ShortTerm Energy Forecasting

1
Short-Term Energy Forecasting

• A Presentation to the Bangladesh Ministry of
  Energy and Natural Resources
• September 20 - 23, 1999
• Tancred C. Lidderdale
• Energy Information Administration
• U.S. Department of Energy

2
Presentation Outline

• Overview of Short-Term Energy Forecast Methods
  and Products
• Techniques for Creating a Bangladesh Short-Term
  Energy Forecast
• Forecasting Energy Prices and Transportation
  Demand
• Forecasting Natural Gas and Electricity Markets

3
Overview of Short-Term Energy Forecast Methods
and Products

• EIA Short-Term Integrated Forecasting System
  (STIFS)
• STIFS-Based Products

4
The Short-Term Integrated Forecasting System
(STIFS)

• National-level model of U.S. energy demand,
  supply, prices
• Data Frequency Monthly data supplied by EIA and
  other sources
• Forecast Horizon 15 to 24 months out
• Structure
• Over 1,000 variables
• 100 estimated series (46 demand, 36 supply, 18
  price)
• Software PROC MODEL (SAS).

5
STIFS-Based Products

• Short-Term Energy Outlook
• Printed Publication - 2 times / year
• Published on Internet - monthly
• PC Model - monthly
• Microsoft Windows version of the SAS-compiled
  STIFS simulation/forecasting model. Updated
  monthly and available for download from the EIA
  FTP site.
• Analysis Reports - periodic
• Impacts of price shocks, energy taxes,
  environmental regulations, etc. of supply, demand
  and prices
• Model Documentation - periodic

6
Short-Term Energy OutlookInternet Page
7
(No Transcript)
8
PC Forecast Model
9
(No Transcript)
10
What Do Our Customers Read?
11
Creating a Forecasting Model
12
Put the Model in Context

• Who is the audience?
• What will the forecasts be used for?
• What resources (staff and computers) will you
  have for developing and maintaining the model?

13
Model Scope and Detail

• Forecast supply as a national total, by regions
  and/or by individual production units (e.g.,
  natural gas field or electric utility plant)
• Forecast demand as a national total, by regions
  and/or by demand sectors
• Measurement units (e.g., physical units, Btus,
  oil equivalent, greenhouse gases, etc.)

14
Formal Model Development

• Identify the concepts to be forecasted
• Derive forecasting relationships
• economic theory, engineering or institutional
  knowledge
• Consider alternative forecasting methodologies
• structural vs. non-structural

15
Structural vsNon-Structural Models

• Structural Models - view and interpret economic
  data through the lens of a particular economic
  theory.
• Non-Structural Models - attempt to exploit the
  reduced-form correlations in observed time
  series, with little reliance on economic theory.

16
Non-Structural ModelsUnivariate Processes

• Autoregressive Process - current value is a
  weighted average of its own past values plus a
  random shock.
• Moving-Average Process - current value is a
  weighted average of current and lagged random
  shocks alone.
• Autoregressive Moving Average (ARMA) Process

17
Non-Structural ModelsMultivariate Processes

• Augmented ARMA Model
• Vector Autoregressive Model

18
Structural ModelsMultivariate Processes

• Multivariate Regression
• Ordinary least squares
• Two-stage least squares
• Three-stage least squares
• Full information maximum likelihood
• Two-stage least absolute deviations
• Neural Networks

19
Recommendation Keep It Simple

• Use forecasts produced by other government,
  academic, industry, or private organizations.
• macroeconomic variables
• capacity variables
• world oil price
• Use simple non-structural equations to start. Add
  structure later, estimating using ordinary
  least squares .

20
Forecast Model Energy Linkages
Coal Demand
Electricity generation from coal
Electricity generation from coal, natural gas and
residual fuel
Electricity Supply/ Demand
Electricity, Natural Gas, and Coal Prices
Natural Gas price to electric utilities
Electric utility demand - distillate and
residual fuel
Natural Gas Stocks and Demands
Natural Gas Supply/ Demand
Natural Gas Prices
Diesel and residual fuel prices
Industrial sector demand for natural gas
Natural gas price to electric utilities
and industrial sector
Petroleum Products Demand
Motor gasoline, distillate and residual fuel
demands
Motor gasoline, distillate residual fuel, and Jet
fuel prices
Petroleum Prices
Natural gas liquids production
Inventories of motor gasoline, distillate, and
residual fuel
Petroleum Products Supply
Motor gasoline, distillate, and Jet fuel prices
21
Forecasting Process

