Stochastic Volatility in Mean Vector Autoregressive models (SVMVARs) are a popular tool for measuring macroeconomic and financial uncertainty and their economic impacts. SVMVARs estimate macroeconomic (financial) uncertainty using a large set of macroeconomic (financial) variables. But what if there is uncertainty regarding whether variables are classified as macroeconomic or financial? We address this question, developing scalable Markov chain Monte Carlo algorithms for classification search in large SVMVARs with unclassified variables. Using time-invariant or time-varying classification, the algorithm determines whether each unclassified variable should be treated as macroeconomic or financial. We show that allowing for data-driven classification improves model fit. Our results also suggest that without data-driven classification, macroeconomic uncertainty, its adverse effects and its contribution to fluctuations
in economic variables tend to be underestimated. Financial uncertainty is also underestimated but its effects on headline macroeconomic variables tend to be overestimated.
|Conference||International Conference on Economic Modeling and Data Science EcoMod 2021|
|Period||7/07/21 → 9/07/21|
- Bayesian VAR
- stochastic volatility
- big data