This was established by estimating a linear multiple regression model that incorporates variables representing the eight assumptions, taking the percentage change in GDP as the variable to be "explained". (The estimated regression equation can be found in the Annex.) An alternative regression analysis that included only the reduction in carbon emissions as an explanatory variable and excluded all the variables representing model assumptions explained only about half as much of the variation in predictions.

• The variable representing the presence or absence of a backstop energy source was specified to affect the curvature of the cost curve, since backstop sources become relevant only at higher energy prices and then limit the rate of cost increase.

emissions reduction) account for fully 80 percent of the variation in predicted economic impacts. 3 This is remarkable because it implies that all the other modeling assumptions— hundreds of assumed parameter values and relationships—are comparatively unimportant. Together, they account for only 20 percent of the differences among predicted impacts. Only a handful of basic assumptions really matters.

This is good news. People don't have to be Ph.D economists to understand the debate over the economic impacts of climate policy. Rather, people can use their own judgment and common sense to decide which of these basic assumptions are more realistic. Having decided that, they can then determine for themselves which predictions are more credible and what the economic impacts of a carbon tax or a climate stabilization policy are likely to be.

To illustrate, we have used the statistical relationship between predictions and assumptions to plot several cost curves in Figure 2. Each cost curve represents a different set of modeling assumptions selected from those listed above, starting from a set of "worst case" assumptions and then successively replacing them, one-by-one, with more favorable assumptions until a set of "best-case" assumptions is arrived at. In the statistical analysis underlying these curves, the slope of the curve connecting GDP change to emissions reduction was allowed to shift with

each of the eight assumptions, but the year for achieving the abatement target was held constant.'

The worst case assumptions are that:

1. there is no non-carbon backstop energy source;

2. the economy does not respond efficiently to policy changes, even in the long-run;

3. the scope for inter-fuel and product substitution is minimal;

4. there is no possibility of joint implementation;

5. revenues are returned through lump-sum rebates;

6. there are no averted damages from air pollution; and

7. there are no averted damages from climate change.

Under these assumptions, many of which are obviously unrealistic, the adverse economic impacts of a carbon tax or equivalent policy would be severe, reaching 6 percent of end-year GDP for a 50 percent reduction in projected baseline emissions by 2020. (See the bottommost curve in Figure 2.)

Scanning Figure 2 from the bottom up reveals the effects on predicted economic impact of changing these worstcase assumptions one-by-one. For

example, assuming that backstop energy sources exist improves the predicted economic impact substantially---by about one percent of GDP for a 50 percent emissions reduction. In Figure 2, what is notable about the predicted impacts on the U.S. economy is that changing only five worst-case assumptions by assuming backstop energy sources, efficient long-run adjustment in the economy, greater substitution possibilities, joint implementation, and recycling of carbon tax revenues by reducing other burdensome tax ratesdramatically alters the predicted economic impacts. Instead of a six percent loss of GDP by 2020, there would be modest positive impact on GDP relative to the business-as-usual scenario. '

Judging from all these simulations using a wide variety of economic models, the doomsday prediction of heavy economic losses if carbon emissions are reduced is implausible. It is more reasonable to predict that with sensible economic policies and international cooperation, carbon dioxide emissions can be reduced with minimal impacts on the economy.

Going further, Figure 2 indicates that if reducing fossil fuel combustion avoids economic damages from climate change or air pollution, then the overall economic impacts could be favorable. The top-most curve in Figure 2 indicates that under all the best-case assumptions, a reduction in CO2 emissions by 2020 would result in a substantial improvement in GDP relative to its business-asusual path." Of course, there is some

degree of emissions reduction beyond which the incremental abatement costs exceed the value of the environmental damages that more pollution would create. This turning point is not adequately reflected in Figure 2, which should not be interpreted to suggest that if some carbon abatement is good, more is necessarily better. Figure 2 does imply, however, that models that take the environmental benefits of carbon taxes into account predict substantially more favorable economic impacts than models that ignore such benefits.

One target that has been analyzed extensively by the Interagency Analytical Team in preparation for the COP-3 meeting in Kyoto in December 1997 is a freeze on carbon emissions at 1990 levels by 2010 and stabilization of emissions thereafter. For the United States, it has been estimated that this target implies about a 26 percent reduction below projected baseline emissions in 2020, if the baseline is calculated on the basis of policies now in place (U.S. EIA, 1996). Figure 3 uses the same statistical analysis to show in detail the range of predicted long-run economic impacts if this target is attained. Under unfavorable assumptions, GDP would be 2.4 percent lower in 2020 than under baseline conditions; under favorable assumptions, GDP would be 2.4 percent higher. Figure 3 also quantifies the relative importance of several modeling assumptions in creating this range of predictions.

Four assumptions stand out in terms of magnitude:

This prediction is consistent with the interpretation of a carbon tax as a corrective tax that reduces a market failure— namely, the unintended effect of carbon emissions on the global climate. Economists agree that, if set at the proper rate, a tax to correct a market failure should improve an economy's produc tivity.

• When air pollution damages are assumed, an expanded measure of GDP in which environmental damages are recorded is the relevant indicator of economic impact.

This analysis cannot encompass short-run transitional impacts predicted by some macroeconomic forecasting models.

