Introduction
 The Residential Demand Module
  The Commercial Demand Module
  Summary
  Appendix A: Lighting Technologies in AEO98

Introduction

In 1990, the Secretary of Energy directed the Energy Information Administration (EIA) to develop the National Energy Modeling System (NEMS), based on recommendations from the National Research Council (NRC) of the National Academy of Sciences.¹ Key features implemented in NEMS include: (a) regional outputs of energy, economic, and environmental activity of the U.S. economy; (b) use of a modular modeling structure to facilitate and enable the model builders to work with particular aspects of the model independently; (c) integration of engineering and economic approaches to represent actual producer and consumer behavior; (d) use of a mid-term projection period spanning 20 to 25 years; (e) involvement of the broader energy analysis community and outside peer groups in the design and update of NEMS. Figure 1 illustrates the modular construction of NEMS and the basic information flows between modules during the solution process.

NEMS was completed at the end of 1993 and was first used to develop the Annual Energy Outlook 1994.² More recently, NEMS has been extended to 2020 and further revised to address electricity restructuring and carbon mitigation issues.³

The primary purpose of NEMS is to analyze the effects of energy policies and other pertinent influences on U.S. energy markets.⁴ Important market influences include, for example, the magnitude of economically recoverable fossil fuel resources, characteristics of the world market for energy and their effects on oil prices, and the rate of development and penetration of new energy related technologies—as well as existing or prospective government policies and actions.

Current and emerging policy questions determine the level of detail required within the structure of NEMS. For example, energy-related environmental issues have taken on a new importance as a consequence of new NO_x and particulate emission regulations issued by the U.S. Environmental Protection Agency and both the Rio Treaty and the Kyoto Protocol on greenhouse gases. In the case of carbon (or carbon dioxide), NEMS constrains national carbon emissions using price. The NEMS electricity sector is designed to measure five emissions (oxides of sulfur, oxides of nitrogen, carbon, carbon monoxide, and carbon dioxide) released in the use of energy products to generate electricity. While NEMS is designed to constrain national carbon emissions using a pricing mechanism, sulfur dioxide constraints are imposed only in the electricity generation market.

The technology representation in NEMS explicitly represents vintaged (time-dependent) energy equipment and structures (e.g., building shells) and tracks vintaged capital stock turnover; thus, NEMS is particularly useful for analyses of carbon mitigation policies. For similar reasons, NEMS contains sufficient detail in the transportation sector to project the use of reformulated fuels or alternative fuels. In addition to environmental concerns, NEMS is designed to account for existing and emerging government policies (e.g., electricity restructuring and renewable portfolio standards). The potential for the development and use of new energy-related technologies, increased use of renewable sources of energy (especially intermittent technologies), and increases in the efficiency of energy use are other features that have been incorporated in NEMS, reflecting the expected scope of present and future analytical activities.

NEMS also allows for different market structures. For example, the Annual Energy Outlook 1998 (AEO98) models California, New York, and New England as competitive electricity generation markets with marginal-cost pricing. The remainder of the U.S. electricity market is modeled with cost-of-service regulation and average-cost pricing.⁵

The representation of energy markets in NEMS focuses on four important interrelationships: (1) interactions among the energy supply, conversion, and consumption sectors, (2) interactions between the domestic energy system and the general domestic economy, (3) interactions between the U.S. energy system and world energy markets, and (4) the interaction between current production and consumption decisions and expectations about the future.⁶

Interactions among domestic energy supply, conversion, and consumption are assured through the representation of simultaneous competitive markets in achieving year-to-year energy-economy equilibrium subject to the equipment constraints imposed by a “bottom-up” approach. This approach begins by modeling the agents at a relatively disaggregated level (e.g., households by housing type and Census division) to determine from their decision rules the relative number of new home and equipment purchases for each housing type. The modeled decisions at the household level are summed to build higher levels of aggregation. The prices paid and quantities demanded for each fuel are balanced with the supply and prices offered through an iterative convergence process between supply and demand.

The general level of economic activity in sectoral and regional detail has traditionally been used as an explanatory variable or “driver” for projections of energy consumption and prices. In reality, energy prices and other energy system activities themselves influence the level of economic activity. NEMS is designed to capture feedback between the domestic economy and the energy system. The macroeconomic component of NEMS is a reduced form of the DRI macroeconomic model.⁷ Changes in energy prices from a DRI reference case cause changes to macroeconomic variables such as disposable income, new car sales, and industrial output. In turn, changes in the macroeconomy cause changes to energy service demands.

