NEMS Forecast Evaluation Methodology
This is a working document
prepared as a job of work (DE-AP01-06EI38129.A000) on behalf of the Energy
Information Administration (EIA) in order to solicit advice and comment on
statistical matters from the American Statistical Association Committee on
Energy Statistics. The topics presented
here will be discussed at EIA's fall 2006, meeting
with the Committee to be held October 5 and 6, 2006.
Summary
The purpose of this note is to initiate the consideration of a methodology for assessing the accuracy of National Energy Modeling System (NEMS) projections. NEMS is configured to project future energy product production and consumption in a fashion that accounts for the wide range of detailed circumstances that result in multi-market economic equilibria for energy products. In addition to market forces NEMS accounts for technological change and the impact of government actions and policies. The evaluation methodology advocated here calls for constructing statistical approximations of important energy market relationships implicit to NEMSS, e.g., crude oil supply. The approximations will be derived via regression analysis of the NEMS solutions prepared in support of the Annual Energy Outlook (AEO). The approximations are to be specified to account for important, explanatory relationships, e.g., the elasticity of crude oil supply to the price of oil. Based on this, “differences” between NEMS projections and the actual values of the variables projected can be partitioned among general uncertainty, “errors” in projecting explanatory variables, structural changes in market behavior, and transitory influences such as the weather. An illustration of the approach is provided for crude oil supply utilizing the versions of NEMS used in support of the 1998 – 2000 AEO’s.
September 2006
NEMS Forecast Evaluation Methodology
Prologue. The purpose of the note below is to identify the
goal of identifying differences between NEMS projections and actual data; and,
given this, to partition the differences found among their major influences.
The sense of the note is to identify what these influences might be and to
identify a practical method for some of them for evaluating their relative
roles in contributing to differences between NEMS projections and the eventual
values of the variables projected. If the method outlined were implemented,
there could be diagnostic value in assessing the variable/parameter
sensitivities in a current version of NEMS compared to past versions. From a
model management perspective, the sources of forecast differences might be used
to support priorities for model component development. As it stands, since the
analysis is not currently conducted, such potential benefits have not been
demonstrated. As a result, as a resource allocation priority compared to other
modeling issues, identifying and partitioning forecast differences in the
manner proposed is problematical for the offices responsible for model
maintenance and development.
An
alternative perspective is that such an analysis is itself an information
product that informs the community for whom the projections of energy
production and consumption are prepared. In brief, that an assessment of
forecast accuracy in the fashion proposed is an item of good professional
practice. Accordingly, an analysis of forecast accuracy and the sources of
differences between forecast and actual values might be prioritized apart from,
rather than within, the process of resource allocation for model development
and maintenance. While issues of method are important, perhaps the most
important issue associated with this note is that of deciding if the analysis
of forecast differences is a necessary activity in association with the production of the
forecasts themselves.
Background. In principle, an evaluation of NEMS forecast
accuracy that accounts for the various sources of differences between forecast
and actual values can be readily accomplished. The form of the model, in brief,
involves projecting a variety of important influences on energy markets, e.g.,
technology and the parameters of consumer behavior. Based on the projections of
these “conditional variables” the model then determines the corresponding
multi-energy-market equilibria and the associated
projections of energy product prices and quantities. When a forecast period
actually occurs, and the actual values of the conditional variables are known,
the model can be re-run using the actual values for model assumptions and the
resulting projections compared to the actual prices and quantities. The impact
of individual conditional variables can be determined via differentially using
the original projections and the actual values, ceteris paribus. Routines within the model might also be amended to
account for changes in government policies, technologies, or consumer behavior.
Extra-model influences such as weather could also be accounted for in the
historical period, independent of NEMS. In general, the size of the model and
need to maintain, and then run,
non-current versions of the model over many years make this approach
impracticable (at least so far for NEMS).
