• the development of a prototypic methodology and modeling framework which can be cost--effectively utilized for integrated assessment of regional economic and energy demand impacts of global climate change, and which can be adapted for use in other regions or for examining other types of large-scale environmental changes;
  
• a major expansion of knowledge of environmental-economic-energy demand interactions resulting from climate change through intensive study of regional impacts, and identification of these interactions and impacts for the Midwestern Great Lakes Basin;
  
• a significant improvement in the ability to predict the regional impacts and economic consequences of global climate change;
  
  the integration of economic, demographic environmental, energy demand, and climate impacts into one model structure;
  
• the development of a framework for developing and evaluating mitigation and adaptation strategies under alternative forecasts of the extent and rate of climate change.

The tasks for the first year were broken down into four basic stages: data collection; initial specification of equations representing specific economic variables (e.g., state employment in agriculture, gross state product); aggregation of individual equations into and fine tuning of the simultaneous equation model; and forecasting. Using a baseline forecast and a climate change--scenario forecast, the resulting climate change impacts for the period from 1995 to 2010 can be derived. In order to derive the baseline forecast, it is assumed that the model's non-climate exogenous variables would continue to change based on past trends. In order to project these trends through 2010, an autoregressive time-series forecast approach provided by the SAS software FORECAST Procedure is utilized. The values of the climate variables are projected as remaining relatively constant over the forecast period. To derive the climate-change scenario, the climate variables are assumed to have an exponential growth rate consistent with doubling the concentration of carbon dioxide in the atmosphere within ten years, and then continuing to grow at an accelerating rate for the remaining years thereafter. This climate-change scenario is intended to illustrate what would occur if the global warming likely during the next 50 to 75 years took place over the next 20 years. This 20 year forecast allows the effects of global climate change to be explored in terms relevant to current economic conditions.

The model structure and the regression estimation results for the primary stochastic equations of the model have been completed. The model is now in its final tuning stage. All independent variables included in these equations are statistically significant at the 0.10 level or better. Thirteen employment sectors have been empirically modeled with employment and wage equations: farming; construction; manufacturing; services; finance, insurance and real estate; retail and wholesale trade; transportation and utilities; federal and state and local government; mining; and forestry and fisheries. The specification of these equations commonly follows the generalized formulation of employment equations in previous research, with the major non-climate explanatory variables being regional wages in the respective sector, national demand as represented via GNP, and local demand as represented by population, total employment, or employment in related sectors. The climate variables in these equations include daily mean annual precipitation, winter precipitation, spring precipitation, and summer precipitation, mean daily annual and winter temperature, and various soil stress indices. In addition to these employment sectors, equations were developed for government revenue and expenditures, total energy consumption, forest use, gross state product, state unemployment rate, personal income, population, and net migration to capture the impacts of climate change on these variables.

Dummy variables were used for each of the states so that the aggregated Great Lakes model could be used for any of the individual states by "turning off" the dummy variables for all but the state of interest. In addition, interaction variables were employed to capture the strong impacts any one state might have on the dependent variable of an equation. As a result, the information presented proxies the environmental-economic linkages in a state's economy as they exist within the larger Great Lakes region. Individual states do not operate in a vacuum but are directly impacted by activities in the other Great Lakes states. The structure of the simultaneous equation model presented here captures these interdependencies. Such state-level results have been produced, to date, for Indiana.

The model's simulation capabilities are represented by the mean absolute percent error (MAPE) for each endogenous equation (i.e., the average error for the forecasts of the endogenous variables over the sample period). Most of these are less than five percent. In fact, 21 of the 44 endogenous variables of the model have MAPEs less than five percent, and another 16 are less than ten percent (see Table 1 for subset of values). This is very good to excellent performance for a regional econometric model. Analysis of the model's tracking behavior also indicates that the model captures the turning points of the economy quite well. In most cases, the large majority of directional changes in the growth of the endogenous variables is captured by the simulation. This ability to capture economic turning points and the relatively low values for the MAPE statistics indicate that the model works well over the simulation period. However, four equations remain problematic, with MAPEs over 20 percent: state forest use, state employment in mining, state total energy consumption, and state wages in agriculture.

The next step in the research is to tune the model so that the final problem equations have acceptable MAPEs. Following this stage, we will convert the model to a two-stage least squares (TSLS) framework to eliminate simultaneous equation bias. The implications of this bias is that the estimators are not consistent, i.e., they will not converge on their true population values regardless of the sample size. The TSLS estimators account for simultaneous equation bias and do not transmit single-equation specification errors throughout the model structure. Although slight improvements in parameter estimates are expected as a result of TSLS, these should not impact the overall performance of the general results of the model as discussed here.

Following these methodological effects, the forecasting and policy analysis elements of the second year of the project will be addressed. This stage should begin within the next few months.