The world oil price (WOP) is a key variable in domestic energy supply and demand decisionmaking. As a result, WOP assumptions have been a key starting point in the development of energy system projections. In fact, the U.S. energy system itself exerts a significant influence on world oil markets, which in turn influence the WOP (another example of a feedback effect). World energy market supply and demand are first specified outside NEMS by a world oil model. NEMS then models the interactions between the U.S. and world oil markets through the use of import crude and product supply curves. Changes in U.S. oil markets affect world supply and demand. As a result, domestic energy system projections and the WOP are made internally consistent.

Production and consumption of energy products today are influenced by past decisions to develop energy resources and acquire energy-using capital equipment. Similarly, the production and consumption of energy in a future time are influenced by decisions made today and in the past. Current investment decisions depend on expectations about future market circumstances. For example, the propensity to invest now to develop alternative energy sources increases when future energy prices are expected to increase. Recognizing that the formation of and response to price expectations in the residential and commercial energy markets differ from those in the electricity generation and industrial sectors, NEMS allows the differential application of foresight assumptions to its individual submodules. This flexibility allows the consequences of different planning horizons and consumer preferences to be incorporated in the NEMS projections.

The NEMS Residential Demand Module (RDM) is a “structural” model of energy demand. That is, its forecasts are built up from underlying projections of demographic variables, the residential housing stock, and the energy-consuming equipment contained in the housing stock. The RDM forecasts energy consumption by Census division for seven marketed energy sources (electricity, natural gas, distillate oil, liquid petroleum gas, kerosene, coal, and wood) plus solar thermal and geothermal energy. For each of the nine Census divisions, three housing types are modeled: single family, multifamily, and mobile homes. Within each housing type, 14 end uses are modeled: space heating, space cooling, water heating, refrigeration, cooking, clothes dryers, freezers, clothes washers, dishwashers, lighting, color televisions, personal computers, furnace fans, and other uses. Of the 14 end uses, the first 10 listed are modeled with underlying technology detail. The remaining end uses are modeled on the basis of trends in energy consumption.

In developing energy consumption projections, the RDM incorporates projections of the effects of four broadly defined determinants: economic and demographic effects, structural effects, technology effects, and energy market effects. Economic and demographic effects include housing starts, population, the number of persons per household, dwelling type (single-family, multifamily, or mobile homes) and location of housing units. Structural effects include changes in the average dwelling size and changes in the mix of desired end-use services provided by energy (new end uses and/or increasing penetration of current end uses, such as the increasing popularity of electronic equipment and computers). Technology effects include changes in the stock of installed equipment caused by normal turnover and replacement of old, worn-out equipment with newer versions (which often are more energy efficient), the integrated effects of equipment and building shell (insulation level) in new construction, and the projected availability of even more energy-efficient equipment in the future. Energy market effects include the short-run effects of energy prices on energy demands, the longer-run effects of energy prices on the efficiency of purchased equipment and the efficiency of building shells, and limitations on minimum levels of efficiency for new purchases of energy-consuming equipment imposed by Federal efficiency standards.

The RDM is initialized with data characterizing housing and appliance stocks for its base year, currently 1993. The RDM relies on EIA’s Residential Energy Consumption Survey (RECS) for a large part of this initial information.⁸ RECS is a nationally representative, stratified sample based on a detailed survey of more than 6,000 households. RECS housing characteristics are derived directly from the survey data. Since no appliance-level metering of energy is performed for the RECS, its end-use consumption data are derived from a statistical analysis of monthly energy bills for the surveyed households. The 1993 edition of RECS is the latest available⁹ and was used for the AEO98. From RECS, the RDM obtains estimates by Census division for the following:

• Housing stock—number and dwelling type
  
• Average dwelling sizes in square feet (stock average and new construction)
  
• General characterizations of equipment stocks (i.e., number and type
  of energy-consuming equipment, but not efficiency)
  
• Number of occupants
  
• Estimated energy consumption by appliance.

Efficiency data for the initial stock are derived from time-series data on equipment shipments, which are generally available from trade groups.¹⁰ These data are used to develop estimates of retirements and retiring efficiency of appliances over the RDM modeling horizon.

