AN ACTIVE MODELING SYSTEM FOR ECONOMETRIC-ANALYSIS

This paper examines the feasibility of developing an "artificially intelligent econometrician" as an active decision support system (ADSS) in the sense articulated by [Manheim, 1988]. We review the system components of an ADSS and then relate them to a modeling system for econometric analysis that we have implemented. We present the query language of the PERM (Progressive EconometRic Modeling) system and offer an extension to the language based on process-oriented constructs for model integration. The query language and its extension correspond to the ADSS's user-directed and computer-directed process managers, respectively. Schemas representing statistical strategies are stored as processes in the extended language and serve as the econometric knowledge base. We suggest an approach to building an inference processor for this system based on experiments to record user query protocols and relate them to the schemas in the knowledge base. The connection between user processes and schemas is implemented by demon constructs in the extended language. Finally, we examine the extent to which our proposed system constitutes an ADSS.