Double clustering for rating mutual funds

Due to the increasing proliferation of mutual funds, in-depth evaluationof the available products for portfolio selection purposes is a difficult task.Hence, classification schemes giving quick information about which fundsare worth to be monitored, are often provided. The aim of this work is toshow an application of clustering methods to the mutual funds historicaldata. Starting from the monthly time series of the Net Asset Values of aspecific style-based category, namely the Large Blend US mutual funds, weapply distance-based clustering methods twice on a set of return, risk andperformance measures: firstly, with the aim of reducing data dimension, andsecondly to cluster funds in homogeneous classes. The adopted procedureclaims the feature of producing a partition of funds that are readily inter-pretable from a financial point of view and it is further possible to rank theidentified groups, thus obtaining a rating of funds that turns out to accountfor different propensities toward the risk exposure.