Speculators and time series momentum in commodity futures markets

Uhl, Björn Article — Published Version Speculators and time series momentum in commodity futures markets Review of Financial Economics Provided in Cooperation with: John Wiley & Sons Suggested Citation: Uhl, Björn (2025) : Speculators and time series momentum in commodity futures markets, Review of Financial Economics, ISSN 1873-5924, Wiley, Hoboken, NJ, Vol. 43, Iss. 2, pp. 213-230, https://doi.org/10.1002/rfe.1228 This Version is available at: https://hdl.handle.net/10419/330168 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Rev Financ Econ. 2025;43:213–230. wileyonlinelibrary.com/journal/rfe | 213wileyonlinelibrary.com/journal/rfe 1 | INTRODUCTION Commodities can offer distinct diversification benefits to investors as they tend not to be highly correlated to traditional asset classes. However, commodities may also exhibit certain features such as a very high volatility or a negative roll yield which need to be considered when constructing a portfolio that includes an allocation to commodities (see FernandezPerez etal.,2016; Levine etal.,2018; Skiadopoulos,2012). Specifically, adding individual commodity markets longonly to an existing portfolio may have limit benefit. For instance, Vinzelberg and Auer(2014) show that Received: 23 July 2024 | Revised: 4 December 2024 | Accepted: 27 December 2024 DOI: 10.1002/rfe.1228 ORIGINAL ARTICLE Speculators and time series momentum in commodity futures markets BjörnUhl Faculty of Business Administration, University of Hamburg, Hamburg, Germany Correspondence Björn Uhl, Faculty of Business Administration, University of Hamburg, Moorweidenstraße 18, 20148 Hamburg, Germany. Email: [email protected] Abstract In this paper, we analyze the relationship between speculators in commodity futures markets and generic time series momentum (TSMOM) traders as well as the impact of this relationship on the subsequent TSMOM strategy performance. We find strong empirical evidence across all commodity markets that speculators in commodity markets tend to trade a TSMOM strategy, which confirms the results found by Boos and Grob (Journal of Financial Markets 64, 100774). On the basis of this result, we also ascertain whether the degree of such alignment has an impact on the performance of the TSMOM strategy. We find that there is weak, but statistically significant and robust evidence to suggest that the higher the degree of alignment between speculators and a generic TSMOM strategy, the lower the realized performance of trading TSMOM in these markets. Albeit we find little evidence that this can be exploited in a dynamic investment strategy, this negative relationship suggests that if a Commodity Trading Advisor (CTA) trades commodity futures markets which are less commonly traded by other CTAs, these markets may not only increase the internal diversification of their fund but these markets may also have a higher TSMOM Sharpe ratio by themselves. Consequently, our analysis provides valuable insights into improving the portfolio construction of CTAs. KEYWORDS commitment of traders, commodities, momentum, speculative crowding JEL CLASSIFICATION G11, Q02 This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2025 The Author(s). Review of Financial Economics published by Wiley Periodicals LLC on behalf of University of New Orleans. 214 | UHL adding crude oil futures does not improve key statistics of typical stocks and bonds portfolios. Therefore, it can prove useful to not invest ‘buy&hold’ into commodities but trade them in an active investment strategy such as alternative risk premia, see e.g. Fuertes etal.(2015), Miffre(2016) and Markwat etal.(2020). One of the most studied alternative risk premia is time series momentum (TSMOM) also known as trendfollowing (Moskowitz etal.,2012),1 which is typically traded by socalled Commodity Trading Advisors (CTA). Essentially, TSMOM dictates to hold a long exposure in an asset if its price has been increasing and vice versa for short positions. Erb and Harvey(2006) argue that whilst holding commodities longonly does not necessarily generate equitylike returns there is evidence to suggest that historically longshort investment strategies such as momentum were profitable. Georgopoulou and Wang(2017) show that much of mutual fund performance in equities and also in commodities can be explained by TSMOM. Clare etal.(2014) find that combining TSMOM with crosssectional momentum for commodity futures yields even higher riskadjusted returns. The popularity of momentum trading is owed to its persistence (e.g. Geczy & Samonov,2016) and resilience (e.g. Foltice & Langer,2015). Consequently, when constructing portfolios, Stadtmüller etal.(2024) argue in favour of a core portfolio which is enhanced by a satellite investment with a fixed allocation to commodity futures momentum. The presence of TSMOM traders in commodity futures markets has been investigated extensively. Regressing the returns of two broad CTA indices on the returns of a generic multiasset TSMOM strategy Hurst etal.(2017) find significantly positive coefficient estimates which suggests that CTAs indeed predominately trade TSMOM. Using futures across all asset classes Fan etal.