greenhouse gas mitigation costing studies is very large. It is difficult to disentangle the various reasons for these disparities in modelling results, given the diversity of tools used to calculate cost estimates, the many and varied assumptions employed, and the disparities in the geographical coverage of different studies.

Historically the debate in the energy field has been framed by the distinction between "top-down" and "bottom-up" studies, a distinction that can be applied in other fields as well. Basically, top-down models analyze aggregated behaviours based on economic indices of prices and elasticities. These models began mainly as macroeconomic models that tried to capture the overall economic impact of a climate policy, which, because of the difficulty of assessing other types of policy instruments, was usually in the form of a carbon tax or, more rarely, tradable permits. Bottom-up models, on the other hand, rely on the detailed analysis of technical potential, focusing on the integration of technology costs and performance data.

In fact, as suggested above, not all models fall neatly into one of these two categories and several "hybrid models" are now available in which analysts have attempted to merge top-down and bottom-up model characteristics. As a result, differences in findings are increasingly the effect of differences in input assumptions rather than differences in model structure. However, the discussion in this text will maintain the dichotomy between these models to the extent that this distinction remains meaningful for understanding some critical policy issues.

The top-down/bottom-up categorization has been portrayed as opposing the optimism of the "engineering paradigm" to the pessimism of the "economic paradigm" (Grubb et al., 1993). From an engineering standpoint the evidence is that the best available technologies have not been adopted so far; this "efficiency gap" is the gap between the energy efficiency of equipment actually chosen by consumers and the energy efficiency of the technology that could theoretically minimize the costs entailed in providing a given amount of energy service. Bottom-up models are able to demonstrate the existence of such an "efficiency gap" and thus they suggest that, thanks to "negative cost measures," substantial emission reductions could be achieved with low taxes and low costs or even net savings.

In response, the professional reflex of many economists has been to call attention to the reasons why consumers do not adopt technologies that appear to be optimal, and to suggest that accounting for these reasons, together with the economic feedbacks of a given policy, would reduce the magnitude of, or eliminate, the efficiency gap that is actually achievable. Top-down macroeconomic models concluded, at least in early analyses, that relatively large carbon taxes resulting in significant economic costs would be required to counter current emission trends.

The methodological difficulties lying behind these differences revolve around how to describe the processes of technology adoption, the decision-making behaviour of

economic agents, and the feedbacks between any public policy measures and the overall economy, and how markets and economic institutions actually operate over a given period of time. From this viewpoint the opposition between top-down and bottom-up methodologies does not fully represent the whole spectrum of critical issues. That is why, before discussing the lessons of the top-down and bottom-up debate, we will sketch the dimensions of a typology of existing models. Given the prominence of energy/economy models in the field, we will concentrate upon them. However, many of the methodological issues are quite general and similar arguments could be applied to models of other sectors, such as forestry.

8.4.2 Critical dimensions of a typology of existing models

Various attempts have been made to categorize the large variety of models that have been used to analyse the costs of reducing greenhouse gas emissions". Instead of providing a new and necessarily arbitrary typology, we will try here to provide a description of the main characteristics that differentiate these models, in order to make clear what each type of model describes and what kind of policy question it can best address. In order to accomplish this, we will focus on those characteristics that have to do with the purposes of the models (i.e., the questions they are meant to address), their structure (i.e., those assumptions that are embedded in the equation system), and their external assumptions (i.e., assumptions expressed in terms of inputs to the models). Because few, if any, individual models represent pure types within these categories, we have not referenced specific models. Instead the goal here is to provide a general framework to aid in interpretation of the specific modelling results presented in Chapter 9.

8.4.2.1 Diversity of models, diversity of purposes

Significant misinterpretations of the results of modelling studies can arise from overlooking the purpose of the analysis they were used for. The meaning of the numerical results of a model will differ depending upon whether a given scenario is used to predict (forecast) the future, to explore it, or as a tool for "backcasting" exercises.

Many models are used to try to "predict" the future and to provide an estimate of the most likely set of future events. This purpose imposes very strict methodological constraints on the modeller. He or she must produce a base case forecast, which amounts to a best-guess projection of the most likely future; to do this requires an endogenous representation of economic behaviour and general growth patterns. This type of predictive exercise attempts to extrapolate the interactions of historical trends into the future, with a minimum of exogenous parameters. This approach has been typical of government and sectoral forecasting activities (e.g., in energy, transport, and heavy industries such as steel) and early climate change scenario analyses. It remains both necessary and convenient for analyzing the short-term impacts of climate policies,

since a number of critical underlying development variables can reasonably be assumed to remain constant for these time periods. Most short-term, econometrically driven macroeconomic models adopt this approach. For the long-term (middle of the next century), the Jorgenson-Wilcoxen, McKibbin-Wilcoxen, and Goulder models are the only models that can be classified in this category. It is noteworthy that none of these models tries to work on the basis of forecasts of explicit technological trends in the engineering

sense.

Because of the difficulty of extrapolating past trends over the long run, the purpose of some modellers is to "explore" the future rather than to predict it and, in so doing, to provide potentially counterintuitive assessments. This leads to a scenario analysis approach, which involves building up different coherent visions of the future (each of which, of course, can undergo sensitivity testing) based on different values for key assumptions about economic behaviour, physical resource endowments, or technical progress, together with assumptions about economic or population growth. The first step in such analyses is the generation of a "reference" or "nonintervention" scenario. This is then contrasted with alternative cases involving an array of policy measures, such as carbon taxes or energy efficiency regulations, giving rise to one or several "policy" or "intervention" scenarios, but the policy analysis is relevant only in the context of each baseline scenario. This approach is increasingly used for climate change analysis and was the basis for the 1992 IPCC scenarios (Legget, et al., 1992). It is shared by both bottomup and top-down models and indeed tends to favour the development of hybrid models (which will be discussed at greater length below). In terms of representing and simulating the behaviour of economic agents, two methodologies predominate. One involves an assumption of least-cost optimization, in which society maximizes the utility of consumption over the long run. The other involves the effort to simulate real world behaviour predictively in terms of technology adoption.

