Marketplace Lending: A New Banking Paradigm (Digest summary)

Lending marketplaces, commonly called peer-to-peer lending platforms, are a significant recent fintech innovation whose transactions totaled one-third of US unsecured consumer loan volume in 2016. Linking borrowers with investors of varying degrees of sophistication, the platforms produce information jointly with the investors on the submitted loan applications to enable funding decisions by those investors. This complementary screening of borrowers is a key feature distinguishing the marketplaces from traditional banks, and it motivates the authors’ research questions:

• Is the screening intensity of borrowers directly related to investor sophistication, and does that relationship result in systematic outperformance by more-sophisticated participants?
• Is any such outperformance related to a platform’s own prescreening process and supplementary information provided on borrowers?
• Given investor heterogeneity, what is the optimal platform design in terms of information processing for maximizing loan volume? How does that design mitigate the potential endogenous adverse selection problem of sophisticated investors capturing the good loans?

Given the absence of any publicly available background on investors and their loan portfolios, the authors use data from an algorithmic third-party robo-advisor, LendingRobot (LR). It fully covers portfolios of both retail and institutional investors on the largest platforms, Lending Club and Prosper. Investor segments are represented across an active/passive continuum, investing either using their custom designed models, relying on prepackaged screening algorithms, or buying and holding while following platform-provided loan classifications. The sample period covers all US transactions by LR users during January 2014–February 2017.

The authors rely on a key unanticipated development: the November 2014 removal by Lending Club of half the tracked variables on borrower characteristics it previously provided to investors. Using difference-in-differences methodology, the authors can measure investor performance effects because of this event.

A platform design model is posited that relates loan quality, pricing offered to borrowers, and investor screening costs. It provides the basis for a series of regressions examining investor screening processes and effects of related cost changes, loan default likelihood by investor type, and time-series evolution of platform prescreening and investor relative performance.

Assessing how borrowers, sophisticated and unsophisticated investors, and the platform interact in funding loans, the authors reach these conclusions:

• More-sophisticated investors, using either their custom models or the LR prescreening algorithm, rely on different loan characteristics than retail investors, with their selection of a given loan predicting its significantly lower probability of default. The approximate reduction of average default risk exceeds 20%. Both factors point to the advantage of sophistication.
• The 2014 Lending Club disclosure change increases screening costs for sophisticated investors and reduces their outperformance by more than half. The move protected more-naive investors who rely on the loan categorization scheme of risk subgrades devised by the platform. Participation by multiple investor types aids borrowers through increased loan volume at better pricing.
• Regarding other potential causes for this relative performance decline, no impact is found on fractional loans (which is the focus of naive investors); rather, the impact is confined to the large (i.e., institution-sponsored) loans. Effects from loan composition or interest rate changes are thereby ruled out.
• Platform prescreening accuracy (i.e., loan risk ratings) improves over time.