(2020) find that speculative pressure as a factor can help to explain the crosssection of returns after controlling for a number of alternative risk premia across all sectors but fixed income. Lutzenberger(2014) shows that in addition to TSMOM predictability also other exogenous variables can help forecasting the returns of commodity futures. More recently, the academic debate has focussed on the relationship between speculators in commodity futures and TSMOM. For instance, Borgards and Czudaj(2022) show that the changes in speculative open interest have forecasting power for the returns of the underlying market as well as their relationship with momentum. In the same direction Boos and Grob(2023) hypothesize that speculators in commodity futures are trendfollowers and show that regressions which explain the changes in net speculative open interest by generic TSMOM positions have substantial explanatory power across commodity markets. They also estimate the average momentum filter weights traded by speculators using a penalized regression. Given the ample evidence that speculators in commodity futures trade a trendfollowing strategy, we address the following question: Does the degree of speculative crowding among CTAs in a commodity market have an impact on the performance of a trendfollowing strategy in that market? Rather than using generic measures speculative crowding e.g. the fraction of speculators in all traders, we use a novel measure which is tailored to the specific problem. We consider how closely aligned the net speculative open interest in any given commodity market is to the positions of a generic trendfollowing strategy. Consequently, our work builds upon Boos and Grob(2023) who empirically show that the changes in net speculative open interest can be explained by the changes in a generic trendfollowing signal. They report high outofsample R2 for all markets and estimate the weights which the average CTA would have applied. Our work is closely related to theirs but we focus predominately on the consequences of such alignment between speculators and TSMOM traders. Essentially, we use the degree of alignment as measure of speculative crowding. The more speculators trade TSMOM the closer their aggregate net positioning should be to a generic trendfollowing position. In contrast to standard metrics, such as e.g. the proportion of speculators to all open interest, this measure is more informative as it specifically test how similar the positions of speculators to the positions of a TSMOM strategy. If speculators were on aggregate aligned with a specific trading strategy such as momentum this could potentially have a more profound impact than merely generic measures of speculative activity because alignment requires different speculators to trade in the same direction at the same time. The underlying hypothesis is that if too many CTAs trendfollow a commodity market in a similar fashion the expected returns of this strategy diminish. The possible reasons could be the increased risk of reversals and tail risks (e.g. Barroso etal.,2022; Brown etal.,2022).2 The consequences of such strategy alignment could both pose a systemic risk but could also impact the performance of trading that strategy. The majority of the literature focusses on the systemic risk aspect. Whether speculators in commodity markets cause price impact has been famously alleged by Masters and White(2008) whose hypothesis that longonly speculators in commodity futures were the main driver behind the 2007–2008 price spike has been tested empirically by Irwin and Sanders(2012) who found no evidence in favour of this hypothesis. Boyd etal.(2018) review the impact of the financialization of commodity futures and conclude that speculators have little impact on price distortions but that they predominately provide liquidity to hedgers. Brooks etal.(2015) investigate extreme price moves in commodity markets and find little, if any, evidence that these were caused by speculative bubbles. On the contrary, using informational efficiency as a measure of market quality, Bohl etal.(2021) find empirical evidence that the higher the speculative activity in a commodity market, the lower on average the quality of that market. Decomposing price shocks to commodity futures into a | 215 UHL permanent and a transitory component, Haase etal.(2019) find no empirical evidence that speculators cause longterm price impact. Haase etal.(2016) summarize a large number of studies on this subject in a metaanalysis and find that there is no conclusive evidence neither for nor against the hypothesis that speculators drive commodity market prices. However, Haase etal.(2017) find some empirical evidence that speculators in commodity markets can Grangercause market volatility, whilst Kim(2015) find that speculative futures trading does not destabilize commodity markets. In another metaanalysis of a large number of published studies Wimmer etal.(2021) find that the null hypothesis of noncausality cannot be rejected. More toward the direction of our work are studies that relate the degree of speculative crowding to strategy performance. For instance, Baltas(2019) investigates the impact of crowding in risk premia strategies including momentum. His results suggest that the impact of crowding can be either beneficial or not depending on whether the investment strategy is convergent (e.g. meanreversion trading) or divergent (e.g. momentum trading). Focussing specifically on trendfollowing Bollen etal.(2021) investigate whether those CTAs that perform more similar to their peers underor outperform them. They find that similarity to the peer group is associated on average with higher performance. Building a heterogenous agent model, He and Li(2015) investigate the relationship between different types of traders. They find that when momentum traders are more active in a market, momentum strategies with a short horizon tend to stabilize the market and may be more profitable. Carter and RevoredoGiha(2023) attribute a degeneration in CTA returns in commodity futures to the financialization of such markets. In contrast to the aforementioned analyses, we focus on the differences between individual commodity markets with regards to the relationship between speculative crowding and trendfollowing performance. This can provide a useful guide to investment managers to adjust their weighting (or even inclusion) of markets in their strategy. In Figure1, an example is given to illustrate the hypothesized relationship between speculators and CTAs. It shows the time series of the (normalized) net speculative position for the Arabica Coffee futures together