NORMAL APPROXIMATIONS FOR SERIAL-CORRELATION STATISTICS

The goodness of fit to normality of the null distributions of several serial correlation statistics is studied empirically. With moderate sample sizes the serial correlation of lag one, even computed in the presence of a moderate amount of missing data, closely follows the normal distribution, but some statistics calculated from serial correlations of multiple lags do not. Critical values at conventional significance levels can be approximated by the normal distribution for multiple serial correlation statistics, but approximations of more extreme percentage points should be avoided.