Essays on Large Bayesian VARs with COVID Volatility

Abstract

Chapter 1 introduces covbayesvar, a Python package for estimating large Bayesian Vector Autoregression(BVAR) models, that account for COVID-induced volatility. The package enables us to estimate the model with hierarchical prior selection when the parameters proliferate, accounting for structural shifts in macroeconomic and financial data during the pandemic. Incorporating various priors, it is versatile enough to answer wide-ranging policy-related questions. With detailed programming examples, I explain how to apply the functions on monthly and quarterly macro and financial data to construct unconditional and conditional forecasts, scenario analysis, assess joint predictive densities of variables, examine structural breaks during and after COVID, and how we can employ entropic tilting to modify the forecast distribution and construct forecasts conditional on long-term targets set by the Federal Reserve Bank. Accessible via PyPI, the package includes extensive documentation and code examples. covbayesvar advances the state-of-the-art in open-source econometric and statistical software, offering researchers a robust tool for analyzing large-scale systems under unprecedented uncertainty. The remainder of the chapters in the dissertation employ the python package detailed in chapter1 to answer a plethora of policy-making questions. For instance, chapter 2 examines the cyclicality of the financial intermediation variables before the 2008 Global Financial Crisis (GFC) using a large BVAR model with 43 variables. To establish stylized facts on the cyclical behavior, I construct reducedform scenario analyses where unemployment rate rises by 1 pp, and illustrate the predictive densities at various quantiles. From the scenario analyses, we observe that M1 behaves counter-cyclically during downturns in business cycles i.e. M1 rises when the real economy contracts such as when the industrial production declines and/ or the unemployment rate spikes. Measures of credit such as real estate loans, real commercial and industrial loans and consumer loans are procyclical. Leverage of securities brokers and dealers, and non-financial businesses are countercyclical in the short run, but procyclical in the medium and long run. On the other hand, tier 1 leverage capital, a measure of capital adequacy, is procyclical in the short run, but countercyclical in the medium and long run. A counterfactual analysis from 2008 to 2024 using a large BVAR with COVID volatility model reveals that financial intermediation variables significantly deviate from historical trends - most prominently evident for monetary aggregates, credit and loan supply, and interest rates. Probing the cylicality of financial intermediary variables is crucial to understand how stress testingscenarios affect the trajectory of financial intermediation, and asses if the responses are consistent with the previously defined stylized facts. The Fed releases stress test projections every February, which contain the projected paths of macro and financial variables for the next 3 years in downturn and severely recessionary environments. Conditioning on these scenarios, they forecast disaggregated data for each bank separately, and sum the values to calibrate the system wide effects. This may miss general equilibrium, spillover and systemic effects. How can we condition the forecasts from a large BVAR on the stress test scenario values for multiple periods at the aggregated level? In chaptr 3, I employ a method known as entropic tilting to alter the forecast distribution to be as close as possible to the scenario paths. I also show how we can extend the entropic tilting method to condition the forecasts of macro and financial intermediary variables on the short and medium-run stress testing scenarios. Then, I show that the forecasts from this novel entropic-tilted approach are very close to the forecasts from the baseline stress testing scenario. How do contractionary monetary policy shocks affect the financial intermediary variables? Answeringthis question is germane in today's environment, where the financial and macroeconomic variables are highly interdependent with some degree of uncertainty around monetary policy actions. Financial intermediaries serve as conduits to effectively transmit monetary policy to the broader economy by not only altering the supply of credit but also impacting asset prices, liquidity, and appetite for risk across sectors. Much of the Bayesian VAR literature has emphasized real macro aggregates such as output, inflation, and employment, overlooking balance sheet variables of financial intermediaries. When the Fed tightens monetary policy, adding the financial intermediary variables is crucial to extricate the elasticities of balance sheets such as loan volumes, leverage, and monetary aggregates. In chapter 4, I examine two scenarios. First, how does a surprise 100bps hike in monetary policy that is structurally recursively identified affect the flow of funds. Second, if agents expect that the Fed will hike the federal funds rate 100 bps 8-12 quarters into the future, how do the responses of the flow of funds compare? Extending the closed-economy analysis of chapter 2, in chapter 5, I broaden the scope to an openeconomicsystem. Now, what is the cyclicality of international economic indicators, and are there any structural breaks? This provides a quick overview of the correlations and co-movements of the international economic variables and sheds light on whether the 2008 GFC altered those patterns. I conclude that there are no structural breaks - treasury securities held by foreign investors and import price index are countercyclical; export price index, real exports and imports are procyclical. Then, with the recent developments in the trade war, I model a few scenarios: What are the implications of the Reverse Greenspan shocks: reduced foreign purchases of US Treasuries? Most importantly, how do the tariffs affect the US aggregate economy? I gauge the effects using a novel approach of changes in the import price index. A cost-push shock, this is a stagflationary scenario analogous to a recursively identified IRF with slow-moving macro and fast-moving financial variables, generating structurally interpretable responses. Then, I extend the analysis using finer-grained sector-specific disaggregated data to evaluate the impact of tariffs on various sectors of the US economy, such as the services, durables, non-durables, retail sector, etc, using a very large dataset of 127 variables.