with the corresponding (normalized) positions of a generic TSMOM strategy. It can be seen that these appear to be closely related. Investigating the trading behavior in futures contracts using the CFTC commitment of traders data has been done in financial markets, too. For instance, using the S&P 500 index future Smales(2016) finds that speculators and small traders reveal some ability to forecast future returns. The contribution of this work can be summarized as follows. First, after replicating some of the key results of Boos and Grob(2023), we introduce the corresponding regression estimates as measure of alignment among speculators. This measure is specifically tailored to the empirical observation that the predominant part of speculators in commodity futures pursue a trendfollowing strategy. Second, we find that the performance of trendfollowing a commodity market is negatively correlated to the degree of the estimated alignment. However, we also find that this effect appears to be predominately contemporaneous though and thus it is unlikely that it can be exploited profitably in a dynamic trading strategy. This negative relationship suggests that if a CTA adds new commodity markets which are less commonly traded by other CTAs relative to other speculators, such markets do not only increase the internal diversification of the portfolio but may also tend to FIGURE 1 Example of similarity between net speculative OI and trendfollowing position. Time series of trendfollowing positions and the % net speculative open interest scaled by their own volatility (fixed) for Arabica Coffee futures. 216 | UHL individually perform superior in a TSMOM strategy. This finding may explain why more and more CTAs launch socalled ‘alternative markets’ programs which trade more exotic and less commonly traded commodity futures markets. This paper is structured as follows. In the next section, we discuss the data and methodology that we use for our analyses. In the third section, some statistics are provided that allow to better understand the relationship between the different categories of commodity traders. The fourth section contains the main empirical results. The fifth section provides some robustness analyses to support the key findings. In the sixth section, we discuss whether the findings can be used profitably in an investment strategy. The final section concludes. 2 | DATA AND EMPIRICAL STRATEGY 2.1 | Data We collect data for n = 26 commodity futures markets. The sample period is from January 2006 to December 2023 and is limited by the historical availability of the disaggregated open interest (OI) data from the Commodity Futures Trading Commission (CFTC).3 All data are retrieved from Bloomberg. The contracts with their basic properties are shown in Table1. Most markets are in the agriculturals sector. With the exceptions of lean hogs and feeder cattle all contracts are physically settled. TABLE 1 The table shows the contract specification for the commodity futures series we consider. Name Ticker Cash settled Exchange Sector Subsector Crude CL1 N New York Mercantile Exchange Energies Energies NatGas NG1 N New York Mercantile Exchange Energies Energies HeatingOil / ULSD HO1 N New York Mercantile Exchange Energies Energies Gasoline XB1 N New York Mercantile Exchange Energies Energies Copper HG1 N Commodity Exchange, Inc. Metals Industrials Gold GC1 N Commodity Exchange, Inc. Metals Precious Silver SI1 N Commodity Exchange, Inc. Metals Precious Platinum PL1 N New York Mercantile Exchange Metals Precious Palladium PA1 N New York Mercantile Exchange Metals Precious Wheat W 1 N Chicago Board of Trade Ags Grains and Oilseeds Corn C 1 N Chicago Board of Trade Ags Grains and Oilseeds Soybean S 1 N Chicago Board of Trade Ags Grains and Oilseeds SoybeanMeal SM1 N Chicago Board of Trade Ags Grains and Oilseeds SoybeanOil BO1 N Chicago Board of Trade Ags Grains and Oilseeds Coffee KC1 N ICE Futures US Softs Ags Softs Sugar SB1 N ICE Futures US Softs Ags Softs Cotton CT1 N ICE Futures US Softs Ags Softs Cocoa CC1 N ICE Futures US Softs Ags Softs Lumber LB1 N Chicago Mercantile Exchange Ags Softs Juice JO1 N ICE Futures US Softs Ags Softs LiveCattle LC1 N Chicago Mercantile Exchange Ags Livestock LeanHogs LH1 Y Chicago Mercantile Exchange Ags Livestock FeederCattle FC1 Y Chicago Mercantile Exchange Ags Livestock HardWinterWheat KW1 N Chicago Board of Trade Ags Grains and Oilseeds SpringWheat MW1 N Minneapolis Grain Exchange Ags Grains and Oilseeds Rough rice RR1 N Chicago Board of Trade Ags Grains and Oilseeds Source: Bloomberg. | 217 UHL For each of the futures contracts, we obtain daily endofday closing prices. As each futures contract expires at a fixed date, we generate a continuous series of prices for each market by rolling every contract either on its last trading date or on the first business day of the expiration month, whichever is earlier. Ratio adjustments are applied backward in history. With these adjustments, there is no pricejump from rolls but the carry return is earned (or paid) continuously (Koijen etal.,2018). Thus, using ratio adjustments for the rolls between contracts allows to compute logarithmic returns simply as difference between the logarithm of the adjusted prices. Futures do not require a full investment but are traded on margin accounts for which only an initial and a variation margin have to be posted. We assume that the futures position is fully collateralized, so that the returns can be interpreted as standard excess returns. The OI data is originally provided by the CFTC on a weekly basis in the ‘Disaggregated Commitment of Traders Report’. The report contains the aggregate long and short positioning of reportable market participants in futures markets which are classified in four categories: swap dealers, producers, money managers and others. In plain terms, ‘swap dealers’ are primarily investment banks who act as market makers; ‘producers’ are all