The last global financial crisis has shown that our current understanding of risk is not sufficient in order to measure risk properly, especially in less liquid pre-emerging equity markets. The consequences of the crisis are transferred like a domino effect to those markets, making them practically illiquid. The paper analyzes the out-of-sample forecasting performance of historical simulation (HS), Risk Metrics (EWMA-based VaR), GARCH based VaR and filtered historical simulation (FHS) models at the usual 1% and 5% quantiles on daily data from Croatian stock market index over the period 05.10.2005 through 5.10.2015. VaR forecasts are backtested using Christofersen's methodology, a sequence of statistical tests aimed to appraise the empirical unconditional coverage of forecasts, first-order independence, and correct conditional coverage, which is a composite hypothesis of the two previous ones. The results indicate that 5% VaR based on symmetric GARCH model with normal distribution of innovations behaves reasonably well in the out-of-sample testing period. Filtered historical simulation at 95% confidence level also gives the good results. Standard VaR models mostly used by financial institutions underestimate risk forecast in Croatian equity market in all market circumstances. The backtests are rejected the suitability of HS and RM at any of the shortfall probabilities over the total sample due to significant underestimation of the downside risk.

Username
Password