jel071308final.dvi
The Economic Effects of Energy Price Shocks∗
Lutz Kilian†
University of Michigan and CEPR
July 12, 2008
Abstract
Large fluctuations in energy prices have been a distinguishing characteristic of the U.S.
economy since the 1970s. Turmoil in the Middle East, rising energy prices in the U.S. and
evidence of global warming recently have reignited interest in the link between energy prices
and economic performance. This paper addresses a number of the key issues in this debate:
What are energy price shocks and where do they come from? How responsive is energy demand
to changes in energy prices? How do consumers’ expenditure patterns evolve in response to
energy price shocks? How do energy price shocks affect U.S. real output, inflation and stock
prices? Why do energy price increases seem to cause recessions, but energy price decreases do
not seem to cause expansions? Why has there been a surge in the price of oil in recent years?
Why has this new energy price shock not caused a recession so far? Have the effects of energy
price shocks waned since the 1980s and, if so, why? As the paper demonstrates, it is critical to
account for the endogeneity of energy prices and to differentiate between the effects of demand
and supply shocks in energy markets, when answering these questions.
KEYWORDS: Crude Oil; Gasoline; Price Shocks; Propagation; Channels of Transmission;
Asymmetry; Elasticity.
JEL: E21, Q43.
∗I thank Paul Edelstein for excellent research assistance. Lucas Davis and Ana-María He
era provided helpful
comments on an earlier draft of the paper, as did three anonymous referees and the editor.
†Department of Economics, University of Michigan, 611 Tappan Street, Ann A
or, MI 48109-1220. E-mail:
[email protected].
1 Introduction
The price of energy is only one of many prices faced by households and firms, yet it attracts a dis-
proportionate amount of attention in the media and from policymakers and economists. A common
perception is that energy price increases are fundamentally different from increases in the prices of
other goods. One reason is that energy prices at times experience sharp and sustained increases not
typical of other goods and services. A second reason is that these price increases matter more than
in the case of other goods because the demand for energy is comparatively inelastic. For example,
most workers have to drive to work every day and thus have little choice but to acquiesce to highe
gasoline prices. Similarly, households have little choice but to endure higher natural gas prices, as
they cannot afford to leave their homes unheated. A third reason that energy prices are perceived to
e different is that energy price fluctuations seem to be determined by forces that are exogenous to
the U.S. economy such as political strife in the Middle East. A fourth reason is that major energy
price increases in the past have often been followed by severe economic dislocations, suggesting a
causal link from higher energy prices to recessions, higher unemployment and possibly inflation.
In this paper, I selectively review the recent literature on the effect of energy price shocks on the
U.S. economy. For a complementary survey of the relationship between oil prices and the macro-
economy the reader is refe
ed to Hamilton (2008). My objective is not to provide a comprehensive
survey of the literature. Rather I wish to highlight some recent methodological developments and to
outline how these developments have altered our perceptions of where energy price shocks originate,
how they are transmitted, and how much they affect the economy. The paper addresses a numbe
of the key questions in the ongoing debate about the economic effects of energy price shocks: Where
do energy price shocks originate? How responsive is energy demand to changes in energy prices?
How do consumers’ expenditure patterns evolve in response to energy price shocks? How do energy
price shocks affect U.S. real output and inflation and the U.S. stock market? Why do energy price
increases seem to cause recessions, but energy price decreases do not seem to cause economic ex-
pansions? Why has there been a surge in the price of crude oil in recent years? Why has this new
energy price shock not caused a recession so far? Have the effects of energy price shocks waned since
the 1980s and, if so, why?
The remainder of the paper is organized as follows. In section 2, I discuss the identification
of exogenous shifts in energy prices with special emphasis on alternative specifications of energy
price shocks and alternative frameworks for estimating the effects of energy price shocks. Section
3 provides an overview of the effects of unanticipated changes in energy prices on U.S. consume
expenditures and firms’ investment expenditures. I discuss the most prominent channels of trans-
1
mission and the empirical evidence in their support. I also address the question of whether there
is an asymmetry in the responses to energy price increases and energy price decreases. Section 3
also contains detailed estimates of the responsiveness of energy consumption to higher energy prices
and, more generally, assesses how consumers’ expenditure patterns evolve in response to energy price
shocks. The section concludes with a discussion of the link between crude oil prices and monetary
policy. In section 4, I address the question of why the effects of energy price shocks have weakened
since the second half of the 1980s. Section 5 illustrates how disentangling demand and supply shocks
in oil markets can help us understand the evolution of energy prices. This section also demonstrates
the differential impact of demand and supply shocks in the global oil market on U.S. real GDP,
consumer prices, and stock prices. The concluding remarks are in section 6.
