As compared to traditional lenders, peer-to-peer platforms (P2P) are supposed to use sophisticated data-intensive proprietary algorithms to assess a borrower’s credit risk. Since an increasing number of loans are now being underwritten by these algorithms, either directly through the platform or via partnerships with traditional banks, investors and regulators now need to understand how the algorithms work in practice. In this post, we evaluate one aspect of the performance of these algorithms, namely, whether they make adequate use of predictors that economic theory suggests are drivers of credit risk. Specifically, we focus on unemployment rates in the area where the borrower resides. Using peer-to-peer consumer loans offered by LendingClub, we find that local unemployment rates can significantly improve the predictive power of their algorithm.
Our analysis is based on 600,000+ 36-month maturity loans originated between 2011 and 2015. The number and dollar amounts of loans issued over this period is shown in Figures 1 and 2.
To examine whether unemployment can boost LendingClub’s algorithm, we compare the default rates of loans that have similar default risk according to LendingClub but were issued in areas with different unemployment rates. For each year of loan origination, we first categorize a loan according to LendingClub’s assessment of its credit risk. Specifically, we categorize a loan as “low risk”, “moderate risk,” or “high risk” based on its interest rate — which is set by LendingClub’s algorithm. Then, we categorize a loan as being in a low, moderate, or high unemployment area based on the unemployment rate in the zip-code where the borrower resides. Finally, we measure the unemployment rate effect as the difference in default rates for loans in the same risk category but different unemployment rate category.
The default rates for loans in different risk-unemployment categories is shown in Figure 3.
This figure plots the average default rates for each risk-unemployment category, averaged across the years of origination. Loans issued to borrowers who live in areas with high unemployment rates default at markedly higher rates. For example, for high-risk loans, the default rate in high unemployment areas is about two percentage points higher in that of low unemployment areas. Furthermore, loans in high unemployment areas default at a higher rate across all years. In Figure 4, we plot the differential in default rates for the high-risk loans for each year of origination.
As seen, the differential is positive across all years in the sample and is about 1.5% even for the most recent years.
Returns on the loans in the high-unemployment category have also been noticeably lower. In Figure 5, we compare annualized rates of return across loans from the different risk-unemployment categories. For example, for the high-risk loans, the returns differential is 60 basis points.
Using a Cox proportional hazard model, an industry-standard framework for analyzing default risk, we confirmed that the unemployment effect described here is statistically significant.
Our analysis on drivers of credit risk for P2P loans is ongoing, and we will update you with our new findings in future posts.
Written by Atay Kizilaslan and Aziz Lookman.