The summer of 2008 saw a spike in crude oil prices to $147/bbl, followed by a steep correction in late 2008/early-2009. A subsequent rebound over the last three years has brought the question of whether it is possible to accurately forecast prices over the medium and longer term.
Making reliable forecasts of oil prices has been of interest to a wide range of economic agents, including policymakers, oil producers and consumers as well as other market participants. However, in a rapidly changing oil market, any forecast of oil prices, especially at longer horizons, is highly uncertain, a fact illustrated by the wide confidence bands around price predictions. Research shows that forecasts based on futures prices, surveys of analyst forecasts, forecasts based on a variety of simple time series regressions and other common forecasting techniques are generally inferior to the random-walk forecast, which implies that the best forecast of crude oil spot prices is simply the current price of oil. Some models improve short-term forecasts up to a year; however, at longer horizons a random walk forecast outperforms all models with the exception of forecasts based on expected inflation due to the large inflation component in the nominal price of oil at longer horizons (See, e.g., Kilian and Baumeister (2012) and Kilian, Alquist and Vigfusson (2012)).
Apart from these simple forecasting techniques, recent research based on more sophisticated VAR-based models showed substantial improvement in short term price forecasting up to a year, and even longer horizon improvement in directional accuracy. These models, using percent change in global oil production, oil price, real global activity and global inventories as their main variables, can be used to not only measure the impact of demand and supply shocks on the price of oil but also to generate projections conditional on different economic scenarios. Although promising, these models also suffer from two distinct limitations. Firstly, they only allow for temporary shocks to global oil supply, such as a disruption in Iranian production, as opposed to a persistent decline in oil supply growth driven by resource constraints, such as finite oil reserves. Secondly, and more importantly, key macroeconomic as well as oil market-specific variables that are the subject of forecasts are not available to the forecaster in real time. Most data are available with a time delay. Furthermore, preliminary data tend to be revised over time. As a result, forecast accuracy is done using ex-post revised data. Therefore, it is not possible to measure the accuracy of forecasts, as opposed to forecasts based on more easily observed futures prices or expected inflation.
Future of Oil Price Forecasting
Recent advances in VAR-based modelling and forecasting are very promising. However, the success of these models relies on timely and accurate data on oil market as well as on other macroeconomic indicators. There is an urgent and growing need for more and better data on oil reserves, production, consumption, refining, exports, imports, inventories from both OECD and, particularly, from non-OECD countries, including data on floating storage and storage costs to fully understand oil price dynamics.
Furthermore, models should be augmented to capture the feedback effects of futures prices on producers’ and consumers’ activities. Recent research using state-of-the-art econometrics finds no systematic, deleterious causality running from so-called ‘speculative’ activity to prices. However, this research fails to measure the impact of activities of financial players on the formation of expectations by physical market players. At the same time, data limitations are not limited to physical oil markets. Energy derivatives markets are also partially opaque. Therefore, transparency in both the physical and financial markets is essential to better understand possible linkages between the two markets, as well as price dynamics in oil markets more generally.
* IEA OMR, September 2012