Evaluation of Value-at-Risk Models in the European Union Candidate States: Evidence from the Serbian Stock Market.

Abstract

The purpose of this thesis is to test how Value-at-Risk (VaR) measures calculated through Historical Simulation (HS) and Monte Carlo Simulation (MCS), which were originally created, suited for and tested in developed stock markets, apply to the volatile and shallow markets of EU candidate states. The data set is spanning from 01.10.2004 to 31.12.2012 and consists from the daily closing prices for the following market indices: S&P 500, DAX 30, CAC 40, NASDAQ and FTSE 100, which represent the developed world markets and are compared with the European Union (EU) candidate state Serbia, which is represented by its stock market index BELEXline. The data is collected from the Belgrade Stock Exchange (BELEX) and Yahoo Finance web pages. The behavior of the Value-at-Risk models with 95% and 99% confidence levels and rolling windows of 50, 100 and 250 days is tested and compared through a number of backtesting procedures.

The conducted tests indicated that the daily returns of BELEXline index are not normally distributed, exhibiting high kurtosis and large positive skewness. The returns are also not independently and identically distributed. The highest percentages of Value-at-Risk violations were observed during highly volatile periods, especially those characterized by steep volatility jumps, which indicates that the Value-at-Risk measures react slowly to sudden and large jumps in volatility, such as those during the Global Financial Crisis (2007-2009). Overall, based on the backtesting results, it can be concluded that the volatile and shallow stock markets of EU candidate states, such as Serbia, require much more complex, time-consuming and computationally demanding Value-at-Risk models, because the simplistic VaR models, which are commonly used in developed stock markets cannot alone capture the true levels of market risk in EU candidate states.