Medeiros, Otávio Ribeiro de2025-01-222020https://hdl.handle.net/20.500.14135/1626This paper is aimed at performing an econometric analysis of the Brazilian stock market at high frequency in order to confirm some of the stylized facts and empirical findings in the high-frequency literature, verifying the impact of outlier treatment on the Duration and volatility models goodness of fit, as well as testing the impact of different series aggregation and volatility estimation methods performed in the fit of high frequency volatility models. Confirmation of the stylized facts reported in the literature regarding kurtosis and return autocorrelation was observed. It was also noticed that the goodness of fit of the ACD models grows as parameter k of the Brownlees & Gallo (2006) algorithm increases, but the other two parameters have little influence on it. Regarding the HARRV models, we observed that the best models were predominantly those estimated based on the ROWCOV measure, with the MIDAS aggregation method and at the highest frequencies of 1 minutes and 30 seconds. Regarding the HEAVY models for BOVA11, the ones that had the best fit were those estimated based on the RKERNELCOV measure for raw data and 10-minute ROWCOV for treated data.Documento textualporAcesso abertoFinançasVolatilidadeMercado de ações - BrasilTese