This paper aims to identify if regime-switching GARCH models perform better than singlestate GARCH models for the Romanian stock market. There will be used two approaches: in-sample and out-of-sample. All estimations are going to be made for the BET Index, which is the most relevant index from the selected market. The results will be ranked based on statistical loss functions for each of the two considered approaches. These rankings should ensure an accurate comparison for models’ performance and they succeeded to return about the same results as in the relevant literature. Hence, for the in-sample evaluation there was no model which performs best for all loss-functions, but one can notice that for the out-of-sample evaluation the regime-switching models performed better especially on short-term (1-day observation period). All of these results were used further to improve some risk management strategies based on VaR, for which the volatility could be estimated through regimeswitching GARCH models, than considering historical volatility.