Abstract :
[en] In this paper, we compare Morningstar ratings with those obtained using the same methodology of rating attribution with a set of commonly used performance measures. We look at three types of investment horizons: 3-year, 5-year and 10-year ratings. Our analysis focuses on Open-End US Mutual Funds available in Morningstar Direct Database from which we create three sets of 16,617, 13,505 and 7,992 funds corresponding respectively to the three investment horizons analyzed.
Our results show that Morningstar ratings are very close (correlation around 80%) to ratings obtained with Sharpe’s alpha, Jensen’s alpha, Four-factor alpha and Excess returns. And less significantly, we also observe that ratings given by the Sortino ratio, Sharpe MVaR, M-squared, Sharpe ratio, One-factor information ratio, Four-factor information ratio, Prospect ratio and Stutzer index are quite similar to Morningstar’s ratings (correlation lying between 70% and 78%). At the other end of the spectrum, however, ratings obtained with Annual return diverge widely from Morningstar ratings.
We also analyse which explanatory variables can explain the differences between ratings computed with Morningstar as compared with the alternative performance measures using a probit regression. We find that Load adjustments, tax and risk included by Morningstar in the computation of MRAR are often determining. Expense ratio, Return Skewness and the three factors of the Fama-French model (Beta, Size load and Book-to-market loading) can be significant determinants depending on the performance measure analyzed and on the selected investment horizon. Fund characteristics such as Age, Fund size, Turnover rate and Manager tenure are not statistically significant in determining the differences in ratings. Besides, we analyze differences between ratings (in terms of number of STARs) and we confirm previous results (i.e. the link between Morningstar’s and the alternative performance measures, but also the explanatory capacity of the load for lots of differences between ratings).
Finally, we test all possible combinations of our set of performance measures, and observe that Sharpe’s alpha, excess return, Sharpe MVaR, Four-factor alpha and Jensen’s alpha are part of the best combinations.
As a conclusion, Morningstar ratings can be replicated using simple and traditional performance measures but the replication is less accurate when tax and loads features are important. Therefore, Morningstar data management and access bring the most of its ratings’ value added.