Abstract
This study examines individual commodity futures price reactions to large one-day price changes, or ‘shocks’. The mean-adjusted abnormal return model suggests that investors in 6 of the 18 commodity futures examined in this study either underreact or overreact to positive surprises. It also detects underreaction patterns in eight commodity future prices following negative surprises. However, after making appropriate systematic risk and conditional heteroscedasticity adjustments, we show that almost all commodity futures react efficiently to shocks.
Notes
1 Note that the identification of shocks is not very sensitive to the estimation window. The use of [−50, −10] or of [−45, −5] has little influence on the number or magnitude of identified shocks.
2 Although the two-stage residual method is commonly used in prior literature, the dummy variable approach is regarded as a more efficient abnormal return estimator (Karafiath, Citation1988; Mazouz et al., Citation2009).
3 There is no clear consensus in the literature as to which factors are likely to influence commodity prices. Clare et al. (Citation2014) use the traditional Fama–French–Carhart four-factor model and the eight hedge fund factors of Fung and Hsieh (Citation2001). Miffre and Rallis (Citation2007) use the prices of equity, bond and commodity indices to predict the commodity futures returns. Since finding the determinants of commodity price returns is not the main objective of this study, and due to the absence of a clear consensus in the literature on the exact factors that would influence commodity futures prices, we have chosen to adopt the approach proposed by Miffre and Rallis (Citation2007), who also examine the pricing efficiency of commodity futures, for our analysis.
4 More details on these results are available upon request.
5 The LM test detects the ARCH effect in almost all of the other commodity futures included in our earlier analysis. Details of these results are available upon request.
6 Most financial economists agree on the presence of asymmetries in asset returns due to the impact of volatility clustering (e.g. Engle et al., Citation1990) and volatility feedback (e.g. Pindyck, Citation1984; French et al., Citation1987). Similarly to the case of GJR-GARCH, EGARCH has been widely used to control the asymmetric impact of positive and negative news on the conditional variance. Using daily returns of Japanese stocks, Engle and Ng (Citation1993) show that GJR-GARCH is the best parametric model for modelling asymmetry. They also show that although EGARCH can also capture most of the asymmetry, it expresses the variability of the conditional variance at a higher than normal level. For robustness purposes, we also use the EGARCH model to verify the validity of our results to alternative conditional variance specifications. The details of the EGARCH results are not reported, as they are quantitatively similar to those of the GJR-GARCH, but they are available upon request.
7 The results obtained from assuming a generalized error distribution and a t-distribution are quantitatively similar to those obtained from the normal distribution assumption. Details are available upon request.