Abstract
Bootstrap methodology is evaluated in heteroscedastic models applied in the prediction of the Lima Stock Exchange Index (S & P / BVL), period 2017 - 2019. Predictions were obtained using the parametric methodology from the algorithm of heteroscedastic GARCH processes with Bootstrap proposed by Pascual (2006) being this generalized for other types of heteroscedastic models and were compared with real values. The performance of both methodologies was evaluated. The models that best fit the series are the ARMA (1,1)-EGARCH (1,1) models with assumptions of the residuals with Normal and t-Student distributions with 5 degrees of freedom. The comparative study showed that the application of the Bootstrap methodology in the series of returns of the Lima Stock Exchange Index, allows prediction intervals with greater and equal amplitudes in some horizons compared to the parametric methodology. The construction of prediction intervals for volatilities were also obtained with good performance, thus, this is an alternative for their construction in heteroscedastic models.