2 Methodological Issues Raised by the Endogeneity of En-
ergy Prices
It is widely accepted that energy prices in general and crude oil prices in particular have been endoge-
nous with respect to U.S. macroeconomic conditions dating back to the early 1970s. Endogeneity
here refers to the fact that not only do energy prices affect the U.S. economy, but that there is re-
verse causality from U.S. and more generally global macroeconomic aggregates to the price of energy.
Clearly, both the supply of energy and the demand for energy depend on global macroeconomic ag-
gregates such as global real economic activity and interest rates (see Barsky and Kilian 2002, 2004).
Thus, a co
elation between energy prices and macroeconomic outcomes does not necessarily imply
causation. One response to this problem has been to apply statistical transformations to the price
of energy to extract the exogenous component of oil prices. The leading example of this approach
is the net oil price increase measure proposed by Hamilton (1996, 2003).1
2.1 Measures of Net Oil Price Increases
Building on work by Mork (1989), Lee, Ni and Ratti (1995) and Hamilton (1996), Hamilton (2003)
suggested that, although the price of crude oil itself is not exogenous with respect to U.S. macro-
economic aggregates, a suitable nonlinear transformation of the price of oil (based on the amount
y which nominal oil prices exceed their maximum value over the previous three years) is. Such
1The net oil price increase is defined as the difference between the cu
ent price of oil and the maximum price
over the previous year (or alternatively the previous three years) if the cu
ent price exceeds the previous maximum,
and zero otherwise. A number of alternative oil price transformations have been suggested by, among others, Lee, Ni
and Ratti (1995). Hamilton (2003) shows that these alternative measures tend to produce results similar to the net
increase measure.
2
transformed regressors have been used widely in studying the impulse responses of U.S. sectoral and
macroeconomic aggregates to oil price shocks. The purpose of the statistical transformation of oil
prices is to isolate the component of the price of crude oil that can be attributed to political events
in the Middle East, which in turn are exogenous to global macroeconomic conditions. In support of
exogeneity of the net oil price measure, Hamilton showed that using the oil price changes predicted
y exogenous oil supply variations as instruments in a regression of real GDP growth on lagged
percentage changes in nominal oil prices results in estimates of the structural regression coefficients
that look remarkably similar to the reduced-form estimates obtained from regressing GDP growth
on net oil price increases instead. Hamilton concluded that the latter reduced-form relationship
effectively represents a causal relationship, lending credence to the practice of treating net oil price
increases as exogenous.2
Although Hamilton’s analysis represents an important step forward in this literature, there is
eason to be skeptical of the exogeneity of the net oil price increase measure because of the nature of
the instruments used by Hamilton. As is well known, weak instruments produce biased instrumental
variable (IV) regression estimators and hypothesis tests with large size distortions (see Stock, Wright
and Yogo (2002) for a review). In the presence of many variables to be instrumented, as in the IV
egressions presented in Hamilton (2003), weak instrument problems may be detected based on the
gmin statistic of Cragg and Donald (1993). Critical values for a formal test of the null hypothesis
of weak instruments have been compiled by Stock and Yogo (2005). Table 1 presents IV regression
estimates of the type presented in Hamilton (2003) in support of the exogeneity of the net increase
measure of the price of crude oil. In addition to the specification favored by Hamilton (2003), which
is shown in column (1), I include a number of alternative specifications involving different measures
of exogenous crude oil supply shocks developed in Kilian (2008a,b), different sample periods, and
different measures of nominal and real oil prices. Table 1 shows that in all cases the estimated
gmin-statistics are far below the least conservative critical value of about 4, suggesting that a weak-
instrument problem cannot be ruled out. The presence of weak instruments would render the IV
estimates inconsistent and standard inference on the dynamic effects of exogenous variations in the
price of oil invalid, invalidating any comparison with the reduced-form evidence. Consequently, we
have no credible evidence that the responses to net oil price increases co
espond to those of truly
causal models.
