By Bill Gilmer, Director of the Institute for Regional Forecasting
For the oil forecasting community, the most recent collapse in oil prices marks one more failure. The long trail of forecast errors includes the market implosions of 1982 and 1986, not seeing the run-up in commodity prices after 2004 and now missing the end of the same commodity boom. For those of us who depend on oil price forecasts, this is a big problem.
Try to forecast the economic outlook for Houston or the Gulf Coast, for example, without a good handle on oil prices. Right now, I am coping with oil price uncertainty by preparing several scenarios for Houston’s economic outlook, mostly conditioned by guessing when and how fast oil prices might recover.
The process took me through a number of current oil price forecasts from banks, investment houses and consultants, and the differences in opinion are wide and discouraging. I was left asking: Why is it so hard to forecast oil prices?
Oil Futures as a Spot Price Forecast
This latest forecasting led me to the crude oil futures market, an often-quoted and much-maligned forecast of oil prices. In principle, it should be a very good predictor. But in fact, using the futures price as a forecast of the spot price of oil is a very small improvement over predicting that oil prices will be the same tomorrow as they are today. That sounds terrible, until you learn that futures market predictions beat all the alternatives, including other financial models, statistical models and expert surveys. Why can’t we do better?
Figure 1 shows prices on the futures strip for NYMEX crude oil on December 31, 2015. At each date, the price is the payment that would be made and received for a barrel of West Texas Intermediate (WTI) delivered at that time. In the 1930s, it was thought that the spot or current price and all futures prices were independent, each determined by economic fundamentals prevailing at that point in time.
In the 1940s, agricultural economist Holbrook Working showed that spot and futures prices were closely linked by the cost of storage. If the 12-month futures price was higher than the spot price plus the cost of 12 months of storage, for example, I should buy inventory today, store it and sell it at a profit later. By the 1970s, economists had worked out how producers, consumers, hedgers and speculators take a history of past prices, inventories and market fundamentals, arbitrage across time, and the market simultaneously solves for the spot price and futures prices.
It also turns out futures prices can be regarded as a forecast of oil prices. For example, the December 2017, futures price in Figure 1 is $48 per barrel, implying that will be the spot price on that date. If you live in Houston, this is a very gloomy outlook. We probably need $65 per barrel to put the fracking industry back to work, and perhaps allow it to grow moderately. Futures don’t see a price near that level before 2020. How seriously do we take this forecast?
Futures as Forecaster
Work on futures prices as a forecasting tool is confusing because it swings back and forth between two concepts of “good forecaster.” One stems from the efficient markets hypothesis, where if we can show futures markets are efficient, then by implication they are good forecasters.
Alternatively, we just ask if the futures price does a good job of forecasting the spot price. If we look at out-of-sample results, does it reproduce the past well? Better than other forecasting tools?
There are two important concepts of efficient markets:
- Weak-form efficient markets reflect all publicly available data on past prices and market fundamentals, and arbitrage eliminates the profit opportunities. In theory, standard forecasting techniques relying on public data cannot improve on the futures price.
- Strong-form efficient markets contain all information, public and private. The weak form properties are subsumed here, but the question now becomes whether there are pools of private information that keep markets from being strong-form efficient. This might be a proprietary model, an analyst with extraordinary insight, or an investment bank that pours tens of millions of dollars into research.
In 1997, William Tomek, a pioneer of futures market research, reviewed decades of work on corn, soybeans, hogs and other agricultural products, and drew the following conclusion:
“The preponderance of evidence suggests that markets are weak form efficient. Thus other publicly available forecasts cannot improve on futures quotes as forecasts. This does not mean, however, that futures quotes or other forecasts have a high degree of accuracy.”
Tomek’s review was based on mature commodity markets that had been operating for decades. The question at hand is whether relatively new energy futures markets, and especially the market for crude oil, would allow us to draw similar conclusions.
Crude oil Futures
Futures markets for grains and cotton were in full swing by the 1870s, but exchanges for crude oil and other energy products weren’t established for another century. Heating oil was the first NYMEX energy product in 1978, followed by WTI crude in 1983 and later by gasoline and natural gas. The delay for crude and oil products was because much of the world’s oil changed hands at posted or official prices until 1986, with the prices negotiated between large national oil producers and major oil companies. The demise of this system allowed today’s futures market for crude to grow and rival the largest exchanges in the world, including commodities such as corn and copper.
