Conditional price volatility, speculation, and excessive speculation in commodity markets: sheep or shepherd behaviour?

Abstract

The present study aims to investigate the dynamics of primary commodity spot prices and the role of speculation for the period 1995–2012. Using a linear and nonlinear Granger causality analysis, the relationship between speculation and GARCH conditional price volatility on the one side, and the linkage between excessive speculation and GARCH conditional price volatility on the other side, is carefully examined with the scope to establish whether volatility drives speculation or speculation drives price volatility, or whether there are no linkages between the two variables. The results show that excessive speculation leads conditional price volatility, and that bilateral relationships often exist between price volatility and speculation. In addition, the lead-lag relationships are not found for the entire sample period, but rather when small sub-periods are taken into account. It turns out, in fact, that excessive speculation has driven price volatility for maize, rice, soybeans, and wheat in particular time frames, but the relationships are not always overlapping for all considered commodities. Generally, the results under linear causality tests are in agreement with those obtained under nonlinear counterparts.

Keywords:

• GARCH conditional volatility
• excessive speculation
• linear and nonlinear Granger analysis

JEL Classifications:

• C10
• E39
• Q14

Acknowledgements

The author is grateful to the Editor, Professor Malcolm Sawyer and an anonymous referee for their insightful suggestions and comments. The author would like to thank Professor Joachim von Braun, Professor Maximo Torero, Professor Carlos Martins-Filho, Professor Feng Yao, Professor Antonio Aquino, Dr Arturo Leccadito, Dr Matthias Kalkuhl for their valuable comments. Financial support from the Federal Ministry for Economic Cooperation and Development, BMZ (Scientific Research Program on ‘Volatility in food commodity markets and the poor’) is grateful acknowledged.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. Volatility can be distinguished as conditional, historical, or implied. Conditional volatility is the standard deviation of a future return that is conditional on known information, such as the history of past returns (Taylor 2007). Conditional volatility uses ARCH-GARCH and extended GARCH models for estimating volatility expectations. Historical volatility is based on the observed movements of prices over the long-term and involves calculating historical average variance or standard deviation of log price returns. Implied volatility is the market’s forecast of the volatility of underlying asset returns. It shows how volatile an asset will be in the future (Brooks 2008). Implied volatility is inferred from the prices of call options using the Black-Scholes model.

2. An explicit relationship between spot and futures prices can be derived from the non-arbitrage theory.

3. The US Commodity Futures Trading Commission (CFTC) identifies three categories of futures traders: ‘commercial traders’, also known as hedgers, who hold position in the underlying commodity and attempt to offset risk exposure through future transactions; ‘non-commercial traders’, the so-called speculators, who hold only positions in futures contracts and do not have any involvement in the physical commodity trade; and ‘non-reportables’, who do not meet the reporting thresholds set by the CFTC. The latter traders are usually small traders, while commercials and non-commercials are reportable traders, i.e. they hold positions in futures and options at or above specific reporting levels set by the CFTC. Traders could take long, as well as short, positions in commodity futures markets depending on whether commodity prices are expected to appreciate (long → buy) or depreciate (short → sell).

4. I also computed 1-month historical volatility as annualised standard deviation of daily log returns. However, given that in historical volatility all past squared return deviations are weighted equally, only conditional volatility has been considered in the Granger test. Indeed, a GARCH model employs weighting schemes in which the most recent squared return deviations receive the most weight and the weights gradually decline as the observation recedes in time. Evidence on volatility clustering and persistence indicates, in fact, that more recent observations should contain more information regarding volatility in the immediate future than older observations (Engle 2004; Poon and Granger 2003).

5. Open interest describes the total number of futures contracts long (purchased contracts outstanding) or short (sold contracts outstanding) for a given commodity in a delivery month or market that has been entered into and not yet liquidated by an offsetting transaction or fulfilled by delivery of the commodity. In analytical terms, the market’s total open interest (TOT OI) is the sum of reporting and non-reporting positions: TOT OI = [NCL+NCS+2*NCSP]+[CL+CS]+[NRL+NRS], where non-commercial open interest is distinguished in long (NCL), short (NCS), and spreading (NCSP), while for commercials (C) and non-reportables (NR) open interest is divided in long (CL and NRL, respectively) and short (CS and NRS, respectively).

6. I have not considered any positions of index traders since the time-frame of the analysis would have been too short. The CFTC has published the Supplemental Commodity Index Traders (CIT) report since 2007.

7. The Ward index is: SL / (HS-HL) if HS ≥ HL; SS / (HL–HS) if HS < HL where SL = long speculation, HS = short hedge (all trading variables measured in contract units out-standing), HL = long hedging, SS = short speculation. All index values of the equation must be equal to or greater than one plus a liquidity factor. Suppose the index were 1.48, then speculation would be 48% over the minimum to offset the net hedged position. Index values greater than a necessary liquidity level (to be estimated subsequently) suggest excessive speculation. Index measures greater than that necessary for market liquidity indicate that groups of speculators are interpreting the same information differently or are utilising market information totally ignored by other speculative groups.

8. Results of the Adjusted Dickey-Fuller and Phillips Perron tests are not reported here for reason of space, but they are available upon request.

9. The random subsampling method has been used to generate multiple sub-samples of the original data-set based upon sampling without replacement.

10. Hiemstra and Jones showed that the asymptotic distribution of the test statistics is the same when one uses the original data or the VAR filtered residuals.

11. The C code for computations is available at the following websites:

http://research.economics.unsw.edu.au/vpanchenko/ http://www1.fee.uva.nl/cendef/whoiswho/showHP/main.asp?pid=6&selected=40.