For purchases of new and replacement equipment, a “menu” of technology characterization data provides the RDM with the equipment available for consideration at any point in the forecast horizon. The menu includes a range of current technologies taken from equipment on the market today. The menu changes in response to efficiency standards by dropping equipment that fails to meet the standard from the menu. For new and improved technologies, the menu reflects combinations of greater energy efficiency and/or lower equipment costs. Future technologies are generally based on the evolution of current technologies, not major innovations.¹¹

Other RDM inputs include estimated housing stock retirement rates, historical heating and cooling degree-days, calibration information for historical and near-term projection years from EIA’s State Energy Data System and Short-Term Energy Outlook, assumed appliance stock minimum and maximum life expectancies, housing shell integrity for new construction, and fuel choices for single- and multifamily housing types (from the Census Bureau’s Characteristics of New Housing, 1994).

The components of the RDM and its interactions with NEMS are shown in Figure 2. NEMS provides the RDM with forecasts of residential energy prices and housing starts by building type and Census division. These inputs are then used by the RDM to develop forecasts of energy consumption by fuel and Census division, which are passed back to the NEMS integrating module.

Figure 2.  NEMS Residential Demand Module

Source:  Energy Information Administration,
Office of Integrated Analysis and Forecasting

Inside the RDM, the Housing Stock Submodule begins the projection cycle for a particular model year by determining the total residential housing stock. The current-year stock is developed by adjusting the previous year’s housing stock for new starts and housing retirements for each of the three building types.

The next action is taken by the Appliance Stock Module, which (1) removes appliances for housing units that were retired from the stock, (2) determines appliances required for new construction, and (3) retires appliances in surviving housing that have reached the end of their useful life. For cases (2) and (3), appliances will need to be purchased. Information on these requirements is passed on to the Technology Choice Submodule, which is discussed in detail below. Briefly, the Technology Choice Submodule combines projected energy costs of an appliance (derived from projected energy prices and appliance energy and operating requirements) with equipment cost and performance data to determine new purchases by efficiency level. The submodule is calibrated to recent shipment efficiency data by adjusting parameters that relate to consumer preferences for energy efficiency.

The Shell Integrity Submodule sets shell improvements, which are of two types: (1) autonomous changes to shell efficiency for new and existing construction (improvements that are independent of price), and (2) energy-price-induced changes to shell efficiency for existing construction. The second effect operates as a “ratchet”— there are no reductions in shell efficiency as energy prices decline. Adjustments to heating and cooling requirements are based on the shell improvements set by the Shell Integrity Submodule.

Finally, the Energy Consumption Submodule determines the end-use energy requirements and returns consumption estimates to the NEMS integrating module. The energy requirements include adjustments for changes to the number of appliances in the equipment stock and the energy efficiency of the stock as provided by the Technology Choice Submodule. The Energy Consumption Submodule also makes a variety of other intensity adjustments for effects that include:

• Changes in real energy prices via the short-term price elasticity of demand
  
• Efficiency rebound effects, which operate similarly to the short-run price elasticity (e.g., increasing thermostat set points for heating when purchases of highly efficient equipment reduce the marginal costs of doing so)
  
• Short-term weather effects (the model needs to be adjusted for base-year weather versus long-term expectations of climate)
  
• Changes in consumption for heating and cooling due to increases in shell integrity
  
• Changes due to demographic effects, such as the effects on heated water consumption of changes in the number of persons per household
  
• Changes for structural effects, such as increasing average floorspace for single-family housing units.

When all these adjustments have been made, the Energy Consumption Submodule computes energy consumption by fuel and Census division and passes the information back to the rest of NEMS for response.

Figure 3 outlines the general workings of the Technology Choice Submodule. The choice of equipment incorporates important residential market attributes—the tendency not to replace equipment until it fails and the tendency to replace equipment with the same technology when it does fail. Technology choice is modeled in two stages in the RDM. The first stage selects fuel and general equipment type (for example, natural gas furnace for space heating). The second stage selects the efficiency of the equipment type selected in the first stage (the furnace).

For appliance purchases in new construction, both stages of equipment selection are required. For replacement equipment purchases, the market is divided into two groups. One group keeps the same equipment (and fuel) and thus skips the first stage of technology choice. The other group considers technology switching, and after adding any fuel switching costs (e.g., running a gas line to a formerly electric home), proceeds through both stages of equipment selection. In both choice stages, the RDM uses a logistical choice function to project the equipment shares.¹² By design, this functional form results in some purchases being made from each type of available equipment.

