Lags, Costs, and Shocks:

An Equilibrium Model of the Oil Industry∗

Gideon Bornstein†, Per Krusell‡ and Sergio Rebelo§
May 2017

Abstract

We use a new micro data set to compile some key facts about the oil market
and estimate a structural industry equilibrium model that is consistent with these
facts. We find that demand and supply shocks contribute equally to the volatility
of oil prices but that the volatility of investment by oil firms is driven mostly by
demand shocks. Our model predicts that the advent of fracking will eventually
result in a large reduction in oil price volatility.

J.E.L. Classification: Q4
Keywords: oil, commodities, volatility.

∗
We thank Wei Cui, Ravi Jaganathan, Ryan Kellogg, Lutz Kilian, Valerie Ramey, and Rob Vig-
fusson for their comments.
†
Northwestern University.
‡
Institute for International Economic Studies, NBER, and CEPR.
§
Northwestern University, NBER, and CEPR.
1 Introduction
The interaction between oil markets and the world economy has been a subject of long-
standing interest in macroeconomics. In this paper, we study this interaction using a
stochastic equilibrium model of the oil industry. Our analysis relies heavily on a new
micro data set that contains information on production, reserves, operational costs, and
investment for all oil fields in the world.
We use these data to guide the construction of our model in three ways. First, we
compile a set of facts about oil markets that a successful model should be consistent
with. Second, we produce micro estimates of two key model parameters: the average lag
between investment and production and the elasticity of extraction costs with respect to
production. Third, we use the generalized method of moments (GMM) to estimate the
remaining model parameters, targeting a set of second moments for oil-related variables.
There is substantial heterogeneity across oil firms along various dimensions. We find
two sources of heterogeneity that are particularly important. The first is the different
behavior of firms that are part of the Organization of the Petroleum Exporting Coun-
tries (OPEC) and those that are not. The second is the difference between hydraulic
fracturing (fracking) and conventional oil production. Our benchmark model features
heterogeneity only in the OPEC/non-OPEC dimension. Our extended model includes
firms that use conventional oil production methods as well as fracking.
We use our model to measure the importance of demand and supply shocks in
driving prices, production, and investment. We find that supply and demand shocks
contribute equally to the volatility of oil prices but that investment in the oil industry is
driven mostly by demand shocks. The reason for this pattern is twofold. There is a long
lag between investment and production and supply shocks are short-lived relative to
demand shocks. So, investment responds much more to demand than to supply shocks.
We also find that the volatility of OPEC production firms is driven primarily by supply
shocks that disrupt the ability of these firms to extract oil. In contrast, both demand

1
and supply shocks are important in explaining the volatility of non-OPEC production.
As discussed above, our data allows us to document two key differences between
hydraulic fracturing (fracking) and conventional oil production. First, it is less costly
for fracking firms to adjust their level of production in the short run, so these firms
are more responsive to changes in prices. Second, the lag between investment and
production in much shorter in fracking operations than in conventional oil production.
We introduce fracking firms into our model to study their impact on the dynamics
of the oil market. We find that their presence leads to a large decline in the volatility
of oil prices. The reason is simple: these firms are more nimble in adjusting production
levels from existing fields and in starting production in new fields, so they can respond
more quickly to price increases.
In Section 2 we review the following six facts about the oil market. First, the
average rate of change of inflation-adjusted oil prices is not significantly different from
zero. Second, oil prices are very volatile, much more volatile than returns to the stock
market or exchange rates. Third, investment in the oil industry is very volatile, roughly
seven times more volatile than U.S. aggregate investment. Fourth, investment in the
oil industry shows a strong, positive correlation with the inflation-adjusted price of
oil. Fifth, there is a low short-run elasticity of the supply of oil from individual oil
fields with respect to price. Sixth, production by OPEC firms is more volatile and less
correlated with oil prices than production by non-OPEC production firms.
We describe our model in Section 3. Section 4 is devoted to estimating model
parameters using both micro data and moments of key aggregate variables for the oil
industry. The estimated version of our model has four main features. First, demand is
relatively inelastic. Second, supply is elastic in the long run because firms can invest
in the discovery of new oil fields.1 Third, supply is inelastic in the short run. This
1
While the amount of oil is ultimately finite, we can think about this investment process as including
new ways of extracting oil as well as the development of oil substitutes, as in Adao et al. (2012). There
has been a large expansion of oil reserves during our sample period. According to the U.S. Energy
Information Administration, proved oil reserves measured in years of production have increased from

2
property results from several model features: there is a lag between investment and
production, there are convex costs of adjusting extraction rates, and there are point-
in-time decreasing returns to oil investment. Fourth, there are shocks to demand, e.g.,
faster growth in China, and to the supply of OPEC firms, e.g., the Iran-Iraq war.
This paper is most closely related to work on models of the response of oil prices
and production to fluctuations in the world economy. Examples of this work include
Backus and Crucini (2000), Kilian (2009), Bodenstein et al. (2011), and Lippi and No-
bili (2012).2 Our paper is also related to the literature that uses VARs to estimates the
importance of demand and supply shocks on the volatility of oil prices (see the com-
prehensive surveys by Hamilton (2005) and Kilian (2008) and the references therein).
Our approach, which is based on a structural model, is complementary to that adopted
in this literature. Finally, our paper is related to a new, emerging literature that uses
micro data to shed new light on key aspects of the oil industry (see, e.g., Kellogg (2014),
Arezki et al. (2016), and Anderson et al. (2017)).

2 Key facts about the oil market

In this section we review six facts about the oil market. Most of our analysis relies on a
proprietary data set compiled by Rystad Energy that includes information on reserves,
production, investment, and operational costs for the universe of oil fields. The data
contains information about roughly 14,000 oil fields operated by 3,200 companies.
We focus our analysis on the period from 1970 to 2015. Until 1972, U.S. regulatory
agencies sought to keep U.S. oil prices stable by setting production targets. This period
of stability ended once the U.S. ceased to be a net oil exporter (see Hamilton (1983)
and Kilian (2014)).
roughly 30 years in 1980 to 52 years in 2015.
2
Earlier work on the impact of oil shocks on the economy generally treats oil prices as exogenous
(see, e.g., Kim and Loungani (1992), Rotemberg and Woodford (1996), and Finn (2000)).

3
 Figure 1: Real price of oil

180

160

Real  Oil  Price  (2010  =  100) 140

120

100

80

60

40

20

0
1900 1920 1940 1960 1980 2000

There is no significant time trend in real oil prices Figure 1 plots the price of
an oil barrel expressed in dollars and deflated by the U.S. consumer price index from
1900 to 2015.3 The average annual growth rate in the real price of oil during this period
is not statistically different from zero. For the period from 1900 to 2015 this average
is 0.01 with a standard error of 0.21.4 For the period from 1970 to 2015 this average is
0.02 with a standard error of 0.27.

Oil prices are very volatile Figure 1 shows that oil prices are quite volatile since
the early 1970s. From 1970 to 2015, the volatility of oil prices is higher than that of
returns to the stock market or exchange rates. The standard deviation of the annual
percentage change in oil prices is 0.28 for nominal prices and 0.27 for real prices. In
3
Between 1900 and 1947, our source for the price of oil is Harvey et al. (2010). After 1947, we use
the price of West Texas Intermediate.
4
The property that average growth rates in real prices estimated over long time periods are close
to zero is shared by many other commodities (Deaton and Laroque (1992), Harvey et al. (2010) and
Chari and Christiano (2014)).

4
contrast, the standard deviation of nominal returns to the S&P 500 is 16 percent and
the standard deviation of changes in exchange rates is 10 percent.
These first two facts were aptly summarized by Deaton (1999) with the phrase
“What commodity prices lack in trend, they make up for in variance.”

Investment in the oil industry is very volatile The annual standard deviation
of the growth rate of real world investment in the oil industry is 0.36 for the period
1970-2015. To put this number in perspective, this measure of volatility is 0.10 for U.S.
manufacturing and 0.07 for U.S. aggregate investment.5
Figure 2 illustrates the high volatility of investment in the oil industry. This figure
plots the linearly-detrended logarithm of two series: real word-wide investment in the
oil industry and real aggregate investment in the U.S.

Figure 2: Detrended investment in the oil industry

2.5
Investment   in   oil   industry
2
U.S.  aggregate  investment
Detrended  log   investment  

1.5

0.5

-­‐‑0.5

-­‐‑1

-­‐‑1.5
1970 1980 1990 2000 2010

5
The only major U.S. manufacturing sector with volatility of investment similar to the oil industry
in the period 1970-2015 is Motor vehicle manufacturing, a sector that has struggled to compete with
foreign manufacturers and had to be bailed out by the Federal government in 2009.

5
Investment in the oil industry is positively correlated with oil prices Figure
3 plots the logarithm of the inflation-adjusted price of oil and the logarithm of inflation-
adjusted world investment in the oil industry. The correlation between the growth rate
of the price of oil and the growth rate of investment is 0.51.

Figure 3: Oil prices and oil investment

3.50
Oil  Price
3.00
Oil  Investment
Log  real  value  (2000  normalized  to  0)

2.50

2.00

1.50

1.00

0.50

0.00

-­‐0.50

-­‐1.00

-­‐1.50
1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

Table 1 reports the correlation between the growth rate of real investment and the
growth rate of the real oil prices for each of the top twenty firms in the oil industry in
descending order.6 We see that the correlation is high for every major oil firm with the
exception of Iran’s NIOC, Total, and Eni, which show a correlation near zero.

