Market Power, Production (Mis)Allocation and OPEC John Asker UCLA and NBER Allan Collard-Wexler Duke and NBER Jan De Loecker KU Leuven, NBER and CEPR 2017 Asker, Collard-Wexler, De Loecker OPEC 2017 1 / 55
Motivation Research Question: production? Impact market power on the misallocation of Approach: Data driven examination of upstream oil industry (Extraction and pre-refinery production) Why is this interesting? Effect of market power is central to IO. Both cartel activity and unilateral market power. Case of aggregate implications of market power in context of misallocation literature. Harberger 1954: When we are interested in the big picture of our manufacturing economy, we need not apologize for treating it as competitive, for in fact it is awfully close to being so. On the other hand, when we are interested in the doings of particular industries, it may often be wise to take monopoly elements into account.? The influence of OPEC on the world market for oil. Asker, Collard-Wexler, De Loecker OPEC 2017 2 / 55
Production Distortion: main approach P MC f MC 1 D q 1 Q = q 1 +q 2 Q SP Asker, Collard-Wexler, De Loecker OPEC 2017 3 / 55
Extending the static (graphical) analysis Oil is an exhaustible resource: we need to take the dynamics of production seriously. Depletion of Reserves. Constraints on extraction speed. When a field gets extracted, not if. Asker, Collard-Wexler, De Loecker OPEC 2017 4 / 55
Literature Borenstein, Bushnell and Wolak 1999 AER. Oil Markets Micro: Kellogg 2014 AER, Covert 2017 WP, Anderson, Kellogg, Salant, 2017 JPE. Macro: Lutz 2009 AER Cartels: Marshall and Marx 2012, Asker AER 2010, Schmitz 2015 WP. OPEC: Cremer and Weitzman 1976 EER, Cremer, Salehi-Isfahani 1991. Misallocation (Harberger 1954, Harberger 1959, Hsieh and Klenow, 2009 QJE, Asker, Collard-Wexler and De Loecker, JPE 2014, + many others ). Asker, Collard-Wexler, De Loecker OPEC 2017 5 / 55
Main Findings Costs of oil production are 10 percent higher due to the OPEC cartel: about a 163 billion dollar welfare loss over a 35 year period. Asker, Collard-Wexler, De Loecker OPEC 2017 6 / 55
Map of Talk Oil and OPEC Data Model Empirical Analysis Conclusion Asker, Collard-Wexler, De Loecker OPEC 2017 7 / 55
Background on Oil Geology and location have a big impact on costs of extraction Exogenous cost variation across production units unrelated to management skill rather: Model (technology): onshore, offshore, shale, etc. Location (geology): bedrock structure, climate, etc. Examples: Asker, Collard-Wexler, De Loecker OPEC 2017 8 / 55
West Texas Asker, Collard-Wexler, De Loecker OPEC 2017 9 / 55
Aasgard Norway Asker, Collard-Wexler, De Loecker OPEC 2017 10 / 55
OPEC Cartel OPEC is Algeria, Angola, Ecuador, Gabon, Indonesia, Iran, Iraq, Kuwait, Libya, Nigeria, Qatar, Saudi Arabia, UAE, and Venezuela. OPEC is an imperfect cartel Asker, Collard-Wexler, De Loecker OPEC 2017 11 / 55
Main Oil Producers Table: Largest crude producers, % of global production 1970-2014 OPEC Non-OPEC Saudi Arabia 11.8% United States 14.4% Iran 5.4% Russia 13.0% Venezuala 3.8% China 4.1% UAE 3.1% Mexico 3.7% Nigeria 2.8% Canada 3.3% Iraq 2.7% UK 2.4% Kuwait 2.6% Norway 2.4% Notes: Global production from 1970-2014 was 1,156 billion barrels. Collectively these 14 countries account for 85.4% of global production. Asker, Collard-Wexler, De Loecker OPEC 2017 12 / 55