market participants who are involved with the physical underlying of the futures contract, i.e. buyers or sellers of the actual commodity; ‘money managers’ are asset management firms who hold positions to speculate on the future price development and thus are subsequently labeled ‘speculators’; finally, ‘others’ are any traders who do not fall into any of the previous categories. The OI data is published on Fridays for Tuesday's holdings. As we are interested in explaining alignment we use the Tuesday timestamps. We use the futuresonly data and ignore options as CTAs tend trade the former. The original OI data is sampled in lots and is mapped to USD positions by multiplying the number of lots with the corresponding contract value.4 Net and gross OI for each commodity market i=1, …,n and each category of trader, cat ∈{prod, swap, spec, other} , are defined as respectively and can be interpreted as the net and gross aggregate positioning of speculative traders in a given market. For comparability we normalize both by the aggregate gross OI, i.e. for net OI we have In order to have welldefined statistics, both OI%net,cat i,t and OI%gross,cat i,t are normalized by the gross OI. 2.2 | Generic momentum signals and positions There are two main types of momentum strategies: time series momentum (TSMOM) which takes directional net exposures, see Moskowitz etal.(2012), and crosssectional momentum (XSMOM) which is generally net cash or risk neutral, see Jegadeesh and Titman(1993). Albeit a number of works have discussed the latter (e.g. Shen etal.,2007) we focus on the former because it is the predominant strategy employed by many CTAs (Hurst etal.,2013). To compare the net speculative positioning with the positions of a generic TSMOM strategy we need to specify a baseline momentum model. We define the TSMOM signal sit at time t=1, …,T for each market i=1, …,n , in accordance with Levine and Pedersen(2016) as the market return over the preceding 260 business days normalized by the corresponding estimate of volatility, i.e. Fan and Zhang(2024) highlight the importance of risk managing the positions in individual commodity markets in risk premia strategies. Thus, the positions of a TSMOM strategy are typically scaled inversely proportional to volatility (see e.g. Harvey etal.,2021), (1) OInet,cat i,t=OI long,cat i,t−OI short,cat i,t and OIgross,cat i,t =OIlong,cat i,t +OIshort,cat i,t , (2) OI %net,cat i,t= OI net,cat i,t ∑ cat∈{prod,swap,spec,other} OIgross,cat i,t . (3) s it =r 1year it 𝜎 1year it . 218 | UHL where we set 𝜎target =10% per annum 5 and we also use the RiskMetrics(1996) standard for 𝜎 it . In the robustness section, we discuss the sensitivity of the key results to these specifications. In particular, we test the impact of applying a forecast function to the signal, e.g. the sign function which is often used in the literature (e.g. Moskowitz etal.,2012), and the impact of using a ‘slower’ position scaling, which may be relevant in practice. 2.3 | Measuring the alignment between speculators and momentum traders We hypothesize that speculators in commodity futures markets trade TSMOM as described by the generic strategy in the previous subsection. To test this hypothesis empirically, we follow Boos and Grob(2023) and estimate a regression where the changes in the net speculative OI are explained by changes in the positions of a generic TSMOM strategy. While we base our primary empirical model on theirs, we deviate in three regards, though. First, in this work, the trendfollowing position is volatilityscaled, which has been highlighted by Kim etal.(2016) as a significantly contributing factor to TSMOM performance and consequently is used in practice by many CTAs, which is relevant to this work. Second, we use weekly data on both sides of the regression as we do not aim to back out average filter weights but explain changes in the speculative positioning. Lastly, we normalize the variables of both sides of the regression by their corresponding standard deviation. This renders both sides of the regressions dimensionless so that the coefficient estimates can be interpreted in terms of standard deviations and are consequently comparable across markets. Thus, the model specification reads where OI%net,spec it is defined in (2) and the positions of the TSMOM strategy pit are defined in (4). Variables which are superscripted by a tilde ⋅ denote the corresponding variable divided by its own volatility, i.e. for instance for the TSMOM positions the normalization is  p it =p it ∕  std ( p it) . In addition, we also use a specification in levels, which entails the issue of high autocorrelation in the regressor but may still provide some additional information. Thus, we also estimate It should be noted that the levels regression (6) uses highly autocorrelated variables which by construction are stationary though. The trendfollowing positions as defined in (4) use a scaled version of the lagged 1year market return. The differencing in (5) has the advantage of removing the high autocorrelation but also removes the level information though. Clearly, both regressions are related but still have their distinct advantages and disadvantages so that we use both in the empirical analysis. A statistically significant estimate for 𝛽Δ 1 or 𝛽1 , respectively, would suggest that the aggregate speculators in that market behave similar to a time series momentum trader as defined in section2.2. We also compute outofsample R2 OS (Campbell & Thompson,2008), which compares the mean squared error (MSE) from the prediction of the regression to the MSE of using the historical average as prediction, i.e. for instance for (5) where yit =Δ  OI %net,spec it for (5) and y it = OI %net,spec it for (6) and the predictors  yt and yt use data up to time t−1 . As the prediction is outofsample a minimum estimation window size 𝜏 for the regression needs to be specified which we set to the equivalent of 4 years of