The short-run elasticity of oil supply is very low Oil producers can respond to
an increase in the market price of oil in two ways. The first is to produce more oil from
oil fields in operation by increasing the extraction rate. The second is to increase the
number of oil fields in operation. We show that the short-run elasticity of the extraction
6
Firms are ranked according to their total oil production in 2015.

6
 Table 1: Investment and price correlation for top 20 firms
Firm Headquarters OPEC corr(∆i, ∆p)
Saudi Aramco Saudi Arabia 3 0.31
Rosneft Russia 7 0.34
PetroChina China 7 0.36
Kuwait Petroleum Corp (KPC) Kuwait 3 0.3
NIOC (Iran) Iran 3 0.06
Pemex Mexico 7 0.27
ExxonMobil United States 7 0.35
Lukoil Russia 7 0.41
Petrobras Brazil 7 0.3
PDVSA Venezuela 3 0.28
Abu Dhabi NOC Abu Dhabi 3 0.14
Chevron United States 7 0.43
Shell Netherlands 7 0.34
BP United Kingdom 7 0.35
Surgutneftegas Russia 7 0.26
South Oil Company (Iraq NOC) Iraq 3 0.19
Total France 7 0.04
CNOOC China 7 0.4
Statoil Norway 7 0.34
Eni Italy 7 0.04

7
rate to an exogenous change in the price of oil is positive but small.7 We also show
that the elasticity of response of the number of oil fields in operation to an exogenous
change in the price of oil is not statistically different from zero.
Table 2 reports panel-data estimates of the elasticity of the extraction rate (the ratio
of production to reserves) for a given oil field with respect to real oil prices.8 These
estimates suggest that a rise in oil prices leads to a very small increase in the supply of
oil from a given oil field.9
Our estimates are obtained by running various versions of the following regression:

ln θit = αi + β ln pt + γXit + εit ,

where θit denotes the extraction rate of oil field i at time t, pt is the real price of oil,
and Xit represents other controls.10 These controls include a time trend, an oil-field
fixed effect, and a fixed effect for year of operation to control for the life cycle of an oil
field.11
Specification 1 in Table 2 is a simple OLS regression. The estimated slope coeffi-
cient that results from this regression can be biased if there is technical progress that
lowers the cost of extraction, raising θit , increasing the supply of oil, and lowering pt .
To address this problem, we instrument the price of oil with our forecast of the cycli-
cal component of world GDP. This forecast is correlated with aggregate demand and
unaffected by aggregate cost shocks.
7
Anderson et al. (2017) estimate this elasticity to be close to zero. The difference between our
results and theirs is likely to reflect differences in data frequency: their data is monthly while ours is
annual.
8
In our data, reserves are proven reserves, which measure the total amount of oil that can be
produced from a given field. Reserves do not change in response to changes in oil prices, so there is
no mechanical impact of oil prices on extraction rates.
9
An oil field generally contains many oil rigs. Production increases can result from the intensive
margin (higher production from existing oil rigs) or from the extensive margin (drilling new oil rigs).
Anderson et al. (2017) use a sample of Texas oil rigs to show that the elasticity of the intensive margin
is close to zero, so production increases result from the extensive margin.
10
Our data includes all oil fields with a positive extraction rate for the period 1971-2015 excluding
the last year of operation.
11
See Arezki et al. (2016) and Anderson et al. (2017) for discussions of this life cycle.

8
 Table 2: Price elasticity of extraction rates

Variable (1) (2) (3) (4) (5)

ln(price) 0.09∗∗∗ 0.12∗∗∗ 0.16∗∗∗ 0.17∗∗∗ 0.16∗∗∗
(0.009) (0.012) (0.017) (0.023) (0.013)

ln(price) × 1OPEC −0.2∗∗∗ −0.19∗∗∗

(0.069) (0.069)

ln(price) × 1Big Firm −0.06

(0.054)

ln(price) × 1|∆ ln(p)|>0.1 −0.17∗∗∗

(0.009)

Oil field FE 3 3 3 3 3
Operation year FE 3 3 3 3 3
Year trend 3 3 3 3 3
IV 7 3 3 3 3
Clusters (oil fields) 12,187 11,479
Observations 173,742 173,034

Oil fields with positive extraction in 1971-2015, excluding last year of operation. Standard
errors in parenthesis. Instrument for price using the forecast of the cyclical component
of world GDP. F-stat > 1000 in (2)-(5). (∗ ∗ ∗) - significant at a 1% level.

Our forecast of the cyclical component of world GDP is obtained as follows. We

HP-filter the series for world real GDP and estimate an AR process for the resulting
cyclical component. Choosing the number of lags using the Akaike information criterion
resulted in an AR(2) process.
Specifications 2-5 use this instrument. Our benchmark specification is regression
2 which yields an estimate for β equal to 0.12. The following calculation is useful for
evaluating the magnitude of this elasticity. The average extraction rate in our sample
is 2.8 percent. A one standard deviation (13.5 percent) increase in price raises the
extraction rate from 2.8 percent to 2.85 percent, resulting only in a 1.8 percent increase
in production.

9
 Specification 3 includes the product of the logarithm of the price and an OPEC
dummy. We see that OPEC firms respond less to changes in prices than non-OPEC
firms. The point estimate for the response of OPEC to oil price shocks is negative
(−0.04) and insignificant (P-value = 0.48).
Specification 4 includes also an interaction term that is the product of logarithm of
price and a dummy for firm size. This dummy is equal to one for firms that produced
more than 0.5 billion barrels of oil in 2015. Total production in 2015 was approximately
27.5 billion barrels of oil so that each large firm according to our definition has at least
a 1.8 percent market share. The idea is to investigate whether large firms behave
differently from small firms. We find no evidence of a firm-size effect.
Specification 5 includes an interaction term that is the product of logarithm of price
and a dummy for price changes that are larger than 10 percent in absolute value. The
idea is to investigate whether firms react more to large oil price changes than to small
price changes. We find that the coefficient on the interaction term is negative (−0.17)
and statistically significant. This finding is consistent with the presence of convex
adjustment costs in the extraction rate, so that the elasticity of response is higher for
small price changes than for large price changes.
These results remain robust when we extend the sample to start in 1900 (see Table
15 in the appendix) with two exceptions. In the long sample, the coefficients on large
firms and large price movements are both negative and statistically significant.
Table 3 reports our time-series estimates of the elasticity of the number of oil fields
in operation with respect to real oil prices. Specification 1 is a simple OLS regression
where the dependent variable is the logarithm of oil fields in operation world wide and
the independent variable is the logarithm of real oil prices. Specification 2 uses our
forecast of the cyclical component of world GDP as an instrument for the logarithm
of real oil prices. Specifications three and four report results for non-OPEC fields and
OPEC fields, respectively. All four specifications yield elasticity estimates that are
statistically insignificant. We also find an insignificant elasticity when we extend our

10
 Table 3: Price elasticity of oil fields in operation

Variable (1) (2) (3) (4)

ln(price) −0.05 −0.02 −0.01 0
(0.06) (0.08) (0.03) (0.06)

Year trend 3 3 3 3
IV 7 3 3 3
Dep. variable All fields All fields Non-OPEC fields OPEC fields
Observations 45 45 45 45

Newey-West standard errors with 5 years lag in parenthesis.

sample to start in 1900 (see Table 14 in the appendix).

Taken together, these results suggest that the number of oil fields in operation does
not respond in the short-run to changes in oil prices.

OPEC and non-OPEC firms behave differently Table 4 shows that OPEC and
non-OPEC firms differ in the volatility and persistence of production and investment,
as well as in the correlation of these variables with real oil prices. The production
of OPEC firms is more volatile and less persistent than that of non-OPEC firms. In
addition, the correlation between investment and prices is higher for non-OPEC firms
than for OPEC firms. These patterns are likely to result from supply shock to OPEC
firms, such as the disruptions in oil markets associated with the Iranian revolution and
the Iran-Iraq war.

11
 Table 4: Correlation table of OPEC and non-OPEC production and investment

∆p ∆in ∆io ∆qn ∆qo std Acorr

∆p 1.00 0.27 -0.03
∆in 0.56 1.00 0.19 0.12†
∆io 0.36 0.67 1.00 0.19 0.31
∆qn 0.03† 0.09† −0.03† 1.00 0.02 0.64
∆qo 0.03† 0.02† −0.23† -0.14† 1.00 0.07 0.21†
† - not significant at a 5% level. See table 7 for standard errors.

3 An industry-equilibrium model
In this section we describe our equilibrium model of the oil industry. Our goal is to
propose a parsimonious model that is consistent with the facts described in Section 2.
We assume that the world demand for oil is given by

−1/ε
Pt = exp(dt )Qt ,

where Pt is the real price of oil, Qt is the quantity consumed, and ε is the price elasticity
of the demand for oil. The variable dt is a stochastic demand shock that follows an
AR(2) process in logarithms

ln (dt ) = ρd1 ln dt−1 + ρd2 ln dt−2 + edt .

We choose this AR(2) specification so that demand shocks can follow a hump-shaped
pattern. This pattern allows an initial shock to contain news about a future rise in the
demand for oil associated, for example, with faster growth in China.

Non-OPEC firms There is a continuum of measure one of non-OPEC firms. These

firms maximize their value (V N ) which is given by
∞
N
X
η
β t Pt θtN KtN − ItN − ψ θtN KtN .
 