This evidence of weak instruments is not surprising, as it has been shown that measures of
exogenous oil supply shocks driven by political events in the Middle East fail to explain much of
2For example, Hamilton (2003, p. 395) writes that nonlinear transformations “. . . filter out many of the endogenous
factors that have historically contributed to changes in oil prices” and “seem in practice to be doing something rathe
similar to isolating the exogenous component of oil price changes” (p. 391).
3
the observed fluctuations in crude oil prices. This result is robust across alternative specifications of
exogenous political oil supply shocks. For example, at most one fourth of the observed increase in
the price of oil in 1973/74 can be explained based on such shocks (see Kilian 2008a). In fact, recent
evidence in Kilian (2008c) suggests that crude oil supply shocks, na
owly defined as unanticipated
changes in world crude oil production, are far less important for understanding crude oil prices than
are shocks to the demand for crude oil. This point will be discussed in more detail in section 5.
The fact that net oil price increases are not measures of exogenous oil price shocks can also be
seen more directly. For example, Kilian (2008a) shows that oil price shocks detected by the nonlinea
transformation proposed by Hamilton (2003) occur at times when exogenous oil supply shocks in
the Middle East were conspicuously absent, and that there are major exogenous events that are not
followed by oil price shocks. Kilian also shows that the same procedure applied to other industrial
commodity prices generates spurious evidence of exogenous price shocks. Hence, oil price series must
e treated as endogenous whether they have been transformed to net oil price increases or not.
The lack of exogeneity of the price of oil is not as much of a problem as it may seem, because
exogeneity is not required for estimating the economic effects of changes in oil prices. A much weake
and more defensible assumption than strict exogeneity of the price of oil is that innovations to the
oil price series (whether transformed or not) are predetermined with respect to U.S. macroeconomic
aggregates. In other words, the price of oil responds to changes in macroeconomic conditions only
with a delay. Under this assumption, recursively identified vector autoregressions with energy prices
ordered first may be used to estimate the dynamic effect of a change in energy prices. Indeed, this
assumption forms the basis of a number of influential papers in the literature including Davis and
Haltiwanger (2001) and Lee and Ni (2002) that focus on the linearly unpredictable component of
the net increase in oil prices. The assumption of predeterminedness typically is inappropriate when
working with annual data, but may provide a good approximation when working with quarterly
and in particular with monthly data. Thus, there are clear advantages to studying the effects of
energy price shocks in a high-frequency time series context compared to panel studies using data at
lower frequencies.3 The implementation of this VAR approach is discussed in more detail in section
2.5. Although the exactly identifying assumption that energy prices are predetermined with respect
to domestic macroeconomic aggregates is not testable within the VAR framework, the working
hypothesis that the feedback from shocks to domestic macroeconomic aggregates to the global price
of oil is negligible within the same month has been regarded as plausible by many researchers.
Ongoing work by Kilian and Vega (2008) tests this hypothesis formally using regressions of daily
3 In the latter case, it becomes necessary to find suitable instruments for energy price changes. For example, Cullen,
Friedberg and Wolfram (2004) use weather data as exogenous instruments for home energy costs.
4
changes in oil prices on U.S. macroeconomic news.
2.2 The Effect of Changes in the Energy Share
It is sometimes argued that regressions of macroeconomic aggregates on unanticipated energy price
changes are potentially misleading in that they fail to account for the declining share of energy in
value added since the 1970s. Indeed this share has been falling from a maximum of 5% in 1981
to a low of 1% in 1998, but by 2005 the share had recovered to its original level of 3.3% in 1977.4
Interestingly, the pattern of these fluctuations seems to reflect primarily changes in the price of
crude oil rather than shifts in energy use. While not trivial, the observed fluctuations in the energy
share in value added are largely immaterial for estimates of energy price shocks because the share
does not fluctuate enough on a quarter-to-quarter basis. Weighted and unweighted quarterly energy
price changes have a co
elation of 99%. Thus, little is lost by studying the effect of unweighted
energy price shocks, as has been demonstrated in Edelstein and Kilian (2008b). Weighting can be
important, however, in constructing measures of the retail energy price shocks faced by households
and firms, as the next subsection illustrates.
