This dissertation investigates several key macroeconomic topics with the use of econometric and machine learning techniques. The dissertation includes three chapters.

In the first chapter, I develop a medium-scale DSGE model with real wage rigidities and financial intermediaries experiencing endogenous capital constraint and liquidity mismatch. Labor market frictions help to mitigate an endogenous labor supply insurance against fluctuations in consumption. Anticipated idiosyncratic bank runs and possibility of direct investment fix cyclical properties of financial leverage and allows explaining banks’ balance sheet dynamics over crisis. The probability of a bank run depends on bank balance sheet and endogenous assets prices. Anticipations of a bank run affect both asset returns and the economy even no crisis periods.

In the second chapter, I investigate nonlinearity in the Effects of Large Oil Shocks on Macroeconomic Activity. There is a broad debate on nonlinearity of oil price shock effects on macroeconomic variables. Previous literature focused attention on the "Net oil price" variable to test for asymmetry of the responses to positive and negative oil price shocks. So far, most of the nonlinear empirical models of oil prices and economic activity are based on oil price changes without identifying the source of oil price change. For linear models, the use of structural vector autoregressions (SVAR) to estimate the effect of oil supply and demand shocks on macroeconomic activity has become common in recent years. However, none of these models consider nonlinear responses to structural shocks. I propose a SVAR model, based on second order Taylor expansion of a general nonlinear model, that allows me to identify structural shocks. To handle the large number of parameters, I apply Bayesian methods with priors on the parameters in the spirit of the Minnesota priors used for linear VARs. For Bayesian inference, I embed a particle-filter-based likelihood. The impulse response analysis suggests no significant asymmetry in the responses for “small” shocks, but a significant asymmetric response in economic activity to large oil shocks, regardless of the source of the oil shock.

In the third chapter, I develop a "human-free" text analysis of the Beige Book that can easily be updated with the new information and used with other macroeconomic data to nowcast current quarter GDP growth. I employ a Latent Dirichlet Allocation to extract topic word groupings from the Beige Book documents; the optimal number of topics is determined by a consensus of four Gibbs sampling methods. GDP growth nowcasting is done by means of supervised learning methods (support vector regression) and a comparison of nowcasting power of text word topics is provided. I find that the Beige Book information about current economic activity is significant for in sample and out of sample forecasting. The prediction power increases with the month of the quarter. Nonlinear kernel SVR analysis show that qualitative information about economic activity has nonlinear nature.

Degree Date

Fall 12-19-2020

Document Type


Degree Name





Nathan S. Balke

Second Advisor

Thomas B. Fomby

Third Advisor

Klaus Desmet

Fourth Advisor

Enrique Martinez-Garcia

Subject Area




Creative Commons License

Creative Commons Attribution-Noncommercial 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License

Available for download on Saturday, December 04, 2021