Do hedging and speculative pressures drive commodity prices, or the other way round?

For example, the former German Federal Minister for Food, Agriculture, and Consumer Protection, Ilse Aigner, put it bluntly in a speech: ‘We want bread—and we do not want speculation that fails to put bread on the tables!’ (Aigner 2012). More recently in March 2013, a group of twelve NGOs supported by about 240.000 signatures submitted a petition to the German Federal Minister of Finance, Wolfgang Schäuble, to tighten regulations of speculation in agricultural commodity markets.

Recent discussions how position changes and money inflows may affect prices in commodity futures markets include, e.g., Irwin et al. (2009), Gilbert (2010a), Pirrong (2010), and Irwin and Sanders (2011). General surveys on the functioning of commodity futures markets can be found in, e.g., Carter (1999) and Garcia and Leuthold (2004).

Other studies include Chang (1985) and Chatrath et al. (1997). Findings of Bessembinder (1992) and Roon et al. (2000), however, were criticized (e.g., Sanders et al. 2004) since their results reflect a strictly contemporaneous relationship and are not sufficient to conclude that commercial traders create hedging pressure on prices. The ‘causal’ relationship could also be the opposite.

More recent studies that focus on the influence of commodity index traders include Stoll and Whaley (2010), Sanders and Irwin (2011), Brunetti et al. (2011), or Irwin and Sanders (2012a). They found very little evidence that index positions influence price movements in commodity futures markets.

On the other hand, there is some indication that traders’ positions may be affected by prices. Results obtained by Wang (2003), Sanders et al. (2004), Röthig and Chiarella (2007), and Buyuksahin and Harris (2011) suggest that price changes precede position changes. Studies on motivations, trading approaches, and profit persistence of speculative traders in commodity markets include Canoles et al. (1998), Reitz and Westerhoff (2007), and Aulerich et al. (2013).

COT reports list open interest position data of market participants classified into commercial (hedgers), noncommercial (speculators), and nonreportable traders (hold positions less than CFTC reporting levels) while CIT reports also include commodity index trader positions.

The subsample split is motivated by the growth of commodity futures and options markets that started around 2004 and the emergence of new financial market participants (i.e., commodity index funds) since that time (Irwin and Sanders 2011, 2012b).

The hog contract was changed from live hog to lean hog, starting with the February 1997 contract. See, e.g., Liu (2005) for further details.

Note that these COT data include futures and futures–equivalent option positions. Futures-only positions are available weekly since October 1992. They were released monthly 1975 through 1991 and bimonthly from January 1991 to September 1992. Prior to 1975, the reports were issued semimonthly.

Spreading positions measure the extent to which each noncommercial trader holds equal long and short futures positions for the same commodity. For example, if a noncommercial trader in corn futures holds 200 long contracts and 150 short contracts, 50 contracts will appear as noncommercial long positions and 150 contracts will appear as noncommercial spreading. These spreading position figures only include intramarket spreads.

That is, not all commercial traders may be acting as hedgers. In practice, hedgers may not always and exclusively hedge positions in the physical markets and also exhibit speculative behavior. On the other hand, because of speculative position limits, speculators may have some incentives to be classified as commericals. Thus, while noncommericals can be assumed to be a relatively accurate classification of speculators, commerical positions may reflect not only pure hedging motives. In addition, there is no information available on the trading motives of nonreporting traders, and it may be assumed that they reflect a range of different hedging or speculative trading of small traders.

The CFTC also started to publish the Disaggregated Commitments of Traders report (DCOT). The DCOT report disaggregates the commercials in the COT report into processors and merchants as well as swap dealers, and noncommercial trader categories into managed money and other reportables. DCOT data are available beginning in June 2006. These classifications, however, do not resolve the uncertainty about swap dealer positions. A valuable discussion on the differences between the different report classifications can be found in Irwin and Sanders (2012a).

The major drawback of the new reports is, however, that they are not publicly available before the emergence of the new financial market participants and the growth of commodity futures markets that began around 2004, and thus, comparisons on relationships among traders’ positions and futures prices are limited. Nevertheless, if structural changes in commodity futures markets and market participants substantially changed the relationships among traders’ positions and prices, then this should also be observable using traditional COT report data.

