The use of an association rules matrix for economic modelling

Knowledge discovery techniques have not been widely used with macroeconomic data despite the social and political importance inherent in accurate economic forecasting. Rather, economic forecasting is currently performed with the use of models that rely heavily on theoretical assumptions. Economic assumptions are invariably contentious and model predictions are often rejected or accepted based on subjective perceptions about assumptions. We present an application of KDD (knowledge discovery in databases) that generates a forecasting model that avoids economic assumptions by focusing entirely on existing data. Although association rules are typically used for finding interesting patterns in data, this is not a strategy we employed. Our approach differs in that all possible association rules between variables representing the current state of an economy in a quarter and the state in the next quarter are generated to form a matrix. A metric based on the support, confidence and expected probability of each rule is then derived. The system has been used to perform economic analyses of existing and future government policies. The system has been developed using data from the US economy over the last 30 years