Projects and Related Studies
Economic Indicators of Motor Fuel Excise Tax Collections
The Pacific Northwest National Laboratory (PNNL) was asked by the Motor Fuel Excise Tax Division of the United States Internal Revenue Service (IRS) to develop structural and statistical models that (1) predict what fuel excise tax revenues should be collected based on economic activity; (2) detect unexpected trends in tax collection that might require further investigation; and (3) detect possible historical under-collection of motor fuel excise taxes. PNNL has prepared a report that documents the PNNL Gasoline Excise Tax Model, which is the first installment of an all fuels model, which will allow the user to understand the underlying causes of changes in fuel consumption and excise tax revenues and to identify and explain unusual changes in consumption levels which could represent potential occurrences of evasion.
The PNNL Gasoline Excise Tax Model was used to construct quarterly estimates of gasoline and gasohol consumption and federal tax liability from 1981 through the fourth quarter of fiscal year (FY) 2001. The model indicates that prior to 1988, there was significant under-collection of federal gasoline excise tax revenue. However, since 1988, when the point of tax collection was moved up the distribution chain to the terminal rack, compliance has improved significantly, with no significant under-collection of gasoline excise taxes evident since that time (See Figure 1).
The trend since 1992, however, indicates that under-collections may be increasing since PNNL forecasts of tax collections have begun to diverge from IRS reported collections. The Federal Highway Administration (FHWA) data appear to corroborate the results of the PNNL model and investigation. This trend needs to be investigated to determine whether it is indicative of a growing compliance problem or due to a short-run problem or prediction error. The trend at this point is not beyond that which could be reasonably explained by variance in the model's prediction error.
Using the PNNL Gasoline Excise Tax Model, quarterly fuel consumption for on-highway use has been estimated for three categories of vehicles: automobiles and motorcycles, light trucks and SUVs, and heavy trucks using three-stage least squares analysis. At this point, only the gasoline sector is completely estimated. Supply and Demand equations for vehicle miles of travel (VMT) for all three vehicle categories are estimated. Demand for VMT is based upon economic variables such as Gross Domestic Product (GDP), Non-farm Employment, and the price of gasoline. Fuel efficiency equations are estimated based on CAFE standards and lagged endogenous variables. Dividing VMT by fuel efficiency by vehicle type provides fuel consumption. Based on data obtained from surveys and tax reports, fuel was divided into gasoline, gasohol and distillate according to vehicle class and allocated to taxable and non-taxable categories to determine excise tax collections.
Model regression results indicate that actual VMT are being estimated reasonably well based upon mean percent error, mean absolute percent errors and Theil statistics. Preliminary Theil forecast error statistics collected for both in and out of sample estimates indicate the model performs well for VMT. Theil relative forecast error statistics also indicate very good performance by the VMT equations. Results also show that heavy truck VMT may be changing structurally during the last two years as the relationship between GDP and heavy truck VMT appears to be diverging.
PNNL's approach differs from previous studies in that although it was developed to project fuel excise tax collections, it was also designed to detect changes in compliance levels. PNNL's approach calculates consumption and then through tax rates and accounting procedures calculates the amount of excise tax due. Most revenue prediction models statistically estimate revenue directly from VMT or consumption, and therefore, are inappropriate for detecting changes in the level of compliance. Furthermore, models that calculate revenue from consumption have not attempted to detect evasion.
Most previous evasion studies try to establish the actual level of evasion occurring using one the following methods: 1) literature review, 2) audit review, 3) analysis of border interdictions, 4) survey of tax administrators, 5) comparison of fuel supply with taxed gallons, 6) comparison of fuel consumption or sales with taxed gallons, or 7) econometric analysis. Previous approaches were clearly flawed as they would need either: 1) a 100-percent sample of consumption by taxable and non-taxable categories, which does not exist; or 2) sample information that correctly characterizes total evasion rather than sample information that has an unknown relationship to the total evasion. Therefore, the PNNL Gasoline Excise Tax Model does not capture the absolute level of evasion, but rather detects historical and current trends in compliance.
Project contact: Mark Weimar