Price and OPEC 100 1 90 0.9 80 0.8 70 0.7 60 0.6 50 0.5 40 0.4 30 0.3 20 0.2 10 0.1 0 0 2014 2010 2006 2002 1998 1994 1990 1986 1982 1978 1974 1970 Asker, Collard-Wexler, De Loecker OPEC 2017 13 / 55
Map of Talk Oil and OPEC Data Model Empirical Analysis Conclusion Asker, Collard-Wexler, De Loecker OPEC 2017 14 / 55
Data Rich Data on oil from Rystad Energy, a Norwegian Energy Consultancy. One of the main data suppliers in the industry (IHS, Wood Gundy). Field Level Information: Gulfaks South versus Ghawar Uthmamiyah. Data on 66K oil fields, of which 19K produce crude oil, of which 13K have reserve data. Information on production, costs, reserves, technology, location. Asker, Collard-Wexler, De Loecker OPEC 2017 15 / 55
Data Sources Globally Representative Data: matches country level flows. Some data is great: Norway s data reported by Statoil. Some data is inferred from modeling: Syria, Iran. Asker, Collard-Wexler, De Loecker OPEC 2017 16 / 55
Summary Statistics Variable mean median 5% 95% Field-year characteristics: Production (mb/year) 3.43 0.22 0.00 10.92 Reserves (mb) 99.49 3.71 0.03 239.78 Discovery Year 1965 1967 1911 1999 Startup Year 1971 1974 1916 2005 Off-shore 0.19 Costs: ($m) Exploration Capital Expenditures 0.61 0.00 0.00 0.41 Well Capital Expenditures 9.10 0.49 0.00 35.32 Facility Capital Expenditures 5.14 0.21 0.00 16.85 Production Operating Expenditures 10.41 0.46 0.00 38.47 Transportation Operating Expenditures 2.27 0.13 0.00 7.01 SGA Operating Expenditures 2.65 0.22 0.00 8.85 Asker, Collard-Wexler, De Loecker OPEC 2017 17 / 55
Reserves, 2014 Reserves reserves Reserves/(Annual production) (mb) (%) (%) Non-OPEC 218,054 50 10 OPEC 220,561 50 19 Saudi Arabia 74,194 17 18 Reserves are measured as the unextracted, but recoverable, quantity of oil remaining in the ground in a field. 1 Descriptive stats: P50 value at an oil price of $70 2 Counterfactual (1970 onward) sum of: i) the actual production history from 1970 to 2014, and ii) the P50 value at an oil price of $70 a barrel in 2014. Asker, Collard-Wexler, De Loecker OPEC 2017 18 / 55
Cost Changes over time: Saudi Arabia black: 95%, grey: 99% and circle: max. 100 90 80 70 60 50 40 30 20 10 0 2014 2012 2010 2008 2006 2004 2002 2000 1998 1996 1994 1992 1990 1988 1986 1984 1982 1980 1978 1976 1974 1972 1970 Asker, Collard-Wexler, De Loecker OPEC 2017 19 / 55
Cost Changes over time: Kuwait 100 90 80 70 60 50 40 30 20 10 0 2014 2012 2010 2008 2006 2004 2002 2000 1998 1996 1994 1992 1990 1988 1986 1984 1982 1980 1978 1976 1974 1972 1970 Asker, Collard-Wexler, De Loecker OPEC 2017 20 / 55
Cost Changes over time: Venezuela 100 90 80 70 60 50 40 30 20 10 0 2014 2012 2010 2008 2006 2004 2002 2000 1998 1996 1994 1992 1990 1988 1986 1984 1982 1980 1978 1976 1974 1972 1970 Asker, Collard-Wexler, De Loecker OPEC 2017 21 / 55
Cost Changes over time: Nigeria 100 90 80 70 60 50 40 30 20 10 0 2014 2012 2010 2008 2006 2004 2002 2000 1998 1996 1994 1992 1990 1988 1986 1984 1982 1980 1978 1976 1974 1972 1970 Asker, Collard-Wexler, De Loecker OPEC 2017 22 / 55
Cost Changes over time: Russia 100 90 80 70 60 50 40 30 20 10 0 2014 2012 2010 2008 2006 2004 2002 2000 1998 1996 1994 1992 1990 1988 1986 1984 1982 1980 1978 1976 1974 1972 1970 Asker, Collard-Wexler, De Loecker OPEC 2017 23 / 55
