How do oil producers respond to giant oil field discoveries?

Accepted Manuscript How do oil producers respond to giant oil field discoveries?

Jochen H.F. Güntner PII: DOI: Reference:

S0140-9883(18)30494-8 https://doi.org/10.1016/j.eneco.2018.12.012 ENEECO 4257

To appear in:

Energy Economics

Received date: Revised date: Accepted date:

13 March 2017 30 November 2018 13 December 2018

Please cite this article as: Jochen H.F. Güntner , How do oil producers respond to giant oil field discoveries?. Eneeco (2018), https://doi.org/10.1016/j.eneco.2018.12.012

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

How do oil producers respond to giant oil field discoveries? Jochen H. F. Gntner∗

PT

Johannes Kepler University Linz

SC

Abstract

RI

December 22, 2018

PT

ED

MA

NU

The classical Hotelling model predicts that the optimal extraction level rises immediately after an unexpected resource discovery, whereas, in reality, there are substantial adjustment costs in petroleum production and an average lag of several years between a discovery and the start of production. Using a large panel of country-level production data and a difference-indifferences identification approach, I show that domestic production levels respond before a newly found oil field comes on line and that this increase is driven by non-OPEC producers, consistent with different responses of OPEC and non-OPEC drilling activity. Offshore fields and exceptionally large “super-” or “mega-giant” fields are also more likely to raise countrylevel production. Given that domestic petroleum consumption rises by less in response to a discovery, at least part of the increase in production seems to go into (net) oil exports.

Keywords: Giant oil field discoveries; Lifetime of reserves; Oil production; OPEC; Proved reserves

AC CE

JEL Classification: C32, N50, Q33, Q41

∗ Jochen G¨ untner is Assistant Professor at the Department of Economics, Johannes Kepler University Linz. Address: Altenberger Strasse 69, 4040 Linz, Austria. E-mail: [email protected], Telephone: +43 732 2468 7360. Fax: +43 732 2468 9679.

ACCEPTED MANUSCRIPT 1

Introduction

The classical Hotelling (1931) model predicts that the optimal extraction level rises immediately after an unexpected resource discovery. However, this ignores the facts that there are substantial adjustment costs in petroleum production (see, e.g., Hamilton, 2009; Kilian, 2009) and that the average delay between a discovery and the start of production from a new field in conventional oil production is four to six years.1 In this paper, I investigate whether and how fast domestic

PT

production and consumption levels respond to a sizeable increase in the country’s reserve base. A comprehensive data set of arguably exogenous giant oil field discoveries compiled by Horn

RI

(2014) enables me to study the behavior of oil producers in a quasi-natural experiment. Aretzki et al. (2017) use giant oil and gas field discoveries as a directly observable measure

SC

of news shocks about future output to explore the effects of such shocks in an open economy. In a large panel of countries, the authors find that the current account and savings decline for the

NU

first five years before rising sharply during the following years, consistent with the predictions of a theoretical two-sector small open economy model with a resource sector. Investment rises

MA

shortly after the discovery, whereas GDP does not increase until after five years. In contrast to Aretzki et al. (2017), which assume a fixed extraction path from the newly found reservoir, however, I do not discard ex ante the possibility of adjustments in production

ED

from existing fields. Focusing on giant oil field discoveries from the same data set and annual panel data on petroleum production starting in 1965, I find that country-level production picks

PT

up soon after the resource discovery, while country-level rig activity increases with a slight delay. Accordingly, at least part of the increase in production seems to come from spare capacity.

AC CE

Using information about the size, type, and location of oil fields, I show that the subsequent increase in petroleum production is mainly due to so-called “super-” or “mega-giant” fields with an estimated ultimate recovery of three billion barrels of oil equivalents as opposed to ordinary “giant” fields, that production increases in response to offshore rather than onshore discoveries, and that non-OPEC producers tend to raise domestic production levels significantly in response to a discovery, whereas OPEC members seem to postpone production from newly discovered as well as existing reservoirs. Consistently, rig activity rises only for non-OPEC producers. Using a large panel of country-level consumption data starting in 1965, I further investigate whether the observed hike in petroleum production is accompanied by a quantitatively similar increase in petroleum consumption or rather destined for raising (net) oil exports, as implicitly assumed in Aretzki et al. (2017). Based on a sample of 172 countries, I find that country-level 1 Larger fields have longer lead times than smaller fields, and lead times for gas tend to be longer than for oil fields (see, e.g., Norwegian Petroleum Directorate, 2011)

1

ACCEPTED MANUSCRIPT consumption increases in response to a giant oil field discovery, albeit by less than production and at a lower level of statistical significance. Interestingly, consumption increases significantly for non-OPEC countries, whereas it falls or remains constant for OPEC members, depending on the exact specification of the dynamic panel distributed lag regression model. My study is closely related to Lepore and Rastad (2011), who use discrete changes in proved reserves and oil production data from the British Petroleum (BP) Statistical Review of World Energy to investigate the response of petroleum producers to oil discoveries during 1980–2009.

PT

Relative to their study, the present paper makes four main contributions. First, starting in 1945 rather than in 1980, my sample adds 35 years of data, which account for the vast majority of

RI

newly discovered oil reserves since 1900, as illustrated in Figure 1. Second, oil discoveries in Horn

SC

(2014) correspond to externally validated and exactly dated findings of giant oil fields rather than to “large changes” in proved oil reserves according to a threshold chosen by the authors. This

NU

is crucial because proved reserves are defined as reservoirs for which “economic producibility is supported by actual production or conclusive formation tests [...] by core analyses and/or electric or other interpretations.”2 In particular, the amount of proved reserves is a function

MA

of the current oil price and available production techniques. As a consequence, the data set in Lepore and Rastad (2011) also contains large reductions in proved reserves, which cannot be explained by depletion due to production and are therefore excluded in one of their robustness

ED

checks. Third, this study draws on the exact size of a newly discovered giant oil field in terms of its estimated ultimate recovery of oil to quantify the economic significance of each discovery,

PT

whereas Lepore and Rastad (2011) define a dummy variable that takes on a value of one in the year of a “large change” in proved oil reserves and zero in all other years, effectively treating the

AC CE

2009 discoveries of the 9 bboe super-giant Libra field in Brazil and the 0.75 bboe giant Tiber field in the U.S. Mexican Gulf as one and the same event. Finally, the use of a panel distributed lag regression model, as in Aretzki et al. (2017), facilitates the computation of dynamic impulse response functions beyond the analysis of selected regression coefficients.

2

Related Literature

In the classic Hotelling (1931) model, the stock of an exhaustible resource is known. Absent extraction costs, the shadow value of the resource should therefore rise over time at the market rate of interest, while all future prices and the optimal extraction path are determined by the initial price. Yet, the theoretical prediction of an upward-sloping price path seems inconsistent with the empirical observation of falling prices for many mineral commodities over much of 2

Compare the “Definitions, Sources and Explanatory Notes” on https://www.eia.gov/dnav/ng/TblDefs/ng_ enr_nprod_tbldef2.asp.

2

ACCEPTED MANUSCRIPT the twentieth century. While there are several candidate explanations for this discrepancy, one obvious candidate is that the reserve base is not known with certainty from the beginning. Instead, repeated downward revisions of the initial price due to unexpected resource discoveries might offset the predicted rise at the market rate of interest (see Arrow and Chang, 1982). The present paper is related to the following three strands of literature.

2.1

Exploration and extraction

PT

First, the theoretical and empirical analysis of oil producers’ optimal exploration and extraction activity. Peterson (1978) develops a model with extractive firms, which select the time paths

RI

of exploration and extraction that maximize their present value. Simulating industry behavior

SC

under competition, monopoly, and central management, the authors find that a monopolist will over-conserve the resource and hold excess reserves, whereas competitors over-explore and over-extract. Pindyck (1980) introduces demand and reserve uncertainty in the model of an

NU

exhaustible resource market by modeling fluctuations in the demand function and the reserve base as continuous-time stochastic processes. With constant extraction costs and risk-neutral

MA

firms, neither uncertainty affects the price, and the Hotelling rule continues to apply both in competitive and monopolistic markets. In Deshmukh and Pliska (1980), the resource base follows a Markov process controlled by the bivariate policy of exploration and consumption

ED

(extraction). Among other things, the authors show that optimal rate of extraction is increasing in the level of reserves. Arrow and Chang (1982) assume that the distribution of a resource across

PT

an unexplored territory follows a Poisson process. At any point in time, the socially optimal rates of exploration and consumption are chosen, assuming that reserves are depleted by consumption

AC CE

and raised by resource discoveries from exploration, which reduces the remaining unexplored land. The authors show that, for a large amount of unexplored land, the shadow values of land and reserves move in random cycles with only a slight upward trend, possibly explaining the empirical failure of the Hotelling rule described above. Deshmukh and Pliska (1983) model the optimal consumption (extraction) of a nonrenewable resource under uncertainty about both the discovery process and the economic environment. In a given environment, the optimal rate of extraction is increasing in the level of resources. The intuition is that a discovery lowers the shadow price of the resource, which raises the optimal extraction rate, as the marginal utility of consumption equals the marginal value of the resource stock in the optimum. Modern empirical research on oil exploration considers reserve additions as the output of a production process with economic and geological conditions and drilling effort as inputs. Since economic theory suggests that the search for resources is affected by prices, much of this literature is concerned with the effect of oil prices on exploration effort and success. Fisher (1964) 3

ACCEPTED MANUSCRIPT decomposes reserve growth into the number of wildcat wells drilled, the rate of commercially successful wells, and the average resource discovery per success, where each variable is modeled as a function of oil price, seismic crews, and drilling cost proxies, to study drilling, success, and recovery rates in different U.S. Petroleum Administration Defence Districts during 1946–1955. Follow-up papers by Erickson and Spann (1971), Pindyck (1974), and others use better data and additional explanatory variables. Pesaran (1990) and Pesaran and Favero (1994) account for the intertemporal nature of exploration and production decisions in the oil discovery process

PT

by deriving theory-consistent exploration and production equations for price-taking suppliers. Applying their econometric frameworks to the U.K. Continental Shelf, significant price effects on

RI

oil production and exploration are found. Due to limited evidence of intertemporal optimization,

SC

subsequent studies for the U.S. oil and gas industry return to period-by-period optimization. Among them, Iledare and Pulsipher (1999) use a region-specific drilling and discovery model

NU

to disentangle the effects of resource depletion and other factors on crude oil reserve additions in the Louisiana onshore basin during 1977–1994. They find that technical progress significantly reduced but did not offset the negative effect of depletion on reserve additions. Similarly, Farzin

MA

(2001) develops a model of reserve additions incorporating the effects of expected resource price, cumulative reserves development, and technological progress. Applying the model to U.S. data for 1950–1995, he finds a statistically significant but small price elasticity of reserve additions.