2.3 On the Choice of the Energy Price Series
Much of the work on energy price shocks has focused on the price of crude oil. This approach reflects
the perception that the bulk of the fluctuations in energy prices since the 1970s has been driven by
distu
ances in global crude oil markets that are reflected in the price of imported crude oil and are
transmitted from the crude oil market to retail energy prices. While it is true that distu
ances in
global crude oil markets are typically reflected in gasoline prices, this is not the only major source of
shocks to gasoline prices. A good example is provided by the effects of Hu
icanes Rita and Katrina
in late 2005. Whereas the reduction of U.S. crude oil supply associated with these exogenous events
was negligible on a global scale, the reduction in refining capacity was not. It constituted both a
decline in the demand for crude oil (associated with a fall of crude oil prices) and a decline in the
supply of gasoline and other refined products, reflected in a sharp increase in gasoline prices. This
example illustrates that from a consumer’s point of view a direct measure of retail gasoline prices
may be more relevant than a measure of crude oil prices.
As of 2002, according to the BEA, gasoline accounts for 48.7% of all energy used by consumers,
compared with 12.3% for natural gas and 33.8% for electricity. Since gasoline is by far the most im-
4See Edelstein and Kilian (2007b). Following Rotemberg and Woodford (1996), the energy share in value added is
approximated by the sum of nominal value added in oil and gas extraction and imports of petroleum and petroleum
products, divided by nominal GDP. No disaggregate value added data are available prior to 1977.
5
portant form of energy consumed in the United States and the one with the most volatile price, little
would be lost by focusing on gasoline prices alone in studying the response of consumer expenditures.
In contrast, in studying firm behavior neither gasoline nor crude oil prices will be representative of
energy prices. For producers, based on 2002 BLS data, electricity makes up 40.3% of energy use,
natural gas 14.5%, and unleaded gasoline only 14%, followed by diesel fuel (11.4%) and jet fuel
(9.7%). The difference in weights has important implications for the magnitude of energy price
shocks. For example, during the Persian Gulf War in 1990, crude oil prices rose by 83%, whereas
the intermediate energy prices faced by firms only rose by 12%. This example illustrates that the
choice of energy price series may matter a great deal. Different questions may require a different
measure of energy price shocks.
2.4 Alternative Specifications of Energy Price Shocks
In studying the effects of energy price shocks, a natural baseline is the hypothesis that firms and
consumers respond proportionately to a percent change in energy prices, regardless of the magnitude
of the change. We will refer to this model as the percent change or C specification. There are othe
ehavioral models, however, that involve nonlinear transformations of the price of energy. One
alternative is the possibility that consumers and firms only respond to large shocks. For example,
the presence of costs to monitoring energy expenditures and of costs of adjusting consumption
patterns might make households reluctant to respond to small energy price changes (see Goldberg
1998). One may allow for that possibility by defining energy price shocks as increases or decreases
that exceed, say, one standard deviation of the percent change in energy prices. We will refer to this
model as the large percent change or LC specification. A third possibility is that consumers and
firms respond only to changes in energy prices that are unprecedented in recent history. This view
calls for a measure of the net energy price increase of the type proposed by Hamilton (1996, 2003).
While Hamilton focuses on net energy price increases, Edelstein and Kilian (2007a,b) extend this
measure to include net energy price declines that are constructed in a similar fashion. They document
statistically significant responses of U.S. macroeconomic aggregates to both net increases and net
decreases. The resulting model will be refe
ed to as the net percent change or NC specification.
Figure 1 illustrates the differences between these three specifications for U.S. retail energy prices.
The example in Figure 1 (as well as all subsequent empirical analysis in this paper) is based on the
eal price of energy, which is the price relevant for resource allocation.5
An important question is how to choose between these alternative behavioral models. Edelstein
5For further discussion of the distinction between nominal and real energy prices see, e.g., Hamilton (2005) and
Kilian (2008a).