The shares of a specific piece of equipment depend on two factors relative to those of other alternatives: the installed equipment cost and the annual operating costs (fuel plus maintenance). These two cost factors, combined via the logit parameters for installed cost and operating cost, determine the share of a particular equipment type relative to others. The parameters are chosen to calibrate the average efficiency choices to those observed for recent shipment data, when available. Approximate discount rates can also be derived from the logit parameters. The discounts rates are referred to as “implicit” discount rates, because they represent the discount rate implicitly used in evaluating purchase decisions based on observed market behavior. The implicit discount rates in the RDM are generally higher than purely financial discount rates and include effects of other factors (uncertainty, institutional barriers, and perceived risks) that influence residential energy-efficiency choices.¹³ The implicit discount rates are specific to general types of equipment (electric heat pumps, gas furnaces, etc.). For AEO98, the implicit discount rates range from 15 to 50 percent for efficiency choices of space heating and central air conditioning equipment. For heat pumps, which are highlighted in the discussion below, the approximate implicit discount rate used in the RDM is 20 percent.

Because the intensity of a particular end use varies considerably across the nine Census divisions while the installed costs do not, the tradeoff between installed cost and maintenance and energy costs varies by Census division. Thus, in divisions with high usage intensity, the RDM selects, on average, somewhat more efficient (and more costly) equipment. Similar differences in choice of efficiency due to usage intensity occur across building types.

Equipment choices are made from a menu of the technologies available to serve a given end use in a particular year. Each end use is potentially served by several technologies, each with multiple efficiency levels available. Generally, two to four different efficiencies are available for a given technology.

The technology menu includes a variety of characteristics for residential appliances:

• The retail cost of equipment plus installation costs (tracked separately)
  
• The energy efficiency of the equipment
  
• The dates available for purchase (for example, efficiency standards may
  limit availability for future years, or technological innovations may cause
  new equipment to become available in future years)
  
• Equipment minimum and maximum lifetimes
  
• Choice function parameters, which vary by appliance type and generally are calibrated to recent shipments.

Table 1 provides data from the AEO98 menu of technologies for several types of equipment.¹⁴

Table 1.  Selected Residential-Sector Technology Cost and Performance Characteristics

Figure 4.  Residential Electricity Prices, 1995-2020

Source:  AEO98 National Energy Modeling system, runs
AEO98B.D100197a (reference case), ELAST98.D042098A (Doubling Case),
and  ELAST98.D042098B (Doubling with Increasing Prices Case).

An important feature of the AEO98 projections is that average real residential electricity prices are projected to decline during the projection period, largely as a result of increasing competition in electricity generation markets, declining coal prices, and relatively stable gas prices. For the reference case, residential electricity prices decline from about 8.5 cents per kilowatthour in 1995 to 7.8 cents in 2000 to 6.8 cents per kilowatthour by 2020. The Doubling Case was constructed with a jump in electricity prices in the year 2000 to just over 15.7 cents per kilowatthour. From 2000 on, prices decline to 13.7 cents per kilowatthour in 2020. This decline is proportional to the decline for the same period in the reference case. For the Doubling with Increasing Prices Case, electricity prices increase from 15.7 cents per kilowatthour in the year 2000 to 17.8 cents per kilowatthour in 2020.

Figure 5 illustrates the prices for heat pump equipment of different efficiencies over the projection horizon. The efficiency ratings in Figure 5 refer solely to the space heating component of performance, represented by the heating season performance factor (HSPF) of the heat pump. The HSPF is typically only about two-thirds that of the air conditioning efficiency rating for a heat pump measured as the seasonal energy efficiency ratio (SEER). Hence, in 2020, while the most efficient unit has an HSPF of 12, the average SEER for the same unit is 18.

Figure 5.  Installed Costs and Efficiencies
of Heat Pumps, 1993-2020

Data Sources

Module Components and Important Interactions with NEMS

Figure 7.  NEMS Commercial Energy Demand Module

Equipment Characterizations and Choice Methodology

Technology Characterization Data

Figure 13.  Lighting Service Demand by Technology,                                     Figure 14.  Lighting Service Demand by Technology,
Doubling Case, 1995-2020                                                                                     Doubling with Increasing Prices Case, 1995-2020

                                                                                                      

Source: AEO98 National Energy Modeling                                                         Source: AEO98 National Energy Modeling System, runs
System, runs AEO98B.D100197A,                                                                        AEO98B.D100197A, ELAST98.D042098A, and ELAST98.D042098B.
ELAST98.D042098A, and ELAST98.D042098B.