We examine predictability for stock mutual funds using risk-adjusted r
eturns, We find that past performance is predictive of future risk-adj
usted performance. Applying modern portfolio theory techniques to past
data improves selection and allows us to construct a portfolio of fun
ds that significantly outperforms a rule based on past rank alone. In
addition, we can form a combination of actively managed portfolios wit
h the same risk as a portfolio of index funds but with higher mean ret
urn. The portfolios selected have small but statistically significant
positive risk-adjusted returns during a period where mutual funds in g
eneral had negative risk-adjusted returns.