Put simply, the empirical results merely reflect the fact that ethanol production increased during the sample period whereas the ratio of gasoline to crude oil prices decreased. These trends make the empirical analysis extremely sensitive to model specification; however, we find that empirical models that are most consistent with economic and statistical theory suggest effects that are near zero and statistically insignificant.

Because ethanol production increased smoothly during the sample period, statistical analysis with this variable is fraught with danger. It is strongly correlated with any trending variable. To illustrate this point, we take the same empirical models in Du and Hayes (2011) and Du and Hayes (2012) and use them to “explain” variables that have no material relationship to US ethanol production: the US price of natural gas and unemployment rates in the US and the European Union.

Our resulting estimates suggest that increases in ethanol production “cause” reductions in natural gas prices but increases in unemployment. The estimates imply that, had we eliminated ethanol in 2010, natural gas prices would have risen by 65 percent and unemployment would have dropped by 60 percent in the US, 12 percent in the EU, and 42 percent in the UK. To further underscore this point, we provide a silly example. Again, using the same empirical models in Du and Hayes (2011) and Du and Hayes (2012), we show that ethanol production “causes” our children to age. Obviously, anyone using these models to advocate eliminating ethanol production to end the Great Recession or make children age more quickly would be greeted by extreme skepticism. We encourage similar skepticism about the estimated effect of ethanol on gasoline prices generated from these models.

—Knittel and Smith

Knittel is the William Barton Rogers Professor of Energy Economics, Sloan School of Management, MIT and Co-director of the Center for Energy and Environmental Policy Research at MIT. Smith is an Associate Professor of Agricultural and Resource Economics at the University of California, Davis.

In their paper, they start with a discussion of the basic economics of how the refining industry and note how ethanol production could influence gasoline prices, emphasizing the importance of the difference between short-run effects and long-run effects.

They then explore in some detail the issues attendant to selecting an empirical model for analysis, including time horizon and trend; choice of dependent variable; controlling for inflation; linearity assumptions; dynamics; and standard errors.

Using the Du and Hayes specifications, Knittel and Smith replicated their results, and presented results from several alternative empirical models that address some of the issues they hard earlier outline. They followed Du and Hayes in the collection of the data used in the analysis.

Noting that Du and Hayes never present the estimated effect of ethanol production on gasoline prices from their crack spread models, Knittel and Smith calculate the ethanol effect from the crack spread models as the implied increase in the crack spread from eliminating all ethanol production. They then assumed that gasoline prices rise by this amount, based on the notion that ethanol reduces the refining margin by relaxing capacity constraints and thereby reduces the prices of the refined products.

...the Du-Hayes crack-spread model produces an estimated ethanol effect of just $0.12 per gallon, a small fraction of the $0.89 estimate trumpeted by the RFA. The estimate drops further to $0.09 per gallon and becomes statistically insignificant when we deflate by the CPI, which is much more defensible than the PPI for crude energy material deflator that Du and Hayes use. When we control for the energy costs of refining using oil and natural gas prices, the estimated effect is $0.13 and statistically insignificant. Finally, the model that includes a lagged dependent variable produces the smallest estimated impact is also statistically insignificant.

We hesitate to endorse any of these models. We only claim that the number reported by the RFA and Secretary Vilsack is (a) inconsistent with the basic economics of the industry, (b) at the high end of the distribution of possible estimates, and (c) outside of the distribution of estimates one obtains when taking the economics of the industry seriously. The smoothness of the ethanol production variable means that it is easily conflated with other trends in the data. We eliminate some of these trends by controlling for the energy cost of refining using oil and natural gas prices. Doing so reduces the estimated effect to statistically insignificant amounts of $0.13 in the crack-spread model. We see these results as representing the most plausible effects, conditional on the modeling approach. However, as we note...this modeling approach does not separate the short- and long-run effects, so it is not surprising that the effect is small.

...More important than our empirical work, however, is our discussion of the basic economics of the industry. The results of Du and Hayes are at odds with the historical levels of either the crack spread or crack ratio and are inconsistent with an equilibrium in the oil refining industry. While an instantaneous surprise elimination of all ethanol sold in the US might raise gasoline prices for a short time period, one cannot assume these instantaneous effects would persist for more than a few weeks. This is precisely what Du, Hayes, the RFA, and Secretary Vilsack have done.

—Knittel and Smith

The Knittel/Smith paper has been released as a Center for Energy and Environmental Policy Research (MIT CEEPR) discussion paper and soon will be released as a UC Center for Energy and Environmental Economics discussion paper.

Resources

• Christopher R. Knittel and Aaron Smith (2012) Ethanol Production and Gasoline Prices: A Spurious Correlation