Value-at-risk estimation with univariate and multivariate models: the evidence from Serbia

Abstract

The main objective of this paper is to compare different empirical performances of various univariate and multivariate Value-at-Risk methods. The methods have been applied to the market data stemming from the Belgrade stock exchange and can be employed by any bank. A constructed hypothetical portfolio comprising of the ten most liquid stocks been constructed for the VaR estimation purposes. Both Normal and Student's-t distributions of the daily returns have been considered and a set of univariate and multivariate conditional and unconditional VaR models has been empirically estimated. Regulatory recommended Basel II test and comprehensive Kupiec and Christoffsen backtesting procedures have been also applied to assess the quality of the VaR forecasts.