■whether the economy will adapt efficiently;

■ whether international joint implementation will be achieved;

■whether carbon tax or permit auction revenues will be recycled by reducing other taxes; and

whether there will be economic benefits from abating pollution.

Most economists believe that the U.S. market economy, with high mobility of capital and labor, can adapt efficiently to moderate the impacts of policy

changes. There is general agreement that a carbon tax that discouraged coal use in electricity generation would have the effect of reducing air pollution, even with current air pollution regulations in place. Whether to use revenue-raising policy instruments to limit emissions and how to dispose of resulting revenues are decisions that the U.S. government must make. Finally, international cooperation in joint implementation of carbon reduction targets is a possibility subject to negotiation. Under reasonable assumptions, the predicted economic impact of stabilizing emissions at 1990 levels would be neutral or even favorable.

The preceding analysis looked broadly at the key assumptions that turn out to determine very largely the predicted economic impacts of climate protection policies. These broad distinctions among the modeling assumptions explain most of the differences among predictions. Nonetheless, other aspects of the key assumptions, which could not be adequately built into the preceding analysis need to be recognized and understood. This section addresses such issues.

A. THE SCOPE FOR REDUCING ENERGY INEFFICIENCIES

Top-down models typically assume that all cost-effective improvements in energy efficiency have already been realized, an assumption contradicted by actual experience (DeCanio, 1993). For example, large companies that joined the Environmental Protection Agency's voluntary Green Lights Program to reduce their energy use found numerous opportunities to save both energy and money in their ongoing operations.

Bottom-up studies have found inefficiencies in energy use that could be remedied through building improvement measures, such as better insulation and low-energy lighting; through

technological advances in transport efficiency; or through conversion of industrial processes. Assessments based on engineering studies suggest that from 20 to 25 percent of existing carbon emissions could be eliminated at an overall cost savings and that substantial further cutbacks could be made at relatively low cost (IPCC, 1996b; National Academy of Sciences, 1991; Office of Technology Assessment, 1991).

Some of these inefficiencies undoubtedly persist because of energy market imperfections, such as the divergence in incentives between tenants and landlords, builders and home purchasers; because of energy subsidies; or because of suboptimal decisionmaking within organizations. However, some reported savings opportunities might be illusory if the management costs of locating and implementing energy investments were overlooked or the differences in product and service characteristics of various energyconversion technologies were ignored. Energy service companies, which seek to find and implement energysaving opportunities on a contractual basis, have not found unlimited business opportunities at current low energy prices.

Top-down models typically assume away energy subsidies that may encourage excessive fuel use and must be financed through higher levels of economically burdensome taxes. Energy subsidies, though not as prevalent in the United States as in some other countries, still include favorable tax and credit treatment for energy producers, below-market provision of power from public sector installations, and federally sponsored research and development. Two recent studies quantify annual federal energy subsidies at between $4.9$14.1bn and $21-$36bn respectively (Alliance to Save Energy, 1993; U.S. EIA, 1992). Some of these subsidies, such as federally subsidized hydropower, actually reduce carbon emissions by replacing fossil fuels with hydroelectricity. Others, such as tax breaks for independent oil drillers, have no effect on U.S. oil consumption or carbon emissions, but merely replace foreign produced oil with domestically produced oil. Nonetheless, a study on the effects of removing U.S. energy subsidies concluded that significant reductions in CO2 emissions could be achieved at no cost to the economy (Shelby et al., 1995).

B. SUBSTITUTION EFFECTS

A carbon tax will raise the price of fuels in proportion to their carbon content, increasing coal prices more than oil or gas prices and having little direct effect on hydro or nuclear power costs. Higher fuel prices will induce firms and households to seek ways to mitigate the cost increases. In particular, they will tend to

■substitute less carbon-intensive fossil fuels, such as gas, for carbonintensive fossil fuels, such as coal (intra-fossil fuel substitution);

■ substitute non-fossil energy sources for fossil fuels (non-fossil fuel substitution);

■substitute other factors of production (materials, labor and capital) for energy; and

■substitute less energy-intensive goods for energy-intensive goods (Cline, 1992).

The easier these substitutions are, the lower the overall burden of reducing CO2 emissions. In addition, all such substitutions become easier as the time for adjustment increases. For example, in the short-term, a firm's production technique will be constrained by its existing equipment, but as new equipment and processes are brought on line, energy use can be reduced more readily. Similarly, consumers need time to replace durable goods and fully adapt purchasing habits to altered prices.

Economic models differ in the degree to which they represent these substitution possibilities. Highly aggregated models, which might have only a single producing sector (i.e., a sector producing a composite commodity called GDP), cannot incorporate the possibility of substituting one product for another. Similarly, models that recognize only two primary fuel sources cannot adequately represent inter-fuel substitution possibilities. More disaggregated models, such as the MarkalMacro model, which recognizes 11 primary fuel sources and dozens of fuel conversion technologies, are potentially better able to deal with such substitution possibilities (U.S. DOE, 1996).

However, this potential may or may not be realized. Despite being more or less disaggregated, models differ in the assumed ease of substitution among products and technologies in response to cost changes. Some models assume only one technology available to produce a given output, with no scope for substituting other inputs for energy. Others assume technologies will switch