Resumen
In this paper, we compare semiparametric additive models with GARCH models in terms of their capability to estimate and forecast volatility during crisis periods. Our Monte Carlo studies indicate a better performance for GARCH models when their functional forms do not differ from that of the specified Data Generating Process (DGP). However, if they differ from the DGP, the results suggest the superiority of additive models. Additionally, we perform an empirical application in three selected periods of high volatility of IBOVESPA returns series, in which both families of models obtain similar results.