"In that purpose, we do not use the fund returns time series, but rather features measuring the sensitivity of these returns with regards to a set of market indexes, as proposed by Sharpe [4] under the name of style analysis. To build an independent classification based on these features, we use Kohonen self-organizing maps. The study is illustrated on the CRSP (Center of Research in Security Prices) database from the Chicago University. In order to compare the results, we use a reference classification from the ICDI (Investment Company Data, Inc.) and Standard & Poor's Fund Services. We model the problem into the following system of equations: 2 2 1 1 where R is the return of fund i, b ki is the sensitivity of fund i return with regard to the indexes returns F k , and e i includes all factors not explained by the set of market indexes. The above analysis is valid only when each indicator rate of return cannot be perfectly explained as a function of other ones."
Cardon, P., Lendasse, A., Wertz, V., de Bodt, E., & Verleysen, M. (2002). Classification of investment funds by self-organizing maps. Proceedings of ACSEG 2002, p. 201-212. https://hdl.handle.net/2078.5/225881