Dr. Greenberg: No. So far, the basis part is working. I have looked at PIES bases, and there are interesting stories that I have, but I don't think there is time to discuss the results here. I think after the speakers are finished, we could go out in the hall, and I could tell you about some of the things I have looked at and made some interesting observations by applying it to PIES. [7] Koopmans, T.C., and A.F. Bausch, "Selected Samuelson, P.A., The Foundations of Economic This report was written to provide an Executive Summary of new developments that will be described in greater detail in a series of Technical Memoranda. I wish to thank George M. Lady for his encouragement and Patricia Green for her typing. Additional copies of this report are available from: Energy Information Administration Clearinghouse 1726 M Street, N.W. There are multiple goals in a detailed model assessment: data validation, software validation, documentation, etc. The most important may be documentation of the assessment process since this permits outside verification of the assessment results. The documentation should be highly detailed, but it must also maintain an overview perspective of its role as an identifier of the fundamental underlying relations which give the model its behavioral character. In another vein, documentation is a very valuable tool for use in inte- An assessment, for example, would uncover the following types of information about a given model: o Dimensionalities, such as region and sector structures; o Time frames; Data sources and definitions; An identification of exogenous and endogenous variables; and, o Some notion of the essential skeletal structure of the model, an understanding of which is critical to the form reduction process so often necessary in model integration. We would like to explore these notions by means of an example of the model integration process, which we recently completed to incorporate the Hirst Residential Model into the DOE Mid-Range Energy Forecasting System (MEFS).1/ This effort was started without benefit of a comprehensive assessment document as a reference tool; thus our contention that assessment work is helpful, if not critical to a model integration process, is borne of real and painful experience. In the next section we would like to discuss in more detail the definition of model integration and present some basic concepts on the integration process. The third section of the paper will address why model integration is of interest and importance in the context of the policy analysis process. The Project Independence Evaluation System (PIES) was the predecessor to the DOE Mid-Range Energy Forecasting System. The component models of PIES have been substantially modified since the initial use of PIES in 1974, so that the original name does not adequately describe the current system. And finally, the fourth section of the paper will attempt to explain how assessment can help by illustrating specific problems and solutions from the Hirst model integration effort. The conclusion will summarize the major results of the study. II. WHAT IS MODEL INTEGRATION? Model integration describes a process by which two or more conceptually complementary models are linked in order to produce a more comprehensive analysis then could be accomplished by the individual models. Typically, models which are complementary describe different but related subsets of a physical or social system. A macroeconomist, for example, might be interested in linking a national income model to an interindustry model of the economy. In so doing, he might explore two areas more fully: (1) the interrelationships between consumption and investment, and (2) changes in factor demands by industry. This analysis would be infeasible using each model individually. Once they are linked, however, they form a powerful tool for understanding integrated analyses. In energy analysis, the operative system is typically the energy market network. Complementary models include those which describe primary supply of coal, oil, gas, transformation activities, and demands. To analyze electric utility capacity and generation decisions in the absence of reasonable electricity demand forecast, for example, is not typically useful. Also further attempts to forecast electricity demand in the absence of a reasonable description of natural gas demand is not meaningful. Conversely, models which are not complementary are competitive. Competitive models describe the very same phenomena (i.e., the same piece of the network) using different techniques or different degrees of detail. The two models being discussed in this paper are obviously competitive, which is why the exercise being described involved replacing one with the other at a particular node in the DOE energy modeling network. Model compatibility is a necessary condition for model integration. If two models are not compatible, their characterizations of their subset of a general system are not fusible in relation to each other. Furthermore, noncompatible models may not have a feasible mechanism for passing information between each model and thus reconciling feedback effects becomes impossible. As such, they can not function in tandem and are unlinkable. Thus, model integration is also a process by which models are modified so they will be compatible with one another. There are two major ways in which models must be altered in order to make them compatible: o Reconcilication in terms of their data definitions and accounting structures; and, o Implementation of common solution algorithms. Data definition refers to the data sources used in a model and the procedures for defining internal variables. Industrial gas demand, for example, includes or does not include refinery demand. It can also include or not include feedstock demand. It can include or not include lease and plant fuel. These concepts become particularly important in the context of a model integration exercise because of the accounting structures imposed on modeling systems. Each model within the system is no longer free to independently define the nature and content of its internal variables, because the system demands material balance integrity. What goes in must come out and nothing can be counted twice. If these rules are violated, the modeling system can produce disastrous distortions. Consider, for example, an energy system in which all supply models utilize the Bureau of Mines (BOM) data definitions, but the industrial demand model observes Census of Manufacturers definitions. Because the primary data collection coverage of the two agencies is different, the Census of Manufacturers is typically between 5 and 30 percent lower than BOM. Thus, industrial demands will end up being significantly understated. Solution of the model will generate a price/quantity equilibrium that may yield a zero level of imports in 1990 because of the understated demands. This may be a tempting way to solve the energy crisis, but it is not science, and it certainly is not policy analysis. The algorithms used to solve systems of linked models are important because they specify the ways in which information is passed among the models. Without the information exchange, of course, feedbacks cannot be reconciled and the integration process is not complete. Typically, the algorithms used to link models are complex because the individual models themselves are of diverse types. In the DOE's MEFS model, for example, engineering, econometric, and optimization-type models are linked within a linear programming/simultaneous equation framework to solve the model. In the generalized equilibrium modeling framework, of which the Gulf-SRI model is the best known example, a network algorithm is employed in which prices and quantities are the only kinds of information passed between models. Individual nodes within the network can be represented by optimization, simulation, or detailed structural models as long as prices and quantities can be extracted from their results. Diversity of inputs utilized in a modeling system demands an algorithmic discipline be imposed on the individual models. Algorithmic structures demand conformity similiar to those required in accounting structures. In essence, the way in which a model is represented to the system is controlled in terms of variables, time frames, and equation specifications. This may not, however, be the way the model is represented in freestanding mode, in such a case, a certain type of alteration known as "form-reduction" or "cartooning" 2/ must be undertaken. This process will be discussed in detail in the next section. The "form-reduction" process is the basic way in which models are made algorithmically compatible for integration. In summary, the question of what model integration is has been answered. In addition, several general principles of the process have been delineated. But the question of why model integration is important is still unanswered. In particular, how is integration important to the policy analysis process, and why is this technique preferable to the separate construction of large multimodal models tailored to each policy question. These issues will be addressed in the next section. III. THE ROLE OF MODEL INTEGRATION IN THE POLICY ANALYSIS PROCESS Model integration is distinct from model building. However, the two activities have identical goals with respect to the policy analysis process. But model integration adds an extra dimension to the modeling facility by allowing greater speed and flexibility in the development of comprehensive analytical tools. Essentially, model integration permits the construction of integrated analysis tools from prefabricated pieces. This creates broad possibilities for specialization of labor in model building, quick and easy redirection of the emphasis of model design and function, and timely improvement and updating of models. These facilities are important to the policy analysis process because of the interrelationship among three factors: the importance of detail, the importance of comprehensiveness, and the importance of producing timely analysis and monitoring an operationally manageable system. Detail Detail is of critical importance in the evaluation of policy decisions precisely because most policy decisions concern themselves with disaggregated areas within the social or economic system. In energy policy, this has become increasingly true as the complexity of 2/ This term is attributed to David Nissen of the Chase Manhattan Bank. |