[ Job Market Paper: latest version, initial version | Code ] Click to expand abstract This paper develops a new Bayesian framework for forecast combination that embeds a Vector Error Correction Model structure within a latent dynamic factor representation for model weights. By imposing cointegration across the latent weight structure, the framework establishes a structured yet flexible regularization on their long-run behavior while allowing for short-term deviations. The specification enables the ensemble weights to evolve around steady states and to adjust rapidly toward a new steady state when the predictive environment changes. Monte Carlo studies and empirical applications to economic growth and inflation forecasting show that the proposed approach delivers substantial improvements in predictive performance over cutting-edge methods. The framework incorporates key features found in state-of-the-art approaches by providing flexible weight dynamics that permit non-convex weights to be negative and allow quick transitions across multiple steady configurations simultaneously.

[ Working Paper | Code ] Click to expand abstract We propose the Markov Switching Dynamic Shrinkage process (MSDSP), nesting the Dynamic Shrinkage Process (DSP) of Kowal et al. (2019). We revisit the Meese-Rogoff puzzle (Meese and Rogoff, 1983a,b, 1988) by applying the MSDSP to the economic models deemed inferior to the random walk model for exchange rate predictions. The flexibility of the MSDSP model captures the possibility of zero coefficients (sparsity), constant coefficient (dynamic shrinkage), as well as sudden and gradual parameter movements (structural change) in the time-varying parameter model setting. We also apply MSDSP in the context of Bayesian predictive synthesis (BPS) (McAlinn and West, 2019), where dynamic combination schemes exploit the information from the alternative economic models. Our analysis provide a new perspective to the Meese-Rogoff puzzle, illustrating that the economic models, enhanced with the parameter flexibility of the MSDSP, produce predictive distributions that are superior to the random walk model, even when stochastic volatility is considered.

[ Working Paper | Code ] Click to expand abstract The first paper integrates the hierarchical Markov-switching model with model-based clustering to address high-dimensional data challenges, aiming to identify bull and bear markets. This approach enhances both identification and forecasting performance.

[ Poster | Code ] Click to expand abstract Assessing the effect of the ACTN3 gene on athletic performance. Developing a widely applicable general model for studying different genetic effects.

[ Working Paper | Code ] Click to expand abstract Extending a demographic model initiated by Salop generates some suggestive theoretical results.