An alternative is to isolate the explanatory variables at issue for each model component to those that correspond to the basic economic forces associated with the energy markets represented by NEMS. Given this, the underlying interrelationships, e.g., price elasticities of supply and demand, can be determined for a current version of the model and then retained. In general, the method would be to selectively change important assumptions, ceteris paribus, and catalogue the corresponding sensitivities. Later, these sensitivities could be applied to actual data to determine the basis for forecast differences without having to archive and re-run the corresponding version of NEMS (This was done in Costello (2006) Reduced Form Energy Model Elasticities from EIA’s Regional Energy Model (RSTEM) , released 5/9/2006 as a one time report, for price and weather effects) A differential accounting for the impact of differences in projected versus actual “conditional variables” could still be approximated from the sensitivities. As before, extra-model, transitory influences on the historical values could also be accounted for. The basis for the approach is to detect the important market sensitivities implicit to NEMS’s representation of energy markets for each version of the model.
Proposed Method. There are basically two methods for extracting the underlying sensitivities implicit to NEMS. One, as already noted, is through comparative statics experiments with the model components themselves. The approach would be to solve the components, relative to a base or reference case, with each important assumption, e.g., the price of oil, changed, individually (with all other assumptions held constant), and compare results. The measure of sensitivity usually brought to comparative statics results is the elasticity (the ratio of the percentage change in the variable value solved for in the model , e.g., crude oil supply, to the percentage change in the assumption, e.g., the price of oil). Forecasts of future energy production and consumption could then be compared to the values of these variables that actually occur. Using the actual values of the conditional variables and the elasticities associated with the model version at issue, the differential impact of errors in projecting the conditional variables could be assessed. Although there have been individual studies of the sensitivities of certain NEMS components using this method, it is not generally used, e.g., in the recently released Annual Energy Outlook Evaluation 2005, DOE/EIA-0640(2006), July 2006, there is no breakout of forecasting error due to errors in projecting conditional variables, or any other specific source, although the sources of error in general are initially enumerated. Extracting the sensitivities using this method would be relatively expensive.
The method proposed here as an alternative is to estimate the market sensitivities implicit to NEMS based upon the solutions prepared each year for the AEO. Among other ways, NEMS solutions are saved in a binary format and can be processed by the PC-based graphic interface Graf2000. Solution data have been saved in this format starting with those prepared for the 1998 AEO. This utility includes a regression component that enables regression analysis to be conducted using resident data; and, a data extraction routine that enables any collection of solution series to be extracted and input to other statistical procedures. Initially, the proposed method entails no additional resource requirements in terms of running NEMS or archiving versions of NEMS for use at a future time.1 Instead, the solution sets for the AEO versions of NEMS can be pooled for the projections to be evaluated at a future time.
The basic approach is to specify the underlying energy market supply and demand relationships in terms of their important, explanatory variables; and, given this, to estimate the relationships based upon NEMS solution data. The results of the estimates provide a description of the NEMS model version in terms of how energy markets are represented. The actual specifications utilized would be guided by the expertise of the EIA staff responsible for developing and maintaining individual model components. Since the solution sets can be readily archived, the actual regression analyses need not be conducted until the time that a model version is to be evaluated, although the outcomes of the regressions can have immediate diagnostic use in NEMS development.2 A demonstration of the general success of representing NEMS components via regression analysis is given in: Buck and Lady, “Approximation of Large, Computer-Based Economic Models,” presented at annual meetings of International Atlantic Economic Association on October 9, 2005 in New York City, New York. A copy of the paper can be downloaded from the link: http://optima-com.com/buck_lady/AES_Paper.htm
It is proposed to configure the means of performing the regression analyses in a fashion that can be routinely conducted and maintained by EIA staff. The goal of the statistical analyses is to enable the errors in EIA forecasts to be explicitly decomposed with respect to the following influences:
Transitory Influences, e.g., weather, strikes, accidents, embargoes not accounted for in the projections.
Institutional Influences, e.g., changes in laws and regulations and changes in data series definitions compared to model assumptions.
