Revealing the impact of index traders on commodity futures markets

Abstract

Commodity futures prices and volatility increased dramatically from 2006 to 2008, following a period during which index traders, a class of large investment funds, took on massive commodity futures positions. This article presents a method to reveal the extent to which index trader trading activity (volume) might have caused increases in futures price volatility. This approach is useful when position-level data are incomplete or confidential, as with index trader position data. The method is applied to leading agricultural commodity futures data. The impact of index traders is identified during their period of greatest activity, that is, 2005 to 2006, using aggregated volume data that are filtered using wavelet transforms. The filtering decision rule is guided by the Commodity Futures Trading Commission's (CFTC) finding that index traders do not engage in short-run trades. A joint model of futures (filtered) volume and (unfiltered) price volatility is estimated by 2SLS to account for the endogeneity of prices and volume. As a robustness check, both log-range and GARCH measures of volatility are used. The evidence provides no support for the claim that index traders have increased price volatility for storable commodities (grains/oilseeds) and only weak support in the case of nonstorable commodities (meats).

Notes

¹To be precise, log-returns should be computed as deviations from the long-run mean, but the empirical literature has found that imposing a long-run mean of 0 provides a substantial efficiency gain at the expense of only a very small bias.

²Before estimating the model, diagnostic tests are computed to establish the stationarity of the sample data. In general, apparent unit roots in commodity prices are caused by structural change or shifts in the mean of the variable in question (Wang and Tomek, 2007). As the existence of both a time trend (deterministic) and a unit root (stochastic) is unlikely in economic time series (Phillips and Perron, 1988), a two-step test procedure is used. First is computed a unit root test under the assumption of no time trend (augmented Dickey–Fuller). If the unit root null hypothesis cannot be rejected, a t-test is computed for the coefficient in a regression of the differenced series on an intercept. This evaluates the presence of a time trend (drift term). If the unit root null is rejected, we may be confident that the data are stationary, in which case the presence of a deterministic trend can be evaluated using a t-test of the slope coefficient in a regression of the data in levels on a time trend vector t = (1, 2, 3, 4, …, T ). Both the price volatility h_t and the natural log of volume v_t are found to be stationary.

³These results are not reported in the article but are available upon request.

⁴For all commodities, a Hausman–Wu test is computed for both the volume and the volatility equations. The test statistic has a null hypothesis of no correlation between the potentially endogenous regressor and the error term. For all commodities and for both equations, we reject the null at the 1% level of significance. Therefore, contemporaneous volume and volatility are endogenous.

⁵The results are not reported here but are available upon request.

⁶Note that full-information estimation methods (3SLS, simultaneous equations FIML) could be used to estimate a joint system of equations (see Hamilton, 1994, pp. 247–53). Although these methods are asymptotically superior, there is in a limited size sample the risk that a specification error will propagate to the entire system of equations. Based on Monte Carlo evidence, it is not clear which approach is preferable (Judge et al., 1985, pp. 646–53).