A New Direction of Fund Rating Based on the Finite Normal Mixture Model

In this paper we try to develop a theoretical framework for fund rating under the assumption that superior funds could have a higher expected return than that of inferior funds, which could arise from the segmented market information or the differentiated ability of mangers to acquire and analyze the information. Under this setting, the funds are rated based on the cross-sectional distribution of all the funds instead of the presetpercentiles as Morningstar. We use the finite normal mixture for rating fund performance with the number of performance groups determined by likelihood ratio test using parametric bootstrap procedures, and we estimate the model with EM algorithm by treating the group information of funds as missing information.