• Update Historical Database
• Copy all data history from original sources
• Manually enter data not available in historical
  databases
• Generate required data transformations (e.g.,
  deseasonalized series)
• Update Model Structure
• Revise equations (if necessary) and re-estimate
  model with new historical data
• Enter Exogenous Forecast Data
• Output and Distribution
• Produce forecast and evaluate results. Adjust
  forecast results using add factors where
  necessary
• Make final runs, save solutions, generate reports

22
Forecast Model Data Flow
23
Software Issues

• Data Management
• Estimation Methods and Analysis
• Simulation Methods
• Output Options

24
PC-Based Software
25
EIA Short-Term Forecast Contacts
General Questions David Costello
(202/586-1468) World Oil Prices/International
Petroleum Doug MacIntyre (202-586-1831) Macroeco
nomic Kay A. Smith (202-586-1455) Energy
Product Prices Neil Gamson (202-586-2418) Petrol
eum Demands Michael Morris (202-586-1199) Petrol
eum Supply Tancred Lidderdale
(202-586-7321) A.H. Payne (214-720-6160) Natural
Gas Evelyn Amerchih (202-586-8760) Hafeez
Rahman (214-720-6160) Coal Elias Johnson
(202-586-7277) Byung Doo Hong (202-426-1126) Elec
tricity Evelyn Amerchih (202-586-2867) Rebecca
McNerney(202-426-1251)
26
Some Demonstrated Model Properties
27
Energy Demand Impacts 10 Colder Winter
6
5
4
3
2
1
0
04/97
07/97
10/97
01/98
04/98
07/98
10/98
-1
Percent Deviation from Base
28
Energy Demand Impacts 1 Percent per Year Higher
Economic Growth
1.5
Total Energy
Demand
Total Petroleum
1.0
Demand
Total Nat. Gas
Demand
0.5
Total Coal
Demand
Total Electricity
Sales
0.0
01/97
04/97
07/97
10/97
01/98
04/98
07/98
10/98
Percent Deviation from Base
29
A U.S. Hydroelectric Power Scenario Continued
High Hydroelectric Availability
35
No Change from Previous Year
30
25
Billion Kilowatt-hours
20
Base
15
10
01/95
07/95
01/96
07/96
01/97
07/97
01/98
07/98
04/95
10/95
04/96
10/96
04/97
10/97
04/98
10/98
30
U.S. Electricity Supply Impacts Continued High
Hydroelectric Availability
8
6
Elec. Utility
Hydro.
4
Generation Bill
Kwh
Elec. Utility Coal
2
Billion Kilowatt-hours
Generation Bill
Kwh
0
04/97
07/97
10/97
01/98
04/98
07/98
10/98
Elec. Utility Nat.
Gas Gen. Bill
-2
Kwh
-4
-6
Deviation from Base
31
U.S. Natural Gas Market Impacts High Hydro Case
2
1
Total Natural Gas
0
In Underground
04/97
07/97
10/97
01/98
04/98
07/98
10/98
Storage
-1
Natural Gas Spot
-2
Price
-3
Total Nat. Gas
-4
Demand
-5
-6
Percent Deviation from Base
32
Evaluating Forecast Error
33
Evaluating the Forecast Model

• Within-sample forecasting error
• Post-sample forecasting error with predicted
  values for right-hand side variables
• Post-sample forecasting error with actual values
  for right-hand side variables.

34
Measurements of Error

• Regression Model Error
• Mean Error
• Mean Square Error (MSE)
• Root Mean Square Error (RMSE)
• Mean Absolute Error (MAE)
• Mean Percentage Error (MPE)
• Mean Absolute Percentage Error (MAPE)
• Thiel Statistics

35
Forecast Uncertainty

• Account for the stochastic nature of model
  equations and exogenous (right-hand side)
  variables
• Use a Monte Carlo procedure
• Requires information on
• regression equation error distributions
• regression equation estimated parameter
  covariances
• probability distribution of exogenous variables
• Repeated simulations following random draws
  from the distributions of the equation error,
  parameters, and exogenous variables

36
(No Transcript)