6
and Kilian (2007a,b) show that the C specification cannot be rejected in favor of the LC specification
in general. Whether the C orNC specification is more appropriate remains an active area of research.
Since the NC specification is not nested in the C specification, it is not straightforward to test these
specifications, or even to compare the magnitudes of the responses implied by the two specifications.
In this paper, I will focus on the C specification, but note that the qualitative results would be
similar under the NC specification, and indeed in most cases for all three specifications.
2.5 Alternative VAR Frameworks for Modeling Energy Price Shocks
Models that treat innovations to energy prices as predetermined with respect to macroeconomic
aggregates at high frequency have been used in the literature for many years (see, e.g., Rotemberg
and Woodford 1996; Davis and Haltiwanger 2001; Lee and Ni 2002; Leduc and Sill 2004; Blanchard
and Galí 2007). The assumption of predetermined energy prices rules out instantaneous feedback
from U.S. macroeconomic aggregates to energy prices, but allows energy prices to respond to all
past information. This assumption permits the consistent estimation of the expected response of
eal U.S. macroeconomic aggregates to an innovation in energy prices. In conjunction with the
assumption that there are no other exogenous events that are co
elated with the exogenous energy
price innovation, these impulse responses can be interpreted as the causal effect of the energy price
innovation (see Cooley and LeRoy 1985).
VAR models based on the assumption of predetermined energy prices are not without limitations,
however, even if the feedback from domestic macroeconomic shocks to energy prices is negligible in
the short run. Notably, these models do not distinguish between energy price innovations driven
y supply shocks and demand shocks in energy markets. The latter distinction can be crucial
ecause different demand or supply shocks in energy markets tend to elicit different responses of
macroeconomic aggregates. In section 5, I will discuss an alternative VAR approach that makes this
distinction apparent. Impulse response estimates derived under the assumption of predetermined
energy prices can still be interpreted as an estimate of the economic effects of an average energy
price shock during the sample period in question. While these response estimates are asymptotically
valid, in small samples they may be sensitive to changes in the composition of the underlying demand
and supply shocks. Moreover, they are potentially misleading when it comes to the interpretation
of specific energy price shock episodes. Finally, such responses also obscure the fact that energy
price shocks driven by demand shocks may proxy for a number of other changes in the economy
associated with the underlying demand shocks. This point will be discussed in more detail in section
5. Nevertheless, estimates of average responses to energy price shocks provide a useful benchmark
7
and allow us to test the implications of various economic models of the transmission of energy price
shocks.
A convenient vehicle for assessing the average economic effects of energy price shocks on a given
macroeconomic aggregate is a recursively identified bivariate vector autoregression (VAR), in which
the percent change in real energy prices is ordered first and the macroeconomic aggregate of interest
is ordered second. For example, we may assess the response of real consumption to a real energy
price innovation by specifying a model in the percent change of the real price of energy and the
percent growth in real consumption. Generalizations to other specifications of energy price shocks
are straightforward. In section 3.1 and 3.2, I use this bivariate workhorse model to quantify in detail
the responses of consumption and investment expenditures to real energy price shocks. These results
will be used in assessing the empirical support for the main channels of transmission discussed in
the literature.
There is no loss of generality from restricting ourselves to a bivariate model under the maintained
assumption of predetermined energy price innovations, if we are only interested in consistently
estimating the effects of energy price innovations on macroeconomic aggregates. If, in addition,
we want to assess the precise nature of the transmission, a bivariate model will not suffice. Fo
example, it is common in the literature to estimate larger-dimensional VAR models of the effects of
unanticipated oil price shocks that include a monetary policy-reaction function (see, e.g., Bernanke,
Gertler and Watson 1997, 2004; Balke, Brown and Yücel 2002; Lee and Ni 2002; Hamilton and
He
era 2004; He
era and Pesavento 2007). Such VAR models can be useful in assessing the extent
to which the overall response of macroeconomic aggregates to unanticipated energy price changes
is driven by the response of the central bank to the change in energy prices. Since these larger-
dimensional structural VAR models require additional identifying assumptions and since they are
typically less precisely estimated, there is no reason to depart from the baseline bivariate VAR model
for the purpose of the analysis in sections 3.1 and 3.2. Models of the reaction of monetary policy to
oil price shocks will be discussed in section 3.4.
