Online Program

Alternative approaches to LGD modeling

Keywords: LGD, loss given default, Basel II, bipolar distribution

Prediction of Loss Given Default (LGD) is an important area of predictive analytics in the financial services and banking industry. It is also a key component of Basel II requirements which lending institution have to meet. Hence it is imperative to have an accurate prediction methodology for LGD.

The challenge with LGD modeling is that it has a bipolar distribution. Most commonly used statistical techniques such as OLS regression rely on a normal distribution or variations of bell shaped distributions of the target variable. There are other GLM techniques that address other frequently observed distributions such as the exponential family of distributions. However, the bipolar distribution is relatively uncommon and requires special techniques.

We present alternative approaches to LGD modeling and show proven ways of achieving high predictive accuracy compared with other methods