Forecasting Net Charge-Off Rates of Banks: A PLS Approach

This chapter relies on a factor-based forecasting model for net charge-off rates of banks in a data-rich environment. More specifically, we employ a partial least squares (PLS) method to extract target-specific factors and find that it outperforms the principal component approach in-sample by construction. Further, we apply PLS to out-of-sample forecasting exercises for aggregate bank net charge-off rates on various loans as well as for similar individual bank rates using over 250 quarterly macroeconomic data from 1987Q1 to 2016Q4. Our empirical results demonstrate superior performance of PLS over benchmark models, including both a stationary autoregressive type model and a nonstationary random walk model. Our approach can help banks identify important variables that contribute to bank losses so that they are better able to contain losses to manageable levels.