Urszula Grzybowska, Marek J. Karwanski

The definition of operational risk applied in banks is based on a very high distribution percentile: 99.95%. Derivation of risk estimators goes along with their high error which in turn is connected with models’ instability. The authors present various approaches to that error, from statistical based on estimation of distribution parameters to data mining that operates on algorithms predicting the responses. Presented results were obtained on both real and simulated data.