ED

Because technology increases monotonically over time and correlates thus with indicators of maturation and depletion, rather than using a time trend, as in Fagan (1997), Cuddington and

PT

Moss (2001) construct an index of the level of technology to analyze the determinants of average finding costs for petroleum reserves in the U.S. during 1967–1990. According to their estimates,

AC CE

technological change prevented a sharp increase in finding costs for natural gas reserves, while the impact on crude oil finding costs has been modest. Similarly, Managi et al. (2005) construct an index that captures both the number and significance of innovations in oil and gas exploration in the Gulf of Mexico during 1947–1998. Their results suggest that technological change offset the adverse effect of depletion over the entire 50-year period. Studying eight global regions for 1981–2009, Lindholt (2015) concludes that technological change outweighed depletion until the mid-nineties, while the opposite was true for the latter decade of his sample. Using data for three separate regions of the Norwegian Continental Shelf for 1965–2004 to estimate error-correction models that capture the longer-term relationships between drilling, geology and technology, Mohn and Osmundsen (2008) find robust evidence of oil price effects on exploration activity in the long run. Mohn (2008) applies a vector error-correction approach and the decomposition of Fisher (1964) to model drilling activity, success rates, and average

4

ACCEPTED MANUSCRIPT discovery size as three simultaneous equations. For the full history of oil and gas exploration on the Norwegian Continental Shelf, he finds that an oil price increase has a positive net effect on reserve additions per effort, as the negative effect on success rates is overcompensated by the positive effects on drilling activity and average discovery size. It is important to note that, with very few exceptions, econometric models of exploration activity are based on the assumption of a price-taking competitive firm and therefore applied to U.S., U.K., or Norwegian data.3

Extraction from existing reservoirs

PT

2.2

Farrow (1985) tests whether the theoretical, privately efficient extraction path is consistent with

RI

the observed behavior of an individual mining firm. Using proprietary firm data, output price

SC

data, and an estimated trans-log cost system to compare changes in the in-situ value with the expected price path, the theoretical model is rejected even when allowing for a time-varying discount rate, an alternative expected price series, or a constraint on the rate of output.

NU

Separating production from existing wells in a known oil field from the drilling of new wells and incorporating geological and engineering motives in an oil supply model, Black and LaFrance

MA

(1998) test the so-called maximum efficiency recovery (MER) hypothesis that production from established fields is invariant to the price of oil. When applying their econometric model to quarterly data from seven onshore oil fields in Montana, the MER hypothesis is strongly rejected.

ED

In contrast, Mauritzen (2017) finds no significant contemporaneous but a modest lagged effect of oil prices on production from active Norwegian offshore fields, after controlling for field-level

PT

production profiles using a semi-parametric approach. Thompson (2001) refers to the “practitioner literature”, which suggests that the owners of

AC CE

an undeveloped resource possess compound options on information acquisition, exploration, and development drilling, whereas the daily production decisions from developed reserves resemble a corner solution. The optimal rate of production is near capacity, primarily because the marginal cost per barrel is constant and far below the market price of oil, while “backwardation” provides the incentive to drill and replace developed reserves.4 Pickering (2008) considers the relationship between extraction rates and remaining reserves. In a simple exhaustible resource model, the slope of a linear extraction rule is determined by the producers’ discount factor, whereas differences in cost and pricing behavior affect the intercept term. The reserves-production relationship is born out by panel data from the world oil industry both across countries and through time. While extraction is characterized by a robust and stable 3

See Dahl and Duggan (1998) and Mohn (2008) for more comprehensive surveys of this literature. Backwardation means that the discounted futures price of a commodity is below the spot price. While this appears to be inconsistent with the Hotelling principle under certainty, Litzenberger and Rabinowitz (1995) argue that backwardation reflects the option value of choosing the timing of production from developed reserves. 4

5

ACCEPTED MANUSCRIPT relationship over large ranges of remaining reserves, the estimated slope is significantly lower for OPEC member states. Pickering (2008) argues that this could be explained by differences in risk aversion, discount rates, and measurement error in the reserves data. H¨o¨ok et al. (2009) also find that OPEC fields tend to exit the plateau phase earlier, entering a longer decline phase with a flatter slope. A possible explanation is that the OPEC quota system has effectively moderated production and maintained the longevity of fields relative to their non-OPEC counterparts. Spiro (2014) shows that, by removing any scarcity considerations of the resource owner, the

PT

assumption of a rolling planning horizon can reconcile the puzzling long-run price dynamics of exhaustible resources such as oil, gas, and metals. As a result, extraction can be non-decreasing

RI

and resource prices non-increasing for a long period of time. A calibration of his model to the

SC

oil market replicates the gradually falling real price of oil after WWII and the sharply increasing price after 1998, suggesting that long-run scarcity might have been more important recently.

Effects of giant oil field discoveries

NU

2.3

In addition to Aretzki et al. (2017), which was already discussed in the introduction section, the

MA

present paper is related to a number of other recent studies that investigate the effects of giant oil (and gas) discoveries as a proxy for an anticipated resource boom on selected economic and political variables. In a descriptive study, H¨o¨ok et al. (2009) find that the average decline rate

ED

of the world’s largest oil fields, which depends on new exploration and production technologies, is increasing over time. Given that these fields represent the most important production base

PT

and that the decline rate of existing giant oil fields is already high and increasing, the authors argue that the world faces an increasing oil supply challenge in the future.

AC CE

Tsui (2009) exploits exogenous variation in the timing and size of giant oil discoveries to identify the impact of changes in oil wealth on democracy and finds that a discovery of 100 billion barrels of crude oil — roughly equal to the initial endowment of Iraq — pushes a country’s level of democracy almost twenty percentage points below trend after three decades. While the effect is larger for fields with higher-quality oil and lower exploration and extraction costs, it is less precisely estimated when the discovery’s size is measured per capita, suggesting that politicians care about the absolute rather than the per-capita value of a country’s oil wealth. Motivated by anecdotal evidence, Lei and Michaels (2014) investigate whether natural resource windfalls, such as from a giant oil field discovery, increase the risk of internal armed conflicts since 1946. On the one hand, a giant oil discovery increases per capita oil production and oil exports by up to 50% after four and ten years, respectively. On the other hand, the incidence of internal armed conflict increases by about 5–8 percentage points. This causal effect is especially high for countries that experienced internal armed conflicts or coups in the decade prior to discovery. 6

ACCEPTED MANUSCRIPT Finally, Harding et al. (2016) estimate the effects of giant oil and gas discoveries on bilateral exchange rates and find that a discovery with a value equal to the country’s GDP leads to a real exchange rate appreciation by 14% within ten years following the discovery that is driven almost exclusively by non-tradable goods inflation. These results are qualitatively and quantitatively consistent with the predictions of a calibrated model with forward-looking behavior and Dutchdisease dynamics.

The Data

PT

3

RI

In order to analyze the response of oil producers to a sizeable change in crude oil reserves, I use a data set of giant oil and gas field discoveries compiled by Myron K. Horn, former President of

SC

the American Association of Petroleum Geologists (AAPG), which contains information about the name, deposit type (i.e. oil or gas), discovery year, hydrocarbon volumes, reservoir depth,

NU

and reservoir location of close to 1,000 oil and gas fields discovered worldwide between 1868 and 2010. In what follows, I focus on the 590 giant, super-giant, and mega-giant oil field discoveries in Horn’s (2014) data. International oil production data are available for a broader sample of

MA

countries and years than international gas production data, and large volumes of crude oil were transported using marine vessels long before natural gas could be liquefied for transportation

ED

and storage on a large scale. Accordingly, the global oil market was essentially integrated during my sample period, whereas regional gas markets remained largely separated.

Giant oil fields of the world

PT

3.1

The research question of this paper is how oil producers respond to an increase in their reserve

AC CE

base and whether the OPEC status of the country or the location (i.e. onshore vs. offshore) and size (i.e. giant vs. super- or mega-giant) of the reservoir matters. Given that the discovery of a giant oil field signals a substantial increase in the country’s oil wealth and production capacity and due to the distribution of giant oil discoveries across countries and over time, each incident can be treated as a country-specific reserves shock. Figure 1 plots the hydrocarbon volumes of giant oil fields cumulated by discovery year against the real price of oil for 1900–2010. Table 1 reports summary statistics by country for the sample period used in this paper. The literature discussed in Section 2.1 challenges the assumption that “the timing of giant oil discoveries is plausibly exogenous” (Aretzki et al., 2017). From a theoretical perspective, past discoveries have two opposing effects on the probability of current and future discoveries. On the one hand, if easily accessible locations are probed first, cumulative discoveries raise the cost of future successful drilling, as in Pindyck (1978). On the other hand, knowledge about

7

ACCEPTED MANUSCRIPT the territory’s geology and the reservoirs’ location might render future discoveries more likely.5 For an earlier vintage of the data set used in this paper, Lei and Michaels (2014) find that giant oil field discoveries in a country’s recent past raise the odds of additional discoveries in its near future. Although controlling for country and time fixed effects reduces the predictive power of past discoveries, it remains statistically significant. Auxiliary panel regressions show that, controlling for country and time fixed effects, countrylevel rig activity from the Baker Hughes International Rotary Rig Count has predictive power

PT

for the incident of a giant oil field discovery, albeit not for its size. Using a four-variable vectorautoregressive model in (log-)differences of world oil production, the real price of oil, cumulated

RI

volumes of oil discoveries, and alternative measures of exploratory effort (e.g. world and U.S. rig

SC

counts, the number of exploratory and development wells, and footage drilled), Granger noncausality from oil price to exploratory effort and from oil discoveries to production is rejected at

NU

the 1% significance level, whereas exploratory effort never Granger-causes any other variable. While one may therefore argue that resource discoveries are more likely in geographical regions with larger known and unknown endowments than others or a history of discoveries, due

MA

to the uncertainty surrounding oil and gas exploration, the exact size of a discovery is unlikely to be predictable — especially for giant oil and gas fields. Since the timing of each discovery has been independently verified and documented based on multiple sources, which are reported

ED

systematically for each incident, Horn’s (2014) data set is largely immune to concerns about accidental or deliberate manipulation of the announcement date of a giant oil field discovery,

Oil production

AC CE

3.2

PT

for example by the government (see Aretzki et al., 2017).

