The aim of this paper is to apply a Gaussian process to decompose the time series of crude default rates into three components: age of the loan, quality of the loan and the exogenous economic environment. This is supported by the empirical result for a mortgage and personal loans portfolio based on five years of historic data. The Gaussian process does not impose an explicit parametric structure to the relationship between the three components and the default rate compared to other methodologies that assume a linear structure. We find that the vintage and economic cycle components are more important drivers of the default rate compared to the age effect and this varies over the economic cycle. The contribution of the economic cycle component to the overall default (for both the mortgage and personal loan portfolios) range from between 20% to 50% depending the position in the economic cycle. In general the economic cycle effect is larger during economic stress.
JEL Classification: C11; C23; C53; E51; G21
Keywords: Credit Risk; Probability of Default; Forecasting; Gaussian process