch for consumer credit at traditional banks reported in the literature (e.g., Drechsler et al., 2021 Table A.2). Only a small portion of the loans in the average portfolio product (0.14%) matures prior to the product’s expiration; in those cases, the proceeds on those loans are reinvested by the platform.

Table 3. Summary statistics, borrowers and lenders.

The table reports summary statistics for borrowers (panel A) and lenders (panel B) on Renrendai, over the period 2010–2017. One observation corresponds to one borrower in panel A, and in panel B respectively to one day for the first two variables, a day-lender for the third, and to one lender for the remaining four. All variables are defined in detail in Appendix A.

The data, moreover, suggest that changes in investor population accompany the growth of Renrendai (and of debt crowdfunding in general). We observe a downward trend among investor portfolios in concentration (with the HHI going from 17% in 2010–2013 to 12% in 2014–2017) and default rates (from 2.4% in 2010–2013 to 0.5% in 2013–2017), driven especially by the Uplan portfolios. That is consistent with the arrival on the platform of investors who are more focused on limiting risk than on seeking yield. These new lenders are less likely to pick individual loans, but prefer to delegate their portfolio choices to Renrendai.

To capture those changes and reflect the increased investor heterogeneity, we focus on the percentage of active lenders on the platform on a given day. We define a lender as active if she is in the top 5% of the distribution of platform use, defined as the number of times she invested up to that date.7 This variable reflects familiarity with the platform and/or laxer financial constraints: because Renrendai requires a minimum investment amount, more frequent investments indicate that the lender has greater financial resources, and should therefore be less liquidity risk-averse. We compute the daily share of active investors as the ratio of active investors to the total number of lenders investing on the platform on a given day. Descriptives for this variable are reported in Table 3.

3. Model

Our model features three players: borrowers, lenders, and a debt crowdfunding platform. Appendix Figure D.1 provides a graphical summary of the model.

3.1. Borrowers

Borrowers post loan applications and, conditional on the loan being funded, make monthly repayments. We treat borrowers as passive agents, which keeps the model tractable and helps us highlight the key drivers in the counterfactuals, where we compare marketplace credit to the alternative lending paradigms, peer-to-peer and bank-like credit. This assumption is justified by three reasons. First, default rates are low (1% on average). This suggests that even if the platform shifts to a bank-like paradigm and attracts a different type of borrower, those borrowers would be at best only marginally safer than under marketplace credit. Second, nearly 80% of loan applications and over 95% of funded loans are made by individuals active on Renrendai only once. These individuals are unlikely to be so familiar with the platform as to condition their decisions on expected lender demand or on the platform’s business model. Moreover, as they typically appear only once, it also unlikely that the platform will face an insolvency–illiquidity spiral, where borrowers struggle to roll over their loans due to lenders’ concerns about rising default rates. Third, borrower characteristics do not exhibit much variation over time, particularly around 2014, when marketplace loan portfolios became the main funding channel.8 This attenuates the possibility that, even though the typical borrower interacts with the platform only once, different types of borrowers may approach the platform in response to a change in the lending model.

Click to read more