A Simulation Model of the U.S. Oil Market

1
A Simulation Model of the U.S. Oil Market

• Alicia K. Birky
• University of Maryland School of Public Affairs
• PhD Dissertation Work in Progress
• November 19, 2003

2
Overview

• Motivation
• Methodology
• Model Description
• Model Results
• Issues

3
Research Question

• Under what conditions can the U.S. transportation
  system transition from conventional petroleum
  while reducing carbon emissions can development
  of a superior alternate technology regime enable
  this transition, or will it only occur as the
  result of a sudden disturbance?

4
Motivation

• The worlds total endowment of oil is fixed
• Transportation accounts for 2/3 of U.S. oil
  consumption
• Many analysts are predicting that half this
  ultimate endowment will be produced by 2020-2030
• Then production will begin to decline, they claim
• Standard economics argues that a transition to
  alternatives will occur via market mechanisms
• What if standard economics is wrong?
• Carbon emissions from fossil fuels are the main
  contributor to climate change
• Will the future fuel for transport also
  contribute?

5
Conventional Economic Analysis

• Rational agents optimize an objective function
  (utility or profit)
• Objective function is exogenous and stable
• Depletion is accounted for in rational
  expectations
• Diminishing returns result in technologies
  sharing the market
• Technological change is exogenously specified

6
Alternative Framework

• Agents are boundedly rational
• Limited cognition and resources
• Unknown or uncertain future
• Preferences evolve endogenously with the social,
  economic and technical environment
• Adaptive preferences and expectations
• Endogenous learning
• Positive feedbacks can lead to lock-in

7
Methodology

• Dynamic simulation model focusing on U.S. highway
  vehicles
• Agents include vehicle manufacturers, vehicle and
  fuel consumers, fuel feedstock producers, and
  fuel refiners
• Fuels include conventional oil, unconventional
  oil, ethanol, and hydrogen
• Positive feedbacks will be modeled
• Bias toward the status quo
• Adaptive expectations
• Evolving preferences

8
Oil Sector Model
U.S. OSM Boundary

• Refiners
• Input level
• Output mix
• Capacity

• Consumers

Product Price
World Oil Price

• Personal income

Finished Products

• Production costs
• Yields
• Product inventory

Domestic Oil Price
Crude Oil

• Domestic Producers
• Production level
• Capacity
• Exploration
• RD expenditures

World Oil Market
World Oil Price

• Reserve Estimates
• Production costs

9
Exogenous to OSM

• World oil price
• Currently only historic data is used
• Will eventually be calculated by iteration to
  clear the world oil market
• Product demand
• Currently represented by a simple regression
  model for gasoline only
• Will eventually include distillates demand by all
  sectors
• GDP and personal income
• Oil price, product price and sales, and vehicle
  price and sales will eventually feed back into
  GDP and income

10
Endogenous to OSM

• Domestic production
• Refinery input
• Product mix
• Gasoline and distillate proportions
• Not currently modeled
• Refinery yield
• Depends on crude quality, regulations, and
  technology
• A measure of production cost
• Not yet modeled
• Net imports refinery input domestic
  production
• Gasoline inventory coverage
• Gasoline price

11
OSM Derivation

• Monthly time-step
• Want higher resolution than the shortest planning
  cycle, which is quarterly
• Seasonal dynamics shape perceptions
• Time series regression models
• Autoregressive structure
• Agents base current behavior on past behavior
• OLS is biased and inefficient, but consistent
• Generally adopted as the most appropriate
  estimator for habit-persistence theory
• Use Cochrane-Orcutt iterative method to account
  for inefficiency

12
Historic Data 1974-2000

• EIA Monthly Energy Review
• Domestic production
• Refinery input
• Net imports
• Gasoline production
• Oil and gasoline price
• Gasoline stock
• BEA
• GDP
• Personal income
• Census Bureau - Population

Problem GDP only available quarterly!
13
Domestic Production

• Domestic production (prod, million bpd) is a
  function of
• prodt-1 Lagged production
• dcRt-1 Lagged real refiner acquisition cost of
  domestic crude, ln(1996 /bbl)
• Grt-1 Lagged GDP growth rate
• rest-1/prodt-1 Lagged reserve estimate/lagged
  total production, years
• dmo dummy for month, 1 or 0, January omitted

14
Domestic Production Results

• Source SS df MS
  Number of obs 315
• ---------------------------------------
  F( 16, 298) 3951.36
• Model 10.0057954 16 .625362213
  Prob gt F 0.0000
• Residual .047162965 298 .000158265
  R-squared 0.9953
• ---------------------------------------
  Adj R-squared 0.9951
• Total 10.0529584 314 .032015791
  Root MSE .01258
• 
  
