After a drop from above $140 per barrel (West Texas Intermediate grade) in July 2008 to almost $30 in December of the same year, oil prices have since started to creep up and have recently topped $85. Higher oil prices on their own are not necessarily bad news, as they can be viewed as a symptom of the recovery (Lippi 2008). Indeed, past increases in world aggregate demand have fuelled oil demand (see among others, Kilian 2009). But further increases in oil prices are a concern for policymakers as they could negatively affect the ongoing global recovery by reducing households and firms’ purchasing power and feeding inflationary pressures. Forecasting oil prices has thus newly become a central policy issue.
Forecasting using futures
For that purpose, it is common practice among policymakers and professional forecasters to exploit the information from oil futures markets. Futures are, however, risky assets – the spot prices at the time of delivery may differ from the price set in the contract. Futures prices therefore incorporate risk premia so that they should be appropriately corrected to obtain a measure of the markets’ expectation of future oil prices.
Building on previous work on risk premia in Fed funds futures (Piazzesi and Swanson 2008), in a recent paper we assess the forecasting performance of oil futures contracts (Pagano and Pisani 2009). Our results suggest that:
- on average, oil futures tend to significantly under-predict the subsequently realised spot price; we interpret the (negative of) forecast error as a measure of the oil futures risk premium;
- this risk premium is larger the longer the forecasting horizon;
- the risk premium varies over the business cycle and is negatively correlated with real-time US business cycle indicators, such as the degree of capacity utilisation in the manufacturing sector;
- an out-of-sample prediction exercise reveals that futures adjusted for this time-varying component (which we dub “risk-adjusted” futures) produce significantly better forecasts than both unadjusted futures and futures adjusted for the average forecast error (“constant-adjusted” futures). Moreover, our prediction exercise provides better forecasts than a random walk, particularly at horizons of more than 6 months. For example, at the 12-month horizon the root mean square forecast error of the risk-adjusted futures is 20% lower than that obtained using either a random walk assumption or the unadjusted futures and 10% lower than that obtained using constant-adjusted futures.
An illustration of the forecasting performance of the different methodologies is presented in Figure 1. In January 1997 (upper panel) the oil spot price hovered around $26 and, according to futures (traded on the New York Mercantile Exchange), oil prices were expected to decline to just over $20 by the following January. Demand was very high and capacity utilisation in manufacturing was running well above its historical average. Accordingly, the risk premium required over the futures would have been very low and, in fact, risk-adjusted futures were virtually indistinguishable from unadjusted futures. By contrast, the constant-adjusted forecast would have signalled roughly constant prices. Indeed, by January 1998 the oil price declined, to just above $16.
In September 2003 (lower panel) oil prices were stable at around $30. Futures pointed to a decline in oil price to just below 26 dollars in the following 12 months. The recovery out of the 2001 recession was not yet firmly established, capacity utilisation was still relatively low and the risk premium was correspondingly sizeable. The risk-adjusted forecast signalled an oil price above $38 dollars by September 2004. Note that failing to take into account the cyclical factor, as in the constant-adjusted forecast, would have yielded a prediction of just slightly increasing oil prices. In the event oil prices did rise, and at the end of the horizon were at around $47.
Turning to more recent history, in Figure 2 we depict oil price movements since their trough in December 2008 and compare them with the forecasts based on unadjusted and risk adjusted futures. The large amount of slack generated by the recession enlarged the risk premium, and, again, our methodology captures much more precisely than the simple unadjusted futures the subsequent behaviour of oil prices.
Where will oil prices presumably go from here? As of mid-April 2010 our forecasts, taking into account the risk premium, indicate that oil prices may climb back to $100 by early next year. According to Hamilton (2010) if oil tops $100 US consumption expenditures on energy goods and services may exceed 6% of total personal consumption spending, a threshold that, if crossed, may dampen consumption spending – all things being equal.
Figure 2. WTI oil spot and futures prices (dollars per barrel)
Disclaimer: The opinions here expressed and the forecasts reported are only the authors’ and do not involve the Bank of Italy.
Hamilton, Jim (2010), “Do rising oil prices threaten the economic recovery?”, econbrowser.com, 10 April.
Kilian, Lutz (2009), “Not All Oil Price Shocks Are Alike: Disentangling Demand and Supply Shocks in the Crude Oil Market”, American Economic Review, 99(3):1053-1069, June.
Lippi, Francesco (2008), "Oil prices: Risks and opportunities", VoxEU.org, 11 June.
Pagano, Patrizio and Massimiliano Pisani (2009), “Risk-Adjusted Forecasts of Oil Prices”, The B.E. Journal of Macroeconomics, 9(1), Article 24.
Piazzesi, Monika and Eric Swanson (2008), “Future Prices as Risk-Adjusted Forecasts of Monetary Policy”, Journal of Monetary Economics, 55:677-691.