The researchers studied different models to see how well they predict extreme events in stock market returns. They looked at data from S&P 500 stocks between 1995 and 2014. The results showed that GARCH models with normal distributions underestimated the risk of extreme events, while models with Student's t distributions were more accurate. GARCH and GJR-GARCH models performed the best in capturing the shape of extreme events in stock market returns.