Like any global commodity, oil price changes are a consequence of shifting supply and demand conditions. Sudden movements – like November’s steep declines – receive ample media coverage, with industry observers providing controversial explanations for their underlying cause (consider President Trump’s recent tweet on oil prices, OPEC and the role of supply).

Although energy economists dissent on the relative importance of different transmission channels, most agree spot prices depend on fluctuations in physical supply, global demand for commodities and shifts in market sentiment (a broad term encompassing speculative changes in demand as well as uncertainty about the future).

We model supply and demand conditions by employing a structural econometric model that allows for dynamic interactions between real (CPI-adjusted) oil prices, global economic activity, the trade-weighted value of the US dollar, oil supply and above-ground crude oil inventories. Figure 1 decomposes year-on-year growth since 1991 into five different components, so that the sum of each yields the observed price change at every point in time. Because our model allows for second-round effects, the bars also capture indirect channels of transmission (e.g. a weaker dollar may lift demand for commodities, so the resulting price increase partly reflects stronger real activity). Our results suggest changes in physical production have a much more limited effect than demand factors. In fact, a variance decomposition over the 1982-2018 period reveals supply shocks accounted for ~5% of price volatility, whereas global activity and the real US dollar contributed close to 50% of the overall variability.

Nevertheless, production shocks do play an important role at specific points in time. Panel A in Figure 2 shows the cumulative breakdown during the first 8 months of 2018. Restrained supply by OPEC members and sharp declines in crude oil stocks in mature markets contributed roughly one-third of the 31% increase. Perhaps more importantly, expectations of future supply disruptions constitute the single most important source of volatility (~46% of overall variability over the period 1982-2018). We can see from panel B that anticipation effects (e.g. to US sanctions on Iranian exports) drove up prices well above their equilibrium level, keeping them elevated even when macro forces started to lessen.

Why should we care about the source of change?

Since oil prices depend on a variety of factors, economic and financial variables may react differently to oil price changes triggered by supply or demand shocks. Figure 3 shows the cumulative responses of selected US indicators to a 10% increase in real oil prices arising from: a) a stronger world economy and b) speculation. The difference in results highlights the importance of analyzing different channels of transmission.

If oil price increases arise from stronger global conditions, household spending on goods rises 0.8% above trend despite a 0.25% increase in consumer prices. Higher expected earnings lift the S&P 500 4.3% above baseline, while the 10-year Treasury bond grows 21 basis points reflecting stronger growth prospects, rising inflation expectations and a portfolio rebalancing effect. The effect is entirely different if triggered by speculation (panel B). In this case, higher prices cause a loss of purchasing power, so consumption falls 0.2% below trend. Reduced growth prospects offset rising inflation expectations, leaving long-term yields unchanged. Equity returns fall 1.2% due to the reduced earnings potential.

But what if oil prices start declining, as we’ve seen over the past three weeks? Although theoretical models typically assume linearity in transmission, there is no guarantee oil price declines will have the same but reverse effect on real activity. Figure 4 shows the effect of a 10% cumulative decline in oil prices driven by the same two factors. Notice the multipliers are considerably smaller. Consider the effects of a 10% drop owing to a weaker global economy. The pass-through to consumer prices is ~20% lower, evidence of sticky prices. The results also suggest that if the current drop is indeed a drag on equities (as has been suggested in the media), then the decline cannot be due market sentiment. Otherwise, oil price declines would be supportive of overall returns (panel B).

That’s not to say oil price movements have similar effects across industries. Clearly, the exposure of energy companies is different from sectors dependent on consumer spending. Figure 5 plots the dynamic response of selected equity indices to both oil market shocks. Each series measures monthly deviations from trend for the S&P 500 (light blue bars) and its energy and consumer staples components. Global macro shocks have a larger and more persistent effect, even causing a level shift in energy stocks. The sentiment shock has a transitory effect on both segments, with the direction of the response reversed. While staples fall 2.3% below baseline after 12 months, returns for US energy companies grow 1.6% as higher oil and gasoline prices temporarily compensate for weaker demand for energy goods.

Monitoring oil price risks

Forecasting oil prices is incredibly hard. In general, macroeconomic forecasts are subject to multiple sources of uncertainty, including model choice, structural change, omitted predictor bias, and the inherent unpredictability of future shocks. In the case of oil prices, even if a forecast model is well specified, prediction errors will be very large due to unexpected shifts in market sentiment. For instance, the 1-year ahead forecast errors for real oil prices from our preferred specification (low bias, high relative accuracy) are as much as 30 times higher than the corresponding prediction errors on headline CPI for advanced economies.

For variables subject to multiple shocks, probability forecasts offer a much more comprehensive assessment of growth prospects. Unlike point forecasts, these are statements of the likelihood of specific events taking place (e.g. prices exceeding a certain range). As a result, they shift the emphasis from central outcomes to the monitoring and evaluation of macroeconomic and financial stress. The difficulty in deriving probability distributions stems from the fact risk varies over the course of the business cycle. Thus, the size and shape of the predictive densities will change over time.

Panel A in Figure 6 plots 1-year ahead probability distributions for real oil prices conditional on all information observed at the time of the forecast. Panel B, in turn, shows the predictive densities for the annualized growth rates. The green bars correspond to 1-year ahead forecasts for September 2017, whereas the blue lines are the 12-month density forecasts for September 2018 (that is, prior to the October declines). The median dollar projection is clearly higher today, as prices have rallied over the past year. But upside risks have greatly diminished, with growth prospects worsening versus 2017. In particular, we can see from panel B that the likelihood of elevated oil price inflation is much lower today than it was in September of last year (the distribution narrows and shifts to the left), even if expected growth rates are close to each other. The results provide a good illustration for the need for probability forecasts in macroeconomic analysis.

This article summarizes the key findings of a longer research paper. For the full document, please click here.


Joaquín Kritz Lara is responsible for developing the Macro Research offering, and for designing and overseeing the firm’s proprietary models for structural analysis and forecasting. He is an empirical macroeconomist specialized in business cycle macro and applied time series econometrics. His research focuses on the study of international macroeconomic dynamics, and the quantification of downside risks and future uncertainty for real activity and asset prices. Joaquín holds a Master’s degree in Economics from University College London (UCL), and a Bachelor’s degree in Economics from McGill University.