For country-level petroleum production in thousand barrels per day (tbpd), I compile data from British Petroleum’s (BP) Statistical Review of World Energy and the U.S. Energy Information Administration’s (EIA) Monthly Energy Review for a total of 80 countries starting in 1965.6 Each oil-producing country is treated as an independent unit of observation, except for former members of the Union of Soviet Socialist Republics (USSR), which are combined in a synthetic “Former Soviet Union” after its dissolution, both in the field discovery and oil production data. Hence, my analysis is based on an unbalanced panel of oil production data for 80 countries, 41 of 5

In Hamilton and Atkinson (2013), for example, finding a resource today raises the cost of future discoveries, while extracting resources yields knowledge that reduces the cost of future discoveries. 6 BP production data includes crude oil, tight oil, oil sands, and NGLs (the liquid content of natural gas where this is recovered separately) and excludes liquid fuels from other sources such as biomass and derivatives of coal and natural gas. EIA production data are for crude oil and lease condensate, excluding natural gas plant liquids.

8

ACCEPTED MANUSCRIPT which experienced at least one giant oil field discovery since 1945.7 All members of the “Former Soviet Union” experienced a total of 81 giant oil field discoveries since 1945. The panel includes 14 current OPEC members, all of which experienced at least one giant oil field discovery since 1945, ranging from exactly one for Gabon to 48 for Saudi Arabia.8 Table A.1 in the appendix reports summary statistics for the petroleum production data by country.

3.3

Oil consumption

PT

For country-level petroleum consumption in thousand barrels per day (tbpd), I compile data from BP’s Statistical Review of World Energy and the EIA’s Monthly Energy Review for a total

RI

of 172 countries starting in 1965, 42 of which experienced at least one giant oil field discovery

SC

since 1945.9 In line with the treatment of oil production and discovery data, oil consumption of former USSR member states is added together in a synthetic “Former Soviet Union”. Given that I can draw on an even broader (unbalanced) panel when analyzing the response of domestic

NU

petroleum consumption, at each point in time, there exists a large control group of countries that has not experienced a giant oil field discovery in the recent past or during the entire sample

MA

period. In fact, the latter applies to the majority of countries in the sample. Table A.2 in the appendix reports summary statistics for the petroleum consumption data by country.

Econometric Methodology

ED

4

Similar to Aretzki et al. (2017), I use a dynamic panel distributed lag regression (DLR) model

PT

to estimate the response of country-level petroleum production to a giant oil field discovery. (1)

AC CE

∆yit = A (L) ∆yit + B (L) dit + αi + µt + it ,

where ∆yit denotes the change in the level of oil production or consumption in thousand barrels per day (tbpd) in country i and year t, αi controls for country fixed effects such as geographical location or the political system, for example, while µt controls for time fixed effects that affect all countries jointly, such as fluctuations in global demand and thus the price of oil. dit denotes the cumulated volume of newly discovered giant oil fields in country i and year t in million 7

The use of 20 lags of giant oil field discoveries in my baseline regressions implies that the relevant sample period for oil discoveries starts in 1945 rather than with the production data in 1965. 8 The list of current OPEC countries includes the five founding members Iran, Iraq, Kuwait, Saudi Arabia, and Venezuela, as well as Qatar (since 1961), Indonesia (1962–2009 and since 2016), Libya (since 1962), United Arab Emirates (since 1967), Algeria (since 1969), Nigeria (since 1971), Ecuador (1973–1992 and since 2007), Gabon (1975–1995 and since 2016), and Angola (since 2007). In what follows, these countries will be treated as OPEC members throughout, as their membership status extends over the majority of the sample period, 1965–2010, with the exception of Angola and Gabon. 9 BP consumption data includes inland demand, international aviation and marine bunkers, refinery fuel and loss as well as consumption of biogasoline (such as ethanol), biodiesel, and derivatives of coal and natural gas.

9

ACCEPTED MANUSCRIPT barrels of oil equivalents (mmboe), while it is a country-year-specific error term. A(L) and B(L) are pth and qth order lag polynomials with p ≥ 1 and q ≥ 0, respectively. In light of the serial correlation in giant oil field discoveries even after controlling for country and year fixed effects (see Lei and Michaels, 2014), it seems crucial to allow for a direct influence of past discoveries on current oil production. Separate regressions for different lags between the discovery and an outcome variable, as in Lei and Michaels (2014), may yield biased coefficient estimates. Serial correlation in (the change of) country-level oil production demands for lagged

PT

terms of the endogenous variable in the dynamic panel DLR model. In order fully trace out the impulse response functions over a horizon of 20 years, I set p = 1 and q = 20 in the benchmark

RI

regression analysis and allow for p = 4 as a robustness check.

SC

Regarding the specification in (1), a few clarifying comments are in order. First, given that the presence of a unit root in production and consumption cannot be rejected at conventional significance levels for the vast majority of countries, I consider first differences of the dependent

NU

variable.10 Second, I do not logarithmize either the dependent variables or the volume of newly discovered petroleum in giant oil fields, because both series contain “true zeros” rather than

MA

missing values. Third, the fact that yit is expressed in tbpd while dit is expressed in mmboe facilitates computing the expected duration until a discrete increase in the reserve base is offset by higher production from newly discovered oil fields or existing reservoirs.11 Fourth, Horn’s

ED

(2014) data set reports three hydrocarbon volumes in mmboe for each discovery: the estimated ultimate recovery (EUR), the estimated ultimate recovery of oil (EURO), and reserves (RSVS),

PT

where EUR ≥ EURO ≥ RSVS. Given my interest in the response of country-level oil production and consumption, respectively, I use the EURO to measure the absolute size of a giant oil field

AC CE

discovery in what follows.

The panel structure of the data allows identifying the impulse responses of oil production and consumption to a giant oil field discovery while simultaneously controlling for country and year fixed effects.12 Accepting that their timing and size are exogenous to changes in country-specific production and consumption levels after controlling for country and year fixed effects, giant oil 10

At the 10% level or better, the null hypothesis of a unit root in country-level production is rejected in favor of a trend-stationary alternative in 14 out of 80 or 17.5% of all cases (see Table A.3 in the appendix). The null hypothesis of a unit root in country-level consumption is rejected in 20 out of 172 or 11.6% of all cases (see Table A.4). Note that the standard caveats about the power of unit root tests in finite samples apply here as well. 11 In contrast to Aretzki et al. (2017), there is no need to derive the net present value of a giant oil field discovery, given that I am only interested in the quantitative response of crude oil production or consumption to an exactly quantifiable discovery. Rather than applying country-specific risk-adjusted discount rates and imposing a certain production profile from the newly found reservoir, the present paper investigates the actual response of petroleum production and consumption at the country level. 12 In theoretical work, Deshmukh and Pliska (1983) model the optimal consumption of a nonrenewable resource under uncertainty about the discovery process and the economic environment. Given that “a general comparison of different environments is not easy.” (Deshmukh and Pliska, 1983, p. 549), it is important to control for “the economic environment” using time and country fixed effects.

10

ACCEPTED MANUSCRIPT discoveries correspond to quasi-natural experiments. Country i belongs to the treatment group, if an oil field has been discovered on its territory in period t, whereas all countries without a discovery on their territories in period t belong to the control group. At each point in time, we can therefore draw on a large number of countries that have not been treated. It is important to note that the dynamic panel specification in (1) also accommodates the fact that a country might have discovered a giant oil field on its territory in period t − l, 1 ≤ l ≤ q. Due to the infrequent incident of a giant oil discovery, the high adjustment costs in petroleum

PT

production (see, e.g., Hamilton, 2009; Kilian, 2009; G¨ untner, 2014), and the gestation lag before a new reservoir can come on line, it is crucial to draw on a sufficiently long and broad panel. By

RI

including the autoregressive term A(L) in the panel DLR model in (1), I control for potential

SC

serial correlation of changes in country-level production or consumption and am able to construct the average impulse response function beyond a forecast horizon of q years as (2)

The response of OPEC producers

MA

4.1

B (L) . 1 − A (L)

NU

Φ (L) =

In order to investigate whether a country’s OPEC membership has any influence on its response to a giant oil field discovery, I introduce an additional regressor in (1) that interacts the dummy

ED

variable opeci ∈ {0, 1} with dit , while again controlling for country and year fixed effects. (3)

PT

∆yit = A (L) ∆yit + B (L) dit + C (L) dit · opeci + αi + µt + it ,

where C(L) is a qth order lag polynomials with q ≥ 0 and it a country-year-specific error term.

AC CE

opeci = 1 for the 14 OPEC member countries in my panel and opeci = 0 else. As argued above, I do not account for the fact that Indonesia, Ecuador, and Gabon temporarily suspended their OPEC memberships during the sample period, while Angola joined only in 2007. If anything, this biases the estimate of behavioral differences between the two groups downwards. From equation (3), I compute the impulse response function of production or consumption in non-OPEC countries as in (2) and the average response of OPEC countries according to Φopec (L) =

B (L) + C (L) , 1 − A (L)

(4)

where C(L)/ [1 − A (L)] equals the marginal response of petroleum production or consumption in OPEC member states to a giant oil field discovery of one mmboe in terms of EURO.

11

ACCEPTED MANUSCRIPT 4.2

Onshore vs. offshore discoveries

Due to the possibility of different gestation lags, in an extension of (1), I also allow for differences in the response to onshore and offshore discoveries (see Aretzki et al., 2017). of f ∆yit = A (L) ∆yit + B (L) don it + C (L) dit + αi + µt + it ,

(5)

of f where don denotes the EURO of giant oil fields discovered onshore and offshore, it and dit

PT

respectively, in country i and year t. Note that the specification in (5) accommodates multiple discoveries in different locations in a given country and year, weighting them by their respective

RI

EURO in mmboe.13 In equation (3), instead, each country either is an OPEC member or not. From equation (5), it is straightforward to compute the impulse responses of country-level

B (L) 1 − A (L)

Does field size matter?

Φof f (L) =

C (L) , 1 − A (L)

(6)

MA

respectively.