observations. All statistical significance tests for these regressions are based on the stationary bootstrap by Politis and Romano(1994). (4) pit = 𝜎target 𝜎 it sit . (5) Δ OI %net,spec it =𝛽Δ 0,i +𝛽Δ 1,i Δ p it +𝜀 Δ it (6)  OI%net,spec it =𝛽 0,i +𝛽 1,i  p it +𝜀 it. (7) R 2 i,OS =1−∑ T t=𝜏�yit − yit� 2 ∑ T t=𝜏� y it −y it� 2,i=1, …, n | 219 UHL 3 | MARKET STRUCTURE 3.1 | Who trades with whom? We first ascertain which categories of traders in commodity futures trade with each other. To this end, we correlate the contemporaneous weekly changes in the corresponding net open interest for each combination of categories. A negative relationship suggests that as one category of traders builds up a net position, the other category reduces its position and thus on aggregate can be interpreted as trading activity between these two categories. Table2 shows the average correlations by sector. In all sectors, the most pronounced correlation is found between speculators and producers, which economically suggests that speculators tend to bet against the hedges from producers. In agriculturals, this correlation has by far the largest magnitude whilst in the other two sectors all correlations are clearly negative which indicates speculators also trade with swap dealers and others traders. Overall, these results are in line with Boos and Grob(2023) who use a variance decomposition and find that producers are the dominant counterparty to TSMOM traders. 3.2 | Positioning and performance by category of trader Next, we investigate the average positioning of each category of traders and the associated performance. Figure2 shows the box plots of the aggregate net OI percentages. On average producers have held short positions and hence have in tendency hedged their existing physical long exposures whilst market makers and speculators have been predominately long and thus taken the opposite side of the hedges. We now estimate the aggregate performance of each category of traders. The net OI of each category serves as proxy for the aggregate positioning which we assume to be approximately constant over the following week due to the lack of higher frequency data. We compare the performance of all categories together with the TSMOM strategy. For a fair comparison we allow the latter to update its positions also only once a week on Tuesdays. The positions of all markets are scaled with an ex ante forecast of volatility to target an annualized level of 10%. The results are shown in Table3. All realized volatilities are well below 10% due to the diversification within each of the portfolios. Overall, the positions of the generic TSMOM strategy outperform all categories of traders. With the exception of energies, the same is true for all individual sectors. Interestingly, speculators actually realize a negative performance overall whilst the other categories of traders and the TSMOM strategy realize positive performances over the sample period. The reason for the poor performance of speculators is a positive bias in the speculators' positions compared to the other categories of traders, see Figure2, and also compared to the TSMOM strategy. An explanation for this bias could be the presence of longonly commodity speculators who hold fairly static long exposures to provide investors with protection against commodity price inflation by investing in or in alignment with commodity indices, for instance the S&P GSCI. By contrast, TSMOM models have held overall short positions on average for each market. This is a sensible result given that most commodities tend to trade in contango, so that the negative roll yield will cause a negative drag in the signal definition(3).6 4 | SPECULATORS AND MOMENTUM TRADERS 4.1 | Are speculators momentum traders? We hypothesize that speculators on aggregate trade a strategy that is similar to the generic TSMOM strategy presented in section2.2, which we test in two steps. First, we test for positive autocorrelation in the changes of the net TABLE 2 Average Pearson correlations of the weekly changes in net open interest among different categories of traders. Sector Prod/swap Prod/spec Prod/other Swap/spec Swap/other Spec/other All −0.04 −0.81 0.04 −0.21 −0.04 −0.38 Ags −0.20 −0.86 −0.02 −0.00 −0.09 −0.32 Energies 0.05 −0.66 0.14 −0.45 0.01 −0.58 Metals 0.44 −0.77 0.15 −0.74 0.05 −0.45 220 | UHL speculative OI, which would imply that the trade flows of speculators in commodity futures are (to some extend) predictable. This is a key feature of a trendfollowing system because TSMOM generates autocorrelated signals and positions by construction as it passes this feature on from the underlying market.7 Figure3 shows the first order autocorrelations of the weekly changes in the net speculation OI for each market with 90%- confidence bands using the stationary bootstrap of Politis and Romano(1994). For all commodity markets, we find that the first order autocorrelation is statistically significantly positive. The average estimate is lowest for energies and highest for agricultural markets. As a second step, we estimate the alignment coefficients in the regressions (5) and (6), which describe the magnitude to which the (changes in the) net speculative positions vary with the (changes in the) positions of a generic trendfollowing system as defined in section2.2. Table4 summarizes the results by market for both weekly changes and levels. The sector averages and an overall average are also shown at the bottom of the table. The estimated betas FIGURE 2 Average net positioning of traders by category. The figures shows ‘boxandwhiskers’ plots the distribution of the proportion of net open interest ( OI%net,cat ) for each category of traders cat ∈{prod, swap, spec, other} . Each of the ‘boxes’ shows the median as well as the first and third quartiles. The ‘whiskers’ are defined by a distance of 1.5× the interquartilerange from the nearest quartile. Outliers, if any, were shown as individual