Structural Analysis and Nowcasting with Bayesian Vector Autoregressions and Sentiment

Details

Supervisors

Monti, Francesca

Faculty

Faculté des sciences économiques, sociales, politiques et de communication

Degree label

Master [120] en sciences économiques, orientation économétrie, à finalité approfondie

Abstract

enThe purpose of this thesis is twofold. First, we estimate macroeconomic effects of monetary policy by Proxy Structural Vector Autoregressions (SVAR). Second, we forecast the US gross domestic product (GDP) by mixed-frequency Bayesian Vector Autoregressions (BVAR) and their combination constructed by Bayesian Predictive Synthesis (BPS). We contribute to macroeconometric literature in two respects. First, we orthogonalize the monetary policy surprises (MPS) with a broader set of sentiments: Beige Books, Federal Reserve System (Fed) speeches, Fed testimonies. Second, we include sentiment scores from the Federal Open Market Committee (FOMC) internal meetings as instrumental variables in the Proxy SVAR. In comparison to quarterly BVAR we find a higher forecasting performance of mixed-frequency BVARs and their combination from the current-quarter to 4-quarter-ahead horizon: (i) 14.98-38.62% by blocking BVAR, (ii) 26.68-40.31% by Cube root BVAR, (iii) 11.01-38.33% by Latent BVAR, (iv) 25.68-47.08% by BPS. With orthogonalized MPS we observe a lower response of consumer price index and industrial production compared to a Proxy SVAR without orthogonalization of MPS. Impulse response functions (IRFs) are robust to sentiment scores from FOMC internal meetings that are included as additional instrumental variables in the Proxy SVAR.