4.3

and

NU

Φon (L) =

SC

production or consumption to an onshore and offshore discovery of one mmboe of EURO as

As a final extension, I Investigate whether oil producers respond differently to the discovery of a

ED

giant as opposed to the discovery of a super- or mega-giant oil field, defined as an oil field with EURO ≥ 3 billion boe or EURO ≥ 3, 000 mmboe. For this purpose, I distinguish discoveries in

PT

the dynamic panel DLR model according to their size class. ∆yit = A (L) ∆yit + B (L) dgit + C (L) dsg it + αi + µt + it ,

(7)

AC CE

where dgit and dsg it denotes the EURO of newly discovered giant and super-/mega-giant oil fields, respectively, in country i and year t. Similar to (5), the specification in (7) accommodates the simultaneous discovery of one (or multiple) giant and super-/mega-giant oil fields in a given country and year, weighting them by their respective EURO in mmboe. The corresponding impulse responses of country-level production or consumption to a giant and super- or mega-giant discovery of one mmboe of EURO can then be computed as Φg (L) =

B (L) 1 − A (L)

and

Φsg (L) =

C (L) , 1 − A (L)

(8)

respectively. 13

For example, Saudi Arabia discovered four giant onshore fields (Jawb, Lughfah, Samin, and Dhib) and one giant offshore field (Hamur) in 1979 but not a single field in 1980 and 1981.

12

ACCEPTED MANUSCRIPT 5

Empirical Results

The key question of this paper is how country-level petroleum production responds to a giant oil field discovery. While newly found reservoirs cannot be tapped without a substantial gestation lag, production from existing fields may nevertheless adjust in response to a reserves discovery, even in the presence of nontrivial adjustment costs, if a country is producing below its potential or increases its drilling activity, for example. In what follows, I therefore analyze whether and

PT

how fast country-level petroleum production adjusts in response to a giant oil field discovery, and whether the associated impulse response function depends on the country’s OPEC member

RI

status or the field’s size and location. In a second step, I consider the response of countrylevel petroleum consumption to investigate whether any adjustment in production is consumed

SC

within the country or exported to raise national income from resource extraction, in line with

5.1

The response of oil production

NU

the narrative in Aretzki et al. (2017).

MA

Panel A of Table 2 reports the coefficient estimates for the DLR model in (1) with p = 1 and p = 4. For either specification, the null hypothesis of jointly insignificant coefficients on current and lagged giant oil field discoveries is strongly rejected (F -statistic=12.52, p-value=0.00 for

ED

p = 1; F -statistic=12.83, p-value=0.00 for p = 4). The test for fixed effects models proposed by Baltagi and Li (1995) indicates no significant serial correlation in the regression residuals (LM statistic=0.76, p-value=0.38 for p = 1; LM -statistic=0.02, p-value=0.90 for p = 4). Inspection

PT

of the probability density estimates of the standardized residuals in Figure A.1 in the appendix does not reveal obvious deviations from normality. Since the autocorrelation function of the

AC CE

squared residuals indicates the presence of conditional heteroskedasticity, heteroskedasticityand autocorrelation-robust (HAC-robust) standard errors are reported in parentheses. Figure 2 plots the cumulated impulse response functions for the entire panel of countries and corresponds thus to the average response of country-level petroleum production in tbpd to a discovery of one mmboe of EURO.14 The broken and dotted lines indicate one- and two-standard-error bootstrap confidence intervals based on the HAC-robust covariance matrix in Newey and West (1987). In response to a giant oil discovery, petroleum production in the treatment group increases within the same year and continues to rise over the subsequent four years relative to the control group. Depending on the lag order p, the corresponding impulse response function is statistically significant at an approximate 5% level for the first 1–3 years. Given a plausible gestation lag of several years between the discovery and the start of production from a new oil field, this finding 14

Recall that the mean EURO of all giant oil field discoveries in my sample of Horn (2014) is 2,265 mmboe.

13

ACCEPTED MANUSCRIPT suggests that producers may raise their output from existing fields.15 Six years after the discovery, petroleum production in the treatment group has increased by 0.04 tbpd per mmboe of EURO relative to the control group, corresponding to an increase of 90.6 tbpd for a giant oil field of average EURO. While seemingly small, this figure rivals the production level of Italy in the 1990s. The increase is even more pronounced about seven years after the discovery, peaking at an additional 0.08 tbpd per mmboe of EURO after 13 years. At this horizon, the relative increase in petroleum production is highly statistically significant,

PT

independent of the lag order p. To put the response of country-level production into perspective, we can also compute the hypothetical lifetime until the newly discovered reserves are exhausted.

RI

Assuming an average increase of 0.08 tbpd or 365 · 0.08 = 29.2 tbpy per mmboe of EURO, they

SC

would last for about 34.25 years. Given that I abstract from any type of heterogeneity, such as in field size and location, this figure should be taken with a grain of salt.

NU

On the one hand, my results for the full panel confirm the common wisdom of a substantial gestation lag before a newly discovered reservoir comes on line. On the other hand, country-level production seems to increase before a new oil field is tapped, indicating that producers raise

MA

extraction rates from existing reservoirs in order to exploit any remaining spare capacity. In light of the finding in Anderson et al. (2018) that production from existing oil wells is physically constrained by reservoir pressure, an increase in country-level oil production should bring about

ED

an increase in investment or drilling activity in order to facilitate production from new wells. Given that country-level data on investment in oil exploration and development is generally not

PT

available, I compile annual data on the number of active oil and gas rigs from the Baker Hughes International Rotary Rig Count for 65 countries, 41 of which experienced at least one giant oil

AC CE

field discovery since 1945.16 The rig count data start in 1975 for the U.S. and Canada and in 1982 for the remaining countries. Former members of the USSR and Russia are not in the data. As a consequence, my panel decreases along both the time and the country dimension. Figure 3 plots the cumulated impulse response functions based on the DLR model in (1), where ∆yit denotes the annual change in the number of oil and gas rigs in operation in country i and year t. For both p = 1 and p = 4, rig activity slightly decreases for about two years following a giant oil field discovery before it increases and remains (partially) statistically significant for the rest of the forecast horizon. For either specification, the impulse response function peaks after 14 years. Given an average EURO of 2265 mmboe in my sample, the point estimates 15 Aretzki et al. (2017) dismiss this possibility by showing that their main results are robust to removing the world’s ten largest oil or gas exporters from the sample, which are arguably more likely to hold spare capacity. 16 Note that the Baker Hughes rig count splits between oil and gas rigs from 1995 onwards only. For 1995–2016, the contemporaneous correlation between (changes in) oil and gas rigs in operation at the country level is 0.04 with a p-value of 0.16. Accordingly, the combination of oil and gas rigs may cause measurement error and less precise coefficient estimates, while it is not expected to induce any directional bias in the results.

14

ACCEPTED MANUSCRIPT in Figure 3 imply an additional 66.1 and 101.5 active rigs for p = 1 and p = 4, respectively. Although it is true that the rise in country-level oil production is accompanied by an increase in drilling activity, the number of active rigs seems to rise with a short delay and fails thus at explaining the production response in the first 1–3 years. Accordingly, the rise in country-level production in response to a giant oil field discovery reflects increased drilling activity in the medium run.17 In the short run, production from existing reservoirs may be stimulated even without increasing the number of rotary rigs in

PT

operation, for example by injecting water into an oil field in order to raise the reservoir’s pressure. In contrast to a demand-driven change in oil prices, which is of uncertain persistence,

RI

a giant oil discovery signals a substantial increase in reserves and production capacity in the

SC

near future. Moreover, high adjustment costs in petroleum production (see, e.g., Hamilton, 2009; Kilian, 2009) may be less relevant over a horizon of several years. For these reasons, the

NU

above finding complements rather than competes with the finding of a near-zero short-run price elasticity of oil supply in G¨ untner (2014), as a country’s production capacity may rise over a

5.2

MA

horizon of several years.

The response of OPEC producers

An obvious candidate for heterogeneity in the response of production to a giant oil field discovery

ED

is a country’s OPEC membership. Of the 490 oil fields in Horn’s (2014) data that were discovered during my sample period (i.e. 1945–2010), 227 are located on current OPEC territories. In this

PT

section, I therefore investigate whether OPEC membership influences the dynamic response of country-level oil production to a discovery of one mmboe of EURO.

AC CE

Panel B of Table 2 reports the coefficient estimates for the DLR model in (3). Both the coefficients on current and lagged giant oil field discoveries in B(L) (F -statistic=3.15, p-value=0.00 for p = 1; F -statistic=2.71, p-value=0.00 for p = 4) and the coefficients on the interaction terms with the OPEC dummy in C(L) (F -statistic=6.40, p-value=0.00 for p = 1; F -statistic=7.65, p-value=0.00 for p = 4) are jointly statistically significant. Baltagi and Li’s (1995) test for panel fixed effects models indicates no significant serial correlation in the regression residuals (LM -statistic=1.24, p-value=0.27 for p = 1; LM -statistic=0.29, p-value=0.59 for p = 4). With regard to the probability density estimates based on the standardized residuals in Figure A.2 in the appendix, deviations from normality seem to be modest, while HAC-robust standard errors are reported in parentheses to address evidence of serial correlation in the squared residuals. The upper panels of Figure 4 plot the cumulated impulse responses of OPEC production, while 17 Note that drilling effort need not be confined to the newly discovered oil field. Instead, additional wells may be drilled in existing reservoirs in order to raise their production capacity and extraction rate in the medium run.