Cost Changes over time: United States 100 90 80 70 60 50 40 30 20 10 0 2014 2012 2010 2008 2006 2004 2002 2000 1998 1996 1994 1992 1990 1988 1986 1984 1982 1980 1978 1976 1974 1972 1970 Asker, Collard-Wexler, De Loecker OPEC 2017 24 / 55
Cost Changes over time: Canada 100 90 80 70 60 50 40 30 20 10 0 2014 2012 2010 2008 2006 2004 2002 2000 1998 1996 1994 1992 1990 1988 1986 1984 1982 1980 1978 1976 1974 1972 1970 Asker, Collard-Wexler, De Loecker OPEC 2017 25 / 55
Cost Changes over time: Norway 100 90 80 70 60 50 40 30 20 10 0 2014 2012 2010 2008 2006 2004 2002 2000 1998 1996 1994 1992 1990 1988 1986 1984 1982 1980 1978 1976 1974 1972 1971 Asker, Collard-Wexler, De Loecker OPEC 2017 26 / 55
Map of Talk Oil and OPEC Data Model Empirical Analysis Conclusion Asker, Collard-Wexler, De Loecker OPEC 2017 27 / 55
Production Distortion P MC f MC 1 D q 1 Q = q 1 +q 2 Q SP Asker, Collard-Wexler, De Loecker OPEC 2017 28 / 55
Competitive Equilibrium Productive Inefficiency Definition Productive inefficiency is the net present value of the difference between the realized costs of production, and the cost of production had the realized production path been produced by firms taking prices as exogenous. In an exhaustible resource industry, the welfare losses come from the welfare effects of when to extract oil given discounting. Asker, Collard-Wexler, De Loecker OPEC 2017 29 / 55
Costs Leontief Production Function: q ft = min {γ ft L ft, α ft K ft } (1) Input Prices (w ft, r ft ) evolve over time: This yields the cost function ( wft C (q ft ) = + r ) ft q ft γ f α f ( w = + r ) µ st q ft γ f α f }{{} c f r ft ω t = µ st r and w ft ω t = µ st w. (2) = c f µ st q ft Asker, Collard-Wexler, De Loecker OPEC 2017 30 / 55
Characterization of Equilibrium From previous slide, marginal cost = c f µ st Homogenous product market δ be the common discount factor. Martingale Assumption on expectation of µ: E (µ st+k µ st ) = µ st Implication: lowest cost fields are extracted first in any competitive equilibrium. Asker, Collard-Wexler, De Loecker OPEC 2017 31 / 55
Sorting Theorem Proposition 1 and corollary 1: lowest cost fields are extracted first in any competitive equilibrium. Sketch: take fields F and F, with c f equal to c and c. By contradiction suppose that F extracted at period 1 and F extracted at period t. Then we have: δ t 1 (P t c) (P 1 c) (3) and δ t 1 (P t c) (P 1 c) (4) Martingale means E(c t c) = c Asker, Collard-Wexler, De Loecker OPEC 2017 32 / 55
Sorting Algorithm for Optimal Extraction Decisions MC 3$ MC 2$ MC 1$ Asker, Collard-Wexler, De Loecker OPEC 2017 33 / 55
Map of Talk Oil and OPEC Data Model Empirical Analysis Conclusion Asker, Collard-Wexler, De Loecker OPEC 2017 34 / 55
Structural Model Use the sorting algorithm to compute counterfactual paths for the industry the competitive path. Notice that, as in the figure, we are looking at changes in costs holding total quantity fixed. We will first present two types of counterfactuals: Static Counterfactual: one period effects of moving to a competitive equilibrium. Dynamic Counterfactuals: long run effects all about when a barrel will be extracted, not if. Asker, Collard-Wexler, De Loecker OPEC 2017 35 / 55