Finally, another possible purpose of models is to assess the feasibility of alternative futures, often defined in terms of desirability rather than likelihood. This contrasts with the two previous approaches, insofar as it involves the development of a vision of a future state of the system being studied and then an analysis of how that future system might be realized. This "backcasting" methodology" allows for identification of major changes as well as discontinuities in present trends that might be required if a desirable future is to be attained (Robinson, 1988, 1990). Two types of research can be carried out in this approach. Most studies involve normative scenarios about desired futures. Many alternative energy studies, for example, belong to this category. But it is also possible to use a backcasting methodology as a purely analytical tool by simply linking bottom-up analyses about the long-term evolution of technology and development patterns to a macroeconomic framework (Hourcade, 1993) so as to be able to assess the economic consistency of different and competing views about the long run.

26

These three different purposes have implications for the types of models that are required, the analytical questions being asked, and the meaning of the results. As

discussed in section 8.3.2.1 above, there has been a historical correlation between attempts to move away from predictive modelling approaches and the early development of bottom-up models; these models were in many cases built precisely in order to undertake simulation and backcasting analyses not possible with the current generation of top-down models (Baumgartner and Middtun, 1987; Robinson, 1982). The greatest emphasis in current climate modelling efforts is on "exploratory" analyses, using a combination of top-down and bottom-up methods, but there continues also to be interest in "backcasting" analyses aimed at exploring quite different future scenarios than would otherwise be examined (FRN, 1987; Goldemberg, et al., 1988; Jäger, et al., 1991; Rothman and Coppock, forthcoming 1996; Robinson, forthcoming 1996).

8.4.2.2 The Structure of Existing Models

A second basis for distinguishing among different models is the nature of the model itself (i.e., those assumptions that are embedded in the mathematical structure of the model). At a very general level, it is possible to characterize some of the main structural differences among existing energy and emissions models in terms of four main dimensions. This description abstracts from a number of more detailed distinctions among models but captures the points that differentiate the models in a way that helps to show the connection between model structure and policy questions (see the next section for a more detailed discussion in the context of the top-down/bottom-up modelling debate). These four dimensions, and some of the related policy questions, are sketched in Table 8.1 for energy/economy models. An equivalent typology could be applied to models focusing on other sectors (e.g., forestry).

[Table 8.1]

Each of the four structural characteristics shown in Table 8.1 represents a spectrum from more to less, and individual models can be located on that spectrum for each dimension. This means that individual models are more or less suited to answering particular policy questions, depending on where they are located on the spectrum for each dimension. Traditionally, top-down models have represented one end of the spectrum on each of these dimensions, and bottom-up models the other, as illustrated in Tabie 8.2.

[Table 8.2]

It is clear from Table 8.2 that the early top-down and bottom-up models represented virtual mirror images of each other, with respect to the four characteristics shown in Table 8.1. Bottom-up models tended to describe the energy system in great detail, with little endogenization of behaviour or description of other parts of the economy. Topdown models tended to have very little detail on the energy sector but explicit treatment

of behaviour and larger economic relationships. As suggested in the previous section, these characteristics led to each type of model being most useful at answering somewhat different questions. Bottom-up models were better at simulating detailed technological substitution potentials ("exploration"), and top-down models were better at predicting wider economic effects ("prediction").

Table 8.2 also shows that this simple characterization of the differences between top-down and bottom-up approaches is increasingly misleading, as more recent versions of each approach have tended to move in the direction of greater detail in those dimensions that were relatively less developed in the past. This is possible because the four dimensions shown in Table 8.1 are independent of each other. Thus any particular model can be located at virtually any point on the spectrum represented by each dimension. It is this independence of these key structural characteristics that makes it so hard to classify the large population of existing models on any single spectrum, whether bottom-up to top-down, or any other. Instead we increasingly have a wide range of models, which, in terms of their structure, occupy different places on each of the four dimensions shown here. Thus, while the differences represented by these four dimensions remain important, no simple classification scheme is adequate.

Tables 8.1 and 8.2 describe the differences among different types of models in very general terms. Coming closer to the structure of actual models, we can distinguish several kinds of modelling procedures.

Among bottom-up models two approaches are usually distinguished: (1) spreadsheet models that resolve a simultaneous set of equations to describe the way a given set of technologies is (or could be) met throughout the economy; and (2) simulation or optimization models, which simulate investment decisions endogenously. Each of these two approaches can be used in two different ways: prescriptively or descriptively.

A prescriptive model examines the effect of acquiring only the most efficient technologies available or of minimizing explicit costs for a given service at a system level (e.g., electric supply, urban transportation, or land use). A descriptive model, in contrast, would try to estimate the technology mix that would result from actual decisions, based on factors such as more complex preferences (people preferring private cars even if the cost per kilometre is higher than railway transportation), intangible costs (differences in cost of acquisition of technologies), capital constraints, attitudes to risk (via higher discount rates for some agents) and uncertainty (actual performance of new technologies), or any kind of market barriers. Such analyses will typically tend to be less optimistic than prescriptive studies about mitigation, unless appropriate policies are assumed to remove existing barriers to the adoption of the best available technologies. Considerable disagreement now exists in the literature about the potential for such policies.