It does this by comparing the prediction glitches of The 2 versions over a particular period. The check checks the null speculation which the two versions have the same overall performance on common, versus the alternative that they do not. Should the check statistic exceeds a vital price, we reject the null speculation, indicating that the differe