Gasoline Prices, Vehicle Spending and National Employment: Vector Error Correction Estimates Implying a Structurally Adapting, Integrated System, 1949-2011 Danilo J. Santini, Phone: (703) 678-7656 Argonne National Laboratory E-mail: dsantini@anl.gov David A. Poyer, Phone: (404) 222-2580 Morehouse College E-mail: dpoyer@morehouse.edu Presented at: 32nd U.S. and International Associations for Energy Economics North American Conference Anchorage, Alaska: July 28-31, 2013 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. Sponsor: J. Ward, Vehicle Technologies Program, U.S. Department of Energy 1 Overview The Vector Error Correction Model (VECM) separates long-term equilibrium relationships from short term deviations from equilibrium. What is called the cointegration relationship tests for the existence of a stable, systematic integrated relationship among levels of variables and estimates correction to deviations from the cointegrated path. Multiple equilibria are possible as the cointegration path is pushed from equilibrium via outside forces. In the model estimated by VECM in this paper, the underlying theory is that reduction of costs of services from highway motor vehicles are a fundamental cause of economic growth. The theory also argues that deviations from equilibrium are caused by impulses to transportation fuel prices and elevated transportation fuel price levels. The VECM is readily capable of testing for the former short term price impulse effect over a few quarters, but not the latter, which is argued to be a more long-term process. This paper reports results emerging from several statistical experiments testing the dynamic relationship among (1) U.S. total employment, (2) real consumption expenditures on motor vehicles and (3) real gasoline price, from 1949-2011. The results are interpreted in the following way ― reduction of transportation sector input costs is a fundamental contributor to long-term economic growth. Positive transport fuel cost deviations from long term reductions are a cause of short-term macroeconomic declines. The more significant transport cost increases are reversed by adoption of more fuel efficient technology and return to a new equilibrium growth path. Transportation technology and economic growth A significant literature on transport economics has been developed over more than two centuries, in parallel with widely taught theories of micro- and macro-economics. Historically, the importance of transportation-related technological change on economic development has not been explicitly considered (Blaug, 1985). However, in general the importance of technological change on economic growth has been long recognized (Solow, 1957; Levinson, 1987). Advances in transportation technologies have been an important contributor to engineering history. Data on the major advances is available. Clearly with the globalization of national economies the importance of transportation-related technological advances has become more evident. Recent research examining 22 industrialized nations from 1962 to 1990 estimates that improvement in container ship technology was as important to growth in world trade as the reduction in trade barriers (Economist, 2013). Induced indirect benefits were argued to be as important as the direct cost reduction effects. Faiz (2013) also finds evidence for the “difficult to trace” causal relationship between rural roads, economic growth and rural prosperity, also finding indirect effects to be pronounced. One major question that we raise in this paper is whether our statistical estimates provide evidence/support that U.S. technological shifts in new light duty vehicle technology that occurred from 1975-1983 helped to insulate the U.S. economy from shocks in energy markets. An important question is whether there is evidence that the implementation of Corporate Average Fuel Economy standards (CAFE) played a role in the dynamics of the adjustment process. 2 As can be seen in Figure 1, between 1973 and 1986, there was a dramatic reduction in the fuel consumption of all vehicles and in the EPA rating-based estimate of fuel consumption of new light duty vehicles. After 1986, despite a dramatic fall in real gasoline prices, there was only a very slow upward trend in fuel consumption per new light duty vehicle. This upward trend was reversed starting in 2004, accompanied by expanding sales of multiple new light duty vehicle powertrain engine technologies – clean diesel, hybrids, and turbocharged and direct injection gasoline equipped cars and light trucks. The share of front-wheel drive vehicles, the technology enabling the sharp drop from 1973-86, also reversed a slow decline from 1990-2004, turning upward in 2005 (U.S. Environmental Protection Agency, 2013). 9 8 Gallons per 100 miles 7 6 5 4 3 All Vehicles (includes heavy trucks) 2 EPA rating of new light duty vehicles (cars and light trucks) 1 2011 2009 2007 2005 2003 2001 1999 1997 1995 1993 1991 1989 1987 1985 1983 1981 1979 1977 1975 1973 0 Figure 1 Comparison of new vehicle on-road fuel use to entire fleet fuel use (per vehicle), 1975-2011 Fuel economy standards vs. gasoline taxes The adoption of fuel economy standards is consistently regarded by economists as a “second best” option to taxation (Blumstein and Taylor, 2013). Jacobsen (2013) argues that the use of fuel economy standards impose an undesirable outcome on consumers of large used-vehicles. However, the potential macroeconomic effects and policy goals of a fuel economy standard ― reduction of overall fleet gasoline consumption and lowering of gasoline prices are not considered. If one included an accounting for the probable CAFE-induced effects on gasoline costs this would reduce the overall cost of ownership of larger vehicles, which would benefit those who prefer such vehicles. The emergence of the U.S. SUV followed the 1985-86 gasoline price collapse and the relatively low gasoline prices (in retrospect) that followed for several years. The SUV emergence occurred with a lag, starting in earnest in the mid-1990s (U.S. Environmental Protection Agency, 2013, Appendix D). The inroads of Japanese imports were halted in the 1990s. However, perhaps more research and scholarly exchange is required in this area. 3 Vehicle technology and gasoline price shocks Gruenspecht, in a 2001 paper, raised concerns about the California “zero emissions vehicle (ZEV)” program. His major criticism is that vehicle performance standards (emissions and fuel economy) are not desirable in part because they are accompanied by a delay in purchases that leave older less technologically sophisticated vehicles on the road longer. We will call this kind of delay the “Gruenspecht effect.” Earlier, Santini (1988) theoretically had addressed the idea that technological change could lead to purchase delay, with a model in which a more capital intensive/fuel-efficient vehicle technology first causes a temporary but sharp drop in sales, though the change later leads to a long-run expansion of the market. This model assumes that capital and energy are substitutes. For the model to be correct, higher spending on vehicles must produce lower fuel consumption by those vehicles. Those policy analysts who evaluate fuel economy standards take this trade-off for