Gasoline Prices, Vehicle Spending and
Vector Error Correction Estimates Implying a Structurally Adapting,
Integrated System, 1949-2011
Danilo J. SantiniSenior EconomistArgonne National LaboratoryArgonne, ILandDavid A. Poyer,ProfessorMorehouse CollegeAtlanta, GA
Failures of macroeconomic models to predict the recent financial and economic crisis are causing a re-examination of macro-economics, generally focusing on failures to understand the financial system. In general responses have addressed issues of over leveraging, the formation of market bubbles, and the effects of the bubble burst. While these financial phenomena are undoubtedly important, the major run up in the real price of gasoline from 2002 to 2008 has not been considered fundamental to the macroeconomic crisis that followed. To what extent did this process affect household budgets and total employment? Were consumer responses to gasoline price increases also a fundamental attribute of the behavior of the macroeconomy, then and before?
In 1983 James Hamilton published statistical evidence that positive oil price shocks “Granger-caused” either GDP declines or unemployment increases (i.e. recessions) about four quarters later, implying that changes in prior oil price values were statistically significant in the determination of changes of either measure of macro activity. Such pairwise or bivariate Granger causality tests have been implemented by use of Vector Autoregression (VAR) models. However, VAR models have also often since been used to develop models of “systems” with many variables.
In 1988 Hamilton stated (but did not statistically test) a hypothesis that oil price shocks cause recessions via a transmission path first to automobiles and then the rest of the economy. Kilian (2008), followed by Ramey and Vine (2010), used VAR models to test whether an index of gasoline expenditures (i.e. final consumer transportation fuel cost) predicted variation in automobile sales over a four decade test period. Kilian used multiple bivariate tests, while Ramey and Vine used trivariate models to examine the effect of alternative gasoline price measures on motor vehicle consumption. When two sub-periods with a mid-1980s break were tested using the same gasoline spending index, both sets of estimates implied that a significant link existed, but there was evidence of a structural weakening of the link after mid-1980s. Though both papers inferred support of a causal chain from gasoline to motor vehicle to macroeconomy, neither estimated a full model of this path. In our paper, we attempt to fill in that gap.
Complex, multiple variable VAR models have been used intensively in past decades to construct the statistical models that underpinned the macroeconomic models that failed to predict the recent financial and macroeconomic crisis. Vector error correction modeling (VECM), an alternative approach, has recently become available to practitioners, along with guidance on when to use it (Hill, Griffiths, and Lim, 2011). It includes a cointegrating relationship where a long term structural link among variables is estimated and plays an important role in the dynamic process. Furthermore, the VECM estimates are more efficient than the basic VAR if there is cointegration (see Hill, Griffiths and Lim).
Since the variables were determined to be cointegrated, VECM was used. Our estimates thus include both a long-term relationship among the log-values of the variables, and short term shock (error) response (correction) to the long-term “cointegrated” relationship. In our opinion, given the theoretical implications of VECM about relationships among log-values of the variables, our research makes a unique contribution to the historical body of research on the energy/macroeconomic relationship.
For the post WWII period (1949 through 2011), we constructed alternative VECMs to test for relationships among quarterly log values of (1) real gasoline prices, (2) real motor vehicle expenditures by households and (3) total U.S. employment. Three different model structures were estimated and statistical comparisons made. First, we tested a single “restricted” model for the full period with all coefficients restricted to be identical over the full period. Next we tested two separate models with a break at 1987/88. We noted that essentially all coefficients other than those for the cointegration vector differed from one period to the next, while the cointegration equation coefficients appeared almost identical. Accordingly, we tested a full period model with the same cointegration vector for the full period, allowing all other coefficients to be unrestricted between 1949-1987 and 1988-2011.
The restricted model, where the cointegrating relationship and error-correction process were assumed the same over the two periods, was statistically rejected as inferior to the third, unrestricted model. The second and third models were estimated to be essentially statistically identical. We could not reject the restriction that the cointegrating equation was the same in the two periods. However, the comparative tests of the three model structures strongly supported the existence of a structural shift for all coefficients other than the cointegration equation coefficients. For the gasoline to motor vehicle path, our results proved to be consistent with Kilian and with Ramey and Vine.
In principle, the purpose of the VECM is to reproduce a historical pattern of relationships among variables. The overall results are generally plotted as “impulse response functions”, in which a one percent change of a given variable in an initial quarter results in a pattern of response of another variable over several following quarters (eight in this case). The model estimates a system of relationships among the included variables, in both directions. Impulse response plots are available for both directions of response – variable X to Y and variable Y to X. Consideration of both directions proves important to interpretation.
