Evans, GeorgeLi, Jungang2021-09-132021-09-132021-09-13https://hdl.handle.net/1794/26643Agent heterogeneity has been a widely discussed topic in the recent decade. However, most of the models that emerged from the literature draw their conclusions from a rational expectations equilibrium. These models impose strong assumptions on what agents know and how much they understand the models operate from one period to the next. Adaptive learning offers a straightforward response to this criticism by assuming agents are econometric learners. My dissertation aims to investigate the implications of combining these two features – agent heterogeneity and adaptive learning – together to see how models behave differently from the traditional models. My research relaxes these rational expectation assumptions in several widely-studied macroeconomic models. In the first chapter of the dissertation, I traduce a novel concept of local rationality in a real business cycle model and with heterogeneous agents. The heterogeneity is introduced through ex-ante identical idiosyncratic income shocks. To understand how heterogeneity plays a role in the result, I implement a series of experiments that include different versions of the model with representative agents and heterogeneous agents. Both rational expectation results and locally rational expectation results are obtained. Both chapters find novel results that aggregate variables behave differently under adaptive learning primarily due to wealth-rich agents’ learning behaviors. The simulations show that the rational expectations equilibrium can be approximated with adaptive learning in these otherwise hard-to-solve models. The last chapter focuses on a different type of heterogeneity with adaptive learning agents – expectational heterogeneity. The agents observe different signals to forecast relevant variables about the future. I show analytically that multiple sunspots can be used by agents in the model simultaneously, and these equilibria near an indeterminate steady state can still be E-stable. The analysis in the model holds for both the linear and the nonlinear versions of the model.Overall, my dissertation makes contributions in the intersection fields of agent- heterogeneity and adaptive learning. The interaction could either be used as a computational method to approximate the rational expectations equilibrium (REE) or introduces extra friction in the model to have different aggregate responses given aggregate shocks.en-USAll Rights Reserved.Electronic Thesis or Dissertation