V = E0 (1)
t=0

12
Here, ItN denotes investment, θtN the extraction rate (the ratio of production to re-

η
serves), and KtN oil reserves. The term ψ θtN KtN represents the costs of extracting
oil. We assume extraction costs are linear in reserves so that aggregate production
and aggregate extraction costs are invariant to the distribution of oil reserves across
firms. This formulation allows us to use a representative firm to study production and
investment decisions. We assume that η > 1, so that these costs are convex, and that
the time t + 1 extraction rate is chosen at time t.
We adopt a parsimonious way of modeling lags in investment by introducing oil
in process, which we denote by Xt . The timing of the realization of shocks and firm
decisions is as follows. In the beginning of the period, the demand and supply shocks
are realized, a fraction λ of the oil in process materializes into new oil reserves, and
production occurs according to the predetermined extraction rate. At the end of the
period, the firm chooses its investment and its extraction rate for the next period. The
law of motion for oil in process is as follows:
N

α
1−α
Xt+1 = (1 − λ)XtN + ItN LN . (2)

Investment adds to the existing oil in process, but only a fraction λ of the oil in
process materializes into oil reserves in every period. Investment requires land (LN )
and exhibits point-in-time decreasing returns (α < 1). Without this feature, investment
would be extremely volatile, rising sharply when prices are high and falling deeply when
prices are low.
One interpretation of equation (2) is as follows. Suppose each firm searches for oil
on a continuum of oil fields containing XtN barrels of oil uniformly distributed across
fields. The probability of finding oil is independent across oil fields and equal to λ. By
the law of large numbers, each firm finds λXtN oil reserves at time t. We pursue this
interpretation when we estimate λ using our micro data.
Oil reserves evolve as follows:
N
Kt+1 N
= (1 − θtN )KtN + λXt+1 . (3)

13
Reserves fall with oil production (θtN KtN ) and rise as oil in process materializes into
N
new reserves (λXt+1 ).
The notion of oil in process embodied in equations (2) and (3) is a tractable way of
introducing time-to-build in investment that might be useful in other problems. This
formulation allows us to introduce a lag between investment and production by adding
only one state variable. The parameter λ allows us to smoothly vary the length of the
lag.12
The problem of the representative non-OPEC firm is to choose the stochastic se-
quences for ItN , θt+1
N N
, Kt+1 N
, and Xt+1 that maximize its value defined in equation (1),
subject to constraints (2) and (3).
N
The first-order conditions for θt+1 is

N

η−1
Et Pt+1 = ψη θt+1 + Et µN
t+1 , (4)

where µN
t is the Lagrange multiplier corresponding to equation (3). The extraction rate

at time t + 1 is chosen at time t so as to equate the expected oil price to the sum of
N

η−1
the marginal cost of extraction, ψη θt+1 , and the expected value of a barrel of oil
reserves at the end of time t + 1, Et µN
t+1 .

The first-order condition for Kt+1 is

η 
µN N N N
)µN

t = Et β Pt+1 θt+1 − ψ θt+1 + (1 − θt+1 t+1 . (5)

N
For a given value of θt+1 , each additional barrel of oil reserves results in additional
N N

η N
revenue equal to Pt+1 θt+1 and increases extraction costs by ψ θt+1 . A fraction 1−θt+1
of the barrel of reserves remains in the ground and has a value of βEt µN
t+1 .

The first-order condition for Xt+1 is

νtN = λµN N
t + β(1 − λ)Et νt+1 , (6)
12
See Rouwenhorst (1991) for a discussion of the large state space and complex dynamics associated
with time-to-build formulations.

14
where νtN is the Lagrange multiplier corresponding to equation (2). The value of in-
creasing oil in process by one unit, νtN , has two components. A fraction λ materializes
into oil reserves and has a value µN
t . A fraction 1 − λ remains as oil in process and has
N
an expected value βEt νt+1 .
The first order condition for It is

α−1
1−α
1 = α ItN LN νtN . (7)

This equation equates the cost of investment (one unit of output) to the marginal

α−1 N
1−α
product of investment in generating oil in process, α ItN L , evaluated at the
value of oil in process, νtN .

OPEC firms There is a continuum of measure one of OPEC firms. The problem for
these firms is to maximize their value (V O ) which is given by
∞
X
η
V O = E0 β t Pt e−ut θtO KtO − ItO − ψ θtO KtO .


(8)
t=0

The key difference between OPEC and non-OPEC firms is that the former are
subject to a supply shock, ut .13 When this shock occurs, production falls for a given
level of the extraction rate. We assume that supply shocks follow an AR(2) process:

ln ut = ρu1 ln ut−1 + ρu2 ln ut−2 + eut ,

and that innovations to demand (edt ) and supply (eut ) are uncorrelated. The laws of
motion for oil in process and reserves are given by

O

α
1−α
Xt+1 = (1 − λ)XtO + ItO LO , (9)
O
Kt+1 = (1 − e−ut θtO )KtO + λXt+1
O
. (10)
13
To simplify, we abstract from supply shocks to non-OPEC producers. While there are supply
shocks to non-OPEC producers (e.g., Canadian wildfires and Gulf of Mexico hurricanes), they seem
small relative to the supply shocks to OPEC producers. This view is consistent with the higher
volatility of production in OPEC relative to non-OPEC.

15
 The problem of the representative OPEC firm is to choose the stochastic sequences
for ItO , θt+1
O O
, Kt+1 O
, and Xt+1 that maximize its value defined in equation (8), subject to
constraints (9) and (10).
The first-order conditions for the problem of OPEC firms are as follows:

η−1
Et Pt+1 e−ut+1 = ψη θt+1 O
+ Et µO −ut+1



t+1 e ,
η
µO Pt+1 e−ut+1 θt+1
O O
+ (1 − e−ut+1 θt+1
O
)µO


t = Et β − ψ θt+1 t+1 ,

νtO = λµO O
t + β(1 − λ)Et νt+1 ,

α−1 O
1−α P
1 = α ItO L νt .

These first-order conditions are similar to those for non-OPEC firms. The key
difference is that production of OPEC firms is scaled by the supply shock e−ut+1 .

Equilibrium In equilibrium, Pt is a function of demand and supply shocks, the ag-

gregate level of reserves in OPEC (KO N
t ) and non-OPEC (Kt ), and the predetermined

aggregate levels of extraction rates in OPEC (θ O N

t ) and non-OPEC (θ t ):

O N
Pt = p(dt , ut , KO N
t , Kt , θ t , θ t ). (11)

Firms maximize their value subject to the laws of motion for reserves and oil in process.
Each firm takes the law of motion for the aggregate levels of reserves, extraction rates,
and oil in process as given and so the price process is exogenous to each individual
firm. These laws of motion for the aggregate variables depend on the six aggregate
state variables included in equation (11) and the aggregate levels of oil in process in
OPEC (XO N
t ) and non-OPEC (Xt ).

The oil market clears, i.e., total oil production equals total oil demand:

−ut O O
θN N
t Kt + e θ t Kt = Qt .

Since there is a continuum of measure one of identical firms within the two groups,
OPEC and non-OPEC. In equilibrium, the values of aggregate variables for each group

16
coincide with the values of the corresponding variables for the representative firm in
each group.

The Hotelling rule The classic Hotelling (1931) rule emerges as a particular case of
our model in which there are no OPEC firms, λ = 0, and η = 1. When λ = 0, investment
does not result in more oil reserves, so oil is an exhaustible resource. Equation (6)
implies that in this case the value of oil in process is zero: νtN = 0. Combining equations
(4) and (5) we obtain
Et Pt+1 − ψ = βEt (Pt+2 − ψ).

This equation is the Hotelling rule: the price of oil minus the marginal cost of production
is expected to rise at the rate of interest in order to make oil producers indifferent
between extracting oil at t + 1 and at t + 2.
For the general case where λ ≥ 0 and η ≥ 1, the marginal cost of production is
ηψθtη−1 and the difference between the price of oil and the marginal cost of production
is given by

η−1 η−1 η



Et Pt+1 − ηψθt+1 = βEt Pt+2 − ηψθt+2 + βEt (η − 1) ψθt+2 .

η
The term βEt (η − 1) ψθt+2 represents the marginal fall in production costs at time
t + 2 from having an additional barrel of oil reserves. When η = 1, this term is zero
and we recover the Hotelling rule.
When λ = 0 (more oil cannot be found), there is no steady state in which Pt and θt
are constant.14 When the extraction rate is constant, production falls over time and,
since demand is downward sloping, the price of oil rises over time. When the price is
constant, production must also be constant and so the extraction rate must rise.
In our model λ > 0 and η > 1. Because it is feasible to find more oil, there is a
steady state in which both Pt and θt are constant. Oil reserves are constant and so
14
When λ = 0 and η > 1 our economy resembles the model proposed by Anderson et al. (2017).
Changes in the extraction rate in our model play a similar role to drilling new wells in their model.

17
the quantity produced is also constant. In the steady state, the marginal decline in
production costs from an additional barrel of oil is such that the difference between
price and marginal cost remains constant:

β (η − 1) ψθη = (1 − β) P − ηψθη−1 .

4 Model solution and estimation

We solve the model using a second-order approximation around its non-stochastic
steady state. In the non-stochastic steady state the ratio of production by OPEC
and non-OPEC firms is equal to the ratio of the land available to each of the two
groups:
LO θO K O
= .
LN θN K N
We calibrate some key parameters using our micro data set as well as information
about the cost of capital in the oil industry. We estimate the remaining parameters
using GMM.
We choose the ratio LO /LN so that in the steady state the ratio of OPEC to non-
OPEC production is equal to the average of this ratio in our data (0.82). We calibrate
the total amount of land (LO + LN ) so that the steady-state extraction rate coincides
with the average extraction rate in our data (2.8 percent).
The parameter ψ matters only for the level of oil prices, so we normalize it to one.
We set 1/β − 1, the real discount rate, to 8 percent which is the real cost of capital
estimated by Damodaran (2017) for the oil industry.