dots outside the whiskers. The asterisks mark the means. TABLE 3 The table shows the (costfree) performance estimates for the categories as well as for a generic TSMOM strategy. Sector Statistic Prod Swap Spec Other TSMOM All Returns (%) 0.11 0.41 −0.44 0.82 1.06 Volatility (%) 4.37 3.54 3.50 2.97 3.58 Sharpe ratio 0.02 0.12 −0.12 0.28 0.30 Agriculturals Returns (%) 0.41 0.03 −0.95 1.15 0.90 Volatility (%) 4.59 4.69 3.85 3.67 3.83 Sharpe ratio 0.09 0.01 −0.25 0.31 0.24 Energies Returns (%) −0.31 0.22 1.80 0.50 1.74 Volatility (%) 6.99 4.83 6.12 6.11 7.42 Sharpe ratio −0.04 0.05 0.29 0.08 0.23 Metals Returns (%) −0.80 1.84 −0.43 −0.02 1.06 Volatility (%) 7.65 5.19 6.94 5.45 6.61 Sharpe ratio −0.10 0.35 −0.06 −0.00 0.16 Note: All markets are equally weighted and volatility scaled to an ex ante annualized volatility of 10%. | 227 UHL Setup and model/sector All Agriculturals Energies Metals Simple regression full sample −0.10 N/A N/A N/A Simple regression annual subsamples −0.63*** −0.70*** −0.39*** −0.63* Fixed effects regression annual subsamples −0.63*** −0.70*** −0.34 −0.66 Fixed and time effects regression annual subsamples −0.63*** −0.70*** −0.23 −0.14 Forecast: binary, volatility scaling halflife: 260 days Simple regression full sample 0.15 N/A N/A N/A Simple regression annual subsamples −0.57*** −0.61*** −0.29*** −0.75*** Fixed effects regression annual subsamples −0.59*** −0.65*** −0.32 −0.74* Fixed and time effects regression annual subsamples −0.59*** −0.63*** −0.43 0.08 Forecast: roll over, volatility scaling halflife: 11.2 days Simple regression full sample −0.07 N/A N/A N/A Simple regression annual subsamples −0.84*** −0.88*** −0.59*** −0.98*** Fixed effects regression annual subsamples −0.89*** −0.93*** −0.52 −1.13** Fixed and time effects regression annual subsamples −0.89*** −0.94*** −0.39 −0.65 Forecast: roll over, volatility scaling halflife: 260 days Simple regression full sample 0.13 N/A N/A N/A Simple regression annual subsamples −0.84*** −0.86*** −0.62*** −0.98*** Fixed effects regression annual subsamples −0.87*** −0.91*** −0.57** −1.04*** Fixed and time effects regression annual subsamples −0.87*** −0.91*** −0.73* −0.56* Note: In Panel A the results for the Δ - alignment betas are reported and in Panel B the results for the levels alignment betas are reported. Only the slope coefficient estimates are reported. Statistical significance is denoted by ‘*’ for pvalues < 10%, ‘**’ for pvalues < 5% and ‘***’ for pvalues < 1%. TABLE 7 (Continued) TABLE 8 Backtest results: We report the Sharpe ratios for an equally weighted TSMOM strategy using only high and only low alignment beta markets. TSMOMsharpe TSMOMsharpe pValue TSMOMsharpe TSMOMsharpe pValue Low 𝚫 - alignment beta High 𝚫 - alignment beta Low levelsalignment beta High levelsalignment beta Lookahead All .55 −.13 .00 .47 −.01 .01 Ags .48 −.14 .00 .48 −.17 .00 Energies .23 .24 .51 .16 .24 .64 Metals .17 .03 .31 .26 .17 .38 Outofsample All .32 .02 .07 .15 .29 .73 Ags .32 −.12 .05 −.00 .26 .83 Energies .20 .11 .38 .02 .29 .90 Metals .22 −.05 .21 −.09 .23 .86 Note: The top rows contain the results for a lookahead analysis where the betas are estimated over the same period over which the portfolio is formed. The bottom rows contain the results for the outofsample analysis where the alignment betas are estimated over the previous year and the portfolio is formed over the next year. The leftmost three columns use the alignment betas from the regression in differences whilst the rightmost three columns use the alignment betas from the levels regression. The pvalues are computed for the null hypothesis that the high alignment beta portfolio has a Sharpe ratio at least as high as the portfolio with the low alignment betas. 228 | UHL performance also exist but are not statistically significant. For the betas from the levels regression, there is no evidence to suggest that the effect can be profitably exploited. Consequently, the effect that the alignment beta negatively correlates with TSMOM performance is predominately a contemporaneous effect and has limited, if any, predictive power. agriculturals may be an exception though. Overall, we believe that the effect cannot be exploited in a dynamic investment strategy. However, the static nature of the effect could possibly be harvested statically by including predominately commodity futures markets in a TSMOM portfolio that are less commonly traded by other CTAs. 7 | CONCLUSIONS In this paper, we investigate the relationship between the aggregate net positions of speculators and the positions of a generic momentum strategy across a broad range of commodity futures markets. We find that the time series of both types of positions are closely aligned across all investigated commodity markets. The evidence prevails for both the levels regression and in differences. The high alignment suggests some level of speculative, strategyspecific crowding in commodity futures markets. Using the estimated alignment coefficients, we also document a weak but statistically significant tendency of time series momentum performance to degenerate when the speculators in the underlying market are more aligned with momentum traders. Albeit the correlations are negative across all sectors, the evidence is strongest in agricultural commodities. In an investment exercise the statistical tests are validated in a historical backtest – both insample and outofsample. Although there is also some evidence to support the latter, the effect appears to be predominately contemporaneous and thus it may not be possible to benefit from this effect in a dynamic trading strategy. However, the negative impact on momentum performances suggests that ‘hardertoaccess’ or less commonly trendfollowed commodity futures markets may not only provide diversification benefits for a CTA program also be more profitable on a singleasset basis. DATA AVAILABILITY STATEMENT The data in this paper are downloaded from Bloomberg and are thus subject to licensing constraints. ORCID Björn Uhl https://orcid.org/0000-0002-8018-4481 Endnotes 1 Throughout this paper we use the terms ‘momentum’ and ‘trendfollowing’ interchangeably. Essentially, the term ‘momentum’ covers both crosssectional momentum and time series momentum, which corresponds to trendfollowing. As we do not use the former, the latter is here also simply referred to as momentum. 