Análise de desempenho de indicadores de volatilidade

Abstract

Volatility measures constitute a concern among scholars and professionals of the financial market. Models of the ARCH/GARCH class from the daily returns produce an indicator of volatility, but do not give the researcher an observable measure of the degree of variability of returns around their expected value. The recent availability of data at frequencies below a trading day allowed the development of indicators of volatility observable through a measurement known as realized volatility. Since then, they can build an observable indicator of daily volatility based on intraday data, so as to represent a more appropriate measure of the riskiness of an asset, and from then estimate volatility through a process of ARIMA family. Provided with the data of a high frequency preferential role of Petrobrás S. A., the present paper therefore proposes to construct a measure of realized volatility by the sum of the squares of the returns obtained at regular intervals (5, 15 and 30 minutes ) during each trading day for the paper PETR4 during 02/01/2007 to 29/10/2010. After the creation of the realized volatility indicator that is supposed to be more appropriate to measure the degree of risk, the intent is to compare the goodness of fit and predictive ability of each of the methods of volatility’s models. The comparison of models based on daily data and intraday give will be through the calculation of the mean square error (MSE) and tests of Diebold and Mariano and Harvey to evaluate the predictive accuracy of models. The results in general showed that the models of the ARIMA class are more suitable for assessing the degree of adjustment and produces predictions more satisfactory than the models of the ARCH/GARCH class.