Zivot, EricMa, Wenqiu2026-04-202026-04-202026-04-202025Ma_washington_0250E_29198.pdfhttps://hdl.handle.net/1773/55489Thesis (Ph.D.)--University of Washington, 2025This dissertation develops a unified Bayesian framework for analyzing and forecasting macroeconomic time series and for measuring price discovery in fragmented financial markets. Across three papers, it proposes flexible yet disciplined methodologies that exploit Bayesian shrinkage, stochastic volatility, and structural identification to handle large systems, noisy data, and time-varying dynamics while delivering finite-sample-relevant inference.The first chapter introduces a large time-varying parameter vector autoregression (TVP-VAR) with stochastic volatility and informative priors that shrink the richly parameterized model toward a parsimonious benchmark. This Bayesian design stabilizes estimation in high dimensions and mitigates overfitting. Using U.S. macroeconomic data of varying dimensions and levels of aggregation, the chapter provides strong empirical evidence that the proposed TVP-VAR improves forecast accuracy—both point and density—relative to standard benchmarks, thereby validating the model as a practical tool for macroeconomic forecasting. The second chapter turns to price discovery and develops a Bayesian vector error correction (BVECM) framework for structural analysis in cointegrated markets. By imposing an economically interpretable decomposition of shocks into permanent and transitory components, the framework yields structural price discovery measures that depend only on permanent innovations, abstracting from microstructure noise and temporary liquidity imbalances. Shrinkage priors tailored to noisy high-frequency data and full posterior inference provide coherent uncertainty quantification. Simulation studies and an empirical application to S&P 500 ETFs demonstrate that the proposed approach delivers robust, interpretable price discovery rankings and credible assessments of cross-market differences. The third chapter extends the analysis to time-varying price discovery by embedding these structural measures in a Bayesian order-invariant VAR with stochastic volatility. Mapping the reduced-form model into a broad family of price discovery statistics, it shows—both in stylized partial-adjustment settings and in applications—that measures grounded in the permanent component are robust to time-varying volatility and noise, while conventional metrics can be severely distorted. Collectively, the three chapters demonstrate how Bayesian inference provides a powerful and flexible toolkit for macroeconomic forecasting and the measurement of evolving information flows in modern financial markets.application/pdfen-USnoneBayesian inferencemacroeconomic forecastingprice discoverystochastic volatilitytime-varying parameterEconomicsEconomicsThesis