15

ACCEPTED MANUSCRIPT the lower panels plot the cumulated impulse responses of non-OPEC production to a giant oil field discovery for p = 1 and p = 4, respectively. The broken and dotted lines indicate one- and two-standard-error bootstrap confidence intervals based on the HAC-robust covariance matrix in Newey and West (1987). Figure 4 reveals both qualitative and quantitative differences in the responses to a giant oil field discovery. On average over the sample period, non-OPEC production increases significantly already during the first five years, whereas OPEC production barely responds until about seven

PT

years after the discovery. Moreover, while production of treated OPEC members peaks at +0.08 tbpd after 13 years, that of non-OPEC producers is up by the same amount within four years

RI

and peaks at +0.11 tbpd relative to the control group after 18 years. Finally, the response of

SC

non-OPEC production is statistically significant at an approximate 5% level throughout, while the response of OPEC production is only significant (at the 5% level) between nine and 15 years

NU

after the discovery. Note that these findings are robust to the lag order p. Our previous findings for the entire panel of countries in Figure 3 and the different production responses in Figure 4 suggest that there might be differences also in the response of OPEC and

MA

non-OPEC producers’ drilling activity to a giant oil field discovery. For this reason, I re-estimate the extended DLR model (3) using the Baker Hughes rig count data described in the previous section. The upper panels of Figure 5 plot the cumulated impulse response functions of OPEC

ED

rigs, while the lower panels plot the cumulated impulse response functions of non-OPEC rigs in operation. Broken and dotted lines indicate one- and two-standard-error bootstrap confidence

PT

intervals. In line with the delayed and statistically insignificant response of OPEC production in Figure 4, the number of active oil and gas rigs is virtually constant over the entire forecast

AC CE

horizon. In contrast, non-OPEC drilling activity starts to increase after three years and becomes (at least partially) statistically significant from year four onwards.18 Regardless of the lag order p, the impulse response function peaks after 14 years. Given an average EURO of 1500 mmboe for non-OPEC discoveries, the point estimates in Figure 5 imply an additional 95.4 and 138.9 active rigs for p = 1 and p = 4, respectively. The rise of non-OPEC production and drilling activity in Figures 4 and 5 is consistent with the theoretical predictions of the classical Hotelling (1931) model and, in particular, its extensions to allow for stochastic resource discoveries (see, e.g., Deshmukh and Pliska, 1980, 1983). From a competitive firm’s perspective, the discovery represents an unexpected increase in its reserve base that shifts the optimal extraction path upwards. In the short run, extraction rates from existing fields can be raised, for example by increasing reservoir pressure through 18

Recall that the Baker Hughes rig count data cumulates oil and gas rigs, which causes measurement error and wider confidence intervals on the impulse response functions in Figures 3 and 5.

16

ACCEPTED MANUSCRIPT waterflooding. In the medium run, additional oil wells can be drilled either in existing reservoirs or in the newly discovered oil field.19 Given the size of the effect on production and the implied hypothetical lifetime of 34.25 years for the average giant oil field in my sample, it seems sufficient that extraction rates are raised for a limited number of existing reservoirs, such as those operated by companies holding interest in the newly discovered field. Accordingly, the observed increase does not necessarily require that such efforts to raise production are coordinated at the national level, which seems less plausible in non-OPEC countries, as oil fields are operated by private

PT

competitive rather than state-owned monopolist companies. On average over the sample period, there is no evidence of strategic behavior among non-OPEC producers that is at odds with the

RI

classical Hotelling model, such as curbing production in response to a windfall resource gain.

SC

At the same time, there are several candidate explanations for the (lack of a) response of OPEC producers to a giant oil field discovery. Given that oil exploration and development

NU

in OPEC member states is often centralized in a state-owned company such as Saudi Aramco (officially the Saudi Arabian Oil Company), for example, decision makers might respond not only to economic incentives but take political considerations into account as well (see, e.g.,

MA

Roumasset et al., 1983). Adherents of the popular theory of OPEC as an effective cartel might argue that its members refrain from producing out of newly discovered reserves or — using the language of Peterson (1978) — “over-conserve the resource.” However, this seems

ED

inconsistent with the observation that the impulse response function of OPEC production is (at least partially) statistically significant after about 10 years. Although the cartel hypothesis

PT

has long been overturned, OPEC members might be reluctant to produce out of a giant oil field discovery in order to sustain a higher oil price.20 Similarly, members of OPEC might

AC CE

postpone production to conserve newly discovered oil reserves, which represent the main or only source of government revenue in these countries. On the contrary, the different impulse responses in Figures 4 and 5 are not due to OPEC producers operating at full capacity and non-OPEC producers operating below capacity. Non-OPEC producers are generally regarded as price takers that tend to produce at or close to full capacity rather than trying to influence market prices by managing their production, whereas OPEC producers — in particular Saudi Arabia — historically held spare capacity for market management. As a result, the observed behavioral differences are more likely due to economic and political considerations. In what 19 Using field-level data on Norwegian offshore oil production, Mauritzen (2017) also finds stronger lagged effects of oil prices on production in declining fields, for example through spare capacity and increased drilling activity, and concludes that policy measures to stimulate production are more effective in mature oil provinces, such as the Norwegian Continental Shelf. 20 This hypothesis is consistent with the finding in Alhajji and Huettner (2000) that target revenue theory (TRT) helps explain production decisions in centrally planned as opposed to free-market economies. According to the TRT, oil exporters’ targeted revenues are effectively determined by their investment needs.

17

ACCEPTED MANUSCRIPT follows, I analyze whether the results could also arise from different geological conditions such as field location and size, for example.

5.3

Onshore vs. offshore discoveries

Aretzki et al. (2017) report qualitative differences in the responses of consumption, employment, saving/GDP, and investment/GDP between giant onshore and offshore oil field discoveries. In this section, I therefore investigate whether the documented heterogeneity with respect to field

PT

location carries over to the response of country-level petroleum production.

The upper panels of Figure 6 plot the cumulated impulse responses to a giant onshore oil

RI

field discovery, while the lower panels plot the cumulated impulse responses to a giant offshore

SC

oil field discovery based on the extended DLR model in (5) for p = 1 and p = 4, respectively. The broken and dotted lines indicate HAC-robust one- and two-standard-error confidence intervals. After a giant onshore discovery, country-level production increases on impact, whereas the

NU

impulse response functions are not statistically different from zero at an approximate 5% level for the rest of the 20-year horizon. Relative to the control group, production of the treated countries

MA

increases by less than six tbpd per mmboe of EURO. In contrast, country-level production does not respond on impact but becomes increasingly positive and statistically significant from about four years after a giant offshore discovery. It is important to note that, for either specification,

ED

the impulse response function peaks at an additional 0.17 tbpd per mmboe of EURO relative to the control group after 16 years, i.e. three times the increase for a giant onshore discovery. This

PT

difference is statistically significant at an approximate 32% level and robust to the lag order p. Of the 490 giant oil fields discovered during 1945–2010, 300 (61.2%) are located onshore.

AC CE

The remaining 190 offshore discoveries (38.8%) account for the significant increase in petroleum production relative to the control group in Figure 2. As a consequence, one may conjecture that field location is correlated with ownership, explaining the similarity of the respective impulse response functions for OPEC and non-OPEC countries in Figure 4 with those for onshore and offshore discoveries in Figure 6. Indeed, of the 227 giant oil fields discovered by OPEC member states, 69 (30.4%) are located offshore, while 121 (46.0%) of the 263 giant oil fields discovered on non-OPEC territory are located offshore. The larger propensity to produce out of offshore reservoirs seems to be linked to the fact that a larger fraction of offshore discoveries is located on non-OPEC territory and thus more likely to be used either directly or by raising production from existing fields.

18

ACCEPTED MANUSCRIPT 5.4

Does field size matter?

In my sample period, the average EURO of giant onshore and offshore discoveries is very similar, regardless of whether they are found on OPEC or non-OPEC territory (see Table 1). At the same time, the average EURO of OPEC fields (3122 mmboe) is about twice as large as that of non-OPEC fields (1559 mmboe). In this section, I therefore investigate whether field size matters. While, in the previous analysis, discoveries are quantified by their EURO, Horn (2014)

PT

also classifies each oil field as a “giant” (89.8%), “super-giant” (9.8%), or “mega-giant” (0.4%). In what follows, I analyze whether oil production responds differently to the discovery of a giant

RI

as opposed to the discovery of a super- or mega-giant, roughly defined as an oil field with an EURO ≥ 3 billion boe (i.e. EURO ≥ 3, 000 mmboe).

SC

The upper panels of Figure 7 plot the cumulated impulse responses to a giant oil discovery, while the lower panels plot the cumulated impulse responses to a super- or mega-giant discovery

NU

based on the extended DLR model in (7) for p = 1 and p = 4. Broken and dotted lines indicate HAC-robust one- and two-standard-error confidence intervals, respectively.

MA

The discovery of an “ordinary” giant oil field does not trigger a statistically significant increase in petroleum production relative to the control group, except for a partially significant increase between one and three years after the discovery. This finding is robust to the lag order

ED

p. Considering the lower panels, the discovery of an exceptionally large super- or mega-giant oil field seems to raise petroleum production in the treatment group by up to 0.09 tbpd per mmboe of EURO after 13 years. Regardless of the specification in (7), this increase is significant at an

PT

approximate 5% level between six and 16 respectively nine and 13 years after the discovery. The results in Figure 7 suggest that the discovery of an “ordinary” giant oil field induces

AC CE

producers to extract more from existing reservoirs in the short run, whereas the discovery of an exceptionally large super- or mega-giant oil field such as the Alaskan Prudhoe Bay and the Saudi Arabian Ghawar, for example, triggers a delayed and persistent increase in petroleum production levels. Conditional on the event of a giant oil field discovery, the likelihood of a super- or megagiant discovery is about twice as large for OPEC member states (13.2%) relative to non-OPEC countries (7.6%). Field size alone is therefore unlikely to explain the heterogeneity in Figure 4. Nevertheless, the above findings reflect effective differences in country-level production behavior conditional on the size of a resource discovery. Apparently, field size matters.

5.5

The response of oil consumption

In this section, I investigate whether the increase in petroleum production in Figure 2 is accompanied by an equally sized increase in domestic consumption or rather destined for raising (net)

19

ACCEPTED MANUSCRIPT exports. Ideally, I would like to use a direct measure of net exports as the dependent variable. However, missing data on crude oil inventories above the ground for most countries implies that {net exports}t = {production}t − {consumption}t − ∆ {inventories}t contains an unknown variable on the right-hand side. For this reason, I substitute petroleum consumption for petroleum production in the DLR models in (1) and (3). The availability of annual consumption data starting in 1965 for a total of 172 countries implies that, at each

PT

point in time, I can draw on an even larger control group of countries without a giant oil field discovery on their territories.

RI

Figure 8 plots the cumulated impulse response functions of ∆yit to a giant oil field discovery

SC

for the entire panel of countries based on the DLR model in (1). This corresponds to the average response of country-level petroleum consumption in tbpd to a discovery of one mmboe of EURO.