Structural Influences, e.g., changes in resource availability or energy use technology compared to model assumptions.*
Errors in Projecting Conditional Variables, e.g., differences in the eventual values of activity drivers and other exogenous factors such as GDP and population.*
Errors in Behavioral Parameters, e.g., changes in consumer price sensitivities compared to those assumed by the forecasting methodology.*
Uncertainty, e.g., the residual error of the projection method.
The methodology for partitioning forecast differences among (such as) the influences outlined above is as follows for the items indicated by “*”, given the availability of actual data for previously forecast values.
The equations derived to represent the important relationships of supply and demand are re-run using the actual values for the explanatory variables. The actual values are substituted for the values used (or solved for) in the original projections, one explanatory variable at a time. This enables the identification of the influence of each explanatory variable separately. The equation is then re-run with all explanatory variables assigned their actual values. For this case, the residual error is due to general forecasting "uncertainty" or other, structural changes. Structural change is the issue of whether or not the values of the coefficients in the forecasting equation have changed for the forecast period compared to the model version to which the equation had been "fit." The equation is re-estimated and the results compared to the outcome of the estimation used to approximate the characteristics of the model. One method to assess if there were significant "differences" between the original, and revised, estimate is the Chow test (Chow, Gregory, "Tests of Equality Between Sets of Coefficients in Two Linear Regressions," Econometrica, 28, (July 1960), pp. 591-605.). Additional methods may be proposed by the ASA committee as the project is ongoing. Events influencing the actual data not accounted for by the forecasting equation will be identified and evaluated by EIA staff as appropriate, e.g., weather effects for consumption variables can be determined independent of the regressions which are based on the NEMS assumption of “normal” weather..
Example for crude oil supply. An illustration of the above procedure is provided here for projections of crude oil (and lease condensate) supply. The market supply of crude oil is dependent upon the size of the developed resource base and the economics of oil extraction. Over time there are the additional considerations of the pace of resource development and discovery. Although (such as) projections of oil reserves and reserve additions are projected by NEMS, acquiring and processing these data was beyond the scope of preparing this initial proposal. Accordingly, for illustrative purposes, a simple specification of oil supply was brought to the NEMS solution data. For this, the explanatory variables were the world oil price (WOP) and crude oil production lagged one year. This “lagged endogenous variable” is a surrogate for such as projections of economic reserves. Accordingly, the example is intended to be illustrative and indicative of the methodology proposed, rather than a definitive example of evaluating NEMS projections of crude oil supply.
The most recent year for
historical data is 2005. For this illustration, NEMS projections of crude oil
production for this year were compared to actual data for the 1998-2000
versions of NEMS.3
The regression specification to be applied to NEMS solution data to represent the implicit crude oil market supply relationship is given by:
Qt = a + bWOPt + cQt-1, (1)
where Qt = crude oil production in year t and WOPt = the world oil price in year t. As noted, the lagged endogenous variable is intended to represent the underlying trends in oil reserves; and also, the associated inter-temporal impact of changes in the WOP. For purposes of comparison, this specification was used to estimate the crude oil supply function using actual data from 1984 – 2005. A plot of the results is given below in Graphic 1. Detailed regression results are given in Appendix A.
Graphic 1: Actual and Regression Results for Crude Oil Supply

As summarized in the tables below, the fit of the supply function to the data was quite good (the average, absolute percent error was 1.81%); however, the statistical significance of the price elasticity was low.
Equation (1) was fit to the solution data for the five principal scenarios for each of the 1998-2000 AEO. The lagged endogenous variable was always significant and played by far the dominant role in the accuracy of the projection. The price effect, while small, was significant for two of the three versions of NEMS considered.
For the three versions of NEMS, the crude oil production projection for 2005 was high for all of them, ranging from 13.68% high in 1998 to 4.7% high in 2000. The equation was re-run for each NEMS version using actual data for both the WOP (adjusted to the monetary units of each NEMSS version) and lagged production, and then with actual data with each individually, using the NEMS projection of the other variable. The “back cast” using actual data for both explanatory variables was more accurate in for two of the three NEMS versions. Inspection revealed that the NEMS WOP projection was significantly low for all three versions of NEMS. Accordingly, use of the actual WOP with the oil supply equation made the resulting projection less inaccurate, i.e., even higher. Alternatively, simply using the correct value for lagged quantity led to a mush more accurate projection for two of the three versions of NEMS with little difference for the third.
