Modelagem e previsão da volatilidade realizada: um estudo comparativo para ativos brasileiros

Abstract

This paper applies the Heterogeneous Autoregressive Model of Realized Volatility (HAR-RV) proposed by Corsi (2004, 2009) to intraday data of the five most traded stocks of the Ibovespa index from 08/02/2016 to 02/02/2018 composed by ABEV3, BBDC4, ITUB4, PETR4 and VALE3, in order of to model and forecast the realized volatility and the jumps of these stocks, comparing the forecasts of the traditional models (HAR-RV and HAR-RV-J) with the models proposed in this paper, which add information from the initial period of market opening for jumps estimation. As for the estimation, unlike the traditional models, in the modified models practically all the parameters are significant confirming the hypothesis of persistence of volatility found in the literature. Furthermore, in the modified models it was found that the addition of the morning jumps and the morning volatility were positive and significant, raising the hypothesis that there may be a transfer of volatility from morning to afternoon, i.e., intra-market volatility transfer. Regarding the forecasts, in terms of Mean Square Error (MSE), the results in-sample and out-of-sample show that traditional models present the worst forecasts. And the addition of jump improved the forecasts in traditional and modified models, corroborating with the literature. It was also verified the addition of information from the market opening period, such as the morning jumps and the morning volatility, significantly improved the forecasts of the modified models, confirming the hypothesis raised in this paper.