Determine Inference Based on Variance to Comp

Once the servicer’s metric performance and its relative performance to the Comp is calculated, an inference is made to determine, within a reasonable degree of statistical certainty, if the servicer performed above Comp or below Comp. To do this, STAR evaluates the variance to Comp for each servicer to first determine whether the servicer’s measured variance is statistically different from zero at a 99% confidence level. If the servicer’s variance to Comp does not meet the threshold of being statistically different, the servicer’s performance is deemed to be “at Comp” for the evaluation period. If the servicer’s variance is statistically different from zero with a 99% degree of certainty, the inference will be determined as “above Comp” if the variance is positive or “below Comp” if the variance is negative.

Figure 3:  Inference Test

Summary

Once a servicer’s Comp is established based on the servicer’s performance in each node of the decision tree, we then compare the servicer’s results to that of its Comp to determine relative performance. That performance is referred to as the servicer’s variance to Comp and a servicer’s relative performance to other servicers measured is represented by a metric score. An inference test is applied to determine whether or not there is a statistically significant difference between the servicer’s performance and that of its Comp.