Longitudinal performance stratification – an iterative Kolmogorov–Smirnov approach

The stratification of entities into statistically distinct levels of performance over time is a problem encountered in a number of research and management settings. Traditional techniques to address this issue (e.g., cluster analysis) often require, either ex ante or ex post, the exogenous specification of the number of groups to be employed in further analysis – and are not especially suited to dealing with distributions over time. The methodology presented here iteratively applies the Kolmogorov–Smirnov two-sample test to identify the number and membership of statistically significantly different performance strata on a longitudinal basis. Monte Carlo simulations compare the new methodology with traditional clustering techniques. An application that stratifies mutual funds by returns illustrates the technique.