MODERN APPROACHES TO FORECASTING FIRM DEFAULT RATES OVER THE SHORT TO MEDIUM TERM: AN APPLICATION TO A PANEL OF POLISH COMPANIES
MODERN APPROACHES TO FORECASTING FIRM DEFAULT RATES OVER THE SHORT TO MEDIUM TERM: AN APPLICATION TO A PANEL OF POLISH COMPANIES

Summary/Abstract: This paper models Polish companies default rates over the period 2000 to 2013, using eight leading classification algorithms – logistic regression, linear discriminant analysis, neural network, k-Nearest neighbors, naïve Bayes classifier, random forest, support vector machine and soft independent modeling of class analogies. Each of the eight methods is applied to five subsamples of data to forecast default rates in a period of one to five years, using 64 financial indicators. Results show that longer term forecasting remains a challenge for all methods, but a random forest algorithm is able to produce satisfactory results for forecast horizons of one or two years. The key drivers behind bankruptcy are then elicited using the relative contribution of variables to the accuracy of the best performing model.

• Details
• Contents