Abstract: | This paper proposes a simple back testing procedure that isshown to dramatically improve a panel data model's ability toproduce out of sample forecasts. Here the procedure is usedto forecast mutual fund alphas. Using monthly data with an OLSmodel it has been difficult to consistently predict which portfoliomanagers will produce above market returns for their investors.This paper provides empirical evidence that sorting on the estimatedalphas populates the top and bottom deciles not with the bestand worst funds, but with those having the greatest estimationerror. This problem can be attenuated by back testing the statisticalmodel fund by fund. The back test used here requires a statisticalmodel to exhibit some past predictive success for a particularfund before it is allowed to make predictions about that fundin the current period. Another estimation problem concerns theuse of a single statistical model for all available mutual funds.Since no one statistical model is likely to fit every fund,the result is a great deal of misspecification error. This papershows that the combined use of an OLS and Kalman filter modelincreases the number of funds with predictable out of samplealphas by about 60%. Overall, a strategy that uses very modestex-ante filters to eliminate funds whose parameters likely deriveprimarily from estimation error produces an out of sample risk-adjustedreturn of over 4% per annum. |