NU

Again, the broken and dotted lines indicate one- and two-standard-error HAC-robust bootstrap confidence intervals. Following a typical discovery, petroleum consumption in the treatment group hardly responds on impact. From period one onwards, the level of consumption starts

MA

to rise relative to the control group, peaking at a cumulated +0.025 and +0.045 tbpd after ten to twelve years for p = 1 and p = 4, respectively. While the impulse response function is only partially significant at an approximate 32% level for p = 1, it is statistically significant at the

ED

5% level for p = 4 around five years after the discovery. A comparison of Figures 2 and 8 reveals that domestic petroleum consumption increases by

PT

less than domestic production. Although the differences in the corresponding impulse response functions are not statistically significant, my findings suggest that at least part of the increase in

AC CE

production raises either the country’s net oil exports or crude oil inventories above the ground. Given that the difference between petroleum production and consumption persist over a horizon of 20 years, a giant oil field discovery eventually translates into higher net exports.21 Now consider Figure 9, which plots the impulse response functions of cumulated changes in OPEC and non-OPEC consumption after a giant oil field discovery based on the DLR model in (3), where broken and dotted lines indicate one- and two-standard-error HAC-robust bootstrap confidence intervals. While the upper-right panel suggests that, for p = 1, an initial increase of petroleum consumption in OPEC member states is followed by a persistent decrease over the remaining forecast horizon that is statistically significant at the 5% level after twelve years, the response for p = 4 in the upper-left panel seems to be virtually flat. In contrast, the lower panels 21

Note that the cumulated impulse response functions in Figures 2 and 8 correspond to flow variables, whereas crude oil inventories are a stock variable. A persistent difference between petroleum production and consumption must therefore translate into a continuous increase in crude oil inventories in response to a giant oil field discovery, unless net oil exports increase.

20

ACCEPTED MANUSCRIPT of Figure 9 show that petroleum consumption in non-OPEC countries increases in response to a giant oil field discovery, peaking at +0.07 tbpd after twelve years. The latter impulse response functions are statistically significant at an approximate 5% level between three and 13 years after the discovery and robust to the lag order p. It is important to note that the differences between the impulse response functions of OPEC and non-OPEC consumption are statistically significant at the 5% (32%) level between ten and 13 years after the discovery for p = 1 (p = 4). Accordingly, the increase of petroleum production

PT

in the treatment group relative to the control group in Figure 8 is driven by non-OPEC countries,

Conclusion

SC

6

RI

whereas the level of consumption in OPEC member states remains constant or even decreases.

In this paper, I investigate how domestic petroleum production and consumption respond to a

NU

sizeable increase in a country’s reserve base. The use of a comprehensive data set of arguably exogenous giant oil field discoveries facilitates analyzing the adaptive behavior of oil producers in a quasi-natural experiment. In an unbalanced panel of oil production and consumption data

MA

starting in 1965 for 80 and 172 countries, respectively, many of which have never experienced a giant oil field discovery, I can control for country and year fixed effects. Each impulse response

ED

function therefore corresponds to the difference between a country treated with a giant oil field discovery in its recent past and all countries in the control group. I find that country-level petroleum production starts to increase before the newly discovered

PT

field has come on line, indicating that the discovery spurs production from existing reservoirs in the presence of spare capacity. At the same time, rig activity increases at the country level.

AC CE

When distinguishing between OPEC and non-OPEC producers, I find that the response of the former is delayed relative to that of the latter and quantitatively less pronounced. Consistently, country-level rig activity increases primarily following a giant oil field discovery on non-OPEC territory. The increase in petroleum production seems to arise mainly from offshore rather than onshore discoveries, which are relatively more concentrated on non-OPEC territory, and from “super-” and “mega-giant” rather than ordinary “giant” oil fields. Finally, I show that country-level petroleum consumption also increases in response to a giant oil field discovery — albeit by quantitatively less and statistically less significantly. Depending on the specification of the panel distributed lag regression model, OPEC consumption decreases whereas non-OPEC consumption increases significantly in response to a giant oil field discovery, suggesting that the increase in reserves is devoted to different aims in the two groups. These findings shed light on the actual response of country-level petroleum production to

21

ACCEPTED MANUSCRIPT a giant oil field discovery rather than imposing a certain production profile and dismissing the possibility of adjusting extraction rates from existing reservoirs, as in Aretzki et al. (2017). By comparing the impulse response functions of country-level production and consumption, I can comment on the likely use of increased domestic petroleum supply. A more rigorous analysis of the effects of giant oil discoveries on the current account requires panel data on net oil exports or crude oil inventories. Given the current lack of such data, this is left for future research. Of the 63 giant oil fields discovered during 2000–2010, 45 (71.4%) are located on non-OPEC

PT

territory and 49 (77.7%) offshore, while the conditional probability of discovering a “super-” or “mega-giant” oil field has decreased from 10.2% during the entire sample period (i.e. 1945–2010)

RI

to a mere 6.3%. Moreover, the average size of a non-OPEC discovery in terms of its estimated

SC

ultimate recovery of oil (EURO) during 2000–2010 was larger than its OPEC counterpart, due to the fact that the latter had fallen more relative to 1945–2010. In light of the results in this

NU

paper, these changes in summary statistics suggest that the scales have tipped in favor of fields that are more likely to come on line, as they are located offshore and/or on non-OPEC territory. The shale oil boom in the U.S. contributes to this development. In the medium to long run,

MA

the different propensities to produce out of (newly discovered) conventional and unconventional reservoirs imply that an increasing share of the world’s remaining oil reserves will be located in OPEC member states. This might eventually give the go-ahead to effective cartelization.

ED

One limitation of Horn’s (2014) data is that the EURO of various oil fields was revised after its initial appraisal.22 If the EURO of a giant oil field discovery is revised downwards after its

PT

discovery, the response of country-level production may be biased upwards due to measurement error. In the more likely event that the EURO of an oil field is revised upwards (e.g. due to

AC CE

technical progress), the coefficient estimates in this paper are biased downwards. Moreover, the absence of small and medium-sized oil field discoveries in Horn’s (2014) data set raises the question of omitted-variable bias. If small and medium-sized discoveries are systematically clustered with giant discoveries, the impulse response functions may be biased upwards. If small and medium-sized discoveries in the control group are clustered with giant discoveries in the treatment group, however, the coefficients are biased downwards. Lei and Michaels (2014) use data from the Oil and Gas Journal Data Book (2008), which records country-year pairs where some oil discoveries — not necessarily giant — were made, to mitigate the concern that country-year observations with oil field discoveries differ from other 22 For example, Prudhoe Bay oil field was originally estimated to contain 9.6 billion barrels of recoverable oil, nearly double the size of the largest field ever previously found in North America. Since the start of production in 1977, it has generated more than 12.5 billion barrels of oil with an estimated 4 billion barrels remaining (see BP Prudhoe Bay factsheet). Similar upward revisions of the EURO are reported for the Saudi Arabian Ghawar — the largest known conventional oil field in the world.

22

ACCEPTED MANUSCRIPT observations not only across but also within countries. Because this robustness check relies on commercial data, its replication is beyond the scope of this paper. Another interesting question is whether the increase in country-level petroleum consumption is driven by private households or firms. Given that only a small amount of crude oil is consumed directly, while the largest part is refined into petroleum products, such as gasoline or heating oil, the EIA uses “petroleum supplied” — the disappearance of products from petroleum refineries, natural gas processing plants, blending plants, pipelines, and bulk terminals — as a

PT

proxy for domestic petroleum consumption. Studying the response of household and industry consumption of gasoline and other petroleum products to a giant oil field discovery is beyond

Acknowledgements

SC

7

RI

the scope of this paper and left for future research.

NU

I thank Christiane Baumeister, Yuriy Gorodnichenko, the editor, and three anonymous referees for helpful comments. The work on this topic started while the author was visiting the University of California at Berkeley and Ludwig Maximilians University Munich. Financial support by the

References

ED

8

MA

DAAD and the European Union is gratefully acknowledged.

Alhajji, A. F. and D. Huettner (2000). OPEC and Other Commodity Cartels: A Comparison.

PT

Energy Policy 28, 1151–1164.

Anderson, S. T., R. Kellogg, and S. W. Salant (2018). Hotelling under pressure. Journal of

AC CE

Political Economy 126, 984–1026. Aretzki, R., V. A. Ramey, and L. Sheng (2017). News shocks in open economies: Evidence from giant oil discoveries. Quarterly Journal of Economics 132, 103–155. Arrow, K. J. and S. Chang (1982). Optimal pricing, use, and exploration of uncertain natural resource stocks. Journal of Environmental Economics and Management 9, 1–10. Baltagi, B. H. and Q. Li (1995). Testing AR(1) against MA(1) disturbances in an error component model. Journal of Econometrics 68, 133–151. Black, G. and J. T. LaFrance (1998). Is Hotelling’s rule relevant to domestic oil production? Journal of Environmental Economics and Management 36, 149–169. Cuddington, J. T. and D. L. Moss (2001).

Technological change, depletion, and the u.s.

petroleum industry. American Economic Review 91, 1135–1148. 23

ACCEPTED MANUSCRIPT Dahl, C. and T. E. Duggan (1998). Survey of price elasticities from economic exploration models of US oil and gas supply. Journal of Energy Finance & Development 3, 129–169. Deshmukh, S. D. and S. R. Pliska (1980). Optimal consumption and exploration of nonrenewable resources under uncertainty. Econometrica 41, 177–200. Deshmukh, S. D. and S. R. Pliska (1983). Optimal consumption of a nonrenewable resource with

PT

stochastic discoveries and a random environment. Review of Economic Studies 50, 543–554. Erickson, E. W. and R. M. Spann (1971). Supply response in a regulated industry: The case of

RI

natural gas. The Bell Journal of Economics and Management Science 2, 94–121.

SC

Fagan, M. N. (1997). Resource depletion and technical change: Effects on U.S. crude oil finding costs from 1977 to 1994. The Energy Journal 18, 91–105.

NU

Farrow, S. (1985). Testing the efficiency of extraction from a stock resource. Journal of Political Economy 93, 452–487.

Energy Economics 23, 271–291.

MA

Farzin, Y. H. (2001). The impact of oil price on additions to US proven reserves. Resource and

Fisher, F. M. (1964). Supply and Costs in the US Oil and Gas Industry: Two Econometric

ED

Studies. Baltimore: John Hopkins Press.

Economics 44, 1–13.

Energy

PT

G¨ untner, J. H. F. (2014). How do oil producers respond to oil demand shocks?

AC CE

Hamilton, J. D. (2009). Causes and consequences of the oil shock of 2007–08. Brookings Papers on Economic Activity 40, 215–283. Hamilton, K. and G. Atkinson (2013).

Resource discoveries, learning, and national in-

come accounting. World Bank Policy Reserach Working Paper 6505 (available at SSRN: https://ssrn.com/abstract=2285342). Harding, T., R. Stefanski, and G. Toews (2016). Boom goes the price: Giant resource discoveries and real exchange rate appreciation. School of Economics and Finance Discussion Paper 1608 (available at https://research-repository.st-andrews.ac.uk/handle/10023/8864). H¨o¨ok, M., R. Hirsch, and K. Aleklett (2009). Giant oil field decline rates and their influence on world oil production. Energy Policy 37, 2262–2272.

