Research on probability of default prediction based on loan company's credit fund trading behaviours

Abstract

The probability of default (PD) is an important parameter to quantify credit risk, which is usually impacted by some less frequent accidents, such as market factors or the macroeconomic climate changes. The traditional approaches to estimate PD, such as expert–method or pattern classification based on a set of financial indicators, are heavily dependent on financial reports offered by borrow–customers, and can lead to unreliability and long–time lags in forecast. According to the business schema of loan–fund expenditure surveillance in some commercial banks in China, this paper proposes a support vector machine (SVM) PD classifier constructed on a set of loan–fund expenditure behaviour features, which can be directly collected from the fund trading databases of the banks in time. For the sake of comparison, both of the Logistic and SVM models are tested in this paper to predict PD, and their classification accuracy can be up to 84.6% and 89.4% respectively.