Inputs into the Dynamic Structural Model Discount rate β = 0.95. Limits on how much oil can be extracted at once (Anderson, Kellogg, and Salant 2017). We cap the extraction rate at 10 percent of reserves. Fields can only be extracted after their discovery date: take the path of new discoveries as exogenous. We do not consider the contribution of fields that do not produce in 1970-2015, likely to understate welfare losses. Simulate out to 2050 until all reserves have been depleted. Demand growth set at 1.3 percent (geometric average over 1970-2015). Forecasted production is optimal after 2015 (end of the data) lower bound on welfare losses. Need to estimate counterfactual costs: what a field would have cost to extract in 1990 using data on costs in 2010. Asker, Collard-Wexler, De Loecker OPEC 2017 36 / 55
Cost Estimate Marginal Costs are given by: c f,t,s = c f µ st exp (ɛ f,t,s ) (5) First estimate µ st by: 1 Compute the average unit cost of production by year t and technology s: ln µ t,s = κ f,t,s ln c f,t,s, (6) f s where κ f,t,s is the quantity weight of a field in a given year s total output, q κ f,t,s = f,t f. s,t q f,t 2 Recover an estimate of field-specific marginal cost shifter µ f, allowing for measurement error, using the following regression: (ln c fts µ st) = ln c f + ɛ f,t,s (7) Asker, Collard-Wexler, De Loecker OPEC 2017 37 / 55
Non-OPEC 0.607 0.212-0.395 Brazil 0.014 0.001-0.013 Canada 0.023 0.006-0.017 China 0.045 0.002-0.043 Asker, Kazakhstan Collard-Wexler, De Loecker 0.010 OPEC 0.000 2017-0.01 38 / 55 Static Distortion: as of 2014 OPEC Table 7: Static counterfactual for 2014: Top 20 producers Country Actual output share Counterfactual output share Share Persian Gulf OPEC 0.258 0.744 0.486 Iran 0.057 0.091 0.034 Iraq 0.029 0.069 0.040 Kuwait 0.030 0.155 0.125 Qatar 0.009 0.015 0.006 Saudi Arabia 0.133 0.414 0.281 United Arab Emirates 0.031 0.075 0.044 Other OPEC 0.135 0.044-0.091 Algeria 0.021 0.015-0.006 Indonesia 0.020 0.002-0.018 Libya 0.025 0.012-0.013 Nigeria 0.028 0.006-0.022 Venezuela 0.041 0.009-0.032
Saudi Arabia 0.133 0.414 0.281 United Arab Emirates 0.031 0.075 0.044 Static Distortion: as of 2014 Not-OPEC Other OPEC 0.135 0.044-0.091 Algeria Table 7: Static0.021 counterfactual for 2014: 0.015 Top 20 producers -0.006 Indonesia 0.020 0.002-0.018 Libya 0.025 0.012-0.013 Nigeria 0.028 0.006-0.022 Country Venezuela Actual output 0.041 share Counterfactual 0.009 output share -0.032 Share Persian Non-OPEC Gulf OPEC 0.607 0.258 0.744 0.212-0.395 0.486 Iran Brazil 0.057 0.014 0.091 0.001-0.013 0.034 Iraq Canada 0.029 0.023 0.069 0.006-0.017 0.040 Kuwait China 0.030 0.045 0.155 0.002-0.043 0.125 Qatar Kazakhstan 0.010 0.009 0.000 0.015 0.006-0.01 Saudi MexicoArabia 0.133 0.023 0.414 0.013 0.281-0.01 United NorwayArab Emirates 0.027 0.031 0.009 0.075-0.018 0.044 Russia 0.144 0.047-0.097 Other United OPEC Kingdom 0.1350.022 0.0440.001-0.091-0.021 Algeria United States 0.021 0.132 0.015 0.013-0.006-0.119 Indonesia 0.020 0.002-0.018 Rest Libya of the World 0.1360.025 0.0440.012-0.092-0.013 Nigeria 0.028 0.006-0.022 Note: Venezuela Reported results are for the top0.041 20 producers between 1970 and 0.009 2014. Initial conditions-0.032 are the state of the global market at the end of 2013. Application of the sorting algorithm gives counterfactual Non-OPEC production for 2014. 