Is there a short-term reversal effect outside the universe of individual stocks? To answer this, we investigate a comprehensive dataset of more than two centuries of returns on five major asset classes: equity indices, government bonds, treasury bills, commodities, and currencies. (Page 2)
The short-term momentum is strongest among assets of high idiosyncratic volatility and in periods of elevated return dispersion. Also, the strategy payoffs display partial commonality across different asset classes. (Page 2)
This anomaly, which dates back to Rosenberg, Reid, and Lanstein (1985), and Jegadeesh (1990) and Lehmann (1990), is a tendency of stocks with a high (low) return in the previous month to underperform (overperform) in the following month. (Page 3)
From the research perspective, its discovery influenced, among other things, the way we measure the momentum effect, i.e., the pattern of stocks which have performed well (poorly) through the past 6–12 months continuing to deliver high (low) returns in the future. (Page 3)
Even across equity anomalies and style strategies, the last month’s performance seems to positively predict the future payoffs (Avramov et al., 2017; Arnott et al., 2018; Gupta and Kelly, 2018). The short-term reversal may be a far less pervasive anomaly than is commonly believed, and the dominant pattern may be—surprisingly—the short-term one-month momentum. (Page 4)
Our study contributes in a few ways. First, contrary to the stock-level evidence, we find no evidence of short-run reversal within and across the tested asset classes. Instead, we document a strong and robust short-term momentum effect. The last month’s payoffs significantly predict future cross-sectional returns in the universes of equities, bills, commodities, currencies, and in a pooled sample of all the asset classes (Page 5)
Third, we show that the short-run momentum strategies in equities, bonds, bills, and currencies display some commonality. The short-term momentum returns in these asset universes are weakly, but significantly, correlated to each other. This suggests the potential existence of some common factor partly contributing to the development of the short-run momentum effect. (Page 6)
Finally, our study provides a few minor insights regarding the short-run return continuation. Similar to classical momentum, we find that it is strongest in the market of high idiosyncratic volatility (Arena, Haggard, and Yan 2008). The payoffs are also particularly elevated in periods of excessive return dispersion (Docherty and Hurst 2018). (Page 6)
Equities. We use GFD USD Stock Return indices representing total returns on 45 different equity markets, encompassing developed, emerging, and frontier countries (Page 8)
Bonds. We consider benchmark total returns on 10-year government bonds from 54 countries (Page 8)
Bills. We utilize GFD USD Bill Return indices. These are total return indices representing profits on short-term treasury bill rates, or—alternatively—bank deposit rates if the bills are unavailable. (Page 8)
Commodities. The dataset includes spot prices of 48 different commodities from different categories: agriculturals, industrials, energy, and precious metals. (Page 9)
Currencies. Our sample accounts for all the currencies from countries which are covered by our sets of equity, bond, or bill indices. (Page 9)
Our basic sample period for the multi-asset sample of monthly returns runs from January 1800 to September 2018. Naturally, not all the assets are available throughout the full study period. On the one hand, the assets enter the sample gradually; on the other hand, some data may be missing during certain months or longer subperiods. (Page 10)
Our basic return predictive variable for short-term momentum, denoted SMOM, is the asset’s return in month t-1. Nonetheless, we also include a range of additional control variables. In selecting them, we impose two mandatory conditions. First, they need to be documented across many asset classes. Second, due to our data limitations, we must be able to derive the variables using only price information, i.e., without any additional use of accounting data, investor positions, credit ratings, etc. (Page 12)
The momentum, MOM, is calculated according to the broadly acknowledged framework dating back to Jegadeesh (1990), i.e., based on the previous 12 months and excluding the most recent month. (Page 12)
REV represents the value effect. Importantly, Asness, Frazzini, and Pedersen (2013) link the value and reversal effects, arguing that the correlation between the two is very strong, and thus the long-run return might be used as a proxy for value. (Page 12)
BETA and IVOL represent beta and idiosyncratic volatility (i.e., the volatility of the error term) from the single factor model based on returns from the period t-60 to t-1. To obtain these variables, we regress excess returns of individual assets on the market (MKT) excess portfolio representing each asset class. The market portfolio returns for the particular asset classes are proxied by the composite asset class indices obtained from GFD (Page 12)
Furthermore, for the multiasset portfolios, we equal-weight the single-asset benchmarks. SKEW is the product moment measure of skewness based on months t-24 to t-1. Finally, SEAS denotes the cross-sectional seasonality calculated as in Keloharju, Linnainmaa, and Nyberg (2016), i.e., the average returns for the same calendar month based on the last 20 years. (Page 13)
To this end, for each month we sort the assets in the sample on SMOM and form equal-weighted quintile portfolios. Moreover, we also build a long-short portfolio which goes long the quintile of assets with the highest returns in the previous month and shorting the instruments with the lowest returns. (Page 13)
The only asset class that does not exhibit long-run momentum is bonds. The uniqueness of this asset class may be linked to its inherent characteristics. Any increase in the price of bonds coincides with a decrease in the corresponding yield to maturity, and thus with a fall in the expected return. This effect should imply a reversal phenomenon, which is—in fact—not visible. The reversal effect is possibly offset by the coexisting short-term momentum effect (Page 14)