2 It should be noted that for crosssectional equity momentum Barroso etal.(2022) find that crowding does not alone explain increased tail risk. 3 Some minor limitations apply to the historical data, see CFTC(2022) for details. 4 Contract value is defined as the quoted price times the corresponding contract multiplier. 5 The actual target 𝜎target does not matter as we consider the aggregate speculators and not one specific CTA. 6 The trendfollowing signals in (3) use total excess returns, which can be decomposed into a spot and a carry component (see Koijen etal.,2018). Thus, if there is no secular trend in the spot component, the negative carry of those commodity markets trading in contango will cause the trendfollowing signals to be short on average. 7 To see this, we use a generalized definition of the TSMOM signal in (3) which is the weighted sum of risk adjusted returns, st=∑swsRt−s , Rt=rt∕𝜎 t and ws≥0 . This definition coincides with the one in Harvey etal.(2021) if one assumes homoscedasticity and and with the signal definition in (3) if one further assumes equal weights ws=1∕S . Now we rewrite the changes in this momentum signal as Δ s t = ∑s w s� R t−s −R t−1−s� . Using the baseline case of ws=1∕S . we get Δ s t = ( R t −R t−S−1) ∕ S . Assuming simple autocorrelation in returns, cov( R t ,R t−s) =𝜌 s , we get cov( Δs t ,Δs t−1) = ( 2𝜌−𝜌 S −𝜌 S+2) ∕S 2 ≈2𝜌∕S 2 noting that 𝜌≫𝜌 S . With var( R t) = 1 we have var( Δs t) =2 ( 1−𝜌 S+1) ∕S 2 and thus corr( Δs t ,Δs t−1) ≈ 𝜌 . 8 These results are not reported here for brevity and are available upon request. 9 The recommended smoothing coefficient for daily data is 0.94 which translates to a halflife of approximately 11.2023 days. 10 In a volatility shock scenario a CTA will absolutely reduce its position in the respective market. The shorter the halflife of the volatility forecast, the more responsive the position scaling and the larger the trade size. Consequently, a CTA will limit its own capacity ceteris paribus if the volatility forecast is faster. | 229 UHL REFERENCES Baltas, N. (2019). The impact of crowding on alternative risk premia investing. Financial Analysts Journal, 75, 89–104. https:// doi. org/ 10. 1080/ 00151 98X. 2019. 1600955 Barroso, P., Edelen, R., & Karehnke, P. (2022). Crowding and tail risk in momentum returns. Journal of Financial and Quantitative Analysis, 57, 1313–1342. https:// doi. org/ 10. 1017/ S0022 10902 1000624 Bohl, M., Pütz, A., & Solewski, C. (2021). Speculation and the informational efficiency of commodity futures markets. Journal of Commodity Markets, 23, 100159. https:// doi. org/ 10. 1016/j. jcomm. 2020. 100159 Bollen, N. P. B., Hutchinson, M., & O'Brien, J. (2021). When it pays to follow the crowd: Strategy conformity and cta performance. Journal of Futures Markets, 41, 875–894. https:// doi. org/ 10. 1002/ fut. 22199 Boos, D., & Grob, L. (2023). Tracking speculative trading. Journal of Financial Markets, 64, 100774. https:// doi. org/ 10. 1016/j. finmar. 2022. 100774 Borgards, O., & Czudaj, R. (2022). Longshort speculator sentiment in agricultural commodity markets. International Journal of Finance and Economics, 28, 3511–3528. https:// doi. org/ 10. 1002/ ijfe. 2605 Boyd, N., Harris, J., & Li, B. (2018). An update on speculation and financialization in commodity markets. Journal of Commodity Markets, 10, 91–104. https:// doi. org/ 10. 1016/j. jcomm. 2018. 05. 005 Brooks, C., Prokopczuk, M., & Wu, Y. (2015). Booms and busts in commodity markets: Bubbles or fundamentals? Journal of Futures Markets, 35, 916–938. https:// doi. org/ 10. 1002/ fut. 21721 Brown, G., Howard, P., & Lundblad, C. (2022). Crowded trades and tail risk. The Review of Financial Studies, 35, 3231–3271. https:// doi. org/ 10. 1093/ rfs/ hhab107 Campbell, J., & Thompson, S. (2008). Predicting excess stock returns out of sample: Can anything beat the historical average? Review of Financial Studies, 21, 1509–1531. https:// doi. org/ 10. 1093/ rfs/ hhm055 Carter, C., & RevoredoGiha, C. (2023). Financialization and speculators risk premia in commodity futures markets. International Review of Financial Analysis, 88, 1–10. https:// doi. org/ 10. 1016/j. irfa. 2023. 102691 CFTC. (2022). Disaggregated commitments of traders report – explanatory notes. https:// www. cftc. gov/ sites/ defau lt/ files/ idc/ groups/ publi c/@ commi tment softr aders/ docum ents/ file/ disag grega tedco texpl anato rynot. pdf Clare, A., Seaton, J., Smith, P., & Thomas, S. (2014). Trendfollowing, riskparity and momentum in commodity futures. International Review of Financial Analysis, 31, 1–12. https:// doi. org/ 10. 1016/j. irfa. 2013. 10. 001 Erb, C. B., & Harvey, C. R. (2006). The strategic and tactical value of commodity futures. Financial Analysts Journal, 62, 69–97. https:// doi. org/ 10. 2469/ faj. v62. n2. 4084 Fan, J., FernandezPerez, A., Fuertes, A.- M., & Miffre, J. (2020). Speculative pressure. Journal of Futures Markets, 40, 575–597. https:// doi. org/ 10. 1002/ fut. 22085 Fan, J., & Zhang, T. (2024). Commodity risk premia and risk management. Journal of Futures Markets, 44, 1097–1116. https:// doi. org/ 10. 1002/ fut. 22507 FernandezPerez, A., Fuertes, A.- M., & Miffre, J. (2016). Commodities for the long run. International Review of Financial Analysis, 46, 219–226. https:// doi. org/ 10. 1016/j. irfa. 2016. 06. 002 Foltice, B., & Langer, T. (2015). Profitable momentum trading strategies for individual investors. Financial Markets and Portfolio Management, 29, 85–113. https:// doi. org/ 10. 1007/ s1140 801502464 Fuertes, A.