The article compares different models for predicting market risk in different volatility states. They tested six indexes from various countries using models like GARCH, EVT, and CAViaR. The best model overall was GARCH(1,1) with a standardized Student 𝑡 distribution. Nonparametric techniques performed well in low volatility but not in turbulent times. They also found that combining the best models gave good results, especially using GARCH(1,1) with a standardized Student 𝑡 distribution and either EVT or CAViaR.