Forecasting Default Rates of Peer-to-peer Lending Using Dynamic Regression Models

Abstract

The recent decade witnesses the exponential growth of online peer-to-peer lending, but default dependence between individual peer-to-peer loans has rarely been studied. This paper examines monthly default rates of loans issued by Lending Club from June 2007 to December 2015. We find significant correlations between monthly default rates of different grades of loans and significant relationships between default rates and concurrent macroeconomic variables. We perform linear regression of monthly default rate against unemployment rate and inflation rate, and use an ARMA model to describe the behavior of residuals. Unemployment rate and inflation rate are modeled as seasonal ARIMA processes. We show that our model produces more accurate and consistent forecast of default rate than other models.