How Do Professionals Forecast Crude Oil Prices?
By Leah Zitter
Economists are hard-pressed to predict oil prices since they are volatile and depend on various situations. Experts use a range of forecasting tools to predict oil prices and depend on time to confirm or disprove their predictions. The five models used most often are oil futures prices, regression-based structural models, time-series analysis, Bayesian autoregressive models and dynamic stochastic general equilibrium graphs. Because economists are still undecided as to which method is most reliable, they use a weighted combination of them all to get the most accurate answer.
Oil Futures Prices
Central banks and the International Monetary Fund (IMF) mainly use oil futures prices as their gauge. Futures prices are used when traders create oil futures contracts where the seller agrees to sell a certain number of barrels of oil to the purchaser at a predetermined price on a predetermined date. The trader estimates the crude oil futures price by two factors: supply and demand and market sentiment. Supply and demand refers to the trader’s speculations on oil supply and the future market demand for that oil. Sentiment refers to the trader’s speculations in the increase, or decrease, of the future price of oil. Oil futures prices can be a poor predictor of the price of oil because they tend to add too much variance to the current price of oil.
Regression-Based Structural Models
Statistical computer programming calculates the probabilities of certain behaviors on the price of oil. For instance, mathematicians may consider forces such as behavior among members of the Organization of Petroleum Exporting Countries (OPEC), oil inventory levels, production costs, or oil consumption and production. Regression-based models have strong predictive power, but scientists may fail to include one or more factors, or unexpected variables may step in to cause these regression-based models to fail.
Some economists use time-series models such as exponential smoothing models and autoregressive models, that include the categories of ARIMA and the ARCH/GARCH, to correct for the limitations of oil futures prices. These models analyze the history of oil at various points in time to extract meaningful statistics and predict future values based on previously observed values. Time-series analysis sometimes errs but usually produces more accurate results when economists apply it to shorter time spans.
Bayesian Vector Autoregressive Model
Statistical computer programs use Bayesian methods to calculate the probability of the impact of certain predicted events on oil. Mathematicians use the standard regression-based model and try to improve upon it by adding calculations of possible change factors to the impacting events. Most contemporary economists like to use the Bayesian vector autoregressive (BVAR) model for predicting oil prices, although a 2015 International Monetary Fund Working Paper noted these models work best when used on a maximum 18-month horizon and when a smaller number of predictive variables are inserted. BVAR models accurately predicted the price of oil during the years 2008-2009 and 2014-2015.
Dynamic Stochastic General Equilibrium Model
Dynamic stochastic general equilibrium (DSGE) models use macroeconomic principles to explain complex economic phenomena; in this case, prices of oil. DSGE models sometimes work, but their success depends on events and policies remaining unchanged, since DSGE calculations are based on historical observations.
Combining the Models
When experts want to predict the price of crude oil, they use a weighted combination of all the models since no one model alone offers an accurate prediction. In 2014, for instance, the European Central Bank (ECB) used a four-model combination to predict oil prices to generate a more accurate forecast. There have been times, however, when the ECB has used fewer or more models to capture best results. Each mathematical model is time-dependent. Unforeseen factors that may alter the calculations include political instability, production costs or natural disasters. It is for this reason that some models work better at one time than another.