Analysing the misclassification of US mutual funds through soft clustering.
2017-02-06T03:16:33Z (GMT) by
This thesis investigates when, how and why the misclassification of US mutual funds occurs. Misclassification is defined as the dislocation between the advertised style of a fund and the actual investment style. Using soft clustering (Fuzzy C-means), over 27% of U.S. equity mutual funds are found to be misclassified from their reported S&P styles. This confirms claims from related literature that traditional styles which rely on self disclosure or third party classifications are prone to misclassification. Misclassified funds seem to experience deliberately gaming of style to the ulterior benefit of their fund managers. The bona fide investment style of a mutual fund, relative to all other funds, is more accurately provided by clustering funds over time in a technique called style-by-clustering. Due to high levels of misclassification, traditional styles provide a poor peer group upon which to form a peer benchmark when evaluating performance. Style-by-clustering provides more accurate peer groups due to the resistance of clustering to misclassification. Peer benchmarks formed from soft clusters are discovered to track returns more accurately than their S&P counterparts and so prove to be superior for performance evaluation. Thus, fund managers may be gaming traditional styles to report better performance. Using Fuzzy C-Means to partition funds into style gamers and non-gamers, the returns of style gamers are statistically different and higher than non-gamers over time. This 'smoking gun' is an economic motive for fund managers to deliberately misclassify themselves and is a key finding that explains why misclassification occurs. Style gamers seem to be deliberately misclassifying style to chase higher returns or to take advantage of an inappropriate peer benchmark to report a superior evaluation of performance.