would eventually support a state-level modeling system, but access to regional detail (especially plant-level) will be critical.
IGEM, like other CGE models, is largely empirically based. Social accounting matrices come from the National Income of Product Accounts (U.S. Bureau of Economic Analysis). Many models also draw from the Global Trade Analysis Project (GTAP), a database containing bilateral trade information for more than 40 countries and 50 sectors. IGEM’s representations of price- and policy-induced technical change are based on observed behavior over 40-50 years, though its managers recognize that past performance is not necessarily indicative of what will happen in the future. IGEM does not represent feedback effects, and so “shocking the system” will not change the parameters themselves.
Marilyn Brown noted that the multiplicity of technology options available can sometimes be a deterrent to action—policymakers want to know what their best option is. Therefore, she wondered how models might be able to offer additional insight into the role of specific technologies, such as different low-carbon fuels, or tradeoffs between all-electric and plug-in hybrid cars. This level of technological detail often confines the timeframe for estimates. For technology-rich models like NEMS, assumptions about specific technologies that go beyond 25 years or so are both difficult to make and highly uncertain. MiniCAM attempts to draw out insights over a longer timeframe (100 years), but Leon Clarke cautioned that technology assumptions are even more influential in its model outcomes. In general, it is more difficult to get technology richness in the more top-down models. ADAGE and other CGE models can try to isolate the electricity sector and include specific technologies directly in the model, or, as EPA has done, it can link results from other modeling frameworks.
Jay Braitsch of DOE noted that technology progress is among the most important of the assumptions made by the DOE in its modeling efforts. IMAGE and several other models use a simple learning-by-doing approach to represent technology learning. In the case of IMAGE, it is affected by resource depletion (including competition for biomass), and in the case of renewable resources, it is also influenced by distance from human settlements. As noted in the discussion on policymakers’ information needs, technology acceptance is also an important consideration that is not always being reflected in models as another uncertainty factor. Dallas Burtraw recalled that in the 1970s, few analysts would have forecast that the nuclear power industry would have such difficulty in siting and building new plants; a similar situation is playing out for many coal plants in the United States, and so the business as usual reference case may no longer apply.
Detailed and complex legislation must be characterized to fit into the models, and while it is not generally a problem to represent the instruments, there is a great deal of uncertainty surrounding what these policies would actually accomplish. Many CGE models capture rich substitution possibilities, based on observed behavior. However, for all models, it is less clear that historical behavior accurately represents the choices consumers would make today or in the future. Forward-looking models, which optimize over time, tend to yield lower macroeconomic costs than do myopic dynamic models, because actors know what will happen in the future and can plan accordingly. The models can use behavioral elasticities to reflect, for example, slow household response to price changes, but capturing explicit distortions is difficult. These substitution elasticities have an important bearing on costs, in both static and dynamic models. John Reilly also pointed out that credit systems like the Clean Development Mechanism (CDM) are not nearly as efficient as a cap-and-trade program, but the credit supply curves are generally based on some sort of cap.