In the case of the relative gasoline price, three estimation methods generate similar results. The
only difference is that standard errors of coefficient estimates get bigger after taking into account
cross-sectional and temporal autocorrelation, which in turn lead to a comparatively lower
significance level for corresponding variables. Crude oil and gasoline inventories, refinery
capacity, short-run supply disruption, and dummy variables for some summer months all
significantly influence the relative gasoline price. Ethanol production has a considerably negative
impact on the gasoline price, which is highly significant at the 1% level in all three estimation
results. This indicates that over the sample period, ethanol has a significant substitution effect on
gasoline. Evaluating at the sample mean, we find that the gasoline price is lowered by 39.50,
28.70, and 34.10 per gallon because of the substitution effect of ethanol.
For the 3:2:1 crack spread, the estimation results of the fixed effect and panel FGLS models are
quite different from that of the pooled OLS regression. In addition, the pooled OLS regression
model generates highly significant estimates for all explanatory variables except the dummy
variables for January, February, and November. As previously mentioned, ignoring cross-
sectional and serial correlation as well as individual heterogeneity typically leads to highly
inaccurate standard error estimation; i.e., the significance estimation results are not reliable.
Hence, we focus on the fixed effect and panel FGLS estimation results.
From these two sets of estimates, all the explanatory variables have intuitively correct signs. First,
the profitability represented by the 3:2:1crack spread presents a strong seasonal pattern. This is
reflected by the fact that the dummy variables for months in the second and third quarters are all
significant at the 1% significance level in the panel FGLS model and at the 5% level in the fixed
effect model. Second, crude oil and refinery product market conditions, refinery capacity,
ethanol production, and unexpected supply disruption significantly affect profit margins. For all
five PADD regions, unexpected supply disruption, measured by dummies for Hurricanes Katrina
and Rita, considerably increased profits in the months right after the occurrence. Gasoline
imports and the HHI are found not to have statistically significant effects on crack spread
nationally. Finally, we find that ethanol production generates negative pressure on crack spread
11