granted. However, economists (correctly) do not, generally treating capital and energy as complements. If vehicle spending and energy consumption are complements, then when one rises, so does the other (and vice versa). Complementarity can certainly be readily recognized to exist at the coarsest level – more vehicles will cause more fuel use. Further, with constant technology, more spending is also usually associated with more fuel use — in particular, the purchase of a V6 engine instead of a four cylinder will cost more and cause more fuel use per mile. If in fact motor vehicle spending and gasoline spending are complements under some circumstances and substitutes in others, it may be difficult to construct a model formulation that will isolate the two effects. Our model construction was not developed with such a goal in mind, but, as it turns out a case does result where vehicle spending increases are predicted to cause gasoline price reductions. This is very difficult to explain if vehicle spending and fuel use are always complementary. Another explanation for gasoline price-shock-induced declines in vehicle sales noted by Kilian (2008) is that changes in energy prices create uncertainty about future prices, causing firms to postpone irreversible investment decisions (Bernanke 1983 and Pindyck 1991). Presumably, this could also apply to individuals, which happens to be the topic of this paper. With regard to oil price shocks, using quarterly data, Hamilton in 1983 statistically estimated that oil price shocks had Granger-caused post WWII recessions up to that time. Recessions followed oil price shocks by about a year. In 1988, Hamilton asserted that the effects of oil price shocks on the macro-economy were likely transmitted through variation in passenger car output. He reiterated this opinion in 1996. Santini (1994), using annual data, also estimated macroeconomic responses of real GDP and unemployment occurred one year after a transportation fuel cost shock, and had Granger-caused both post- and pre- WWII recessions (except the Great Depression) with a year lag. The hypothesis that the automobile sector is the key sector responding to oil price shocks as stated by Hamilton in 1988 has not yet been widely accepted. Jones, Leiby and Paik (2004) lumped this argument into a broad, generic “sectoral shifts” hypothesis, concluding that continued research on the hypothesis would not support any single cause of macroeconomic responses to oil price shocks. In 2008 Kilian focused precisely on Hamilton’s actual assertion about the central mechanism, with three key modifications. He argued that gasoline prices were more relevant than oil as a specific cause of fluctuation in motor vehicle output, that if the hypothesis were correct the determination of gasoline 4 prices should be endogenous, and that if motor vehicles are the key sector, then subsector responses to gasoline price changes are important. Using bivariate vector autoregression (VAR), he illustrated that positive gasoline price changes were statistically significant causes of immediate monthly declines in domestic new vehicle sales, while new imported automobile sales actually rose initially, declining more modestly a few months after domestic vehicle sales. Kilian noted that when his 1970-2006 sample was broken into two parts at 1987/88 the response of consumption of new motor vehicles to a gasoline price increase had dropped sharply after 1987. Responses by unemployment, and other categories of consumption were also diminished. Kilian attributed the diminished response of motor vehicles to a change in mix of production capability of domestic and foreign vehicles, resulting in a better ability of domestic producers to provide smaller more fuel efficient cars in the latter period. Also in 2008, we estimated with a bi-variate VAR test that the dollar value of domestic motor vehicle output had been a statistically significant predictor of the rest of GDP from 1970 to 2005. Kilian’s 2008 work supported the existence and importance of the first causal leg of Hamilton’s specific hypothesis that the automobile sector is the key mechanism in transmitting effects of oil price shocks to the macroeconomy, while our paper supported the importance of the second leg. Kilian argued that one of two things had reduced the effect of gasoline price shocks on motor vehicle consumption. One was that the importance of the motor vehicle industry to the economy had declined. The other was that by the late 1980s and 1990s the difference between domestic and foreign auto producers had been greatly reduced, as domestic auto manufacturers offered small and energy efficient cars, while foreign manufacturers were beginning to branch out into the market for jeeps, SUVs, vans and pickup trucks. Thus, the U.S. auto industry became relatively less vulnerable to energy price increases than in the 1970s (Kilian, p. 25). Automotive fuel standards should also have contributed to this closure of the difference between domestic and foreign automakers. This explanation was not offered by Kilian. Levels vs. impulses In this paper we use the Vector Error Correction Model instead of Vector Autoregression (VAR) method we used in 2008. “With a cointegrated system, one should include lagged levels along with lagged differences” (Hamilton, 1994). The VAR method does not include use of lagged levels of variables, only lagged differences. In the VECM, the cointegration equation is used to retrospectively predict — via a cointegrated “Vector” — what values in a given year should be. These are deducted from the actual values, constructing a set of “errors” for each time period. Coefficients are estimated for a series of these lagged error estimates. The coefficients are supposed to predict how the system reacts to these errors ― how it corrects for the past errors. Thus, the term “error correction” is used. 5 The VECM software that we use in this paper does allow us to test for the existence of cointegration, and to construct “impulse response functions” that take into account effects of both lagged levels and lagged errors. To the best of our knowledge, this is the first time that the VECM has been used within this context. In Figure 2, real crude oil prices at point of first purchase (or initial sale), real gasoline prices, and an estimate of real annual expenditure to operate an average vehicle in 2005 dollars are shown. The highest Post WWII historical unemployment rates have occurred a bit after the price/cost peaks in 1981 and 2008. However, fortuitously, unemployment rates have actually dropped while real gasoline costs reached a new peak in 2011. In the case of the first price peak in 1980 there were “double dip” closely spaced recessions essentially at that point that led to the deepest decline in real GDP seen since WWII — until the 2008 case. The 2008 peak was followed immediately by what is now called the “Great Recession”. In both cases, the recessions halted the price increases, so in a certain sense saying that the recessions occurred at the peak is deceptive, since the fuel demand drop from the recession causes the end of the price increase. From 1986 to 2001 the annual cost of operating the average motor vehicle was probably at the lowest level of