Futures crude oil prices as predictors of spot prices: lessons from the foreign exchange market

Abstract

Thanks mainly to the work of Post Keynesian economists, it is no longer universally accepted that the forward exchange rate is an accurate, unbiased and efficient forecaster of the spot exchange rate. However, the proposition that futures prices of crude oil can be used to forecast spot prices seems to be accepted without much scrutiny. This proposition is challenged both theoretically and empirically, suggesting instead that futures prices have nothing to do with forecasting. Since spot and futures prices are related contemporaneously, futures prices are as good or as bad forecasters as spot prices, in which case it is not sound to use the futures price as a forecaster and the spot price as a benchmark. The results show that spot and futures prices are not as good forecasters as they are portrayed to be. While futures prices produce small forecasting errors, because they are related contemporaneously to spot prices, they fail to capture turning points and exhibit signs of biasedness and inefficiency. Adjusting the random walk and the unbiased efficiency equations, by including a time-varying risk premium or a drift factor, does not make the models better in terms of predicting turning points.

Keywords:

• Forecasting
• futures contracts
• unbiased efficiency

Notes

1 A modified version may include a term representing the risk premium or another reflecting the effect of speculation.

2 It is not clear why in theory the forward rate “should be” an unbiased forecaster when the UEH is a “hypothesis that lacks theoretical plausibility” (Lavoie 2000; Moosa 2004b). If the correct model depicts a contemporaneous relation between spot and forward rates, a model whereby the two rates are related with a lag is likely to be misspecified.

3 This problem, however, is not unique to the testing of the UEH. As Hansen and Hodrick (1980) put it, this is a problem that “plagues much of time series analysis”.

4 Unbiased efficiency is a testable hypothesis, a theory that may or may not be supported by empirical evidence. Why is it a puzzle that this particular theory is not supported by empirical evidence? How about interpreting the results to mean that the theory is not valid and does not represent the facts on the ground? Claiming that failure to find supportive evidence for the UEH is a puzzle is tantamount to confirmation bias at its best (or worst).

5 The studies that found deviations from CIP (hence unexploited risk-free arbitrage opportunities) are typically plagued by measurement errors in the sense that reported rather than transaction data are used for the purpose of empirical testing. Taylor (1987) used high-frequency data recorded personally in the London foreign exchange market and found results overwhelmingly supporting CIP (in the sense that he detected no deviations). Committeri, Rossi, and Santorelli (1993) went even further, arguing that a proper test of CIP should be based on data that have three properties: (i) market agents could have actually dealt with them, (ii) sampled at the same instant in time, and (iii) collected from independent sources. By using data satisfying these criteria, they found unambiguous results supportive of CIP. Earlier, this point had been made by Coulbois and Prissert (1974) who suggested that reported data can differ from transaction data, thereby explaining why there sometimes appear to be deviations from CIP. For example, Treasury bill rates may be used (incorrectly) instead of Eurocurrency rates.

6 Bopp and Lady (1991) correctly argue that apot and futures prices have the same information content but this sounds inconsistent with the finding that they differ with respect to forecasting power.

7 The reason why these graphs are shown for the whole period is simple. If the price range is wide, as it is during the whole sample period, the graph will not show the turning points clearly. However, the same kind of behaviour will be observed because the spot and futures price are not independent.

8 The correlation coefficients in Figure 3 were calculated from daily data on the spot and one-month futures prices over the period 12 December 2001 to 19 January 2016.

Additional information

Notes on contributors

Imad A. Moosa

Imad A. Moosa is at School of Economics, Finance and Marketing, RMIT, Melbourne, Victoria, Australia.