I`m trying to forecast some forex returns of currencies couples. I build up my ARIMA model and test for normality of distribution after the arima is applied. I get different results from the Jarque - Bera Normality Test which p-value is 0.004 < 0.05 and then I should reject the H0 for normality. But when I plot and hist the residuals I see perfectly normal distribution. Where the difference comes from? My data is from 250 observations.