24

ACCEPTED MANUSCRIPT Horn, M. K. (2014). Giant oil and gas fields of the world. Data set retrieved from Energy Data Exchange (EDX) on September 27, 2018, https://edx.netl.doe.gov/dataset/ aapg-datapages-giant-oil-and-gas-fields-of-the-world. Hotelling, H. (1931). The economics of exhaustible resources. Journal of Political Economy 39, 137–175. Iledare, O. O. and A. G. Pulsipher (1999). Sources of change in petroleum drilling productivity

PT

in onshore Louisiana in the US, 1977–1994. Energy Economics 21, 261–271.

RI

Kilian, L. (2009). Not all oil price shocks are alike: Disentangling demand and supply shocks

SC

in the crude oil market. American Economic Review 99, 1053–1069.

Lei, Y.-H. and G. Michaels (2014). Do giant oilfield discoveries fuel internal armed conflicts?

NU

Journal of Development Economics 110, 139–157.

Lepore, J. J. and M. Rastad (2011). Do oil producers extract more as they become richer? A natural experiment using unanticipated reserve discoveries. Working Paper (available at

MA

SSRN: http://dx.doi.org/10.2139/ssrn.1927652).

Lindholt, L. (2015). The tug-of-war between resource depletion and technological change in the

ED

global oil industry 1981–2009. Energy 93, 1607–1616. Litzenberger, R. H. and N. Rabinowitz (1995). Backwardation in oil futures markets: Theory

PT

and empirical evidence. Journal of Finance 50, 1517–1545.

AC CE

Managi, S., J. J. Opaluch, D. Jin, and T. A. Grigalunas (2005). Technological change and petroleum exploration in the Gulf of Mexico. Energy Policy 33, 619–632. Mauritzen, J. (2017). The effect of oil prices on field production: Evidence from the Norwegian Continental Shelf. Oxford Bulletin of Economics and Statistics 79, 124–144. Mohn, K. (2008). Efforts and efficiency in oil exploration: A vector error-correaction approach. The Energy Journal 29, 53–78. Mohn, K. and P. Osmundsen (2008). Exploration economics in a regulated petroleum province: The case of the Norwegian Continental Shelf. Energy Economics 30, 303–320. Newey, W. K. and K. D. West (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica 55, 703–708.

25

ACCEPTED MANUSCRIPT Norwegian Petroleum Directorate (2011). Petroleum resources on the Norwegian Continental Shelf. The Resource Report 2011 . Oil and Gas Journal Data Book (2008). Petroleum Publishing Company, Tulsa, Oklahoma. Pesaran, H. M. (1990). An econometric analysis of exploration and extraction of oil in the U.K. Continental Shelf. The Economic Journal 100, 367–390.

PT

Pesaran, H. M. and C. A. Favero (1994). Oil investment in the north sea. Economic Modelling 11, 308–329.

SC

Environmental Economics and Management 5, 236–251.

RI

Peterson, F. M. (1978). A model of mining and exploring for exhaustible resources. Journal of

Pickering, A. (2008). The oil reserves production relationship. Energy Economics 30, 352–370.

NU

Pindyck, R. S. (1974). The regulatory implications of three alternative econometric supply models of natural gas. The Bell Journal of Economics and Management Science 5, 633–645.

MA

Pindyck, R. S. (1978). The optimal exploration and production of nonrenewable resources. Journal of Political Economy 86, 841–861.

Economy 88, 1203–1225.

ED

Pindyck, R. S. (1980). Uncertainty and exhaustible resource markets. Journal of Political

PT

Roumasset, J., D. Isaak, and F. Fesharaki (1983). Oil prices without OPEC: A walk on the supply-side. Energy Economics 5, 164–170.

AC CE

Spiro, D. (2014). Resource prices and planning horizons. Journal of Economic Dynamics and Control 48, 159–175.

Thompson, A. C. (2001). The Hotelling principle, backwardation of futures prices and the values of developed petroleum reserves – the production constraint hypothesis. Resource and Energy Economics 23, 133–156. Tsui, K. K. (2009). More oil, less democracy: Evidence from worldwide crude oil discoveries. The Economic Journal 121, 89–115.

26

ACCEPTED MANUSCRIPT Table 1: Summary statistics of country-level giant oil field discoveries (# and EURO)

— 1025 975 — 1597 850 801 — 525 486 — 1008 500 — 1000 925 640 2427 — 625 1000 — 533 517 6692 737 2192 — 967 9681 500 — — 500 — 3885 696 631 943 725 — 1944 3213 1431 2078

# 5 14 2 1 25 14 17 4 4 1 2 8 1 1 2 5 4 26 24 1 1 8 25 3 10 22 12 6 2 38 1 2 1 1 2 14 12 28 12 2 2 74 197 243 440

Giant EURO

PT

0 13 2 0 26 2 3 0 4 1 0 7 1 0 2 2 3 9 0 1 1 0 1 3 4 12 15 0 2 13 1 0 0 1 0 11 12 21 5 2 0 8 69 121 190

RI

938 500 — 475 590 872 2056 1499 — — 618 500 — 1000 — 507 800 3791 4091 — — 2794 995 — 1478 496 — 841 — 4845 — 750 708 — 820 4540 — 2575 1527 — 900 1684 3122 1559 2383

Offshore # EURO

SC

5 1 0 1 1 12 16 4 0 0 2 1 0 1 0 3 1 22 31 0 0 10 24 0 10 10 0 6 0 35 0 2 1 0 2 8 0 8 8 0 2 73 158 142 300

MA

938 988 975 475 1560 869 1858 1499 525 486 618 944 500 1000 1000 674 680 3395 4091 625 1000 2794 977 517 2967 627 2192 841 967 6155 500 750 708 500 820 4161 696 1167 1302 725 900 1709 3150 1500 2265

ED

5 14 2 1 27 14 19 4 4 1 2 8 1 1 2 5 4 31 31 1 1 10 25 3 14 22 15 6 2 48 1 2 1 1 2 19 12 29 13 2 2 81 227 263 490

AC CE

Algeria Angola Australia Austria Brazil Canada China Colombia Congo (Brazzaville) Denmark Ecuador Egypt Equatorial Guinea Gabon Ghana India Indonesia Iran Iraq Italy Ivory Coast Kuwait Libya Malaysia Mexico Nigeria Norway Oman Qatar Saudi Arabia Sierra Leone Sudan Syria Trinidad and Tobago Tunisia United Arab Emirates United Kingdom United States Venezuela Vietnam Yemen Former Soviet Union OPEC Non-OPEC Total

Onshore # EURO

NU

Discoveries # EURO

PT

Country

938 988 975 475 1195 869 1226 1499 525 486 618 944 500 1000 1000 674 680 1470 1596 625 1000 1802 977 517 858 627 1064 841 967 1018 500 750 708 500 820 1371 696 663 1078 725 900 1062 1144 951 1037

Supergiant # EURO 0 0 0 0 2 0 2 0 0 0 0 0 0 0 0 0 0 5 7 0 0 2 0 0 4 0 3 0 0 10 0 0 0 0 0 5 0 1 1 0 0 7 30 20 50

— — — — 6125 — 7223 — — — — — — — — — — 13407 12643 — — 6763 — — 8242 — 6700 — — 25673 — — — — — 11973 — 15273 3987 — — 8556 16322 8179 13064

Note: OPEC includes Angola, Algeria, Ecuador, Gabon, Iran, Iraq, Indonesia, Kuwait, Libya, Nigeria, Qatar, Saudi Arabia, UAE, and Venezuela. The sample period is 1945–2010. 27

ACCEPTED MANUSCRIPT Table 2: Effect of current and lagged giant oil field discoveries on country-level oil production Panel A. DLR model in (1) p=1 p=4

2,743

.2855∗∗∗

2,500

28

(.0697)

PT

.0163∗∗ (.0065) .0118∗∗ (.0057) ∗∗ .0108 (.0043) .0240∗∗∗ (.0042) .0054 (.0050) .0024 (.0043) –.0005 (.0049) ∗ –.0057 (.0030) –.0038 (.0055) –.0001 (.0039) .0046 (.0044) .0028 (.0056) –.0044 (.0057) .0043 (.0041) .0029 (.0034) .0052 (.0046) .0026 (.0039) .0051 (.0045) –.0011 (.0029) –.0063 (.0052) –.0052 (.0053) .0109 (.0182) –.0161 (.0156) –.0251 (.0214) –.0253∗∗∗ (.0083) –.0100 (.0128) .0052 (.0116) .0077 (.0110) .0065 (.0081) .0228∗∗ (.0093) .0198∗∗∗ (.0078) –.0159 (.0116) –.0060 (.0122) .0121 (.0133) .0009 (.0075) ∗ –.0191 (.0095) –.0103 (.0100) .0007 (.0071) –.0189∗∗ (.0088) –.0104 (.0125) .0050 (.0088) .0104∗ (.0056) 2,743

RI

(.0822) (.0952) (.0766) (.0619) (.0106) (.0092) (.0120) (.0089) (.0105) (.0053) (.0070) (.0049) (.0074) (.0048) (.0091) (.0076) (.0094) (.0049) (.0094) (.0081) (.0067) (.0067) (.0122) (.0069) (.0050)

SC

(.0081) (.0074) (.0100) (.0054) (.0067) (.0044) (.0066) (.0046) (.0070) (.0045) (.0082) (.0064) (.0077) (.0043) (.0073) (.0067) (.0043) (.0063) (.0092) (.0052) (.0032)

Panel B. DLR model in (3) p=1 p=4

NU

.0187∗∗∗ .0051 –.0035 .0058 –.0053 .0060 .0045 –.0026 .0127∗ .0136∗∗∗ –.0078 –.0015 .0052 .0038 –.0121∗ –.0003 .0034 –.0104∗ –.0108 –.0028 .0033

.2918∗∗∗ .0486 –.0191 –.0720 .0209∗∗ .0014 –.0071 .0136∗ –.0062 .0096∗ .0021 –.0030 .0138∗ .0158∗∗∗ –.0109 .0005 .0058 .0070 –.0166∗ –.0020 .0048 –.0084 –.0129 –.0054 .0058

MA

(.0721)

