Higher moments and industry momentum returns

ABSTRACT

Motivated by the demonstrated profitability of industry momentum strategies and the established explanatory power of higher moments for momentum returns at the firm level, we investigate the relation between higher moments and industry momentum returns. We use January 1970 to December 2021 data on 48 US industry return series from Kenneth French’s data library to calculate realised skewness and kurtosis and to construct double-sort trading strategies over dynamic trading horizons up to 24 months. The results reveal that the lowest-skewness group earns the strongest industry momentum returns over most trading horizons, while kurtosis is positively related to industry momentum returns. Additional testing reveals that these relations are uncaptured by the market, size, value, investment, profitability, and firm-level momentum risk factors in common asset pricing models.

KEYWORDS:

• Market anomalies
• industry momentum
• higher moments
• skewness
• kurtosis

JEL CLASSIFICATION:

• C22
• G11
• G12
• G14

Funding declaration

This research received no specific funding from the public, commercial, or not-for-profit sectors.

Notes

1 We also apply the double-sorting strategy on longer trading horizons over 24-month estimation or holding periods. Compared with the results in Section 4, the relation between skewness/kurtosis and industry momentum returns is much weaker over much longer trading horizons, indicating that the relation survives only over the short and medium terms. The results for longer trading horizons are unreported but available upon request.

2 We also apply other commonly used models, such as the Fama and French three- and five-factor models, and find very similar results to those in Tables 3 and 5.

3 In addition to the t-statistics, we obtain p-values of the Wilcoxon rank-sum test for addressing the possible issue of using non-normally distributed data. The unreported results (available upon request) show that the p-values of the Wilcoxon rank-sum test are consistent with the t-statistics reported in Table 2 for skewness and Table 4 for kurtosis, suggesting the reliability of the t-statistics.

4 We also use a bootstrap method to generate alphas using 2 000 replications and obtain z-statistics for addressing the possible issue of non-normality. The unreported results (available upon request) provide very similar results to those in Table 3 for skewness and Table 5 for kurtosis

5 The CAPM-adjusted results, not reported in Table A1, are very consistent with those adjusted by the Carhart and Fama and French models.

6 The CAPM-adjusted results, not reported in Table A2, are very consistent with those adjusted by the Carhart and Fama and French models.