Not accounted for by the analysis is the inter-temporal impact of the WOP on oil resource development. The historical time series of the WOP for the two decades before 2005 shows a flat, and significantly lower profile than the price in 2005, as given below.
Year
Nominal WOP $2004 WOP
1984 28.54 46.02
1985 26.67 41.73
1986 16.16 24.74
1987 17.65 26.31
1988 14.08 20.29
1989 17.68 24.55
1990 21.13 28.25
1991 18.02 23.28
1992 17.75 22.41
1993 15.72 19.4
1994 15.18 18.35
1995 16.78 19.87
1996 20.31 23.61
1997 18.11 20.71
1998 11.84 13.39
1999 17.23 19.21
2000 27.53 30.04
2001 21.82 23.25
2002 23.91 25.04
2003 27.69 28.42
2004 36.07 36.07
2005 49.34 48.00
How this price profile is accounted for in the NEMS crude oil supply projection methodology might be accounted for more explicitly in a more detailed regression specification that was used here. Finally, hurricane Katrina resulted in a supply disruption sufficient to account for some the NEMS projection differences. A recent analysis concluded that the oil supply disruption due to Katrina was .647 quads, or 5.9% of measured total production.4
In summary, the source of NEMS
projection differences for crude oil supply is not significantly in the
immediate-run impact of oil prices, although, compared to the regression
results using actual data, the WOP-effect on oil supply is generally larger and
more significant. The explicit accounting for projections of reserves would
provide a more satisfactory partition of the basis for projection differences
from actual data. Significantly, when the effect of Katrina is accounted for,
actual production (10.84+.647=11.487) falls within the NEMS high and low oil
supply cases for each versions of NEMS. For each version of NEMS the residual
“uncertainty” not accounted for by the influences measure is small. The results
of the regression analyses are provided below in Tables 1-4. Next, notes
defining the basis for the table entries and elaborations of some of the points
above are provided. Finally, in an Appendix, the results of the regression
analyses are provided for the historical data and for the 1998-2000 versions of
NEMS.
Table 1: Domestic Crude Oil/Lease Condensate Production in 2005 (Quads)5
|
AEO Year |
1998 |
1999 |
2000 |
History |
|
Actual |
10.84 |
10.84 |
10.84 |
10.84 |
|
NEMS High Case |
13.31 |
12.83 |
11.71 |
|
|
NEMS Base Case (%∆) |
12.32 (13.68) |
12.31 (13.56) |
11.35 (4.7) |
n/a |
|
NEMS Low Case |
10.88 |
11.54 |
10.58 |
|
|
NEMS Sim (%∆) |
12.33 (13.75) |
12.32 (13.67) |
11.43 (5.44) |
11.26 (3.9) |
|
R2 |
.981 |
.998 |
.991 |
.981 |
|
Average Absolute % SIm
Difference |
.162 |
.163 |
.366 |
1.81 |
|
Backcast Actual All (%∆) |
11.50 (6.17) |
11.40 (5.12) |
12.07 (11.36) |
n/a |
|
Backcast Actual WOP only (%∆) |
12.46 (14.95) |
12.34 (13.87) |
12.06 (11.23) |
n/a |
|
Backcast Actual Lag only (%∆) |
11.38 (4.97) |
11.37 (4.92) |
11.44 (5.57) |
n/a |
Table 2: World Oil Price (WOP) in 2005 ($/Barrel)5
|
AEO Year |
1998 |
1999 |
2000 |
History |
|
$Units |
$1996 |
$1997 |
$1998 |
$2004 |
|
Actual |
41.40 |
41.98 |
42.45 |
48.01 |
|
NEMS (%∆) |
20.19 (-51.1) |
19.24 (-54.15) |
20.49 (-51.73) |
n/a |
|
NEMS Sim
Elasticity (t-statistic) |
.011 (2.24) |
.002 (.54) |
