• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site

Working paper

Modelling Probability of Default of Russian Banks and Companies Using Copula Models

Department of Economics and Management Working Paper Series. DEM. University of Pavia, 2015. No. 113.
Ханьков И. О., Penikas H. I.
Research is devoted to examination of the classifier, based on copula discriminant analysis (CODA). Performance of the classification of this algorithm was assessed. On samples, modelled with some typical features of corporate default data, sensitivity of the classifier was tested, to sample size, to default rate and to different patterns of variables’ interdependence. Alternative copula families’ selection method is proposed based on certain performance metric optimization. Difference in classification performance of different algorithms are investigated. On real data of Russian corporate defaults, CODA classifier was built. It was supported by single factor analysis, based on discriminant analysis too. Final model demonstrates better classification performance than Linear Discriminant Analysis and Random Forest algorithm, and is comparable to Quadratic Discriminant Analysis. Another experiment was set on data of Russian banks. Single factor analysis was assessed via standard procedure. CODA performance appeared to be lower than of Random Forest here, it was similar to QDA.