3 The Effect of Energy Price Shocks on the Economy
The standard approach to modeling energy price shocks has been to focus on the effects of an
exogenous increase in the price of imported crude oil. Such terms-of-trade shocks traditionally have
een thought to matter for the domestic economy through their effects on production decisions (see,
e.g., Kim and Loungani 1992; Backus and Crucini 2000). In this view, oil is treated as an intermediate
input in domestic production. How imported oil enters the production function for domestic value
8
added is one of the most studied and least resolved issues in empirical macroeconomics (Backus and
Crucini 2000).
There are well-known problems in explaining a decline in real GDP based on this intermedi-
ate input cost or supply channel. The first problem is that the interpretation of crude oil as an
intermediate input in the value added production function is questionable if we think of oil as an
imported commodity. Under standard assumptions, imported oil enters the production function of
domestic gross output, but it does not enter the production function of domestic value added (see,
e.g., Rotemberg and Woodford 1996). Since gross output is separable in value added and imported
energy, holding capital and labor fixed, oil price shocks do not move value added. Hence, oil price
shocks by definition cannot be interpreted as productivity shocks for real GDP (see Barsky and
Kilian 2004). The second problem is that, to the extent that oil prices affect domestic output, unde
standard assumptions their impact should be bounded by the cost share of oil in gross domestic
production, which is known to be very small. Indeed standard production-based models of the
transmission of energy price shocks are not capable of explaining large fluctuations in real output.
There are three proposals in the literature for dealing with this problem. All three involve
modifications of the baseline supply-shock model to generate quantitatively important effects of oil
price shocks on real GDP. The first proposal is Rotemberg and Woodford’s (1996) model which
elies on large and time-varying markups to generate large effects of oil price shocks on real GDP.
The second proposal is Atkeson and Kehoe’s (1999) putty-clay model which appeals to capital-
energy complementarities in production. The third proposal is due to Finn (2000). Finn establishes
that in a perfectly competitive model, in which energy is essential to obtaining a service flow from
capital, there may be a large effect of oil price shocks on real GDP. In all three models, the supply
channel of the transmission of energy price shocks may be quantitatively important, yet there is
no consensus which, if any, of these models has empirical support. For example, it remains to be
shown that mark-ups in the U.S. economy are as large and as time-varying as required for the
Rotemberg and Woodford model to generate large effects on value added. Likewise, it remains to
e shown that changes in capacity utilization in response to oil price shocks are indeed as important
and pervasive in the real world as they are in Finn’s model. Similarly, the microeconomic evidence
on the existence and quantitative importance of capital-energy complementarities is mixed at best.
A second unresolved issue is whether these models can account for a large share of business cycle
fluctuations in real GDP. A third issue is that all three models postulate that oil prices follow
an exogenous stochastic process, an assumption that is at odds with both the data and standard
economic models of the oil market, as discussed in section 2.
These caveats are important because in the absence of an empirically supported model of the
9
supply channel, there is no reason to expect global oil price shocks to exert major effects on the
domestic economy. In part in response to these challenges, another
anch of the literature has
developed that focuses on the reduction in the demand for goods and services triggered by energy
price shocks rather than treating energy price shocks as aggregate supply shocks for the U.S. economy
(or as productivity shocks for U.S. domestic production). In this alternative view, the primary
channel of transmission is on the demand side of the economy. For example, in a recent survey on
the effects of energy price shocks, Hamilton (2008) stresses that a key mechanism whereby energy
price shocks affect the economy is through a disruption in consumers’ and firms’ spending on goods
and services other than energy. This view is consistent with anecdotal evidence of how oil price
shocks affect U.S. industries. Most U.S. firms perceive energy price shocks as shocks to the demand
for their products rather than shocks to the cost of producing these products (see Lee and Ni 2002).
This view is also shared by many policymakers. There is a widespread perception that an increase
in energy prices slows economic growth primarily through its effects on consumer spending (see, e.g.,
Bernanke 2006a). In the remainder of section 3, I outline the economic rationale for the demand
channel of transmission and assess its empirical support. I focus on the response of real consumption
first, before considering real investment expenditures in section 3.2.