.054 (5.78) |
.004 (.221) |
|
Long Run Elasticity |
n/a |
n/a |
.527 |
.136 |
Table 3: Domestic Crude Oil/Lease Condensate Production in 2004 (Quads)
|
AEO Year |
1998 |
1999 |
2000 |
History |
|
Actual |
11.503 |
11.503 |
11.503 |
11.503 |
|
NEMS (%∆) |
12.44 (8.18) |
12.42 (7.94) |
11.49 (-.14) |
n/a |
|
NEMS Sim
Elasticity (t-statistic) |
1.02 (115) |
1.05 (128) |
.898 (31.2) |
.996 (30.3) |
Table 4: Sources of NEMS Base Case Projection Differences
|
AEO Year |
1998 |
1999 |
2000 |
|
Total % Difference |
+13.68 |
+13.56 |
+4.7 |
|
% Due to Price Projection |
-1.2 |
-.20 |
-5.79 |
|
% Due to Lag Projection |
+8.78 |
+8.75 |
-1.4 |
|
% Due to Weather |
+5.9 |
+5.9 |
+5.9 |
|
% Residual |
+.20 |
-.89 |
-3.41 |
Notes
1. A fair number of NEMS scenarios are run in support of each AEO, e.g., around forty were run for the AEO2006, in order to reveal important sensitivities and uncertainties. As the use of regression analysis of solution data is developed for the purposes proposed here, some number of additional runs might be formulated to facilitate the isolation of important influences for the ultimate evaluation of NEMS projections.
2. Large changes in year to year implicit sensitivities could detect anomalies in model development.
3. NEMS solutions for “early” years depend all, or in part, upon projections utilizing EIA’s short term forecasting system. Accordingly, for the illustration here, NEMS projections are assessed only when more than five years out. Since the NEMS solutions for an AEO are finalized in the fall of the year prior to publication, this makes the solutions for the 2000 AEO the most recent for assessment. Solution data of the sort proposed for processing are only available starting for the solutions prepared in support of the 1998 AEO.
4. In a
5. “NEMS Sim” refers to the projections provided by the regression equation. For this and the tables below: Actual crude oil production was from the AER 2005, Table 1.2 as posted on the EIA website on 7/27/06; the value for the actual WOP used was that quoted in the August 2006 MER for 2005 as “Landed Cost of Imported Oil;” the price index used to adjust monetary units to those used in each of the nine versions of NEMS was the “GDP Chain-Type Price Index,” as taken from Table B-3 (Appendix B) of the 2006 Economic Report of the President. The percentage base for all differences is the actual value of the variable. The elasticities are evaluated at the means of the variables used for the forecasts, e.g., the WOP elasticity is equal to b(WOP/Q) where “b” is the estimated regression coefficient and {WOP,Q} are the means of the variables. The “long run” elasticity accounts for the impact of a change in the WOP being transmitted to future years via the lagged endogenous variable. The effect is convergent for 0 < c < 1, and multiplies the short run elasticity by (1/(1-c)). In the table, “n/a” identifies cases for c > 1 or b < 0. All of the regressions using NEMS solution data were run for data pooled from all five principal scenarios prepared for the AEO using the projected values for the years 2005-2020 inclusive. The high and low oil production cases are, respectively, from the high and low world oil price scenarios.
