Cross-validation of the economic significance of factors in security returns

Recently, researchers have indicated that factor analysis of security returns produces only a single statistically significant factor, but they did not investigate the economic significance of the derived factors. Using datasets constructed so as to minimize serial correlation, the cross-validation technique of Conway and Reinganum (1988) is extended to examine both the statistical and economic significance of extracted factors. This procedure also allows consideration of the reliability of the tests. Economic significance occurs more often than expected by chance for both the single statistically significant factor and for higher order factors, but the tests are highly inconsistent between original and holdout samples. The results imply that tests of the economic significance of individual estimated factors are unreliable and should be interpreted cautiously.