Essays on Asset Prices, Expectations Formation and Nonlinear Macroeconomic Effects
My dissertation's main objective is to estimate the effects of asset prices and expectations over the business cycle, considering potential nonlinear effects. Moreover, I am interested in quantifying the impact of the financial sector, both domestic and from abroad, on the economy. In this tenor, in Chapter 1, I study the effects of house prices on the economy by introducing adaptive learning expectations formation in a DSGE model with housing. This framework provides flexibility in beliefs to match the non-rational behavior of house price expectations. Additionally, I can capture the evolving effects of extrapolative expectations on house prices on the economy. The results suggest that the feedback from house price beliefs into the economy was more severe around the period of the housing price bubble and continued to exist, in a lower magnitude, around the Great Recession. Meanwhile, in Chapter 2, we further consider the effects of house prices on credit markets' conditions, for which we introduced a banking sector in the model. The results suggest that agents' expectations amplify the credit supply during episodes of asset bubbles. Additionally, we find that macro-prudential policies may lessen the response of financial intermediaries to the housing shocks, magnified by the learning dynamics. In this regard, I found it essential to consider the effects of agents' sentiment over asset prices and the business cycle. Moreover, Adaptive Learning allows me to analyze this behavioral element jointly with other factors to separate their effects. Finally, I have also explored the asymmetry in the business cycle, intending to create a more efficient estimator. Given the documented asymmetries in business cycles, it is vital to consider nonlinear DSGE models to better approximate the data. In this context, the "occasionally binding constraints" is one avenue used to address the nonlinearity estimation challenge. In Chapter 3, I revisit this issue with an MCMC algorithm based on a mixture model. By carefully defining the sampling scheme, I can make most of the draws directly from their conditional distribution with a Gibbs sampler step. As a result, the algorithm features fast convergence and low inefficiency factors.