0.607 In every other respect the baseline 0.212 specification is used: -0.395 a field Brazil extraction rate of 10 percent of reserves 0.014is imposed in the counterfactual, 0.001 the p50 measures-0.013 of reserves Canadaare used where needed and0.023 a demand growth rate of 1.3 percent 0.006 per year after 2014-0.017 is assumed. China 0.045 0.002-0.043 Asker, Collard-Wexler, De Loecker OPEC 2017 39 / 55
Doing the welfare accounting Comparing the sorting algorithm to the data is too strong: encapsulates any distortion, and also measurement error, model misspecification and such. MC 3$ MC 2$ MC 1$ Asker, Collard-Wexler, De Loecker OPEC 2017 40 / 55
Doing the welfare accounting Comparing the sorting algorithm to the data is too strong: encapsulates any distortion, and also measurement error, model misspecification and such. MC 3$ MC 2$ MC 1$ Asker, Collard-Wexler, De Loecker OPEC 2017 41 / 55
Doing the welfare accounting Comparing the sorting algorithm to the data is too strong: encapsulates any distortion, and also measurement error, model misspecification and such. MC 3$ MC 2$ MC 1$ Asker, Collard-Wexler, De Loecker OPEC 2017 42 / 55
Doing the welfare accounting Comparing the sorting algorithm to the data is too strong: encapsulates any distortion, and also measurement error, model misspecification and such. MC 3$ MC 2$ MC 1$ Asker, Collard-Wexler, De Loecker OPEC 2017 43 / 55
Doing the welfare accounting Comparing the sorting algorithm to the data is too strong: encapsulates any distortion, and also measurement error, model misspecification and such. MC 2 +t 2$ MC 3$ MC 2$ MC 1$ Asker, Collard-Wexler, De Loecker OPEC 2017 44 / 55
Doing the welfare accounting Comparing the sorting algorithm to the data is too strong: encapsulates any distortion, and also measurement error, model misspecification and such. MC 2 +t 2$ MC 3$ MC 2$ MC 1$ Asker, Collard-Wexler, De Loecker OPEC 2017 45 / 55
Doing the welfare accounting Comparing the sorting algorithm to the data is too strong: encapsulates any distortion, and also measurement error, model misspecification and such. MC 2 +t 2$ MC 3$ MC 2$ MC 1$ Asker, Collard-Wexler, De Loecker OPEC 2017 46 / 55
Welfare accounting: implementation Nested Set of Constraints: Hold production in each field fixed (actual). Hold production in each country fixed. Hold production inside and outside of OPEC constant Table: Static Distortion: Production Cost in 2014 in Billions of Dollars Actual 240 Optimal Country 203 Optimal OPEC 154 Optimal 121 Also, can look at cartel inefficiency at intensive and extensive margin. Asker, Collard-Wexler, De Loecker OPEC 2017 47 / 55
Note: Static distortions for each year are presented in 2014 dollars (left vertical axis), with the total heig Asker, Collard-Wexler, De Loecker OPEC 2017 48 / 55 Static Distortion over Time Figure 6: Decomposing Static Distortions 250 100 80 200 60 150 40 100 20 0 50-20 0-40 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014
Almost all the production in the 1970s is accounted for by a couple of fields: Ghawar Uthmaniyah, Greater Burgan, Ghawar Shedgum. Asker, Collard-Wexler, De Loecker OPEC 2017 49 / 55 Dynamic Counterfactual Simulate from 1970 to 2015: NPV starting in 1970. 100 80 60 40 20 1970 1980 1990 2000 2010 year OPEC Share -- Actual OPEC Share -- Counterfactual