Note that the construction of a weekly price series based on rolling from contract to contract could potentially result in ‘jumps’ in the price series. Some studies have analyzed different approaches to construct a price series by rolling from contract to contract (e.g., Ma et al. 1992; Carchano and Pardo 2009); however, it is not clear what the effect might be (if any) on Granger-causality testing. To assess whether test results are sensitive to the used price series, Granger-causality tests are additionally conducted on differenced price series using only differences of weekly prices of always the same contracts, and they are alternatively applied on open-interest-weighted futures prices based on the nearby and the next three deferred contracts.

One problem of the T index is how to deal with nonreporting positions. Following the literature for the calculation of the T index (e.g., Rutledge 1977; Sanders et al. 2010), nonreporting positions are simply allocated to the commercial and noncommercial categories in proportions as observed for reporting traders since there is no information available on the trading motives of nonreporting traders.

However, Buyuksahin and Harris (2011) used Granger-causality tests in the framework of Dolado and Lütkepohl (1996) that is robust to integrated and cointegrated data.

Another approach would be to test before for cointegration and then, conditional on the test results, choose either the VAR or VECM for the causality testing (e.g., used in Röthig 2011), Granger-causality in VECM being tested with a test setting such as presented by Mosconi and Giannini (1992). However, this approach is an example of ‘preliminary test testing,’ and significance levels and the power of the causality test will be distorted. Results obtained by Clarke and Mirza (2006) suggest that an approach such as proposed by Toda and Yamamoto (1995) is preferred to the practice of pretesting for cointegration. In a recent paper, Bauer and Maynard (2012) indicate that the lag-augmented approach can provide robust Granger-causality tests not only in the case of nonstationarity, but also for problems such as long-memory and certain (unmodeled) structural breaks. Note that there are also similar approaches given by Dolado and Lütkepohl (1996) and Saikkonen and Lütkepohl (1996). A further discussion on Granger-causality tests in the presence of integrated or cointegrated data can also be found in Toda and Phillips (1994) and Zapata and Rambaldi (1997).

To conserve space, summary statistics and correlations are reported in the supplementary material as an online resource. Further, please note that all statistical tests and data analyses are conducted in the statistical software package R (2012).

As found in previous studies (e.g., Bryant et al. 2006; Buyuksahin and Harris 2011), ADF-GLS test results generally imply that commodity futures price time series appear to be I(1) while variables of hedging and speculative activities tend to be I(0). However, no variable appears to have a greater order of integration than one.

To determine the lag order for each bivariate VAR, the BIC is used (with a maximum lag order of 10 lags). Since the maximum order of integration in any bivariate relationship appears to be one, one additional lag, \(d_{max}=1\), is included in the VAR model. Note again that the null hypothesis of no Granger-causality is only tested on the first \(p\) lag coefficients. In addition, the potential for heteroscedasticity is considered, and tests for heteroscedasticity (Breusch–Pagan tests) are performed for VAR models. Heteroscedasticity-consistent standard errors are used to correct the standard errors when necessary.

Furthermore, since position data and futures prices started to increase around 2004, constants and linear time trend coefficients in Eq. 9 will not be stable over the full sample period, and thus, the full sample may not represent a consistent time series process.

Some previous studies have found that, in particular, speculative positions react positively to prices. Sanders et al. (2004) and Sanders et al. (2009) for energy and agricultural markets, respectively, conclude that speculators tend to be trend followers while evidence for trend following or contrarian trading behavior of hedgers and nonreporting traders varies for markets. In addition, analyzing only three agricultural futures markets, Röthig and Chiarella (2007) found that speculation responds positively to previous price changes while significant results could not be reached for hedging positions. In an analysis of nonlinearities in the response of speculative positions to price changes, they found that speculative trading activity induced by price changes is stronger during price expansions.

The results of hardly any lead–lag relationship between commodity index traders and prices are in particular interesting given generally positive contemporaneous correlations between changes of index trader long positions and price returns. This indicates that relationships are either only apparent within a weekly time frame (which is unlikely given the long-term investment strategy of index funds) or there is no lead–lag relationship at all and index trader long positions and prices tend to respond contemporaneously to another factor.

Note that CIT reports, available starting January 2006, have always included futures plus option–equivalent futures positions, and futures-only CIT positions are not reported.

Note that only differences of weekly prices of always the same contracts are used. That is, when rolling from the nearest contract to the second nearest contract, the price difference of the second nearest contract is used.

To conserve space, all these alternative results are not presented in this paper. They are included in the supplementary material as an online resource.