the entire period from 1949-2011. This general time period, stretched a few additional years, has come to be called “the Great Moderation”, due to its relatively infrequent, relatively mild recessions (Bernanke, 2004). The longest period between post-WWII recessions was 1991-2002; the second longest from 1961-69 (~ 9 years), and the third longest from 19821990 (~ 8 years). The 1961-69 mini moderation also followed two closely spaced recessions, and began with small car competition from a foreign manufacturer (the Volkswagen Beetle), with a compact car response by each of the big 3 domestic manufacturers. Based on the Fig. 2 plot, it was also a period with relatively low gasoline prices. So, just as the domestic auto industry pushed back against fuel efficient competition in the 1960s (VW Beetle), leading to a long period without recession, the process was repeated and sustained during the Great Moderation, with Japanese makes suffering from the required competition. The Great Moderation period was preceded by a far greater reduction in fuel consumption than the 1960s case. Thus, anecdotally, periods of low gasoline prices and low consumer costs of driving experience less frequent recessions than when prices are high. Perhaps high gasoline price levels cause an urgent technological response by industry and government. Then if reactions to frequent changes in technology during high gasoline price periods cause fluctuations in vehicle spending, recessions would become more frequent at that time. The fruits of development of and introduction of more efficient technology during the periods of high prices could then help push not only operating costs, but also vehicle costs downward. As Fig. 1 shows, the implementation of enhanced efficiency in new cars is followed with a lag by improved efficiency in the entire fleet. Once a pulse in efficiency has been put into place by construction of new factories producing improved vehicles, then automakers can “coast” and work on process efficiency rather than product efficiency, bringing the cost of vehicles down. The periods with high gasoline (or generically transportation fuel) prices would be periods in which substitution of capital for energy was required to move to a period when technology could once again remain constant for a number of years and energy and capital could complement one another in supporting widespread economic growth. The quiescent periods when new levels of efficiency are established may also lead to indirect benefits as productivity is improved throughout the economy via 6 3.5 All gasoline grades $/gallon except as noted 3.0 Unleaded regular gasoline Leaded regular gasoline 2.5 Per vehicle annual $/1000 Domestic crude 1st sale 2.0 1.5 1.0 0.5 2009 2006 2003 2000 1997 1994 1991 1988 1985 1982 1979 1976 1973 1970 1967 1964 1961 1958 1955 1952 1949 0.0 Year Fig. 2 Real gasoline and oil prices and annual vehicle operating costs, 1949-2011 an improved transportation system, in a virtuous cycle such as illustrated earlier by the container ship and rural roads examples. Turning to the consideration of impulses, Kilian plotted monthly energy prices from 1970-2006 and noted that volatility was greater in the 1987-2006 period. Fig. 3 illustrates quarterly price changes, which exhibit the same behavior. Kilian noted anecdotally that it was not likely that there was a stable relationship between energy price impulses and vehicle output, since the variability of vehicle sales was less in the later period, despite the increase in energy price volatility. Fig. 4 illustrates this point visually. Fig. 5 plots changes in total U.S. employment. Concerning volatility before and after Kilian’s 1987/88 cut point, the employment pattern in Fig. 5 appears much more consistent with Fig. 4 than with Fig. 3. Thus, although the visual comparison can be misleading, it nevertheless seems more intuitive to expect a strong relationship between impulses of motor vehicle spending and total employment, than impulses of gasoline price and motor vehicle spending Although Kilian stated that his 1987/88 cut point was half way through his sample, ours is not. In terms of the entire sample, the choice of cut point in terms of the path of all vehicle fleet fuel efficiency in Fig. 1 is close to the 1991 point when the sharp new vehicle fuel consumption reductions from 1977-82 have worked their way into the entire fleet. In the future, a later cut point of 1991 would be more consistent with the implications of Fig. 1. However, we do have the advantage of a degree of comparability to Kilian’s estimates. Further, the choice of 1987/88 properly leaves the gasoline price collapse of 1985-86 in the sample that is estimated to have been affected by the dramatic technological shift to front-wheeldrive in conjunction with the unibody. 7 .2 0 1987q4 -.4 -.2 The Great Recession 1950q1 1960q1 1970q1 1980q1 Quarter 1990q1 2000q1 2010q1 .4 Fig. 3. Proportionate quarterly changes in real gasoline price, 1949-2011. 0 .2 1987q4 -.2 The Great Recession 1950q1 1960q1 1970q1 1980q1 Quarter 1990q1 2000q1 2010q1 Fig. 4. Proportionate quarterly changes in real personal consumption of motor vehicles, 1949-2011. 8 .02 0 .01 1987q4 -.02 -.01 The Great Recession 1950q1 1960q1 1970q1 1980q1 Quarter 1990q1 2000q1 2010q1 Fig. 4. Proportionate quarterly changes in total U.S. employment, 1949-2011. Model Estimates We use the Vector Error Correction methodology (VECM) to construct our first tri-variate model, to more appropriately test the theories discussed, within an estimated dynamic system. The estimates are not only a test of the theories discussed, but also a test of the value of VECM in their examination. The estimate of, and test for co-integration is our first. Our variables were downloaded from EconStats. The total national employment variable was the Bureau of Labor Statistics “Employment Level” quarterly variable LNS12000000Q. Real vehicle spending was obtained by deflating the Bureau of Economic Analysis (BEA) “Personal Consumption Expenditures by Major Type of Product” accounts variable expenditures on “Motor vehicles and parts” (DMOTRC1) by the price index for Motor vehicles and parts (DMOTRG3). This was multiplied by 100. In addition to DMOTRG3, information to develop real gasoline price involved use of two of the BEA “Price Indexes for Personal Consumption Expenditures by Major Type of Product and by Major Function”. These were the indexes for (1) gasoline & other energy goods (DGOERG3) and (2) total personal consumption expenditures (DPCERG3). The former was divided by the latter. The logarithmic form of each variable was used in the estimated models. 9 Four experiments were conducted and five models estimated: 1. Restricted assuming that real gasoline price does not belong in the cointegration equation – over the 1949q2 to 2011q3 period (one-period model) 2. Unrestricted normalizing on employment and testing for cointegration with motor vehicle spending and real gasoline price – over the 1949q2 to 1987q4 and 1988q1 to 2011q3 periods (two-period model) 3. Semi-restricted: same cointegrating structure for the two-period models, but unrestricted error coefficients over the 1949q2 to 1987q4 and 1988q1 to 2011q3 periods. 