Notes: The NPV of costs from 1970 to 2014, and to 2100 (exhaustion of all fields), are reported in Asker, billions Collard-Wexler, of 2014 De Loecker dollars (assuming a 5 percentopec discount rate). Results are for the baseline specifi- 2017 50 / 55 Full dynamic model: results Table 6: Dynamic counterfactual results (NPV of costs in billions of 2014 dollars) Timespan 1970-2014 1970-2100 Actual (A) 2184 (125) 2499 (130) Counterfactual (C) 1268 (76) 1756 (79) Total distortion (A - C) 916 (124) 744 (112) Decomposition of total distortion Within country (non-opec) 329 (80) 284 (41) Within country (OPEC) 192 (46) 157 (72) Across country (within non-opec) 163 (18) 139 (17) Across country (within OPEC) (X) 85 (22) 58 (21) Between OPEC and non-opec (Y) 148 (29) 105 (25) Production distortion due to OPEC market power Upper bound (X+Y) 233 (42) 163 (38) Lower bound (Y only) 148 (29) 105 (25)
Notes: The units are billions of 2014 dollars. Results correspond to the 1970-2100 (exhaustion) timespan. All specifications use time invariant field level costs c f. Specifications are: (1) the baseline specifi- Asker, Collard-Wexler, De Loecker OPEC 2017 51 / 55 Robustness of Counterfactuals Table 9: Dynamic counterfactual results, conditional on inferred wedges Specification (1) (2) (3) (4) Actual (A) 2829 2837 2670 2596 Counterfactual (C 1 ) 1917 1906 - - Counterfactual (C 2 ) - - 1825 2452 Total distortion (A - C 1 ) 912 931 - - Second-Best distortion (A - C 2 ) - - 845 144 Distortion due to OPEC - - 845 144 Upper bound 217 203 - - Lower bound 133 124 - -
Conclusions Significant misallocation aligned with known OPEC mechanism. Countries with clear market power: Gulf OPEC members. Most of impact comes from timing of Ghawar (SA), Burgan (KW) and Kirkuk (IQ) extractions. Misallocation rises when OPEC is known to be holding down productions and prices spike. Very large welfare loss, due to productive inefficiency: 160 billion USD. No discussion of the role of distortionary taxes or carbon externalities in this market. Asker, Collard-Wexler, De Loecker OPEC 2017 52 / 55
Conclusions Asker, Collard-Wexler, De Loecker OPEC 2017 53 / 55
100 Conventional Oil 100 Offshore Shelf Cumulative Cost 80 60 40 20 0 0 20 40 60 80 100 Cumulative Production Cumulative Cost 80 60 40 20 0 0 20 40 60 80 100 Cumulative Production 100 Offshore Deepwater 100 Shale, Oil Sands Cumulative Cost 80 60 40 20 0 0 20 40 60 80 100 Cumulative Production Cumulative Cost 80 60 40 20 0 0 20 40 60 80 100 Cumulative Production Asker, Collard-Wexler, De Loecker OPEC 2017 54 / 55
Figure: Observed and Predicted Marginal Cost Ghawar Uthmaniyah (SA) 10 8 Cost Per Barrel 6 4 2008 2 10000 20000 30000 40000 50000 60000 70000 Cumulative Production in Thousands of Barrels Observed mc Predicted mc Notes: Observed and predicted marginal cost, using the cost specification in equation??, is plotted against cumulative production. The vertical line indicates the proven reserves, and we insert the production year 2008, the year with the highest oil price in the sample period 1970-2014. Asker, Collard-Wexler, De Loecker OPEC 2017 55 / 55