Three Essays in Macroeconomics, Forecasting, and Monetary Policy

This dissertation comprises three essays exploring the monetary policy risk-taking channel, quantitative easing under commitment monetary policy, and macroeconomic variables forecasting evaluations. In chapter one, I evaluate the performance of various machine learning algorithms in forecasting macroeconomic variables and compare them with baseline econometric models. The variables of interest are industrial production, employment, inflation, and stock index. The findings can be summarised in three points. First, some machine learning models, such as kNN(uniform)-DI and AdaBoost-DI, perform better than the baseline model in forecasting real variables (industrial production and employment) but perform worse than the baseline models in forecasting nominal variable inflation and financial variable stock index. Second, machine learning models can provide better forecasts in extreme circumstances like the COVID-19 pandemic. Third, the dimension reduction technique improves the forecasting performance of machine learning models to some extent but is not a game changer.

In chapter two, I study the welfare gains from quantitative easing (QE) under commitment monetary policy at the binding zero lower bound (ZLB). First, I provide an analytical solution to the equilibrium paths under commitment policy when the natural rate shock is deterministic. Second, I investigate QE at the ZLB under discretionary and commitment policies and quantify welfare improvements in both cases. This chapter finds that the central bank choosing two instruments (QE and policy rate) optimally under commitment policy yields the highest welfare among the scenarios considered, and it implies an extra two periods of ZLB duration, overshooting of policy rate after lifting from ZLB, and reducing balance sheet size in the mid-term. Specifically, welfare is 92% higher from adopting two instruments under commitment than only having one instrument (policy rate). Finally, I conduct robustness analyses on important parameters and constraints to corroborate the findings of QE's welfare improvements at the ZLB.

In the final chapter, I examine broker-dealer leverage and monetary policy risk-taking channel in three steps. First, I examine the risk-taking channel proposed by Bruno and Shin (2015) by extending the dataset to a longer sample and find there is a large time-variation of the channel over time. Second, I shift the focus of the international standpoint in Bruno and Shin (2015) to the US context and look into the impact on real activity. Third, I apply a time-varying parameter VAR model to investigate the time-varying risk-takling channel. I find that both time-varying coefficients and stochastic volatility are important drivers of this variation.