4.1 Estimating λ and η with micro data

To estimate λ, we compute the lag between the first year of investment and first year
of production (Ti ) for every oil field in our data set. If the arrival of production occurs
according to a Poisson process, the lag between investment and production follows a

18
geometric distribution with mean λ. The maximum likelihood estimator for λ is:
N
λ̂ = PN = 0.085,
i=1 Ti

where N denotes the number of oil fields.

This estimate implies that the average lag between investment and production is 12
years. Figure 4 shows the empirical distribution of this lag together with the implied
geometric distribution for our estimate of λ̂.15

Figure 4: Empirical distribution of lags between investment and production

We also use our micro data to estimate η, the parameter that controls the convexity
of the extraction costs. We run the following regression:
 
C(θit , Kit )
ln = γi + η ln (θit ) + εit ,
Kit
15
Our estimate of the average production lag is higher than that reported in Arezki et al. (2016).
This difference occurs because we estimate the lag between initial investment (which includes seismic
analysis and drilling wells to discover and delineate oil fields) and production. Arezki et al. (2016)
estimate the lag between oil discovery and production, which is shorter.

19
where C(θit , Kit ) denotes extraction costs. The potential presence of cost shocks, either
field specific or aggregate, creates an endogeneity problem. Suppose it becomes more
costly to extract oil, so that firms reduce their extraction rates. This correlation between
the cost and the rate of extraction biases downward in our estimate for η. To address this
problem, we instrument the extraction rate with our forecast of the cyclical component
of world GDP. This forecast is correlated with aggregate demand and unaffected by
field-specific cost shocks.
Our data includes all oil fields with positive extraction rates between 1971 and 2015.
We exclude the last year of oil field operation since the data for this year includes the
costs of shutting down the field, which are not related to the rate of extraction.
Table 5 contains our slope estimates. Specifications 1 through 4 include fixed effects
for oil field and operation year. Specification 1 includes all the oil fields in our sample.
Specification 2 includes only non-OPEC firms. Specification 3 includes only OPEC
firms. While our instrument is independent of oil-field-specific cost shocks, it may be
correlated with aggregate supply shocks. The Iran-Iraq war, for example, may have
caused a slowdown in world GDP at the same time as interrupting the supply of oil
by the two countries. As a result, the point estimate is negative and statistically
insignificant. We use specification 2 as our benchmark and set η equal to the point
estimate (9.3). The fact that η > 1 is consistent with the fact that the elasticity of
response of production to prices is higher for small price increases than for large price
increases (see regression 3 in Table 2).

GMM estimation We estimate , α, the parameters of the AR(2) processes for

demand and supply shocks, together with the variance of the two shocks using GMM.
Table 6 reports our results. Our estimate for  is 0.135 with a standard error of 0.015.
This point estimate implies that demand is very inelastic. A 1 percent increase in
production reduces the price by 7.4 percent.

20
 Table 5: Extraction rate adjustment costs regression

Variable (1) (2) (3)

ln(extraction) 11.49∗∗∗ 9.30∗∗∗ −36.65
(1.60) (1.14) (41.06)

Oil field FE 3 3 3
Operation year FE 3 3 3
Sample All Non-OPEC OPEC
IV 3 3 3
1st stage F-stat 43 53 1
Clusters (oil fields) 11,527 9,969 1,558
Observations 174,339 146,879 27,460

To see why demand needs to be inelastic, it is useful to write the demand function:
1
ln Pt = dt − ln qt .

The model embeds two mechanisms that generate price volatility. The first mechanism
is a low value of  that makes low production volatility consistent with high price
volatility. The second mechanism is volatile demand shocks. Since oil prices are highly
persistent (changes in prices are roughly i.i.d.), demand shocks must also be highly
persistent. Volatile, persistent demand shocks lead to high volatility in investment and
production. The observed volatility of investment and production help determine the
importance of the two channels and therefore identify .
Table 6 shows that the standard errors associated with our parameter estimates are
generally small. The only parameter that is imprecisely estimated is alpha. This impre-
cision results from the small impact of local changes in the value of α on the moments
implied by the model. Our point estimate of α is close to zero. However, it is impor-
tant that α be strictly positive since, when α = 0, the model has the counterfactual
implication that there should be no investment in the oil industry.

21
 Table 6: Estimated parameters

Parameter Estimate (s.e.)

 0.135 (0.015)
α 3 × 10−6 (0.978)
ρd1 1.761 (0.075)
ρd2 -0.775 (0.073)
ρu1 1.428 (0.15)
ρu2 -0.51 (0.11)
var(edt ) 0.02 (0.007)
var(eut ) 0.002 (0.001)

Table 7 compares the estimated moments targeted by our GMM procedure with the
population moments implied by the model. We see that the fit of the model is good
with most of the model population moments inside the 95 percent confidence interval of
data moments. One exception is the correlation between changes in prices and change
in quantities for OPEC firms which is negative in the model and close to zero in the
data.
One interesting property of the model is that it is consistent with the high corre-
lation between prices and investment. In the literature on the cattle and hog cycles
(e.g., Ezekiel (1938) and Nerlove (1958)) this positive correlation is often interpreted
as suggesting that expectations have a backward-looking component. Investment rises
when prices are high, sowing the seeds of a future fall in prices. In our model, the high
correlation between the price of oil and investment results from the rational response of
forward-looking firms. A positive demand shock raises the price of oil above its steady
state level. As a result it is profitable to invest in oil to expand production and take
advantage of the high oil prices. This supply expansion brings the price back to its
steady state level.
Another interesting property is that both in the data and in the model changes in
the price of oil are close to being i.i.d. So, in the short run, the stochastic process for
oil prices is well approximated by a random walk. A related result is that the volatility

22
 Table 7: Data and model moments

Moment Data (s.e.) Model

(1) std(∆pt ) 0.273 (0.028) 0.227
(2) std(∆int ) 0.192 (0.24) 0.213
(3) std(∆iot ) 0.193 (0.27) 0.211
(4) std(∆qtn ) 0.022 (0.003) 0.025
(5) std(∆qto ) 0.069 (0.011) 0.054
(6) corr(∆pt , ∆int ) 0.557 (0.147) 0.709
(7) corr(∆pt , ∆iot ) 0.362 (0.109) 0.657
(8) corr(∆pt , ∆qtn ) 0.031 (0.069) 0.049
(9) corr(∆pt , ∆qto ) 0.030 (0.122) -0.572
(10) corr(∆int , ∆iot ) 0.673 (0.096) 0.997
(11) corr(∆int , ∆qtn ) 0.087 (0.094) -0.001
(12) corr(∆int , ∆qto ) 0.023 (0.112) -0.103
(13) corr(∆iot , ∆qtn ) -0.034 (0.145) 0.003
(14) corr(∆iot , ∆qto ) -0.226 (0.153) -0.036
(15) corr(∆qtn , ∆qto ) -0.141 (0.125) 0.014
(16) corr(∆pt , ∆pt−1 ) -0.027 (0.088) -0.018
(17) corr(∆int , ∆int−1 ) 0.119 (0.135) 0.008
(18) corr(∆iot , ∆iot−1 ) 0.311 (0.096) 0.009
(19) corr(∆qtn , ∆qt−1
n
) 0.643 (0.113) 0.334
o o
(20) corr(∆qt , ∆qt−1 ) 0.213 (0.211) 0.326

23
of one-year futures prices implied by our model coincides with the volatility of one-year
futures in the data, 0.2.16
Figure 5 depicts the impulse response function for a one standard deviation demand
shock. The shock follows a hump-shaped pattern with a peak in year seven. On impact,
firms cannot change their extraction rates so the price increases one to one with the
demand shock. In the following periods, the price of oil increases but the magnitude
of this increase is moderated by a rise in the extraction rate. Production rises and
reserves are depleted. Since the shock is very persistent, investment rises to increase
future reserves and production in order to take advantage of the extended period of
high oil prices.

Figure 5: Impulse response to a demand shock

Demand (d) Price (P ) Production (q)
50 30 3
% dev. from ss

40 25
2
20
30
15 1
20
10
0
10 5
0 0 -1
0 20 40 0 20 40 0 20 40

Oil (K) Extraction (θ) Investment (I)

0.5 4 25
Non-OPEC
% dev. from ss

20 OPEC
0 3
15
-0.5 2
10
-1 1
5

-1.5 0 0
0 20 40 0 20 40 0 20 40

Figure 6 depicts the impulse response function for a one standard deviation supply
16
Our data for oil futures covers the period from 1986 to 2015. We constructed annual real future
oil prices by averaging all the future contracts with one-year maturity within year t and deflating the
average by the time t consumer price index. See Alquist et al. (2013) for a discussion of the properties
of oil price futures.

24
shock. The shock follows a hump-shaped pattern with a peak in year 2. Compared
to the demand shock, the supply shock is less persistent and is smaller in magnitude.
Extraction rates and production rise in non-OPEC firms and fall in OPEC firms. Non-
OPEC firms increase their investment to boost reserves so they can raise production.
But since the shock is short lived, the rise in investment is much smaller than the one
that occurs in response to a demand shock. OPEC firms also raise their investment but
not as much as non-OPEC since in the short run OPEC firms have a lower extraction
rate than non-OPEC firms.