Appendix Crude Oil Supply
Regression Results
History
Endogenous
Variable:
Table
#1 Total Energy Supply and Disposition Summary (quadrillion Btu, unless
otherwise noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate
Exogenous
Variables:
#
1) Table #1 Total Energy Supply and Disposition Summary (quadrillion Btu,
unless otherwise noted)
Supply, Disposition, and Prices: Prices
(2004 dollars per unit): World Oil Price
($ per bbl) 10/
#
2) Lagged Table #1 Total Energy Supply and Disposition Summary (quadrillion
Btu, unless otherwise noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate
Exogenous
Variable Mean Coefficient Elasticity t-statistic
Variable#
1 25.56688
.002101 .003739 .221085
Variable#
2 14.74595 .970695 .996464 30.2608
Constant -.00292
Endogenous Mean SER R-sq LR-Multiplier
Variable 14.36462 .341296 .980845 34.12386964681
Data
pooled for the years
1985 to 2005 for
Historical Data
Table
#1 Total Energy Supply and Disposition Summary (quadrillion Btu, unless
otherwise noted)
Supply,
Disposition, and Prices: Production:
Crude Oil & Lease Condensate
Year Actual Estimated %abs((Est-Act)/Act)
1985
18.992 18.38042 3.22
1986
18.376 18.4845 0.59
1987 17.675 17.88984 1.215
1988
17.279 17.19674 0.476
1989
16.117 16.8213 4.37
1990
15.571 15.70112 0.836
1991
15.701 15.16068 3.441
1992
15.223 15.28505 0.408
1993
14.494 14.81474 2.213
1994
14.103 14.10488 0.013
1995
13.887 13.72855
1.141
1996
13.723 13.52672 1.43
1997
13.658 13.36143 2.171
1998
13.235 13.28296 0.362
1999
12.451 12.88458 3.482
2000
12.358 12.14631 1.713
2001
12.282 12.04177 1.956
2002
12.163 11.97176 1.572
2003
12.026 11.86335 1.353
2004
11.503 11.74644 2.116
2005
10.84 11.26384 3.91
Average
Absolute Percent Error = 1.808953

AEO 1998 Simulation Results
Endogenous
Variable:
Table
#1 Total Energy Supply and Disposition Summary (Quadrillion Btu per Year,
Unless Otherwise Noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate.
Exogenous
Variables:
#
1) Table #1 Total Energy Supply and Disposition Summary (Quadrillion Btu per
Year, Unless Otherwise Noted)
Supply, Disposition, and Prices: Prices
(1996 dollars per unit): World Oil Price
($ per bbl) ...
#
2) Lagged Table #1 Total Energy Supply and Disposition Summary (Quadrillion Btu
per Year, Unless Otherwise Noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate.
Exogenous
Variable Mean Coefficient Elasticity t-statistic
Variable#
1 21.09386 .006141 .011359 2.2424
Variable#
2 11.52222 1.012341 1.022852 115.76361
Constant -.390142
Endogenous Mean SER R-sq LR-Multiplier
Variable 11.40381 .055787 .998362 -81.03071063933
Data
pooled for the years
2005 to 2020 for the
solutions given below:
aeo98b.ran hmac98.ran lmac98.ran hwop98.ran lwop98.ran
Estimation Results for aeo98b.ran
Table
#1 Total Energy Supply and Disposition Summary (Quadrillion Btu per Year,
Unless Otherwise Noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate.
Year Actual Estimated %(Est-Act)/Act
2005
12.32254 12.33081 0.067
2006
12.20534 12.20866 0.027
2007
12.10194 12.09089 0.091
2008
11.99377 11.98713 0.055
2009
11.89241 11.87854 0.117
2010
11.78969 11.77682 0.109
2011
11.69125 11.67369 0.15
2012
11.57265 11.57496 0.02
2013
11.44692 11.45567 0.076
2014
11.25463 11.32913
0.662
2015
11.09442 11.13527 0.368
2016
10.9388 10.97394 0.321
2017
10.78883 10.81739 0.265
2018
10.65493 10.6666 0.109
2019
10.54534 10.53207 0.126
2020
10.42591 10.4224 0.034
Average
Absolute Percent Error = .1623125

AEO 1999 Simulation Results
Endogenous
Variable:
Table
#1 Total Energy Supply and Disposition Summary (Quadrillion Btu per Year,
Unless Otherwise Noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate.