- M., Miffre, J., & FernandezPerez, A. (2015). Commodity strategies in based on momentum, term structure and idiosyncratic volatility. Journal of Futures Markets, 35, 2530–2548. https:// doi. org/ 10. 1002/ fut. 21656 Geczy, C., & Samonov, M. (2016). Two centuries of pricereturn momentum. Financial Analysts Journal, 72, 32–56. https:// doi. org/ 10. 2469/ faj. v72. n5. 1 Georgopoulou, A., & Wang, J. (2017). The trend is your friend: Timeseries momentum strategies across equity and commodity markets. Review of Finance, 21, 1557–1592. https:// doi. org/ 10. 1093/ rof/ rfw048 Haase, M., Zimmermann, Y. S., & Zimmermann, H. (2016). The impact of speculation on commodity futures markets? a review of the findings of 100 empirical studies. Journal of Commodity Markets, 3, 1–15. https:// doi. org/ 10. 1016/j. jcomm. 2016. 07. 006 Haase, M., Zimmermann, Y. S., & Zimmermann, H. (2019). Permanent and transitory price shocks in commodity futures markets and their relation to speculation. Empirical Economics, 56, 1359–1382. https:// doi. org/ 10. 1007/ s0018 101713872 Haase, M. M., Zimmermann, Y. S., & Zimmermann, H. (2017). Commodity returns and their volatility in relation to speculation: A consistent empirical approach. Journal of Alternative Investment, 20, 76–91. https:// doi. org/ 10. 3905/ jai. 2017. 20.2. 076 Harvey, C., Rattray, S., & van Hemert, O. (2021). Strategic risk management: Designing portfolios and managing risk. Wiley. He, X.- Z., & Li, K. (2015). Profitability of time series momentum. Journal of Banking and Finance, 53, 140–157. https:// doi. org/ 10. 1016/j. jbank fin. 2014. 12. 017 Hurst, B., Ooi, Y. H., & Pedersen, L. H. (2013). Demystifying managed futures. Journal of Investment Management, 11, 42–58. Hurst, B., Ooi, Y. H., & Pedersen, L. H. (2017). A century of evidence on trendfollowing investing. Journal of Portfolio Management, 44, 2–16. https:// doi. org/ 10. 3905/ jpm. 2017. 44.1. 015 Irwin, S., & Sanders, D. (2012). Testing the masters hypothesis in commodity futures markets. Energy Economics, 34, 256–269. https:// doi. org/ 10. 1016/j. eneco. 2011. 10. 008 Jegadeesh, N., & Titman, S. (1993). Returns to buying winners and selling losers: Implications for stock market efficiency. Journal of Finance, 48, 65–91. https:// doi. org/ 10. 2307/ 2328882 230 | UHL Kim, A. (2015). Does futures speculation destabilize commodity markets? Journal of Futures Markets, 35, 696–714. https:// doi. org/ 10. 1002/ fut. 21716 Kim, A., Tse, Y., & Wald, J. (2016). Time series momentum and volatility scaling. Journal of Financial Markets, 30, 103–124. https:// doi. org/ 10. 1016/j. finmar. 2016. 05. 0036 Koijen, R., Moskowitz, T., & Pedersen, L. (2018). Carry. Journal of Financial Economics, 127, 197–225. https:// doi. org/ 10. 1016/j. jfine co. 2017. 11. 002 Kwon, K. Y., Kang, J., & Yun, J. (2020). Weekly momentum in the commodity futures market. Finance Research Letters, 35, 101306. https:// doi. org/ 10. 1016/j. frl. 2019. 101306 Levine, A., Ooi, Y., Richardson, M., & Sasseville, C. (2018). Commodities for the long run. Financial Analysts Journal, 74(2), 55–68. https:// doi. org/ 10. 2469/ faj. v74. n2. 4 Levine, A., & Pedersen, L. H. (2016). Which trend is your friend? Financial Analysts Journal, 72, 51–66. https:// doi. org/ 10. 2469/ faj. v72. n3. 3 Lim, B., Zohren, S., & Roberts, S. (2019). Enhancing timeseries momentum strategies using deep neural networks. The Journal of Financial Data Science, 1(4), 19–38. https:// doi. org/ 10. 3905/ jfds. 2019.1. 015 Lutzenberger, F. (2014). The predictability of aggregate returns on commodity futures. Review of Financial Economics, 23, 120–139. https:// doi. org/ 10. 1016/j. rfe. 2014. 02. 001 Markwat, T., Quist, J., & Zomerdijk, C. (2020). Value investing for commodities. Journal of Alternative Investment, 23, 127–140. https:// doi. org/ 10. 3905/ jai. 2020.1. 107 Masters, W. M., & White, A. (2008). The accidental hunt brothers: How institutional investors are driving up food and energy prices. Technical document. Miffre, J. (2016). Longshort commodity investing: a review of the literature. Journal of Commodity Markets, 1, 3–13. https:// doi. org/ 10. 1016/j. jcomm. 2016. 01. 001 Moskowitz, T. J., Ooi, Y. H., & Pedersen, L. H. (2012). Time series momentum. Journal of Financial Economics, 104, 228–250. https:// doi. org/ 10. 1016/j. jfine co. 2011. 11. 003 Politis, D., & Romano, J. (1994). The stationary bootstrap. Journal of the American Statistical Association, 89, 1303–1313. https:// doi. org/ 10. 1080/ 01621 459. 1994. 10476870 RiskMetrics. (1996). RiskMetrics – Technical document (4th ed.). JP Morgan/Reuters. Shen, Q., Szakmary, A., & Sharma, S. (2007). An examination of momentum strategies in commodity futures markets. Journal of Futures Markets, 27, 227–256. https:// doi. org/ 10. 1002/ fut. 20252 Skiadopoulos, G. (2012). Investing in commodities: Popular beliefs and misconceptions. Journal of Asset Management, 13, 77–83. https:// doi. org/ 10. 1057/ jam. 2011. 35 Smales, L. (2016). Trading behavior in s&p 500 index futures. Review of Financial Economics, 28, 46–55. https:// doi. org/ 10. 1016/j. rfe. 2015. 11. 001 Stadtmüller, I., Auer, B., & Schuhmacher, F. (2024). Coresatellite investing with commodity futures momentum. Journal of Asset Management, 25(3), 261–287. https:// doi. org/ 10. 1057/ s4126 002400352 - 5 Vinzelberg, A., & Auer, B. (2014). Do crude oil futures still fuel portfolio performance? Review of Financial Economics, 39, 402–423. https:// doi. org/ 10. 1002/ rfe. 1127 Wimmer, T., GeyerKlingeberg, J., Hütter, M., Schmid, F., & Rathgeber, A. (2021). The impact of speculation on commodity prices: A metagranger analysis. Journal of Commodity Markets, 22, 100148. https:// doi. org/ 10. 1016/j. jcomm. 2020. 100148 How to cite this article: Uhl, B. (2025). Speculators and time series momentum in commodity futures markets. Review of Financial Economics, 43, 213–230. https://doi.org/10.1002/rfe.1228