Factor Investing in Commodities
Assessing Common Approaches to Predicting Commodity Returns
Commodity prices have experienced a round trip since 1990, with excitingly positive returns through 2006 and terrible returns since. Given that buy-and-hold has been a disappointing strategy of late, we wanted to explore the popular commodity trading strategies to see what returns these active overlays might have generated.
The three most common and well-documented predictors of commodity returns are trend, carry, and mean reversion, which we define below.
Trend. Trend-following strategies are rooted in the assumption that short-term past performance will persist for some time into the future. Trend-following strategies typically go long an asset with positive trailing returns and go short an asset with negative trailing returns. This strategy and its application to commodities is well-documented, most notably in Erb and Harvey’s 2005 paper.
Carry. The price of a commodity futures contract can be above or below the actual commodity spot price, which means that the futures return will differ from the underlying spot return upon expiry of the futures contract. This difference is known as the “carry” or “roll yield,” which accumulates as the investor maintains their position by periodically “rolling” it from the near-to-expiration (“front-month”) contract to a more distant expiration (“back-month”) contract. Whether the carry is positive or negative will depend on the structure of the futures curve: the carry is negative when the curve is upward sloping (“in contango”) and prices of front-month contracts are lower than prices of back-month contracts; the carry is positive when the curve is downward sloping (“in backwardation”) and prices of front-month contracts are higher than prices of back-month contracts. A simple strategy is to go long commodities with an expected positive carry, betting that the “prospective long-only excess return will, on average, be positive.” The reverse strategy, shorting commodities with an expected negative carry, is also common.
Mean reversion. Historically, prices in most asset classes have displayed a tendency to revert to their long-term real mean. High commodity prices, for instance, encourage exploration and production, which lead to increased supply and lower prices, and vice versa. Investors who wish to employ a simple mean reversion strategy can go long commodities trading below their long-term average prices and short commodities trading above their long-term average prices.
There are two ways to think about using the above factors to predict returns in commodities. The first way is to focus on each individual commodity and determine the extent to which their returns can be predicted by one or more of the three factors. The second way is to look at it cross-sectionally, with each commodity as a potential asset in a portfolio of commodity futures. We conducted both individual and cross-sectional regressions to test the predictive power of these factors on a basket of representative commodities comprising oil, gold, copper, and corn (the highest dollar-weighting commodities in each category of the GSCI) from 1990 to 2020.
Factors Predicting Returns of Individual Commodities
To test the predictive power of carry, trend, and mean reversion on individual commodities, we regressed each commodity’s six-month futures returns against each factor from 1990 to 2020. We show the T-stats and R-squared for these regressions in Figure 1, highlighting the statistically significant relationships in green.
Figure 1: Regression Statistics for 6M Commodity Futures Returns (1990–2020)
Source: Bloomberg, Verdad
Our results paint a bleak picture for common market timing strategies when it comes to predicting futures returns of individual commodities: in two-thirds of our tests, carry, trend, and mean reversion were not shown to be statistically significant predictors of commodity futures returns, and the R-squared failed to exceed 5% across all cases.
Interestingly, the predictive power of these fundamental factors increases materially when testing the period between 1990 and 2006, indicating that their statistical significance faded (with notable exceptions in gold and oil) when taking into account the last decade and a half of muted commodity returns.
We also note that the statistical significance of carry, trend, and mean reversion improves dramatically when using it to predict spot returns instead of futures returns. However, because futures returns comprise both spot returns and roll returns, the futures return is often materially different from the spot return, so predicting spot returns is often of little use to the commodity futures investor.
Factors Predicting Returns of Commodities Cross-Sectionally
The second way to think about factor investing in commodities is through a cross-sectional lens. Instead of looking at how these factors predict returns for each commodity individually, we combined the returns for a basket of representative commodities into a single dataset and regressed those returns against each of the factors from 1990 to 2020.
Figure 2: Regression Statistics for 6M Cross-Sectional Regression (1990–2020)
Source: Bloomberg, Verdad
In the cross-sectional regression, carry and trend are approaching significance while mean reversion exhibits a very high level of significance. Following the same pattern as in the individual regressions, the statistical significance of trend in the cross-sectional regression is within rounding of significance pre-2006 but fades when including the years since then.
There are myriad potential explanations or forces behind the changes in behavior of commodity futures pre- and post-2006. Perhaps it had to do with a dramatic fundamental change in key commodity markets (e.g., the US Shale Boom), or maybe it was all just noise in the first place, or perhaps the release of Gorton and Rouwenhorst’s paper—and all the capital that surged into commodities as a result—brought commodities to the competitive forefront, where simple strategies revolving around the significance of these factors were quickly arbitraged away.
Conclusion
The results of our analyses remind us that a constantly changing investing landscape requires us to consistently reassess the viability of back-tested strategies and the veracity of observed relationships between variables. The literature on commodity futures might have one believe that commodities are an undeniably attractive asset class to hold in investor portfolios and that common market timing strategies can easily improve returns for commodity investors. Alas, we did not find that those results held robustly when taking into account the last decade of data.
On an individual basis, commodities appear to be largely unpredictable. And even though the significance of the mean regression factor improves dramatically in the cross-sectional analysis, there remains a lot of noise within the other factors, as their signs flip from positive to negative and they go from statistically significant to insignificant on different time frames. Market timing strategies based on fundamental factors in commodities do not appear to be as reliable as factor-based strategies in stocks and bonds.
Commodities, we argue, are connected to the business cycle and thus are better treated as macro assets, with trends and tendencies that are best observed over long stretches of time. Investors are probably better served by understanding when it is an optimal time in the business cycle to own commodities, which will be the subject of our next papers in this series. In preview, we have found that oil and gold may offer tangible diversification benefits during periods of rising inflation. The challenge is to achieve balance and reap these benefits without getting trapped by colossal drawdowns and without being fooled by weak market timing signal.
Acknowledgment: This piece was co-authored by three of our term-time interns, David Balass, James Patton, and James Lockowandt. David previously studied finance and economics and is now finishing a double-degree in common and civil law (JD/BCL) at McGill University. James Patton is a freshman at Harvard currently pursuing a concentration in neuroscience with a secondary in economics. James Lockowandt is a sophomore at Harvard on an exchange year at the University of Oxford where he studies mathematics and philosophy. Both Patton and Lockowandt are actively seeking internship opportunities in finance for next summer, while Balass is seeking full-time employment at a value-focused hedge fund upon graduating law school in December.