Parametric versus Nonparametric Methods in Risk Scoring
Accurately assessing risk is key to providing appropriately priced loans to rural producers. This paper examines non-parametric techniques for risk scoring to avoid the erroneous rejection of credit-worthy loan applicants. Both parametric and non-parametric techniques were tested against simulated data and then evaluated on microfinance loan applicants in Peru. Because non-parametric techniques impose fewer modeling assumptions, they are able to better predict default.
The main innovation of this scoring methodology is the use of non-parametric methods in risk scoring. Non-parametric methods do not impose a functional form that relies on a distribution. Instead they allow the data to reveal the best functional form. By imposing fewer assumptions on the model this reduces the risk of rejecting a credit-worthy loan applicant.