Exogenous
Variables:
#
1) Table #1 Total Energy Supply and Disposition Summary (Quadrillion Btu per
Year, Unless Otherwise Noted)
Supply, Disposition, and Prices: Prices
(1997 dollars per unit): World Oil Price
($ per bbl) ...
#
2) Lagged Table #1 Total Energy Supply and Disposition Summary (Quadrillion Btu
per Year, Unless Otherwise Noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate.
Exogenous
Variable Mean Coefficient Elasticity t-statistic
Variable#
1 21.38648 .000966 .001807 .546316
Variable#
2 11.55995 1.038382 1.049795 127.881358
Constant -.590025
Endogenous Mean SER R-sq LR-Multiplier
Variable 11.43428 .049681 .997553 -26.05387942264
Data
pooled for the years
2005 to 2020 for the
solutions given below:
aeo99b.ran hmac99.ran lmac99.ran hwop99.ran lwop99.ran
Estimation
Results for aeo99b.ran
Table
#1 Total Energy Supply and Disposition Summary (Quadrillion Btu per Year,
Unless Otherwise Noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate.
Year Actual Estimated %(Est-Act)/Act
2005
12.31041 12.32166 0.091
2006
12.25971 12.21235 0.386
2007
12.1361 12.16043 0.2
2008
12.02927 12.03221 0.024
2009
11.9313 11.9214 0.083
2010
11.83284 11.8198 0.11
2011
11.74482 11.71769 0.231
2012
11.6603 11.62642 0.291
2013
11.55284 11.53878 0.122
2014 11.43066 11.42729 0.03
2015
11.29543 11.30053 0.045
2016
11.1615 11.16023 0.011
2017
11.02854 11.02131 0.066
2018
10.89726 10.88341 0.127
2019
10.71355 10.74726 0.315
2020
10.50622 10.55669 0.48
Average
Absolute Percent Error = .16325

AEO 2000 Simulation Results
Endogenous
Variable:
Table
#1 Total Energy Supply and Disposition Summary (Quadrillion Btu per Year,
Unless Otherwise Noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate.
Exogenous
Variables:
#
1) Table #1 Total Energy Supply and Disposition Summary (Quadrillion Btu per
Year, Unless Otherwise Noted)
Supply, Disposition, and Prices: Prices
(1998 dollars per unit): World Oil Price
($ per bbl) ...
#
2) Lagged Table #1 Total Energy Supply and Disposition Summary (Quadrillion Btu
per Year, Unless Otherwise Noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate.
Exogenous
Variable Mean Coefficient Elasticity t-statistic
Variable#
1 21.09438 .028584 .054042
5.77773
Variable#
2 11.16395 .897522 .898054 31.192915
Constant .53448
Endogenous Mean SER R-sq LR-Multiplier
Variable 11.15733 .068051 .990887 9.758192002185
Data
pooled for the years
2005 to 2020 for the
solutions given below:
aeo2k.ran hmac2k.ran lmac2k.ran hwop2k.ran lwop2k.ran
Estimation
Results for aeo2k.ran
Table
#1 Total Energy Supply and Disposition Summary (Quadrillion Btu per Year,
Unless Otherwise Noted)
Supply, Disposition, and Prices:
Production: Crude Oil & Lease
Condensate.
Year Actual Estimated %(Est-Act)/Act
2005
11.35001 11.42934
0.699
2006
11.24389 11.30978 0.586
2007
11.13607 11.21767 0.733
2008
11.04106 11.12372 0.749
2009
10.98465 11.04159 0.518
2010
10.95652 10.99378 0.34
2011
10.95853 10.97136 0.117
2012
10.96388 10.9763 0.113
2013
11.00002 10.98392 0.146
2014
11.00318 11.01949 0.148
2015
11.00747 11.02547 0.163
2016
11.11085 11.03214 0.708
2017
11.14443 11.12806 0.147
2018
11.16455 11.16102 0.032
2019
11.16817 11.18222 0.126
2020
11.12987 11.18829 0.525
Average
Absolute Percent Error = .365625