In any case, it is worth to highlight that though the short-term momentum is not visible in government bonds, even in this asset class we find no evidence of short-run reversal, as documented by Jegadeesh (1990) and Lehmann (1990) in the equity universe. (Page 15)
In our case, the underlying concept of the time-series spanning test is to check whether the portfolios formed on short-term momentum lie outside the mean-variance frontier spanned by other quantitative strategies, namely value, momentum, idiosyncratic volatility, skewness, and seasonality. (Page 15)
Not surprisingly, we observe no abnormal return on the short-term momentum strategy for the bond asset class, which has not produced reliable profits, as shown in Table 2. We record significant alphas on three asset classes: equities – 1.27% (t-stat = 5.05), bills – 0.50% (t-stat = 2.32), and commodities – 1.42% (t-stat = 4.37). The two strategies implemented in the multi-asset framework also produce very high and significant positive intercepts. The alphas on the short-term momentum portfolios, based on asset class combination (Combination) and implemented within the pooled set of assets (All assets), amount to 1.27% and 0.85% (t-statistics equaling 9.22 and 5.88), respectively. For none of these datasets are the established factor strategies able to explain the abnormal short-run momentum returns. (Page 17)
Interestingly, the factor strategies considered do seem to explain the profitability of the short-term momentum in currencies. The strategy exhibits significant exposure to the market, momentum, reversal, and seasonality factors. Having accounted for these, the short-term momentum in currencies produces no reliable alpha. (Page 17)
Even after accounting for MOM, SMOM still significantly predicts future returns in the cross-section. Panel C shows the results of multiple regressions, including all the return predictive variables considered. SMOM remains a reliable predictor of performance for equities, bills, commodities, and in the multi-asset framework. Importantly, it loses its power in currencies, corroborating our observations in Table 3. (Page 19)
As in Panel C, the SMOM is significant for equities, bills, commodities, and across multiple assets, and also displays a positive and significant coefficient for bonds. The currency returns—identically as in Panel C— cannot be reliably forecasted. (Page 19)
Furthermore, the short-term momentum effect is stronger among the assets of high idiosyncratic volatility (Panel D). This finding also matches the behavioral story of momentum, assuming that the market anomalies are manifestations of investors’ limited rationality that cannot be easily arbitraged away. (Page 21)
The insights from Section 4 suggest that the short-run momentum effect might be linked to behavioral factors. In such a case, one might expect that there is some cross-asset sentiment-related component that drives the short-run returns. (Page 22)
The results in the table yield several interesting insights. As could have been expected, the short-term momentum strategies in equities, bonds, bills, and currencies display significant exposure to other asset class returns. For instance, the equity short-term momentum appears to be linked to bond short-term momentum. The striking exception is commodities, showing no clear exposure to any other asset class. (Page 23)
The subperiod analysis indicates an interesting pattern: the short-run momentum is not present in the first two subperiods, i.e., in the years 1800–1880. However, in the later years, it is strong and robust across virtually all of the asset classes in all of the subperiods. The reason might lie in the small number of assets available in the early years. Fewer different assets result in smaller return dispersions, which is one of the key determinants of momentum returns (Docherty and Hurst, 2018) (Page 25)
Wang and Xu (2015) indicate that market volatility has strong predictive power for time variation in momentum payoffs. Hence, we calculate trailing 60-month market volatility and verify the short-run momentum performance within above-median and below-median periods. (Page 25)
Complementarily, we also research the returns in bull and bear months, i.e., in the months of high and low returns. All these tests confirm the robustness of the short-term momentum across asset classes. (Page 25)
Next, we verify two alternative implementations of the short-term momentum strategy. First is the time-series momentum of Moskowitz, Ooi, and Pedersen (2012). In this approach, we simply go long (short) the assets with a positive (negative) excess return in the previous month. Importantly, a recent article by Huang et al. (2018) points out that the time-series momentum studies usually concentrate on the 12-month horizon, and it would be valuable to examine such a strategy at other time horizons. Second, we loosely follow Dudler, Gmür, and Malamud (2015), van Zundert (2017), and Zaremba, Umutlu, and Maydybura (2018), and we implement the volatility-adjusted momentum strategy. In this framework, we scale the SMOM signal with the trailing 60-month volatility. Both approaches yield consistent results and lead to significant short-term momentum profits. We find no major qualitative difference in the results (see Table A10 in the Online Appendix). (Page 26)
This study examines the short-run momentum effect in five different asset classes: equity indices, bonds, bills, commodities, and currencies. Using a comprehensive dataset spanning from 1800 to 2018, we demonstrate that there is no short-term reversal effect outside the stock-level equity data. The prior month’s return positively predicts the returns in the cross-section in all asset classes with the exception of government bonds, as well as across the asset classes. Save for currencies, the phenomenon is not explained by the broadly acknowledged predictors of asset returns, such as value, momentum, beta, idiosyncratic volatility, skewness, and seasonality. (Page 27)
Moreover, the findings of this study could be employed to optimize implementation of the momentum strategies, both in individual asset classes and in a multi-asset framework. For instance, if skipping the last month in the momentum measurement is not dictated by some practical aspect of trade execution, this treatment may be redundant. In other words, it may not be necessary to drop the most recent month in the momentum signal calculation. (Page 28)