• --------------------------------------------------
  ----------------------------
• lnprod Coef. Std. Err. t
  Pgtt 95 Conf. Interval
• -------------------------------------------------
  ----------------------------
• lnprod1 .9825311 .0071591 137.243
  0.000 .9684423 .9966198
• lndcR1 .0089638 .0020314 4.413
  0.000 .0049661 .0129614
• Grate1 .3545127 .2015847 1.759
  0.080 -.0421972 .7512225
• lnrp1 .0001805 .0116465 0.015
  0.988 -.0227393 .0231003
• dxlnrp1 .0017712 .0009053 1.957
  0.051 -.0000103 .0035527
• feb .0090832 .0043204 2.102
  0.036 .0005809 .0175856
• mar -.0016321 .00349 -0.468
  0.640 -.0085003 .0052362
• apr -.0003315 .0037717 -0.088
  0.930 -.0077539 .007091
• may -.0009574 .0036614 -0.261
  0.794 -.0081629 .0062481

15
Refinery Input

• Refinery input (million bpd) as a function of
• reft-1 Lagged refinery input
• invgt-1 Lagged gasoline inventory coverage
  (inventory/consumption, days)
• ccRt-1 Lagged real refiner acquisition cost of
  crude, composite of domestic and import, (1996
  /bbl)
• Irt-1 Lagged personal income growth rate
• yldt-1 Lagged total refinery yield
  (gasolinedistillate production/input, unitless)
• dmo dummy for month, 1 or 0, January omitted

16
Refinery Input Results

• Source SS df MS
  Number of obs 316
• ---------------------------------------
  F( 16, 299) 400.04
• Model 2.5610934 16 .160068337
  Prob gt F 0.0000
• Residual .119638012 299 .000400127
  R-squared 0.9554
• ---------------------------------------
  Adj R-squared 0.9530
• Total 2.68073141 315 .008510258
  Root MSE .02
• 
  
• --------------------------------------------------
  ----------------------------
• lnrefine Coef. Std. Err. t
  Pgtt 95 Conf. Interval
• -------------------------------------------------
  ----------------------------
• lnref1 .8452734 .0216908 38.969
  0.000 .8025875 .8879593
• lninvg1 -.0929428 .015881 -5.852
  0.000 -.1241956 -.0616901
• lnccR1 -.0069287 .0039028 -1.775
  0.077 -.014609 .0007517
• Irate2 .5123249 .2136754 2.398
  0.017 .0918267 .9328231
• lnrefty1 -.2282638 .0449093 -5.083
  0.000 -.3166422 -.1398854
• feb .0136456 .0062379 2.188
  0.029 .0013698 .0259214
• mar .0231774 .0058262 3.978
  0.000 .0117119 .0346429
• apr .0274634 .0061472 4.468
  0.000 .0153661 .0395607
• may .0365577 .006004 6.089
  0.000 .0247422 .0483731

17
Gasoline Price

• Real gasoline price (1996 /gal), all grades, as
  a function of
• gpRt-1 Lagged price
• icR Real refiner acquisition cost of imported
  crude, (1996 /bbl)
• dsh Dummy for price shocks and Gulf Wars
• dcR Real refiner acquisition cost of domestic
  crude, (1996 /bbl)
• invgt-1 Lagged gasoline inventory coverage
  (inventory/consumption, days)
• refu Refinery capacity utilization rate,
  percentage points
• dmo dummy for month, 1 or 0, January omitted

18
Gasoline Price Results

• Source SS df MS
  Number of obs 316
• ---------------------------------------
  F( 19, 296) 392.71
• Model 2.15203438 19 .113264967
  Prob gt F 0.0000
• Residual .08537115 296 .000288416
  R-squared 0.9618
• ---------------------------------------
  Adj R-squared 0.9594
• Total 2.23740553 315 .007102875
  Root MSE .01698
• --------------------------------------------------
  ----------------------------
• lngpR Coef. Std. Err. t
  Pgtt 95 Conf. Interval
• -------------------------------------------------
  ----------------------------
• lngpR1 .5646027 .0311939 18.100
  0.000 .5032127 .6259927
• lndcR .1023198 .0206559 4.954
  0.000 .0616688 .1429707
• pshlndcR -.0464461 .0321457 -1.445
  0.150 -.1097092 .0168171
• lnicR .1083551 .0159321 6.801
  0.000 .0770005 .1397096
• pshlnicR .0882323 .0267616 3.297
  0.001 .0355653 .1408993
• lnginv1 -.0680625 .0203612 -3.343
  0.001 -.1081335 -.0279915
• lnrefu .1214082 .0363829 3.337
  0.001 .0498063 .1930101
• pshocks -.3167965 .1161874 -2.727
  0.007 -.5454546 -.0881384
• feb .0122338 .0046646 2.623
  0.009 .0030539 .0214137
• mar .0143259 .0051619 2.775
  0.006 .0041672 .0244847

19
Historic Simulation Results
20
Historic Simulation Results
21
Historic Simulation Results
22
Historic Simulation Results
23
Historic Simulation Results
24
Historic Simulation Results
25
Further Work

• Resolve GDP issue for domestic production
  regression
• Inclusion of omitted variables to improve fit
• Environmental regulations (fuel formulation)
• Tax laws
• Weather forecasts (heating/cooling fuel demand)
• Counter-historic simulations and predictions
• Add
• Refinery yield
• Refinery mix
• Capacity additions and retirement
• Exploration
• Move on to other sectors!