Early studies of crude oil futures as a forecast of spot prices were deeply divided. From one study to another, the markets were/were not efficient, or futures prices were/were not good forecasters. Many of these studies were premature, as it takes years to accumulate the data needed for good studies. To get around the lack of data, early studies too often relied on prices from the fixed-price regime of the 1970s and 1980s.
To see what we know about these markets today, I found four relatively recent studies of crude oil and oil product markets; none of them used data from before 1990. This brief summary sounds very much like Tomek’s conclusions for agricultural products.
- Crude oil markets are probably weak-form efficient. Three of the four studies support the notion across all the futures horizons studied.
- The studies typically show that the futures price forecast can beat a random walk, i.e., it is better than a naïve forecast that says tomorrow will be the same as today.
- But futures are rarely better than a random walk by statistically significant margins. We can’t be 90% or 95% sure futures are better.
- Both futures prices and a random walk predict spot prices better than other financial or statistical models. For example, the study from the IMF looked at two alternative financial models and six alternative time-series models. Once more, futures beat out the random walk by a small margin, but the accuracy of other models fell far short of either futures or a random walk.
Why oil prices are hard to predict
We have dug ourselves into a pretty deep hole. Futures prices are a poor predictor of spot prices, barely beating a random walk, but standard statistical models are even worse. Since futures markets are weak-form efficient, no financial model, statistical technique or subjective survey based on public data should do better.
Why are all the forecasts so poor? It is because the world will not stand still. All of the evaluations of crude futures markets assume that on a particular day the market takes past prices, inventory data and other fundamentals to produce a set of spot and futures prices. We write down the 12-month futures price, for example, then wait a year and check the spot market to see if the forecast was right.
But that forecast was completely predicated on information available a year ago. We can all think of moments that have suddenly and unexpectedly turned oil markets on their head: the Arab oil embargo, the fall of the Shah or the invasion of Kuwait. An efficient market scrapes together all available data and uses it to look forward, but no one should pretend it can somehow divine the future.
And it doesn’t take big headlines to upset the forecast. The global crude oil market depends on the politics of dozens of producing countries, economic cycles in consumer countries and a vast infrastructure of pipes, ships and refineries. Even if we account for the known issues correctly, we could list 1,000 or more low-probability events that could push our forecast off course.
Suppose that each of these events has a probability of one in a 1,000 over the next 12 months. There is no reason to incorporate any of these possibilities into our forecast or even list them as a risk. But if these events are independent of each other, the chance that at least one will significantly and unexpectedly affect the oil market within a year is 1-(.999)1000 or 63.2%.
When I opened the newspaper December 31 and looked at the futures prices in Figure 1, what was I reading? Was the 12-month futures contract at $44 telling me what the spot price of crude oil will be a year from now? Probably not very accurately, because it is not clairvoyant; unanticipated events in crude markets over the next 12 months – those constantly changing facts – leave the futures price barely more capable than a random walk.
When important new information changes the December 31 outlook, has the futures forecast failed? No, the world changed and the futures market quickly updated its forecast to include new data – efficiently, as far as we know. As long as the world does not stand still, neither will the futures price.
But if on December 31, you wanted the best oil price forecast possible based on the facts available that day, you wanted the crude futures prices. The forecast is available daily, updated continuously and all for the price of a newspaper.
 William G. Tomek, “Commodity Futures Prices as Forecasts,” Review of Agricultural Economics, 19, #1 (Spring-Summer, 1997), p. 24.
 S. Abrosedra and H. Baghestani, “On the Predictive Content of Crude Oil Futures Prices,” Energy Policy, 32 (2004), pp 1389-1393; M. Chinn, M. Le Blanc, and Olivier Coibion, The Predictive Content of Energy Futures: An Update on Petroleum, Natural Gas, Heating Oil, and Gasoline, Working Paper #1103, National Bureau of Economic Research, January 2005; T.A, Reeve and R.J. Vigfusson, “Evaluating the Forecasting Performance of Futures Prices,” Working Paper #1025, Board of Governors of the Federal Reserve System, August 2011; D.A, Reichsfeld and S.K. Roache, “Do Commodity Futures Help Forecast Spot Prices?” Working Paper 11/254., International Monetary Fund, November 2011.
Bill Gilmer is director of the Institute for Regional Forecasting at the University of Houston’s Bauer College of Business. The Institute monitors the Houston and Gulf Coast business cycle, analyzing how oil markets, the national economy and global expansion influence the regional economy.