3.1 How Do Consumer Expenditures Respond to Higher Energy Prices?
3.1.1 The Channels of Transmission
The literature has focused on four complementary mechanisms by which consumption expenditures
may be directly affected by energy price changes. First, higher energy prices are expected to reduce
discretionary income, as consumers have less money to spend after paying their energy bills.6 All
else equal, this discretionary income effect will be the larger, the less elastic the demand for energy,
ut even with perfectly inelastic energy demand the magnitude of the effect of a unit change in
energy prices is bounded by the energy share in consumption. Second, changing energy prices may
create uncertainty about the future path of the price of energy, causing consumers to postpone
i
eversible purchases of consumer durables (see Bernanke 1983, Pindyck 1991). Unlike the first
effect, this uncertainty effect is limited to consumer durables.7 Third, even when purchase decisions
6 Implicit in this view is the assertion that higher energy prices are primarily driven by higher prices for imported
energy goods, and that at least some of the discretionary income lost from higher prices of imported energy goods is
transfered a
oad and is not recycled in the form of higher U.S. exports. In the case of a purely domestic energy price
shocks (such as a shock to U.S. refining capacity), it is less obvious that there is an effect on aggregate discretionary
income. First, the transfer of income to the refiner may be partially returned to the same consumers in the form
of higher wages or higher stock returns on domestic energy companies. Second, even if the transfer is not returned,
higher energy prices simply constitute an income transfer from one consumer to another that cancels in the aggregate.
7Alternatively, one might expect durables consumption to fall in response to a positive energy price shock, as
consumers wait for more energy-efficient technologies to become available. As shown in section 3.1.4 that explanation
10
are reversible, consumption may fall in response to energy price shocks, as consumers increase thei
precautionary savings. This response may arise if consumers smooth their consumption because
they perceive a greater likelihood of future unemployment and hence future income losses. By
construction, this effect will embody general equili
ium effects on emplyment and real income
otherwise ignored by the demand channel of transmission. In addition, the precautionary savings
effect may also reflect greater uncertainty about the prospects of remaining gainfully employed, in
which case any unexpected change in energy prices would lower consumption. Finally, consumption
of durables that are complementary in use with energy (in that their operation requires energy)
will tend to decline even more, as households delay or forego purchases of energy-using durables.
This operating cost effect is more limited in scope than the uncertainty effect in that it only affects
specific consumer durables. It should be most pronounced for motor vehicles (see Hamilton 1988).8
These four direct effects have in common that they imply a reduction in aggregate demand in
esponse to unanticipated energy price increases. In addition, there may be indirect effects related
to the changing patterns of consumption expenditures. A large literature has stressed that shifts in
expenditure patterns driven by the uncertainty effect and operating cost effect amount to allocative
distu
ances that are likely to cause sectoral shifts throughout the economy (see, e.g., Davis (1987)
and Hamilton (2008) for a review). For example, it has been argued that reduced expenditures on
energy-intensive durables such as automobiles may cause the reallocation of capital and labor away
from the automobile sector. As the dollar value of such purchases may be large relative to the value
of the energy they use, even relatively small changes in energy prices (and hence in the purchasing
power of consumers) can have large effects on output and employment (see Hamilton 1988). A
similar reallocation may occur within the same sector, as consumers switch toward more energy-
efficient durables (see Hamilton 1988; Bresnahan and Ramey 1993). In the presence of frictions in
capital and labor markets, these intersectoral and intrasectoral reallocations will cause resources to
e unemployed, thus causing further cutbacks in consumption and amplifying the effect of highe
energy prices on the real economy. This indirect effect could be much larger than the direct effects
listed earlier, and is considered by many economists to be the primary channel through which energy
price shocks affect the economy (see, e.g., Davis and Haltiwanger (2001) and Lee and Ni (2002) and
the references therein). Concerns over reallocation effects also help explain the preoccupation of
policy makers with the effects of energy price shocks on the automobile sector (see, e.g., Bernanke
2006b).
is unlikely to be empirically relevant.
8This last effect need not involve a reduction in the number of automobiles sold. It can also take the form
of consumers substituting small energy-inefficient automobiles for large energy-inefficient automobiles. If the latte
automobiles tend to be lower priced, aggregate real consumption of automobiles may fall, even when the number of
automobiles sold does not (see Bresnahan and Ramey 1993).