Figure 6: Impulse response to a supply shock

Supply Shock (u) Price (P ) Production (q)
8 20 2
% dev. from ss

15
6 0
10
4 -2
5
2 -4
0

0 -5 -6
0 20 40 0 20 40 0 20 40

Oil (K) Extraction (θ) Investment (I)

1.5 2 3
Non-OPEC
% dev. from ss

0 2 OPEC
1
-2 1
0.5
-4 0
0
-6 -1

-0.5 -8 -2
0 20 40 0 20 40 0 20 40

We can use our model to answer a classic question: what is the role of demand and
supply shocks? Table 8 illustrates the role played by demand and supply shocks in the
performance of the model. Eliminating demand shocks leads to a small decline in the
volatility of prices and to a large decline in the volatility of investment. Eliminating
supply shocks leads to a small decline in the volatility of prices, to a large decline in
the volatility of OPEC production, and implies a perfect correlation between OPEC

25
 Table 8: Moments sensitivity to demand and supply shocks

Moment Data (s.e.) Bench. No demand shocks No supply shocks

std(∆pt ) 0.273 (0.028) 0.227 0.165 0.156
std(∆int ) 0.192 (0.024) 0.213 0.028 0.211
std(∆iot ) 0.193 (0.027) 0.211 0.011 0.211
std(∆qtn ) 0.022 (0.003) 0.025 0.018 0.017
std(∆qto ) 0.069 (0.011) 0.054 0.051 0.017
corr(∆qtn , ∆qto ) -0.141 (0.125) 0.01 -0.303 1.00

and non-OPEC production levels.

Table 9 provides a variance decomposition for the key variables in our model: price,
production and investment. We see that demand and supply shocks contribute roughly
equally to the volatility of prices. This result reflects the fact that demand and supply
shocks tend to have a similar short-run impact on the price of oil (see Figures 5 and 6).
This result is consistent with the importance of macroconomic performance in driving
oil prices emphasized in Barsky and Kilian (2001) and Barsky and Kilian (2004).
Table 9 also shows that the volatility of investment is predominantly driven by
demand shocks. These shocks are long-lived and so they elicit a large response of
investment (see Figure 5). In contrast, supply shocks have a much lower impact on
investment because these shocks are less persistent (see Figure 6).

Table 9: Variance decomposition

Shocks
Moment
Demand Supply
std(∆pt ) 47.2% 52.8%
std(∆int ) 98.3% 1.7%
std(∆iot ) 99.7% 0.3%
std(∆qtn ) 48.3% 51.7%
std(∆qto ) 10.1% 89.9%

Demand and supply shocks contribute equally to the volatility of production by

26
non-OPEC firms. In contrast, the volatility of production by OPEC firms is dominated
by supply shocks.

4.2 Modeling OPEC as a cartel

Our model assumes that oil firms are competitive. It would be interesting to study
the properties of a version of our model in which OPEC firms behave as a cartel and
non-OPEC firms are a competitive fringe. Stiglitz (1976) and Hassler et al. (2010) solve
for an equilibrium in which the oil market is controlled by a monopolist that faces a
demand with constant elasticity. A version of our model in which OPEC behaves as
a cartel is significantly more challenging to solve for three reasons. First, there is a
sizable competitive fringe, so the cartel faces a residual demand that is endogenous and
does not have constant elasticity. Second, in our model the extraction decision has a
dynamic element because the marginal cost function of oil at time t is a function of
all past investment decisions. Third, and most importantly, a version of our model in
which OPEC is a cartel involves a complex dynamic game. In particular, the cartel can
influence its future competition by manipulating today’s oil prices in order to influence
the investment decisions of its competitors. It is challenging to solve the resulting game
even if we assume perfect commitment on the part of the cartel. We leave this task for
future research.
Here, we take two modest steps towards investigating the potential impact of non-
competitive behavior by OPEC on our results. First, we solved a version of the model
in which OPEC firms behave as a cartel for one period and behave competitively there-
after. Non-OPEC act as a competitive fringe in all periods. We find that the response
of this economy to demand and supply shocks is similar quantitatively and qualitatively
to the impulse response functions we discuss in the previous section.
Second, we re-estimate the model allowing demand and supply shocks to be cor-
related. One interpretation of this correlation is that it might be induced by non-
competitive behavior on the part of OPEC firms; for example, OPEC firms can be

27
thought to change their supply so as to smooth oil prices. We present the results in
two tables included in the Appendix. Table 12 reports our parameter estimates and
Table 13 the implications of this version of the model for the moments targeted in the
estimation. The estimated correlation between innovations to demand and supply (εd
and εu ) is −0.64. This negative correlation is consistent with the smoothing hypothesis:
OPEC firms cut production when demand is low. The fit of this version of the model
is very similar to the fit of the version with independent shocks.

4.3 Robustness
In this section we discuss the results of two robustness exercises. First we consider two
alternative instruments for the price of oil. Second, we exclude from the sample two
countries for which the data might have larger measurement error (Saudi Arabia and
Venezuela). The tables with our robustness results are included in the appendix.
In our benchmark results we instrumented the price of oil with the forecast of the
cyclical component of world GDP. Here we consider two alternative instruments for oil
prices: copper prices, as in Newell et al. (2016), and the IMF’s metals price index. We
deflate both indexes by the U.S. consumer price index. Tables 16 and 21 show our
estimates of the elasticity of the extraction rate with respect to prices. These estimates
are still quite low (0.26 and 0.20 instrumenting with real copper and metals prices,
respectively) but are higher than our benchmark estimate (0.12).
Tables 17 and 22 report the elasticity of the extensive margin (number of oil fields
in operation) with respect to the real price of oil. As in our benchmark results we find
that this elasticity is not significantly different from zero.
Tables 18 and 23 contain our estimates of η obtained using real copper prices and
real metals prices as instruments. Both instruments yield a lower value of η. Our
estimates of η for non-OPEC are 9.3 in the benchmark case, 4.1 with real copper prices
as an instrument, and 4.4 with real metals prices. We re-estimated the model with
these values of η and report the results in Tables 19, 20, 24, and 25. The fit of the two

28
alternative models with η = 4.1 and η = 4.4 is only slightly worse than the fit of the
benchmark model. The parameter estimates are similar across the three models. The
main difference is that demand shocks are more persistent in the models with η = 4.1
or η = 4.4 than in the benchmark model.
Next we redo our analysis excluding Saudi Arabia and Venezuela from the sample.
Our motivation is the possibility of larger measurement error for these two countries.
Table 26 shows the estimates of the elasticity of the extraction rate with respect to prices
obtained using this restricted sample. This estimate (0.15) is similar to our benchmark
estimate (0.12). Table 27 reports the elasticity of the extensive margin (number of oil
fields in operation) with respect to the real price of oil. As in our benchmark results
we find that this elasticity is not significantly different from zero. We re-estimated
our structural model excluding Saudi Arabia and Venezuela from the countries used to
compute the data moments. The point estimates of the data moments are very similar
to those of the full sample and, as a result, the estimated parameters and model fit are
quite similar to those obtained in the benchmark specification (see Tables 29 and 30).

5 The impact of hydraulic fracturing

The advent of hydraulic fracturing (fracking) is transforming the oil industry, making
the U.S. once again one of the world’s top oil producers.17 We study the quantitative
impact of fracking using an extended version of our model that incorporates fracking
firms. Since our dataset includes the universe of oil fields in operation, it allows us
to compare the properties of conventional oil fields with the properties of oil fields ex-
plored using fracking. We find that fracking operations differ greatly from conventional
oil production in terms of production flexibility and lags between investment and pro-
duction. Our model implies that an expansion in the share of fracking production in
total oil production will result in a sizable decline in the volatility of oil prices.
17
See Kilian (2016) and Gilje et al. (2016) for a discussion of the impact of fracking on oil and
gasoline markets.

29
 Figure 7: Empirical dist. of lags between investment to production - fracking fields

There are two important differences between fracking and conventional forms of
oil production. First, the lag between investment and production is much shorter for
fracking operations. Second, it is much less costly to adjust the extraction rate in
fracking operations than in conventional oil operations.18
Figure 7 shows the distribution of the lag between investment and the first year of
production in non-conventional oil fields. Our maximum likelihood estimator of λ is
1.13, so the average lag between investment and production is about one year. Recall
that this lag is 12 years for conventional production.
Table 11 reports our estimates of η for fields explored with fracking obtained using
the cyclical component of world real GDP as an instrument. We see that for non-
conventional oil fields our estimate of η is roughly 1.7. In contrast, when we include all
18
One piece of evidence that suggests that fracking operations are highly flexible is the data compiled
by Baker Hughes on the number of oil rigs in operation in the U.S. In January 2009, there were 345
rigs in operation. By September 2014, this number had risen to 1,600. Most of the new rigs are likely
to be used in fracking. Between September 2014 and February 2016 oil prices plummeted from 93 to
30 dollars per barrel. Between September 2014 and September 2015, the number of oil rigs declined
from 1,600 to 641. By the end of February 2016, the number of rigs in operation was reduced to 400.

30
 Table 10: Extraction rate adjustment costs regression - fracking fields

Variable (1) (2)

ln(extraction) 1.68∗∗∗ 9.30∗∗∗
(0.34) (1.14)

Oil field FE 3 3
Year of operation FE 3 3
Sample Fracking fields All non-OPEC
IV 3 3
1st stage F-stat 9.1 53
Clusters (oil fields) 952 9,969
Observations 4,940 146,879

oil fields in our sample in the regression, we obtain an estimate of η = 9.3.

