Volatility forecasting has been of great interest both in academic and
professional fields all over the world. However, there is no agreement about
the best model to estimate
volatility. New models include measures of
skewness, changes of regimes and different distributions; few studies,
though, have considered different distributions. This paper aims to
investigate how the specification of a distribution influences the
performance of volatility forecasting on Ibovespa intraday data, using the
APARCH model. The forecasts were carried
out assuming six distinct
distributions: normal, skewed normal, t-student, skewed t-student,
generalized and skewed generalized. The results evidence that the model
considering the skewed t-student distribution offered the best fit to the
data inside the sample, on the other hand, the model assuming a normal
distribution provided a better out-of-the-sample performance
forecast.