PhD student Emily Tallman has been awarded a prestigious NSF Graduate Research Fellowship. Emily summarized the research she plans to do for the Fellowship as follows.
My research proposal focused on new methodology in the intersection of Bayesian multivariate forecasting and decision analysis. With my advisor Mike West, I am developing a new, decision-focused approach to evaluating, comparing, and combining multiple forecasting models— methodology referred to as Bayesian Predictive Decision Synthesis (BPDS). BPDS extends recent theoretical and practical advances in both Bayesian Predictive Synthesis and empirical goal-focused model uncertainty analysis. In contrast to traditional model uncertainty analysis like Bayesian Model Averaging (BMA), my new framework explicitly integrates Bayesian decision perspectives to emphasize the intended uses of models in prediction and decision problems. This leads to an innovative model weighting and combination methodology for comparing forecasting models, while also providing information on the relative performance of models with respect to multiple decision goals. This opens up a new Bayesian approach to multi-criteria decision analysis and defines new model aggregation methods for forecasting. Specific applications include portfolio optimization, experimental design, and reinforcement learning.