Conditional skewness modelling for stock returns

Two approaches to modelling conditional skewness in a nonlinear model for stock returns are studied. It is found that a normal distribution can be rejected. A log-generalized gamma distribution with one time-varying density parameter, and a Pearson IV specification with three parameters are better supported by data. While the log-generalized gamma indicates that time-varying skewness is an important feature of the daily composite returns of NYSE, the Pearson IV model suggests that excess kurtosis rather than skewness should be accounted for.