This paper shows that out-of-sample forecast comparisons can help prevent data mining-induced overfitting. The basic results are drawn from simulations of a Monte Carlo design and a real data-based design similar to those in Lovell (1983) and Hoover and Perez (1999). In each simulation, a...