ED

.3077∗∗∗

AC CE

∆yi,t−1 ∆yi,t−2 ∆yi,t−3 ∆yi,t−4 dit di,t−1 di,t−2 di,t−3 di,t−4 di,t−5 di,t−6 di,t−7 di,t−8 di,t−9 di,t−10 di,t−11 di,t−12 di,t−13 di,t−14 di,t−15 di,t−16 di,t−17 di,t−18 di,t−19 di,t−20 opeci · dit opeci · di,t−1 opeci · di,t−2 opeci · di,t−3 opeci · di,t−4 opeci · di,t−5 opeci · di,t−6 opeci · di,t−7 opeci · di,t−8 opeci · di,t−9 opeci · di,t−10 opeci · di,t−11 opeci · di,t−12 opeci · di,t−13 opeci · di,t−14 opeci · di,t−15 opeci · di,t−16 opeci · di,t−17 opeci · di,t−18 opeci · di,t−19 opeci · di,t−20 Observations

PT

∆yit

.2716∗∗∗ (.0816) .0473 (.0982) –.0323 (.0796) –.0718 (.0686) .0124∗ (.0065) .0100 (.0073) ∗∗ .0122 (.0053) .0253∗∗∗ (.0057) .0068 (.0057) .0040 (.0060) .0010 (.0065) –.0058 (.0046) –.0033 (.0070) .0015 (.0050) .0054 (.0044) .0026 (.0058) –.0050 (.0061) .0038 (.0044) .0033 (.0038) .0056 (.0048) .0025 (.0045) .0053 (.0046) –.0010 (.0034) –.0059 (.0052) –.0039 (.0039) .0181 (.0196) –.0244∗ (.0133) –.0380∗∗ (.0192) –.0170 (.0192) –.0251 (.0194) .0081 (.0153) .0047 (.0141) .0102 (.0091) .0265∗∗∗ (.0102) .0202∗∗ (.0095) –.0274∗ (.0146) –.0021 (.0151) .0134 (.0162) .0097 (.0084) ∗ –.0271 (.0142) –.0111 (.0138) .0012 (.0115) –.0178∗ (.0095) –.0138 (.0168) –.0025 (.0107) .0120 (.0076) 2,500

ACCEPTED MANUSCRIPT Adjusted R2

.2109

.2115

.2439

.2556

Note: Point estimates with HAC-robust standard errors in parentheses. The sample period is 1965–2010. ∗∗∗ /∗∗ /∗ indicates statistical significance at the 1/5/10% level. Coefficient estimates for the DLR models in (5) and (7) and for country-level petroleum consumption are available from the author on request. Figure 1: Cumulated hydrocarbon volumes in mmboe of giant oil fields by their year of discovery 4

x 10

120

8 6

80 60

SC

mmboe

100

RI

10

2

1910

1920

1930

1940

1950

1960

1970

MA

0 1900

NU

4

2016 US$

EUR EURO RSVS RPO

PT

12

40 20

1980

1990

2000

0 2010

PT

ED

Note: Cumulated hydrocarbon volumes i.t.o. estimated ultimate recovery (EUR), estimated ultimate recovery of oil (EURO), and reserves (RSVS) from Horn (2014). Real price of oil (RPO) deflated using the U.S. consumer price index for 2016

Figure 2: Impulse response of cumulated changes in country-level petroleum production to a giant oil field discovery based on the dynamic panel DLR model in (1)

0.1

tbpd

0.08 0.06 0.04

0.2 0.15 0.1 tbpd

0.12

Production (p=4, q=20)

AC CE

Production (p=1, q=20)

0.14

0.05

0.02

0

0 −0.02 0

5

10 15 years after discovery

−0.05 0

20

5

10 15 years after discovery

20

Note: Point estimates with one- and two-standard-error HAC-robust confidence intervals based on 1,000 bootstrap replications

29

ACCEPTED MANUSCRIPT Figure 3: Impulse response of cumulated changes in country-level rotary rig activity to a giant oil field discovery based on the dynamic panel DLR model in (1) Rig activity (p=4, q=20) 0.1

0.06

0.08 rotary rigs/mmboe

0.02 0

0.06 0.04 0.02

−0.02 −0.04 0

0

5

10 15 years after discovery

−0.02 0

20

5

PT

0.04

10 15 years after discovery

RI

rotary rigs/mmboe

Rig activity (p=1, q=20) 0.08

20

NU

SC

Note: Point estimates with one- and two-standard-error HAC-robust confidence intervals based on 1,000 bootstrap replications

OPEC production (p=1, q=20) 0.25 0.2

MA

Figure 4: Impulse response of cumulated changes in OPEC and non-OPEC production to a giant oil field discovery based on the dynamic panel DLR model in (3)

0.05

tbpd

ED

0.1

0

AC CE

−0.05 −0.1 0

0.2 0.15

PT

tbpd

0.15

OPEC production (p=4, q=20) 0.25

5

10 15 years after discovery

0.1 0.05 0 −0.05 −0.1 0

20

Non−OPEC production (p=1, q=20)

tbpd

0.15

10 15 years after discovery

20

Non−OPEC production (p=4, q=20) 0.2 0.15 0.1 tbpd

0.2

5

0.1

0.05

0.05

0 0

0

5

10 15 years after discovery

−0.05 0

20

5

10 15 years after discovery

20

Note: Point estimates with one- and two-standard-error HAC-robust confidence intervals based on 1,000 bootstrap replications

30

ACCEPTED MANUSCRIPT

OPEC rig activity (p=1, q=20)

SC

rotary rigs/mmboe

0.005

0

NU

0

5

10 15 years after discovery

Non−OPEC rig activity (p=1, q=20)

ED

0.15

0.05

AC CE

0

5

10 15 years after discovery

−0.005

−0.01 0

5

10 15 years after discovery

20

Non−OPEC rig activity (p=4, q=20) 0.2

0.15

PT

0.1

−0.05 0

20

rotary rigs/mmboe

−5 0

rotary rigs/mmboe

OPEC rig activity (p=4, q=20)

0.01

MA

rotary rigs/mmboe

x 10

RI

−3

5

PT

Figure 5: Impulse response of cumulated changes in OPEC and non-OPEC rotary rig activity to a giant oil field discovery based on the dynamic panel DLR model in (3)

0.1 0.05 0 −0.05 0

20

5

10 15 years after discovery

20

Note: Point estimates with one- and two-standard-error HAC-robust confidence intervals based on 1,000 bootstrap replications

31

ACCEPTED MANUSCRIPT

PT

Figure 6: Impulse response of cumulated changes in production to giant onshore and offshore oil field discoveries based on the dynamic panel DLR model in (5) Onshore discovery (p=4, q=20)

0.1

0.05

0.05 tbpd

NU

0

0

5

10 15 years after discovery

Offshore discovery (p=1, q=20)

ED

0.3 0.25 0.2

0.1

−0.05 0

AC CE

0.05

5

10 15 years after discovery

−0.1 0

5

10 15 years after discovery

20

Offshore discovery (p=4, q=20) 0.3

0.25 0.2

PT

0.15

0

20

tbpd

−0.1 0

−0.05

MA

−0.05

tbpd

SC

0.1

tbpd

0.15

RI

Onshore discovery (p=1, q=20) 0.15

0.15 0.1 0.05 0 −0.05 0

20

5

10 15 years after discovery

20

Note: Point estimates with one- and two-standard-error HAC-robust confidence intervals based on 1,000 bootstrap replications

32

ACCEPTED MANUSCRIPT

PT

Figure 7: Impulse response of cumulated changes in production to giant and super-/mega-giant oil field discoveries based on the dynamic panel DLR model in (7) 0.3

0.15

SC

0.2

0.1

tbpd

0.1

0

NU

tbpd

0.05

0

−0.05 −0.1

−0.1

5

10 15 years after discovery

MA

−0.15 −0.2 0

20

−0.2 0

Super−/Mega−giant discovery (p=1, q=20)

10 15 years after discovery

20

Super−/Mega−giant discovery (p=4, q=20)

0.15

PT

tbpd

0.1

0.2

tbpd

0.15

AC CE

0.05

−0.05 0

5

0.25

ED

0.2

0

RI

Giant discovery (p=4, q=20)

Giant discovery (p=1, q=20) 0.2

5

10 15 years after discovery

0.1 0.05 0 −0.05 −0.1 0

20

5

10 15 years after discovery

20

Note: Point estimates with one- and two-standard-error HAC-robust confidence intervals based on 1,000 bootstrap replications

33

ACCEPTED MANUSCRIPT Figure 8: Impulse response of cumulated changes in country-level petroleum consumption to a giant oil field discovery based on the dynamic panel DLR model in (1) Consumption (p=1, q=20)

Consumption (p=4, q=20)

0.08

0.1 0.08

0.06

0.06 tbpd

tbpd

0.04 0.04

0.02 0

0

5

10 15 years after discovery

−0.02 0

20

RI

−0.02 0

PT

0.02

5

10 15 years after discovery

20

NU

SC

Note: Point estimates with one- and two-standard-error HAC-robust confidence intervals based on 1,000 bootstrap replications

OPEC consumption (p=1, q=20) 0.015 0.01

MA

Figure 9: Impulse response of cumulated changes in OPEC and non-OPEC consumption to a giant oil field discovery based on the dynamic panel DLR model in (3)

−0.005

tbpd

ED

0

−0.01

AC CE

−0.015 −0.02 0

0.015 0.01

PT

tbpd

0.005

OPEC consumption (p=4, q=20) 0.02

5

10 15 years after discovery

0.005 0 −0.005 −0.01 −0.015 0

20

Non−OPEC consumption (p=1, q=20)

0.12 0.1

tbpd

0.08

20

Non−OPEC consumption (p=4, q=20)

0.1 0.08

0.06 0.04

0.06 0.04 0.02

0.02

0

0 −0.02 0

10 15 years after discovery

0.12

tbpd

0.14

5

5

10 15 years after discovery

−0.02 0

20

5

10 15 years after discovery

20

Note: Point estimates with one- and two-standard-error HAC-robust confidence intervals based on 1,000 bootstrap replications

34

ACCEPTED MANUSCRIPT Highlights • Giant oil field discoveries indicate a sizeable increase in a country’s reserve base. • Oil production at the country level increases before the new oil field comes on line. • The increase is driven by non-OPEC producers, while OPEC production barely responds.

PT

• Offshore fields and super- or mega-giant fields are also more likely to come on line.

AC CE

PT

ED

MA

NU

SC

RI

• Part of the production hike goes into exports, as domestic consumption rises by less.

35