To study the impact of fracking, we include in our model a third type of firm that
produces oil using fracking. These firms have extraction costs that are less convex
(η F = 1.7 < 9.3), no lag between investment and production (λF = 1), and no lag
in the adjustment of the extraction rate. We also assume that fracking firms are not
subject to supply shocks.
There is a continuum of measure one of fracking firms. The problem of the repre-
sentative firm is to maximize its value (V F ):
∞
F
X
t
h
F F F F

F Fi
F η
max V = E0 β Pt θt Kt − It − ψ θt Kt , (12)
{ItF ,θt+1 t+1 }
F ,K F ,X F
t+1 t=0

subject to

F

α
1−α
Xt+1 = (1 − λF )XtF + ItF LF (13)
F
Kt+1 = (1 − θtF )KtF + λF Xt+1
F
. (14)

Here, ItF denotes investment, θtF the extraction rate, XtF oil in process, KtF oil reserves,
and LF the land available to the fracking firm.
We assume that the convexity of extraction costs is lower for fracking firms: η F < η.
This assumption does not imply that the average and marginal cost of extraction for

31
fracking firms are different than those of the traditional firms, as this comparison also
depends on ψ F and on the equilibrium levels of reserves and extraction rates. The
optimality conditions for fracking firms are identical to those for non-OPEC firms.
Total production is now given by:

−ut O O
θN N
t Kt + e θ t Kt + θ Ft KFt = qt ,

where KFt and θ Ft denote the aggregate reserves and aggregate extraction rate of frack-
ing firms, respectively.
These new forms of oil production are only feasible in some parts of the globe.
Rystad estimates that they will represent 20 percent of oil production by 2050. We
calibrate ψ F and the amount of land available for fracking to satisfy two properties.
First, in the steady state fracking represents 20 percent of total oil production. Second,
the steady-state extraction rate for fracking coincides with the one in our data (1.2
percent).
In equilibrium, the marginal return to investing in fracking is equal to that of in-
vesting in conventional oil production. Our calibration implies that the steady state
marginal cost of extracting a barrel of oil is 19 percent higher for fracking firms than for
conventional production. This higher production cost is compensated by the shorter
lag between investment and production associated with fracking.
Figures 8 and 9 show the impulse response for a demand and supply shock, respec-
tively. The panels for the price response also include the price response in our bench-
mark model without fracking firms. We see that fracking firms respond much more to
the shock, in terms of production, extraction and investment, than non-fracking firms.
The result is a much lower increase in the price of oil.
Comparing the response of conventional firms in a world with (Figures 8 and 9) and
without fracking (Figures 5 and 6), we see that conventional firms respond much less
to shocks in a world with fracking firms. Since fracking firms are more nimble, they
respond rapidly to shocks, moderating the equilibrium price increase and making it less

32
 Figure 8: Impulse response to a demand shock - fracking
Demand (d) Price (P ) Production (q)
50 30 16
45 w/ fracking 14
40
25 No fracking
12
% dev. from ss
35
20 10
30
8
25 15
6
20
10 4
15
10 2
5
5 0
0 0 -2
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Oil (K) Extraction (θ) Investment (I)

1 18 25
16 Non-OPEC
0
20 OPEC
14 Fracking
% dev. from ss

-1 12
15
10
-2
8
10
-3 6
4 5
-4
2
-5 0 0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Figure 9: Impulse response to a supply shock - fracking

Supply Shock (u) Price (P ) Production (q)
8 16 8

7 14 w/ fracking 6
No fracking
12
% dev. from ss

6 4
10
5 2
8
4 0
6
3 -2
4
2 2 -4

1 0 -6

0 -2 -8
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Oil (K) Extraction (θ) Investment (I)

1.2 8 5
1 6 Non-OPEC
4 OPEC
0.8 Fracking
% dev. from ss

4
0.6 3
2
0.4
0 2
0.2
-2
0 1
-0.2 -4
0
-0.4 -6

-0.6 -8 -1
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

33
 Table 11: Implication of fracking on key aggregate moments

Moment Benchm. model Model w/ fracking

(1) std(∆pt ) 0.23 0.08
(2) std(∆it ) 0.21 0.12
(3) std(∆qt ) 0.03 0.02
(4) corr(∆pt , ∆it ) 0.69 0.7
(5) corr(∆pt , ∆qt ) -0.47 0.6
(6) corr(∆qt , ∆it ) -0.06 0.53

profitable for conventional firms to change their investment and extraction rates.
Table 11 compares the implications of versions of the model with and without frack-
ing for some key moments. We find that the main impact of fracking is to reduce the
volatility of oil prices by 65 percent from 0.23 to 0.08. Investment and production
are also less volatile in the version of the model with fracking. In addition, there is
also a higher correlation between prices and quantities, and between investment and
quantities, reflecting the response of fracking firms to high-frequency movements in
prices.

6 Conclusion
Our paper reviews some key facts about the oil market and proposes a simple industry
equilibrium model that is consistent with these facts.
We leave four interesting projects for future research. The first is to develop a richer
model of firm heterogeneity. In our model there are only two types of firms, OPEC and
non-OPEC. In practice, firms in both groups differ in their attributes which results in
different choices of investment and extraction rates. The second is to study a version of
our model in which OPEC firms act as a cartel while non-OPEC firms are a competitive
fringe. The third is to introduce the possibility of above-ground inventories that can
be used by commodity speculators to respond to high-frequency changes in oil prices.19
19
See Deaton and Laroque (1992), Deaton and Laroque (1996), Kilian and Murphy (2014), and

34
The fourth is to combine our model of the oil market with a fully-fledged model of the
world economy. Such a combined model would allow us to study the effect of energy-
saving technical change and evaluate the impact of solar, wind, and other alternative
energy sources on the dynamics of the oil markets and the world economy.

References
Adao, B., Narajabad, B., Temzelides, T. et al. (2012). Renewable technology
adoption and the macroeconomy. Tech. rep.

Alquist, R., Kilian, L., Vigfusson, R. J. et al. (2013). Forecasting the price of
oil. Handbook of economic forecasting, 2 427–507.

Anderson, S. T., Kellogg, R. and Salant, S. W. (2017). Hotelling under pres-

sure. Journal of Political Economy.

Arezki, R., Ramey, V. A. and Sheng, L. (2016). News shocks in open economies:
Evidence from giant oil discoveries. The Quarterly Journal of Economics.

Backus, D. K. and Crucini, M. J. (2000). Oil prices and the terms of trade. Journal
of international Economics, 50 185–213.

Barsky, R. B. and Kilian, L. (2001). Do we really know that oil caused the great
stagflation? a monetary alternative. NBER Macroeconomics annual, 16 137–183.

Barsky, R. B. and Kilian, L. (2004). Oil and the macroeconomy since the 1970s.
The Journal of Economic Perspectives, 18 115–134.

Bodenstein, M., Erceg, C. J. and Guerrieri, L. (2011). Oil shocks and external
adjustment. Journal of International Economics, 83 168–184.
Olovsson (2016) for a discussion.

35
Chari, V. and Christiano, L. (2014). The optimal extraction of exhaustible re-
sources. Tech. rep., Federal Reserve Bank of Minneapolis.

Damodaran, A. (2017). Cost of equity and capital. Tech. rep., New York University.

Deaton, A. (1999). Commodity prices and growth in africa. The Journal of Economic
Perspectives, 13 23–40.

Deaton, A. and Laroque, G. (1992). On the behaviour of commodity prices. The

Review of Economic Studies, 59 1–23.

Deaton, A. and Laroque, G. (1996). Competitive storage and commodity price

dynamics. Journal of Political Economy, 104 896–923.

Ezekiel, M. (1938). The cobweb theorem. The Quarterly Journal of Economics, 52

255–280.

Finn, M. G. (2000). Perfect competition and the effects of energy price increases on
economic activity. Journal of Money, Credit and banking 400–416.

Gilje, E., Ready, R. and Roussanov, N. (2016). Fracking, drilling, and asset
pricing: Estimating the economic benefits of the shale revolution. Tech. rep., National
Bureau of Economic Research.

Hamilton, J. D. (1983). Oil and the macroeconomy since world war ii. Journal of
Political Economy, 91 228–248.

Hamilton, J. D. (2003). What is an oil shock? Journal of econometrics, 113 363–398.

Hamilton, J. D. (2005). Oil and the macroeconomy. The new Palgrave dictionary of
economics, 2.

36
Hamilton, J. D. (2013). Historical oil shocks. In Routledge Handbook of Major Events
in Economic History (R. E. Parker and R. Whaples, eds.). New York: Routledge
Taylor and Francis Group, 239–265.

Harvey, D. I., Kellard, N. M., Madsen, J. B. and Wohar, M. E. (2010). The

prebisch-singer hypothesis: four centuries of evidence. The Review of Economics and
Statistics, 92 367–377.

Hassler, J., Krusell, P. and Olovsson, C. (2010). Oil monopoly and the climate.
The American Economic Review, 100 460–464.

Hotelling, H. (1931). The economics of exhaustible resources. Journal of Political

Economy, 39 137–175.

Kellogg, R. (2014). The effect of uncertainty on investment: evidence from texas oil
drilling. The American Economic Review, 104 1698–1734.

Kilian, L. (2008). The economic effects of energy price shocks. Journal of Economic
Literature, 46 871–909.

Kilian, L. (2009). Not all oil price shocks are alike: Disentangling demand and supply
shocks in the crude oil market. The American Economic Review, 99 1053–1069.

Kilian, L. (2014). Oil price shocks: causes and consequences. Annu. Rev. Resour.
Econ., 6 133–154.

Kilian, L. (2016). The impact of the shale oil revolution on us oil and gasoline prices.
Review of Environmental Economics and Policy, 10 185–205.

Kilian, L. and Murphy, D. P. (2014). The role of inventories and speculative trading
in the global market for crude oil. Journal of Applied Econometrics, 29 454–478.

