We would thus reject the null hypothesis of normality for log(pt/pt−1). Further, the quantile–quantile plot is shown in Fig. 3. In the plot, there was a clear bend down on the left and bend up on the right that meant the log-transformed daily returns might not be Gaussian, a fact that was consistent with the conclusions from the three nonparametric tests. The Box–Cox plot in Fig. 4 also presented possible heavy tails for the trading days during the week. It is well known that parametric methods are sensitive to model assumptions, our results indicated that nonparametric model without a strong distribution assumption
may be better and more robust in assessing the day of the week effect on the log-transformed returns from the Shenzhen
Stock Exchange.