4. ϋber-restricted: same cointegrating structure from the two-period model imposed on the restricted model (error coefficients as well as the cointegrating equation restricted to be the same throughout the sample period) All models were estimated using maximum likelihood procedure. Each model estimated coefficients for lagged deviations from the cointegrating equation prediction (vector errors) for four quarters. The aggregate dynamic response of the system, including both vector and error components, is called an “impulse response function”. Details on statistical results regarding parameter estimates are available on request from the authors. We concentrate in this paper on presentation and interpretations of the impulse response functions, which are readily produced by the software used and are presented after an overview discussion of the four models tested and compared. The model in experiment number one excluded gasoline prices from the cointegrating equation. This would be consistent with an assumption that gasoline prices are entirely exogenous and not influenced by U.S. actions, at least with respect to employment and real vehicle spending. Results were not at all consistent with theories and hypotheses reviewed. Experiment number two differed in two ways. First, the sample was broken into two subsamples at the 1987/88 break point and second, real gasoline prices were allowed into the cointegration equation, with normalization on total employment. In comparison to model 1, this is essentially a test of the Kilian hypothesis that gasoline prices are endogenous. Hamilton notes the potential importance of the normalization decision. “… choosing which variable to” normalize on “might end up making a material difference … for the evidence one finds for cointegration among the series.” (Hamilton, 1994, p. 589). We used the full information maximum likelihood estimation method proposed by Johansen to “avoid this normalization problem” (Hamilton, 1994, p. 590). Once experiment number two was conducted there was an amazing similarity in the empirical estimate of the cointegrating equation parameters for the two periods. Given the similarities of the cointegration system parameter estimates when the time periods were separately estimated, in the third model the same cointegrating structure was imposed for the entire period — gasoline prices were endogenous to the cointegrating relationship normalizing on employment. The coefficients of the deviations from the cointegrating equation predictions (errors) were allowed to differ between the first and second period. 10 Finally, in the fourth model, both the cointegrating equation and coefficients of the deviations from the cointegrating equation predictions (errors) were required to be the same throughout the full sample period. Tests were conducted under the following null conditions: 1. Overall structure is the same for the Post-War period (1949q2 to 2011q3) 2. The cointegrating equation is the same in the two-period model. 3. The structure for cointegrating equation from the two-period model on the one-period model Chi square test was applied to a comparison of model 2 against model 4. This test rejected the hypothesis that model 2 offered no improvement over model 4, rejecting the null hypothesis that the same dynamic and cointegrating structure was an adequate representation of the behavior of the system over the entire post-war period. When model 2 and model 3 were compared there was no evidence of a difference between the two models, which was supportive of the existence of the same cointegrating relationship over the entire period. Model 2 and model 3 are essentially identical. Between these two essentially identical models, the most readily produced impulse response functions are from the separate subperiod tests in experiment 2. Both quarterly and cumulative impulse responses are presented. Although the models were estimated with four quarters of errors, we chose to plot the impulse response functions over a period of eight quarters. Historical estimates have found that recessions follow energy price impulses in about four quarters. Therefore, if we had not plotted eight quarter impulse responses we might not have adequately understood the dynamics of the system with respect to the anticipated relationship between real gasoline price impulses and employment. The Vector Error Correction Model (VECM) addresses some of our theoretical questions. It separates long-term equilibrium relationships from short term deviations from equilibrium. The cointegration relationship tests for the existence of a stable, systematic integrated relationship among levels of variables and estimates correction to deviations from the cointegrated path. Multiple equilibria are possible as the cointegration path is pushed from equilibrium via outside forces. In the model estimated by VECM in this paper, the underlying theory is that reduction of costs of services from highway motor vehicles are a fundamental cause of economic growth. The theory also argues that deviations from equilibrium are caused by (1) impulses to transportation fuel prices and (2) elevated transportation fuel price levels. The VECM is readily capable of testing for the latter short term price impulse effect over a few quarters, but not the latter, which is argued to be a more long-term process. The test for structural differences between the 1949-87 and 1988-20011 period show a unique behavior of responses of real gasoline prices to real vehicle spending that is difficult to explain unless vehicle spending and fuel consumption were substitutes in the first period. Cointegration equation results Parameter estimates for cointegration relationships are shown in Table 1.The estimated cointegration equation(s) for experiments 3-4 consistently predicted a stable, positive long-term relationship among the levels of all three variables. The results strongly support Kilian’s hypothesis that real gasoline prices are 11 endogenous and that real motor vehicle spending and national employment have been cointegrated in a stable long-term fashion over the entire estimation period. There is no evidence of a diminishing importance of the automobile industry in the estimates and tests behind this system of equations. Impulse response results Impulse responses of the other two variables to a 1% change in the specified variable are plotted in Fig. 5. Values of six cumulative impulse responses for each of the two subperiods in experiment 2 are shown in Table 1. When any of the estimates of the error correction coefficients are statistically significant with a p-value of 0.10 or less, the quarter number for which the significant coefficient was estimated is noted, with the sign of the coefficient in parentheses. While our focus for the VECM estimates is the net impulse response functions, examination of the pattern of error correction coefficients across periods and for the full sample can reveal which of the relationships changed most dramatically, and which relationships remained fairly stable. Variable names used in the table are: RMVE (real motor vehicle expenditures); N (employment); Gas (real gasoline price). Motor vehicle spending on employment Error response coefficient estimates of the effect of motor vehicle spending on employment