Combining Kohonen maps and prior payment behavior for small enterprise default prediction

This study aims to verify the potential of combining corporate prior payment behavior and Kohonen maps for small enterprise default prediction. Logistic regression, discrete-time hazard models, and Kohonen maps were applied to a sample of 1200 Italian small enterprises, and two categories of prediction models were calculated: one exclusively based on financial ratios and the other based also on payment behavior-related variables. The main findings are as follows: (1) Kohonen map-based trajectories give significantly higher prediction accuracy rates compared to both logistic and hazard models; (2) the longer the forecast horizon and/or the smaller the firm’s size, the greater are the improvements in prediction accuracy obtainable through Kohonen maps; (3) accuracy rates are higher when company payment behavior-related variables are added to financial ratios as default predictors; and (4) the smaller a firm, the greater is the increase in accuracy obtainable by adding payment behavior-related variables.