Data-snooping biases in tests of financial asset pricing models

Tests of financial asset pricing models may yield misleadinginferences when properties of the data are used to constructthe test statistics. In particular, such tests are often basedon returns to portfolios of common stock, where portfolios areconstructed by sorting on some empirically motivated characteristicof the securities such as market value of equity. Analyticalcalculations, Monte Carlo simulations, and two empirical examplesshow that the effects of this type of data snooping can be substantial.