11
Unlike the discretionary income effect, the uncertainty effect and the reallocation effect neces-
sarily generate asymmetric responses of macroeconomic aggregates to unanticipated energy price
increases and decreases. The asymmetry arises because these effects amplify the response of macro-
economic aggregates to energy price increases, but reduce the co
esponding response to falling
energy prices. In sections 3.1 and 3.2, I will deliberately abstract from the possible presence of such
asymmetric effects. As will be discussed in section 3.3, there is no compelling statistical evidence
against the symmetry hypothesis, and the symmetric model appears to fit the real consumption
data rather well. The historical reasons for the prominence of asymmetric models in empirical and
in theoretical work on oil price shocks, and the reasons why the apparent evidence of asymmetries
is likely to be spurious, are discussed in section 3.3.
3.1.2 Estimating the Energy-Price Elasticities of Energy Demand
A central question in the literature is how price-elastic the demand for energy is. Such estimates
play an important role in assessing the potential impact of a ca
on tax, for example, in assessing
alternative regulatory policies, and in understanding the transmission of energy price shocks. The
esponse of real consumption may be estimated by applying the bivariate regression model described
in section 2 to various forms of energy consumption. All energy prices have been weighted by the
nominal expenditure share on energy to obtain a measure of the gains and losses of households’
purchasing power associated with energy price fluctuations (see Edelstein and Kilian 2007a). The
sample period is 1970.2-2006.7. The results are expressed as elasticities with respect to the price of
energy, evaluated at the average energy share. The upper panel of Table 2 shows that consumption
of all forms of energy declines in response to energy price increases. The elasticity estimate for total
energy consumption is -0.45 with e
or bounds of -0.27 and -0.66, but there are important differences
across different forms of energy.
The strongest responses are observed for gasoline and for heating oil and coal. Contrary to
the conventional wisdom, gasoline consumption responds immediately to unanticipated energy price
increases reaching an elasticity of -0.48 after one year. The strikingly large response of -1.47 fo
heating oil and coal is likely due to households’ ability to store heating oil in tanks. This storage
feature allows households to delay purchases of new heating oil when the price of heating oil is high
and to fill the tank completely when prices are low.9 In contrast, electricity and natural gas are
inherently unstorable, and gasoline may not be stored for safety reasons beyond the tank capacity
of a car. Indeed, the declines in electricity consumption and in natural gas use are smaller and not
statistically significant.
9For a discussion of this storable goods feature see Dudine, Hendel, and Lizzeri (2006).
12
It is useful to put these estimates into perspective by comparing them with estimates based
on other methodologies. Using a structural model, Reiss and White (2005) a
ive at an estimate
of the short-run electricity price elasticity of electricity demand of -0.39. The point estimate in
Table 2 is only -0.15 after twelve months, but the 95% confidence interval for the elasticity estimate
includes -0.39. Dahl and Sterner (1991), in a comprehensive survey, report estimates of the short-run
gasoline price elasticity of gasoline demand between -0.08 and -0.41. It is not clear, however, how
well identified their estimates are. Our point estimate of -0.48 is larger with 95% e
or bands of -0.32
and -0.69, respectively.10 These estimates suggest that the demand for energy is not as unresponsive
to energy prices as is sometimes believed. Further work at a more disaggregate level is likely to
prove useful in refining these estimates.
3.1.3 Estimating the Energy-Price Elasticities of Non-Energy Consumption
The lower panel of Table 2 summarizes the co
esponding elasticities of major non-energy real
consumption aggregates. All estimates have the expected negative sign and most are statistically
significant. Table 2 demonstrates that the overall elasticity of -0.15 is driven mainly by a reduction in
vehicles purchases. The elasticity of demand for vehicles is -0.84. It can be shown that this estimate
in turn primarily reflects reduced consumer demand for cars. There is much weaker evidence of
educed demand for other durables.
It is useful to put these results in perspective. The sharp rise in gasoline prices in recent years
has renewed interest in the question of how much higher energy prices affect consumer expenditures.
Our analysis allows us to assess the overall effect...