Efficient pooling of cross-section and time series data using Bayesian machine learning with two econometric applications

In this project, we adapt a Bayesian modelling strategy, namely the minimum message length principle, to the problem of efficient partitioning of economic units, such as firms or countries, into groups whose behavioural patterns are similar within each group but distinct across groups. This methodology can incorporate the requirements of economic theory. The resulting software will be developed for the Web. We consider two specific applications, namely modelling gasoline demand in OECD countries, and finding the foreign factor with the most predictive power for the growth rate of the Australian economy. The second application is of considerable national interest.