are unstable (Table 1). Apparently, the net error correction effect is not large and the consistent cointegration relationship dominates. The implication of the impulse response coefficients is that motor vehicle spending remained just as important to total national employment in the second period as the first. Perhaps this is a result of compensating effects. Although direct employment in the motor vehicle industry has clearly declined dramatically, indirect multiplier effects must remain strong. The loss of direct employment may be offset by the domestic motor vehicle industry vehicle mix composition effects noted by Kilian. Since fewer losses of domestic production to imports occurred in the second period, this could have offset the direct employment losses otherwise occurring in the auto industry when consumers Table 1 Summary of selected statistical results of Experiment 2 and Experiment 4 Coefficient type Variables & direction (if applicable) 1st period 8 qtr CIRF Coefficients significant in 1st period @ 10% N/A Yes Yes 2(-) 1(+),2(-),4(-) Noneb 4 (-) 4(+) none 2nd period 8 qtr CIRF Coefficients significant in 2nd period @ 10% N/A Yes Yes 2(+), 3(+) Nonec 1(-) 1(+), 2 (+) 1(+) 2(+) 4th Experiment Coefficients Cointegration N 1.0 1.0 N/A Cointegration RMVE -0.38a -0.41 a Yes Cointegration Gas -0.19 a -0.21 a No Error response RMVE on N 0.37^ 0.31^ None Error response N on RMVE 6.82^ 5.08^ 1(+),2(-),4(-) Error response Gas on RMVE -6.23^ -1.96^ 1(-) Error response Gas on N -0.26^ -0.05^ 4 (-) Error response N on Gas 20.28^ 36.96^ none Error response RMVE on Gas -0.97^ 5.16^ 3(-) ^Cumulative Impulse Response over 8 quarters a Though negative, these “cointegrating vector” coefficients imply a positive net relationship among variables – see Hamilton, pp. 586-590. Normalization (restricting N=1) is discussed there. b The p-value for the negative 1st quarter coefficient is 0.11, and the coefficient is highest of all quarters c The p-value for the positive 1st quarter coefficient is 0.22, and the coefficient is highest of all quarters 12 Figure 5 Estimated Quarterly Impulse Response Functions (Response to 1% Impulse) reduced motor vehicle spending. When consumers reduced spending on vehicles in total, they did not reduce spending on domestic vehicles as much as in the past, so when total spending was reduced, downward pressure on direct domestic vehicle employment was less in the second period. While this is inferred, further statistical testing can confirm its probable validity. 13 Employment on motor vehicle spending Error response coefficient estimates of the effect of employment on motor vehicle spending consistently have the same sign for the three largest (in absolute value) coefficients (quarters 1, 2 and 4). The second time period effects are smaller than for the first and are statistically insignificant (Table 1). In all tests illustrated in Table 1 the first quarter coefficient is largest in absolute value and is positive. The second and fourth quarter coefficients are consistently negative, tending to wash out the short term positive error correction effect. The implications are that new jobs and immediate automobile purchases are strongly associated with one another. The long-term cointegration relationship between employment and motor vehicle spending is consistent and positive. The cumulative impulse response obtained by adding each of the individual quarter impulse responses (Fig. 5) is very much the same in the first and second period. The cumulative two year response of motor vehicle spending to employment is 7% for each 1% increase in employment in the base quarter in the first period, and 5% in the second (Table 1) Gasoline price changes on motor vehicle spending The largest absolute value of the coefficient for the response of vector errors is consistently for the first quarter. The coefficient is negative and is statistically significant in two of the three cases, and is nearly significant in the third case (period 1). Considering both cointegration relationship and error responses, the estimated response of motor vehicle spending to a 1% gasoline price impulse is consistently immediately negative and remains negative in all quarters (Fig. 5). For the first period there is evidence that a recovery might begin beyond two years, though all that is happening after 8 quarters is a reduction of negative effect. In the second period there is a sharp reduction in the two year cumulative effect of a 1% gasoline price impulse on motor vehicle spending. In the first period, a cumulative two year inverse effect of 6% is predicted (Table 1). This drops to just 2% in the second period. Interestingly, however, this cannot be due to the cointegration equation, which is about the same in both periods, nor to the cumulative effect of the four error response coefficients, which sum to -0.25 in the first period, and a slightly larger -0.31 in the later period. The VECM system is complex and chains of effects can be difficult to sort out. As in Granger causality VAR models, lagged effects of the transformations of the dependent variable (not presented here) are also estimated. Statistically, in this model, support for the hypothesis that positive gasoline price impulses cause negative responses of motor vehicle spending is stronger since 1987 than before. Gasoline price changes on employment Error response coefficient estimates of the effect of gasoline price changes on employment are unstable (Table 1). Those coefficients that are estimated to be statistically significant both have a different lag point and a different sign across the two periods (Table 1). Nevertheless, the pattern of the response of a 1% impulse of gasoline price is temporally consistent (Fig. 5). There is a very small percentage change in the second year, after negligible positive change in the first year. The axis choice on the plot is chosen to visually emphasize this. Thus, there is no evidence that the initial reaction of motor vehicle sales after a gasoline price impulse could be caused by the strong immediate link of employment to motor vehicle spending. Delayed negative employment effects in the second year may compound a motor vehicle 14 spending problem that has already arisen in response to a gasoline price impulse, but it is not possible that the mechanism of cause of first year motor vehicle spending decline is a first year decline in employment. Employment changes on gasoline prices Error response coefficient estimates of the effect of employment on gasoline price are positive for three of four quarters and the four quarter sums are also consistently positive. The subperiod estimates each have one statistically significant positive coefficient, but the timing of these is as far apart as possible (Table 1). The significant coefficient for the first quarter in the second period is 4.7, while the significant coefficient for the fourth quarter in the first period is 0.89. Since the cointegrating equation is the same, the large difference in early vs. late impulse response in Fig. 5 is clearly dominated by the coefficients for the error terms. The overall cumulative impulse response of gasoline price to a 1% change in employment is large, at 21% in the first period, and 37% in the second. Employment — particularly new jobs — must have a relatively large effect on fleet fuel consumption. Given the results for effects of employment on motor vehicle spending, along with these results for employment on gasoline prices, it appears that newly employed individuals must be considerably more likely to purchase vehicles (in order to reliably get to and from the new job) and to use those vehicles to enjoy the new increase in disposable income (driving for the purpose of recreation and entertainment), leading to an unusually large increment in fuel consumption, thus driving up demand for gasoline. It may also be true that the majority of vehicle purchases associated with new jobs are purchases of used cars, which have been less fuel efficient than new cars during most of the sample period. Considering both these results and the cointegration equation results, it certainly appears that Kilian’s contention that gasoline prices are endogenous was correct. Motor vehicle spending on gasoline price This is the only case where the signs of the impulse response predictions are different from the first to second period. Error response coefficient estimates are only statistically significant in two of twelve cases among those examined for Table 1. In those two cases, different quarters are involved and the coefficients have different signs. The third quarter coefficient is negative in all three models and is estimated to be statistically significantly negative in the fully constrained full period model. The estimated cumulative two year response to a 1% increase in motor vehicle spending in the first period is estimated to be a 1% reduction in gasoline price. In the second period the sign is reversed and the magnitude of the effect quintuples (Table 1). Nevertheless, the response of gasoline price to increased motor vehicle spending in this system of equations is far less than its percentage response to a 1% increase in employment. If the only operative effect were complementarity between vehicle and gasoline spending, one would think that the estimated response of gasoline price to vehicle spending would be larger than its response to employment. As noted earlier, gasoline use can be either complementary to vehicle spending or can be substituted by additional vehicle spending, when switching technologies (diesel or hybrid for conventional gasoline, front-wheel drive unibody vehicles for rear-wheel drive). Since both effects exist within the sample, then the response of gasoline price to fuel demand changes associated with vehicle purchases could logically be less than for employment induced increases. In addition to contributing to purchases of additional vehicles, a portion of employment increases should also increase utilization of 15 presently owned vehicles as commuting to work becomes necessary. So, while increases in spending on vehicles can reduce fuel use when a new more efficient vehicle is used to replace an old inefficient vehicle, increased employment without a vehicle replacement must inevitably increase fuel use. Since vehicle spending is accounted for in the system of equations, the only estimated gasoline price effects of employment would arise from increased fuel use/demand. Thus, the 5% response of gasoline price to a 1% increase in vehicle spending, compared to its 37% increase in response to a 1% increase in employment is probably also evidence that the fuel savings substitution effect exists when new vehicles replace old. Though the decline is small in the second period, fleet fuel consumption nevertheless goes down, so some efficiency enhancing substitutions are implied (Fig. 1). Fig. 1 shows that the first period involved adoption of considerably more fuel efficient new vehicles than did the second. The first period was one during which capital and energy became substitutes when industry response to CAFE and high gasoline prices essentially forced 1980s consumers, on average, to purchase a much more efficient new technology when replacing the old. It was also argued that vehicle and fuel capital can be substitutes only when alternative powertrain technology is available to allow the switch. Fig. 1 shows that a dramatic switch was made in the 38 year long first estimation period, but not in the 24 year long second period. Sharp increases in fuel economy required by the new round of CAFÉ standards did not start until the very end of the second estimation period (Edmunds, 2013). Perhaps in 14 years an estimate of such a model for 1988 to 2025 will result in estimation of a negative reaction of gasoline price to vehicle spending. As far as estimation of the long run effects of CAFE standards is concerned, the first period is the only one for which such a test is possible. Fig. 1 shows some small fuel use reduction potential in new vehicles since 2004, but little in the entire highway fleet so far. The consumer use weightings in the new vehicle ratings may not match with the way vehicle owners actually drive. While the important question is not necessarily whether a CAFE standard accomplishes an absolute reduction in gasoline prices, it is imperative that it accomplish a reduction compared to what prices would have otherwise been. We do not have to rely on this argument, however, since the estimated impulse response function for the first period does imply that additional vehicle spending decreased gasoline prices. Any analysis of a fuel consumption standard that fails to estimate the benefits of reduced fuel prices (Jacobsen, 2013) is inaccurate. Standards that require new vehicle purchasers to purchase more efficient vehicles than they otherwise would ultimately benefit all drivers and the economy as a whole by reducing gasoline prices. Analysts of the desirability of such standards routinely consider the initial cost increases in comparison to the net benefits of reduced lifetime cost, but they seldom (if ever) incorporate an estimate of effects of reduced gasoline price, nor of potential indirect benefits resulting from greater availability of disposable income for purchases other than gasoline. Implications and Interpretations At the outset, the primary motivation was to test whether the estimated model would produce dynamic results consistent with a causal path from real gasoline price to motor vehicle spending, to national employment. In addition to the question of the magnitude of impulses and their period-to-period stability, the timing of impulses was important. If recessions tend to start about four quarters after significant 16 gasoline price impulses, then national employment effects should show up after about four quarters and later. This is what the model simulates. If motor vehicle spending is a cause of the employment change in the recession, the response of motor vehicle expenditures must begin before the response of employment. Fig. 5 shows that this is what happens. This is consistent with Kilian’s estimates for monthly unit sales. Reactions of vehicle spending to a gasoline price impulse start immediately, then employment impacts follow. The evidence from the impulse response predictions is that a decline in employment initially induced by a decline in vehicle spending can compound the initial vehicle spending effect by causing a second round