For model number two (the two sub-period models), the impulse response functions (IRFs) are shown in Figure 1. There are a number of implications drawn from these results.
One overall interpretation regarding the structural shift across the two tested periods is that there has been a growing decoupling of energy use and economic activity due to growth in overall energy efficiency within the economy, with a major contribution being made by long term transportation fuel cost savings from a shift to greater new motor vehicle efficiency during the 1975-81 period.
Our conclusions are as follows:
The final conclusion also considers this paper’s strengthening and repetition of Santini and Poyer 2008 VAR estimates of a significant bi-directional link between motor vehicle output and the rest of GDP. Those 1967-2008 VAR results also separated the sample into two sub-periods at 1987/88. Cointegration was estimated to exist, but no VECM was estimated at that time. Statistical tests for significance of four quarters considered jointly supported the direction of cause from motor vehicle output to the rest of the economy for 1967-87 and 1988-2008 estimated separately, but results for individual quarters in 1967-87 did not. In the more recent paper, the adjustment parameter for the deviations of motor vehicle spending from the cointegrated path in 1949-1987 is easily statistically significant, more strongly supporting the direction of cause from motor vehicle-spending-to-macroeconomy before 1988. In contrast, each coefficient for the four lagged quarters of errors remains statistically insignificant or is significantly negative (second quarter). In other words, in our recent paper, the much stronger support for the pre-1988 vehicle-spending-to-macroeconomy path of the IRF results from the cointegration equation portion of the VECM that has no analogy in VAR. In the post-1987 period the opposite is true, as the coefficients of the four lagged quarters of errors consistently have the anticipated signs and are statistically significant for the second and third quarter lags, while the cointegration equation adjustment parameter is insignificant and of the wrong sign. Thus, the visually similar aggregate vehicle-spending-to-employment IRFs for the two sub-periods result from internal dominance by different processes within the VECM.
Figure 1. Estimated Quarterly Impulse Response Functions (Response to 1% Impulse)
In summary, the impulse response functions for the best models (i.e. Fig. 1) support an interpretation that high transport costs are problematic for the macroeconomy via effects on vehicle sales, while low costs are beneficial. We noted that the three longest recession free periods in the 1949-2011 interval occurred during low and declining real gasoline prices and/or declining real share of spending on transportation fuel (Figure 2). Further, although the estimated IRF of gasoline price on motor vehicle consumption was less after 1987, the major gasoline price increase from trough to peak was greater.
Fig. 2 Real gasoline and oil prices and annual vehicle operating costs, 1949-2011
Transport fuel cost increases have immediate negative effects on vehicle sales, but other aggregate effects of those fuel cost increases on the macroeconomy lag for about four quarters. Vehicle sales respond promptly both to changes in employment and gasoline prices. It seems at least as important to focus on interactions of vehicle spending and the macroeconomy as gasoline price and macroeconomy. We believe that variable gasoline prices are only one cause of the dramatic fluctuation of vehicle spending. Other factors, such as interest rates, household balance sheet adjustments, and dramatic technological change are probably also very important short term factors and may, like gasoline price changes, precede important vehicle sales and related employment effects.
Hamilton, J.D. (1983). Oil and the macroeconomy since World War II. J. of Political Economy, 91, 228-248.
Hamilton, J.D. (1988). A neoclassical model of unemployment and the business cycle. Journal of Political Economy, 96(3), 593-617.
Hill, R. C., W. E. Griffiths, and G. C. Lim (2011) Principles of Econometrics, John Wiley & Sons, Hoboken, NJ.
Kilian, L. (2008). "The Economic Effects of Energy Price Shocks." Journal of Economic Literature, 46(4): 871-909.
Ramey, V.A., and D.J. Vine (2010). Oil, automobiles and the U.S. economy: How much have things really changed? National Bureau of Economic Research Working Paper 16067. Cambridge, MA.
ACKNOWLEDGMENTS & DISCLAIMERS
This paper is a condensation of a presentation at the 32nd USAEE/IAEE North American Conference, Anchorage, Alaska, July 2013.
This work was supported by the Vehicle Technology Program of the Office of Energy Efficiency and Renewable Energy of the United States Department of Energy under contract # DE-AC02-06CH11357.
The submitted manuscript has been created by Argonne National Laboratory, a U.S. Department of Energy laboratory managed by UChicago Argonne, LLC, under Contract No. DE-AC02-06CH11357. The U.S. Government retains for itself, and others acting on its behalf, a paid-up, nonexclusive, irrevocable worldwide license in said article to reproduce, prepare derivative works, distribute copies to the public, and perform publicly and display publicly, by or on behalf of the Government.