THE DESIGN OF BELIEF NETWORK-BASED SYSTEMS FOR PRICE FORECASTING

ARCO1 is a knowledge-based system that models the crude oil market as a belief network and uses scenario generation and Monte-Carlo analysis to forecast oil prices. From an applications-oriented viewpoint, its purpose was to incorporate recent advances in artificial intelligence and decision analysis into ARCO's forecasting procedures. From a basic research viewpoint, its primary objective was to investigate the ability of knowledge-based technology to strengthen and broaden the tools used by professional forecasters. ARCO1's development raised a variety of issues concerning the relationship between diagnosis and forecasting, the relative propriety of domain-based and shell-based representations in an AI system, and the role that AI can play in strengthening the existing tools of heavily automated domains (as opposed to its traditional role of introducing automation to exclusively human domains). This paper outlines these issues, and provides a case study in the design of intelligent systems whose knowledge bases are structured as belief networks.