of declines in vehicle spending, etc. Consistent with the classic formulation of a multiplier effect, the reductions in employment caused by the initial gasoline-priceimpulse-induced vehicle spending reduction cause a second round reduction of vehicle spending manifested by the prompt reaction of vehicle spending to employment. In the meanwhile, the initial negative response of vehicle spending to the gasoline price impulse persists for many quarters. There is no evidence of a diminishing importance of the automobile industry with respect to national employment levels. These results are strongly supportive of the specific sectoral shifts hypothesis advocated by James Hamilton, which focuses on motor vehicles. With the benefit of the important post-2004 statistical evidence (Jones, Leiby and Paik, 2004) provided by another recession, along with proper focus of the investigation, we think that the fundamental argument of Hamilton’s specific sectoral shifts hypothesis can soon be proven. A question for econometrics is how important is the use of VECM instead of VAR in conducting investigations of the sectoral shifts hypothesis and/or correctly revealing the facts of the theoretical relationship. In addition to isolating the vehicle sector as the overall sector that shifts most dramatically in response to gasoline price increases, Kilian has suggested that sub-sector compositional effects are also important, as consumers shift from larger vehicles to smaller vehicles. Thus, Kilian suggests a corollary to Hamilton’s sectoral shifts hypothesis, what we term a sector composition hypothesis. Kilian asserted that the domestic motor vehicle industry was protected (through 2008) from the early 21st century gasoline price impulses by its ability to domestically produce relatively more small, fuel efficient motor vehicles than in the past. We believe this is correct, and believe that policies that promoted domestic production of smaller cars provided a cushion that mitigated effects of the Great Recession. However, this should be analyzed by estimation of appropriate models rather than simply asserted. Our model does not contain the needed subsector detail. Even though public policy with regard to the composition effect may have reduced the magnitude of decline of sales of motor vehicles in the Great Recession, the second period statistical evidence is that the motor vehicle sector’s shifts, transmitted through personal consumption declines, caused a significant portion of the collapse into the Great Recession. We note that the more robust recovery from the double dip recessions after 1980 was associated with a dramatic drop in gasoline price. Such a drop has not yet been realized in support of the Great Recession recovery. 17 Macroeconomists have proven unable to predict recessions. Apparently the response preferred at the moment is to improve the characterization of the financial sector which is said to be helpful “when studying how an economy in ‘equilibrium’ responds to things like a spike in the price of petrol” (Economist, 2013b, p. 75). We agree with the quote of Hyun Song Shin, that “macro is an empirical subject” that “cannot forever remain impervious to the facts.” (Economist, 2013b, p. 75) If Hamilton’s 1988 sectoral shifts hypothesis is correct, better understanding the financial sector alone will not be enough. It may be the wrong sector to start with. In statistically evaluating his sectoral shifts hypothesis Hamilton has highlighted the impact of oil price shocks, then unanticipated oil price shocks on the automobile industry, because the industry is seen as a transfer mechanism for the effects of this exogenous event. Kilian asserted that the correct hypothesis would use gasoline price impulses as the starting point, and that gasoline prices are probably endogenous. Though we doubt that endogenous feedbacks alone explain most gasoline price shocks, our methods and results strongly support both elements of Kilian’s opinion. Our results imply that future analyses should attempt to isolate exogenous and endogenous components of gasoline price shifts. Though we wished to give the sectoral shifts hypothesis a fair test, our perspective on the role of transportation vehicles is actually fundamentally broader. We assume that variations in the domestic output of motor vehicles are a fundamental cause of isolated recessions, double dip recessions, Great Recessions and (in Santini’s case [1988,1989]) Depressions before WWII. Oil or gasoline prices are only one of many possible causes of the fluctuation in output of motor vehicles. Transportation cost shocks, broadly conceived, are the primary cause of dramatic fluctuation in output of motor vehicles. In this paper we add discussion of society’s desire for increased capital spending to obtain lesser fuel consumption as a cause of reduction and fluctuation of motor vehicle production during the time the transition takes place. In the case of this paper, agreements between industry and government to pursue exploitation of this trade-off (Edmunds, 2013) for the long-term interests of both (CAFE) are a core topic. The conceptual problem of technology shock, however, does not require such agreements. If industry and motor vehicle purchasers relatively suddenly decide that they want new technology to be introduced, then declines in sales and production capacity of the old technology may create sharp overall declines in motor vehicle output as the transition is accomplished. So, in addition to recommending continued study of gasoline price impulses as the cause of sharp declines in motor vehicle spending and output, we caution that there are many different candidate causes of changes in sales of motor vehicles, including: Purchase delay effects, including: Gruenspecht effects (standards imposing safety, emissions, and/or fuel efficiency technology) Bernanke and Pindyck uncertainty effects Santini capital for energy substitution effects Depreciation allowances Subsidies Taxes and fees Vehicle and network performance capabilities Interest rates 18 Credit rating systems Strikes Natural disasters Income Employment Emergence of delayed evidence on vehicle reliability and safety Fuel reformulation or switching Fueling infrastructure technology Vehicle operating network infrastructure funding and design Enabling technology feasibility shocks (such as battery technology, fuel cells, steam engines, electric street railways, internal combustion engines) All of these are studied in the transportation economics and/or engineering literature. However, they are seldom considered as potential causes of sharp fluctuations in motor vehicle output when they dramatically change. Regarding transportation economics, we are sympathetic to economic historian Blaug’s 1985 puzzlement that the economics discipline had not folded the assessment of transport cost into mainstream economic theory. Given the recent research on growth impacts of container ships, we are emboldened to ask why a discipline that has for multiple centuries emphasized the value of low cost movement of goods (tariff reduction in support of “free” trade) has failed to consider the methods by which engineers, politicians, lawyers, and corporate managers repeatedly develop and implement the