where ABCDEFG is estimated using the standard error from the market model regressions (or sample standard deviation in the case of the constant-mean-return model). The estimated variances for the CARs are then adjusted for the number of days in each cumulative abnormal average return of interest. The normal distribution allows for convenient construction of standard normal test statistics that are used to examine our hypotheses regarding the potential effects of rating changes on security returns. We construct CARs for various subperiods prior to day 0, around day 0, and after day 0 in order to test for information leakage prior to the event, control for delays in the announcement of the new effective rating, and test for persistent effects of an event, respectively.