Quantifying Extreme Risks in Stock Markets : A Case of Former Yugoslavian States

One of the reasons why investors were not prepared for heavy losses in the stock markets that occurred after the beginning of sub-prime mortgage crisis in the U.S. lies in the curious fact that many practitioners were led to believe that there are so many independent agents participating in the stock markets that surely they must act according to Central limit theorem i.e. according to Gaussian distribution. As it turns out the paradigm of normality has let us down once again and reputation of VaR based risk measurement is seriously damaged. An alternative measure that looks very strong at these dire times and quantifi es the losses that might be encountered in the tail is the conditional VaR (CVaR). While VaR represents a loss one expects at a determined confidence level for a given holding period, CVaR is the loss one expects, provided that the loss is equal to or greater than VaR. In this paper the testing of CVaR models is performed on stock indexes from Slovenia, Croatia, Bosnia and Herzegovina, Serbia, Montenegro and Macedonia. Error statistics show that CVaR models are quite successful at capturing extreme losses that occurred in these markets, especially models based on Generalized extreme value distribution and a proposed Hybrid historical simulation CVaR model