37
Kim, I.-M. and Loungani, P. (1992). The role of energy in real business cycle models.
Journal of Monetary Economics, 29 173–189.

Leduc, S. and Sill, K. (2004). A quantitative analysis of oil-price shocks, systematic

monetary policy, and economic downturns. Journal of Monetary Economics, 51 781–
808.

Lippi, F. and Nobili, A. (2012). Oil and the macroeconomy: a quantitative structural
analysis. Journal of the European Economic Association, 10 1059–1083.

Nerlove, M. (1958). Adaptive expectations and cobweb phenomena. The Quarterly

Journal of Economics, 72 227–240.

Newell, R. G., Prest, B. C. and Vissing, A. (2016). Trophy hunting vs. manu-
facturing energy: The price-responsiveness of shale gas. Tech. rep., National Bureau
of Economic Research.

Olovsson, C. (2016). Oil prices in a real-business-cycle model with precautionary

demand for oil.

Rotemberg, J. J. and Woodford, M. (1996). Imperfect competition and the effects

of energy price increases on economic activity. Journal of Money, Credit and Banking,
28 549–577.

Rouwenhorst, K. G. (1991). Time to build and aggregate fluctuations: A reconsid-

eration. Journal of Monetary Economics, 27 241–254.

Stiglitz, J. E. (1976). Monopoly and the rate of extraction of exhaustible resources.

The American Economic Review, 66 655–661.

38
 Appendix
A Additional tables

Table 12: Estimated parameters when shocks are correlated

Parameter Benchm. model (s.e.) Model w/ corr. shocks (s.e.)

 0.135 (0.015) 0.084 (0.013)
α 3 × 10−6 (0.978) 1 × 10−7 (0.674)
ρd1 1.761 (0.075) 1.759 (0.089)
ρd2 -0.775 (0.073) -0.773 (0.09)
ρu1 1.428 (0.15) 1.719 (0.072)
ρu2 -0.51 (0.11) -0.733 (0.063)
var(edt ) 0.02 (0.007) 0.05 (0.018)
var(eut ) 0.002 (0.001) 0.002 (0.001)
corr(edt , eut ) - - -0.639 (0.105)

39
 Table 13: Data and model moments when shocks are correlated

Moment Data (s.e.) Benchm. model Model w/ corr. shocks

(1) std(∆pt ) 0.273 (0.028) 0.227 0.232
(2) std(∆int ) 0.192 (0.24) 0.213 0.205
(3) std(∆iot ) 0.193 (0.27) 0.211 0.216
(4) std(∆qtn ) 0.022 (0.003) 0.025 0.022
(5) std(∆qto ) 0.069 (0.011) 0.054 0.072
(6) corr(∆pt , ∆int ) 0.557 (0.147) 0.709 0.749
(7) corr(∆pt , ∆iot ) 0.362 (0.109) 0.657 0.53
(8) corr(∆pt , ∆qtn ) 0.031 (0.069) 0.049 0.003
(9) corr(∆pt , ∆qto ) 0.030 (0.122) -0.572 -0.418
(10) corr(∆int , ∆iot ) 0.673 (0.096) 0.997 0.946
(11) corr(∆int , ∆qtn ) 0.087 (0.094) -0.001 0.008
(12) corr(∆int , ∆qto ) 0.023 (0.112) -0.103 0.027
(13) corr(∆iot , ∆qtn ) -0.034 (0.145) 0.003 0
(14) corr(∆iot , ∆qto ) -0.226 (0.153) -0.036 0.251
(15) corr(∆qtn , ∆qto ) -0.141 (0.125) 0.014 -0.221
(16) corr(∆pt , ∆pt−1 ) -0.027 (0.088) -0.018 -0.125
(17) corr(∆int , ∆int−1 ) 0.119 (0.135) 0.008 0.025
(18) corr(∆iot , ∆iot−1 ) 0.311 (0.096) 0.009 0.022
(19) n
corr(∆qtn , ∆qt−1 ) 0.643 (0.113) 0.334 0.46
o o
(20) corr(∆qt , ∆qt−1 ) 0.213 (0.211) 0.326 0.671

40
B Robustness Tables
B.1 Sample period starting at 1900

Table 14: Price elasticity of extraction rates (1900–2015)

Variable (1) (2) (3) (4) (5)

ln(price) 0.06∗∗∗ 0.09∗∗∗ 0.12∗∗∗ 0.14∗∗∗ 0.16∗∗∗
(0.006) (0.008) (0.009) (0.01) (0.011)

ln(price) × 1OPEC −0.21∗∗∗ −0.2∗∗∗

(0.022) (0.023)

ln(price) × 1Big Firm −0.06∗∗∗

(0.018)

ln(price) × 1|∆ ln(p)|>0.1 −0.15∗∗∗

(0.01)

Oil field FE 3 3 3 3 3
Operation year FE 3 3 3 3 3
Year trend 3 3 3 3 3
IV 7 3 3 3 3
Clusters (oil fields) 13,811 13,372
Observations 351,442 351,003

Oil fields with positive extraction in 1901-2015, excluding last year of operation. Standard
errors in parenthesis. Instrument for price using one year lagged price (F-stat > 1000 in
(2)-(5)). (∗ ∗ ∗) - significant at a 1% level.

41
 Table 15: Price elasticity of oil fields in operation (1900–2015)

Variable (1) (2) (3) (4)

ln(price) −0.4 −0.51 −0.18 −0.32
(0.27) (0.32) (0.13) (0.19)

Year trend 3 3 3 3
IV 7 3 3 3
Dep. variable All fields All fields Non-OPEC fields OPEC fields
Observations 114 114 114 114

Newey-West standard errors with 5 years lag in parenthesis.

42
B.2 Instrument with real copper prices

Table 16: Price elasticity of extraction rates

Variable (1) (2) (3) (4) (5)

ln(price) 0.09∗∗∗ 0.26∗∗∗ 0.31∗∗∗ 0.34∗∗∗ 0.22∗∗∗
(0.009) (0.018) (0.019) (0.021) (0.018)

ln(price) × 1OPEC −0.25∗∗∗ −0.24∗∗∗

(0.041) (0.041)

ln(price) × 1Big Firm −0.1∗∗∗

(0.032)

ln(price) × 1|∆ ln(p)|>0.1 −0.03∗∗∗

(0.005)

Oil field FE 3 3 3 3 3
Operation year FE 3 3 3 3 3
Year trend 3 3 3 3 3
IV 7 3 3 3 3
Clusters (oil fields) 12,187 11,479
Observations 173,742 173,034

Oil fields with positive extraction in 1971-2015, excluding last year of operation. Standard
errors in parenthesis. Instrument for price using real copper price prediction (F-stat
> 1000 in (2)-(5)). (∗ ∗ ∗) - significant at a 1% level.

43
 Table 17: Price elasticity of oil fields in operation

Variable (1) (2) (3) (4)

ln(price) −0.05 −0.44 −0.13 −0.31
(0.06) (0.29) (0.10) (0.19)

Year trend 3 3 3 3
IV 7 3 3 3
Dep. variable All fields All fields Non-OPEC fields OPEC fields
Observations 45 45 45 45

Newey-West standard errors with 5 years lag in parenthesis.

Table 18: Extraction rate adjustment costs regression

Variable (1) (2) (3)

ln(extraction) 4.54∗∗∗ 4.05∗∗∗ 15.19∗∗
(0.22) (0.18) (7.72)

Oil field FE 3 3 3
Operation year FE 3 3 3
Sample All Non-OPEC OPEC
IV 3 3 3
1st stage F-stat 276 198 3.4
Clusters (oil fields) 11,527 9,969 1,558
Observations 174,339 146,879 27,460

44
 Table 19: Estimated parameters

Parameter Estimate (s.e.)

 0.077 (0.014)
α 0.062 (0.079)
ρd1 1.991 (0.003)
ρd2 -0.991 (0.003)
ρu1 1.729 (0.077)
ρu2 -0.748 (0.072)
var(edt ) 0.001 (0.000. . . )
var(eut ) 0.001 (0.000. . . )

Table 20: Data and model moments

Moment Data (s.e.) Benchm. model Model w/ copper IV

(1) std(∆pt ) 0.273 (0.028) 0.227 0.195
(2) std(∆int ) 0.192 (0.24) 0.213 0.228
(3) std(∆iot ) 0.193 (0.27) 0.211 0.217
(4) std(∆qtn ) 0.022 (0.003) 0.025 0.025
(5) std(∆qto ) 0.069 (0.011) 0.054 0.034
(6) corr(∆pt , ∆int ) 0.557 (0.147) 0.709 0.621
(7) corr(∆pt , ∆iot ) 0.362 (0.109) 0.657 0.484
(8) corr(∆pt , ∆qtn ) 0.031 (0.069) 0.049 -0.051
(9) corr(∆pt , ∆qto ) 0.030 (0.122) -0.572 -0.357
(10) corr(∆int , ∆iot ) 0.673 (0.096) 0.997 0.977
(11) corr(∆int , ∆qtn ) 0.087 (0.094) -0.001 0.228
(12) corr(∆int , ∆qto ) 0.023 (0.112) -0.103 -0.066
(13) corr(∆iot , ∆qtn ) -0.034 (0.145) 0.003 0.238
(14) corr(∆iot , ∆qto ) -0.226 (0.153) -0.036 0.095
(15) corr(∆qtn , ∆qto ) -0.141 (0.125) 0.014 -0.198
(16) corr(∆pt , ∆pt−1 ) -0.027 (0.088) -0.018 -0.108
(17) corr(∆int , ∆int−1 ) 0.119 (0.135) 0.008 0.295
(18) corr(∆iot , ∆iot−1 ) 0.311 (0.096) 0.009 0.324
(19) corr(∆qtn , ∆qt−1
n
) 0.643 (0.113) 0.334 0.591
o o
(20) corr(∆qt , ∆qt−1 ) 0.213 (0.211) 0.326 0.609

