Default prediction for SME using discriminant and survival models, evidence from Polish market

Original by A. Ptak-Chmielewska & A. Matuszyk, 2014, 13 pagesHamster_gagarin_linkedin
hamster writter This summary note was posted on 11 January 2017, by in Credit risk Finance #, #, #, #, #, #

Quantitative methods in economics, Vol. XV, No2,pp 369-381

  • Paper from Glennon and Nigro (2005) shows that the default behaviour of the loans is time sensitive. The likelihood of default is high at the beginning, peaks in the second year and declines thereafter.
  • According to Moon and Sohn (2011) scorecards are often filled-in based on the evaluator’s goal perception rather than the individuals’ score of which the scorecards are built.
  • Foreign models appear to not be successful in the Polish conditions of political and economic changes thus the need to use a domestic model
  • Use Altman Z’-score as benchmark
  • Make use of Survival analysis for modelling
  • Censored data is defined as default out of the window and closing of account
  • Survival function is the probability of x to stay x when t is passed T
  • The most used survival model is the Cox hazards regression model
  • Use pseudo likelihood (Cox 1072): first include only information about parameters and second including information about parameters and hazard function
  • For checking the proportionality assumption the easy way is to include the interaction with time, the significance of these parameters confirms that the proportionality assumption is violated
  • Use 2004-2012 defaults history: 1053 good and 494 bads
  • SME: companies with turnover between 2-35 mio EU
  • Macroeconomic variables , from the end of the calendar year are included: GDP, unemployment rate, CPI
  • Basel II: need at least 5 years of data coming from at least one source regardless whether the source is internal or external
  • Assumes AUC must be between 0.75-0.80
  • Macroeconomic variables do increase the effectiveness of the model