The article examines different models for predicting Expected Shortfall (ES) in stock market data. By comparing various approaches, the researchers found that models based on extreme value theory (EVT) provide more accurate ES forecasts for 1- and 10-day periods. They also discovered that using asymmetric probability distributions for return innovations is more suitable. During a crisis period, all models underestimated risk levels. Combining EVT and Filtered Historic Simulation (FHS) yielded the most precise ES forecasts.