technologies that enable efficient low cost trade of final goods and supply of inputs that in turn support long-term growth of the macroeconomy, even at the expense of short term difficulties. Descriptively, we note that the three longest periods between recessions since WWII were each preceded by a shift of production of domestic motor vehicles toward smaller compact cars that aided the industry in reducing market share of competing small imported cars. The latter widely spaced pair of recessions during what has been called the Great Moderation followed a reversal of market share of Japanese vehicles. The first of the three longest stretches between Post WWII recessions followed industry response to the success of the German Volkswagen Beetle. Kilian’s sector composition hypothesis argues that matching foreign competitor’s small and efficient vehicle strengths reduces losses to domestic competition when gasoline prices rise. This diminishes the odds of collapse of domestic vehicle production and therefore, of sub-sector-shift-induced recessions. It may also reduce the odds of recessions because the mix of vehicle sizes tilts toward smaller more fuel efficient vehicles than otherwise would have been sold, putting downward pressure on average vehicle purchase and operating costs, gasoline prices, and thus needed expenditure per vehicle for productive services rendered (commuting to work, delivering goods and services). Conclusions Clearly the ability to steadily reduce cost of ownership and use of motor vehicles year-by-year is consistent with enhanced economic efficiency, a fundamental contributor to economic growth. Periodic displacements from the equilibrium path induced by transportation costs occasionally rising significantly relative to the long-term cointegrated path require structural adjustment to return to the path. These 19 adjustments are accomplished via introduction of new efficiency-enhancing transportation system technology. Thanks to its ability to test data available via years of dedicated collection efforts by numerous government employees supporting the National Income and Product Accounts, the VECM seems to capture the post WWII nature of this process quite well. The results are supportive of a focused, specific sector shift hypothesis that assigns a response of the economy to transportation fuel cost increases acting via their influence on the equipment that uses the fuel, motor vehicles. They strongly support Hamilton’s 1988 hypothesis that the transfer mechanism for effects of oil price shocks on the macroeconomy passes through motor vehicles. Kilian’s 2008 contention that if motor vehicles are important gasoline prices are more appropriate is also supported, as is his contention that gasoline prices should be endogenous. Though additional empirical clarification is desirable, the results also support Santini’s argument (1988, 1989) that fuel efficiency enhancing transportation technology shifts are both macro-economically problematic in the short term, but also foundational to economic growth in the long term. References: Bernanke, B.S. (1983), “Irreversibility, Uncertainty, and Cyclical Investment,” Quarterly Journal of Economics, 98, 85-106. Bernanke, B. (2004), “The Great Moderation.” Presented at the meetings of the Eastern Economic Association, Washington, DC (Feb. 20) Blaug, M. (1985). Economic Theory in Retrospect (4 ed) Cambridge, England: Cambridge University Press. Blumstein, C. and M. Taylor (2013). Rethinking the Energy Efficiency Gap: Producers, Intermediaries and Innovation. Energy Institute at Haas working paper EI@Haas WP 243 (May). https://ei.haas.berkeley.edu/pdf/working_papers/WP243.pdf. Economist (2013a). Free Exchange: Containers have been more important for globalization than freer trade. May 18th, p. 82. Economist (2013b). Economics after the crisis: New Model Army. Finance and Economics, Jan. 19th, p. 75. Edmunds, D. (2013). New Corporate Average Fuel Economy Standards http://www.edmunds.com/fuel-economy/faq-new-corporate-average-fuel-economy-standards.html EconStats: U.S. National Income and Product Accounts (NIPA) http://www.econstats.com/nipa/nipa_2__3___5y.htm. Faiz, A. (2013). Rural Roads: Harbingers of Opportunity, Prosperity, and Livability in Developing Countries. TR News, 285, pp. 32-35. Gruenspecht (2001). “Zero Emission Vehicles: A Dirty Little Secret.” Resources, Issue 142. Winter. Pp. 7-10. 20 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. Hamilton, J.D. (1994). Time Series Analysis. Princeton University Press, Princeton, NJ. Hamilton, J.D. (1996). Analysis of the Transmission of Oil Price Shocks through the Macroeconomy. Paper presented at the DOE Conference, “International Energy Security: Economic Vulnerability to Oil Price Shocks, Washington, D.C., October. Hamilton, J. D. (2003) “What Is an Oil Shock?" Journal of Econometrics, 113, pp. 363-398. Jacobsen, Mark R. (2013). “Evaluating U.S. Fuel Economy Standards in a Model with Producer and Household Heterogeneity.” American Economic Journal; Economic Policy 5(2): 148-187. Jones, D.W., P. N. Leiby, and I. K. Paik (2004). “Oil Price Shocks and the Macroeconomy: What Has Been Learned Since 1996”? The Energy Journal 25 (2). pp. 1-32. Kilian, L. (2008). "The Economic Effects of Energy Price Shocks." Journal of Economic Literature, 46(4): 871-909. Levinson, A. (1987). Associated Press. Nobel Prize Not the Only Citation for Solow. Oct. 21. http://www.apnewsarchive.com/1987/Nobel-Prize-Not-The-Only-Citation-For-Solow/idc256f1d8e45a89ba4be1446067ea5906 Pindyck, R.S. (1991), “Irreversibility, Uncertainty and Investment,” Journal of Economic Literature, 29, 1110-1148. Santini, D.J. (1988). A model of economic fluctuations and growth arising from the reshaping of transportation technologies. The Logistics and Transportation Review, 24(2), 121-151 (June). Santini, D.J. (1989). Interactions among transportation fuel substitution, vehicle quantity growth, and national economic growth. Transportation Research A, 23(3), 183-207 (May). Santini, D.J. (1994). Verification of energy's role as a determinant of economic activity. In Advances in the economics of energy and resources, Vol. 8, Energy prices and production. ed., J.R. Moroney, Greenwich, CT: JAI Press, pp. 159-194. Santini, D.J., and D. Poyer (2008a). Motor Vehicle Output and GDP, 1968-2007. Atlantic Economic Journal. Online www.springerlink.com/content/g252004447w40868/ (Aug.) Santini, D.J., and D. Poyer (2008b). Motor Vehicle Output and Other GDP. The 66th International Atlantic Economic Conference, Montreal, Canada (Oct. 9-12, 2008). 21 Solow, Robert (1957), “Technical Change and the Aggregate Production Function,” Review of Economics and Statistics, pp. 312-320. U.S. Department of Commerce, Bureau of Economic Analysis (2012). http://www.bea.gov/ (accessed July). U.S. Department of Commerce, Bureau of Economic Analysis (2012). Gross Domestic Product (GDP), 2012. http:// www.bea.gov/national/txt/gdp-srce.txt - 33k (accessed Aug. 31). U.S. Environmental Protection Agency (2013). Light-Duty Automotive Technology, Carbon Dioxide Emissions, and Fuel Economy Trends: 1975 Through 2012. Office of Transportation and Air Quality Report EPA-420-R-13-001. Ann Arbor, MI (March 15.). http://www.epa.gov/otaq/fetrends.htm 22