45
B.3 Instrument with real metal prices

Table 21: Price elasticity of extraction rates

Variable (1) (2) (3) (4) (5)

ln(price) 0.09∗∗∗ 0.20∗∗∗ 0.24∗∗∗ 0.26∗∗∗ 0.07∗∗∗
(0.009) (0.016) (0.018) (0.02) (0.014)

ln(price) × 1OPEC −0.22∗∗∗ −0.21∗∗∗

(0.04) (0.041)

ln(price) × 1Big Firm −0.06∗

(0.032)

ln(price) × 1|∆ ln(p)|>0.1 −0.12∗∗∗

(0.006)

Oil field FE 3 3 3 3 3
Operation year FE 3 3 3 3 3
Year trend 3 3 3 3 3
IV 7 3 3 3 3
Clusters (oil fields) 12,187 11,479
Observations 173,742 173,034

Oil fields with positive extraction in 1971-2015, excluding last year of operation. Standard
errors in parenthesis. Instrument for price using real metal price index prediction (F-stat
> 1000 in (2)-(5)). ∗ ∗ ∗ [∗] - significant at a 1% [10%] level.

46
 Table 22: Price elasticity of oil fields in operation

Variable (1) (2) (3) (4)

ln(price) −0.05 −0.08 −0.02 −0.06
(0.06) (0.07) (0.02) (0.06)

Year trend 3 3 3 3
IV 7 3 3 3
Dep. variable All fields All fields Non-OPEC fields OPEC fields
Observations 45 45 45 45

Newey-West standard errors with 5 years lag in parenthesis.

Table 23: Extraction rate adjustment costs regression

Variable (1) (2) (3)

ln(extraction) 4.72∗∗∗ 4.26∗∗∗ 11.73∗∗
(0.22) (0.18) (4.15)

Oil field FE 3 3 3
Operation year FE 3 3 3
Sample All Non-OPEC OPEC
IV 3 3 3
1st stage F-stat 271 286 3.5
Clusters (oil fields) 11,527 9,969 1,558
Observations 174,339 146,879 27,460

47
 Table 24: Estimated parameters

Parameter Estimate (s.e.)

 0.077 (0.013)
α 0.067 (0.087)
ρd1 1.991 (0.003)
ρd2 -0.991 (0.003)
ρu1 1.728 (0.076)
ρu2 -0.746 (0.071)
var(edt ) 0.001 (0.000. . . )
var(eut ) 0.001 (0.000. . . )

Table 25: Data and model moments

Moment Data (s.e.) Benchm. model Model w/ metal IV

(1) std(∆pt ) 0.273 (0.028) 0.227 0.199
(2) std(∆int ) 0.192 (0.24) 0.213 0.229
(3) std(∆iot ) 0.193 (0.27) 0.211 0.217
(4) std(∆qtn ) 0.022 (0.003) 0.025 0.025
(5) std(∆qto ) 0.069 (0.011) 0.054 0.035
(6) corr(∆pt , ∆int ) 0.557 (0.147) 0.709 0.621
(7) corr(∆pt , ∆iot ) 0.362 (0.109) 0.657 0.484
(8) corr(∆pt , ∆qtn ) 0.031 (0.069) 0.049 -0.041
(9) corr(∆pt , ∆qto ) 0.030 (0.122) -0.572 -0.372
(10) corr(∆int , ∆iot ) 0.673 (0.096) 0.997 0.977
(11) corr(∆int , ∆qtn ) 0.087 (0.094) -0.001 0.244
(12) corr(∆int , ∆qto ) 0.023 (0.112) -0.103 -0.076
(13) corr(∆iot , ∆qtn ) -0.034 (0.145) 0.003 0.238
(14) corr(∆iot , ∆qto ) -0.226 (0.153) -0.036 0.087
(15) corr(∆qtn , ∆qto ) -0.141 (0.125) 0.014 -0.202
(16) corr(∆pt , ∆pt−1 ) -0.027 (0.088) -0.018 -0.111
(17) corr(∆int , ∆int−1 ) 0.119 (0.135) 0.008 0.293
(18) corr(∆iot , ∆iot−1 ) 0.311 (0.096) 0.009 0.323
(19) corr(∆qtn , ∆qt−1
n
) 0.643 (0.113) 0.334 0.587
o o
(20) corr(∆qt , ∆qt−1 ) 0.213 (0.211) 0.326 0.599

48
B.4 Excluding Saudi Arabia and Venezuela from the sample

Table 26: Price elasticity of extraction rates

Variable (1) (2) (3) (4) (5)

ln(price) 0.11∗∗∗ 0.15∗∗∗ 0.17∗∗∗ 0.18∗∗∗ 0.18∗∗∗
(0.009) (0.012) (0.015) (0.02) (0.013)

ln(price) × 1OPEC −0.17∗∗∗ −0.17∗∗∗

(0.07 ) (0.041)

ln(price) × 1Big Firm −0.05

(0.05)

ln(price) × 1|∆ ln(p)|>0.1 −0.18∗∗∗

(0.001)

Oil field FE 3 3 3 3 3
Operation year FE 3 3 3 3 3
Year trend 3 3 3 3 3
IV 7 3 3 3 3
Clusters (oil fields) 11,893 11,195
Observations 167,848 167,150

Oil fields with positive extraction in 1971-2015, excluding last year of operation. Standard
errors in parenthesis. Instrument for price using real metal price index prediction (F-stat
> 1000 in (2)-(5)). ∗ ∗ ∗ [∗] - significant at a 1% [10%] level.

49
 Table 27: Price elasticity of oil fields in operation

Variable (1) (2) (3) (4)

ln(price) −0.03 0.01 −0.01 0.02
(0.08) (0.1) (0.03) (0.09)

Year trend 3 3 3 3
IV 7 3 3 3
Dep. variable All fields All fields Non-OPEC fields OPEC fields
Observations 45 45 45 45

Newey-West standard errors with 5 years lag in parenthesis.

Table 28: Extraction rate adjustment costs regression

Variable (1) (2) (3)

ln(extraction) 9.97∗∗∗ 9.3∗∗∗ 19.77
(1.20) (0.18) (12.03)

Oil field FE 3 3 3
Operation year FE 3 3 3
Sample All Non-OPEC OPEC
IV 3 3 3
1st stage F-stat 56 53 2.5
Clusters (oil fields) 11,243 9,969 1,274
Observations 168,388 146,879 21,509

50
 Table 29: Estimated parameters

Parameter Estimate (s.e.)

 0.122 (0.017)
α 7 × 10−7 (0.930)
ρd1 1.765 (0.077)
ρd2 -0.779 (0.075)
ρu1 1.529 (0.094)
ρu2 -0.585 (0.092)
var(edt ) 0.002 (0.001)
var(eut ) 0.022 (0.008)

Table 30: Data and model moments

Benchmark No Saudi Arabia and Venezuela
Moment Data (s.e.) Model Data (s.e.) Model
(1) std(∆pt ) 0.273 (0.028) 0.227 0.273 (0.028) 0.226
(2) std(∆int ) 0.192 (0.024) 0.213 0.192 (0.024) 0.221
(3) std(∆iot ) 0.193 (0.027) 0.211 0.208 (0.031) 0.218
(4) std(∆qtn ) 0.022 (0.003) 0.025 0.273 (0.028) 0.024
(5) std(∆qto ) 0.069 (0.011) 0.054 0.076 (0.016 0.049
(6) corr(∆pt , ∆int ) 0.557 (0.147) 0.709 0.557 (0.147) 0.767
(7) corr(∆pt , ∆iot ) 0.362 (0.109) 0.657 0.36 (0.12) 0.708
(8) corr(∆pt , ∆qtn ) 0.031 (0.069) 0.049 0.031 (0.069) 0.061
(9) corr(∆pt , ∆qto ) 0.030 (0.122) -0.572 -0.025 (0.104) -0.53
(10) corr(∆int , ∆iot ) 0.673 (0.096) 0.997 0.626 (0.108) 0.995
(11) corr(∆int , ∆qtn ) 0.087 (0.094) -0.001 0.087 (0.094) 0.001
(12) corr(∆int , ∆qto ) 0.023 (0.112) -0.103 -0.065 (0.112) -0.145
(13) corr(∆iot , ∆qtn ) -0.034 (0.145) 0.003 -0.067 (0.138) 0.004
(14) corr(∆iot , ∆qto ) -0.226 (0.153) -0.036 -0.357 (0.246) -0.062
(15) corr(∆qtn , ∆qto ) -0.141 (0.125) 0.014 -0.126 (0.089) 0.002
(16) corr(∆pt , ∆pt−1 ) -0.027 (0.088) -0.018 -0.027 (0.088) -0.019
(17) corr(∆int , ∆int−1 ) 0.119 (0.135) 0.008 0.119 (0.135) 0.009
(18) corr(∆iot , ∆iot−1 ) 0.311 (0.096) 0.009 0.252 (0.11) 0.01
(19) corr(∆qtn , ∆qt−1
n
) 0.643 (0.113) 0.334 0.643 (0.113) 0.381
o o
(20) corr(∆qt , ∆qt−1 ) 0.213 (0.211) 0.326 0.349 (0.154) 0.401

51