• recursive learning
• rational expectations
• heuristics and bounded-rationality
• forecasting models
• complexity
• heterogeneity
• agent-based computational economics

This thesis consists of 3 chapters.
Chapter 1 briefly introduces rational and extrapolative expectation models and the empirical evidence on the testable implications of such models is reviewed. Rational expectations are robustly rejected by using both survey data and controlled experiments in the analysis of expectations and forecasting.
Chapter 2 compares adaptive learning “rational” expectations with the “Wilderness of bounded-rationality” by means of an agent-based model. Specifically, heuristics and recursive least squares learning are introduced in the K+S model. Agents are allowed to switch across expectation rules, according to their past performance. We find that: (i) agents rationally do not systematically choose to adopt “rational expectations”; (ii) the macroeconomic and individual level performance worsens, when sophisticated expectations are introduced.
Chapter 3 provides the main contribution of this work: a new form of expectations is advanced for agent-based models. Analogously to Muth’s hypothesis, agents perform forecasting by using “the true model” of the economy. The pseudo rational ABM expectations are implemented in the K+S model and compared with static naïve expectations. However, expectations are subject to the “infinite regress in expectations problem”, since the model is populated by heterogeneous interacting agents. In order to break the circularity above we propose the following exercise: only one agent is rational whereas all the other ones have adaptive expectations and the rational one knows it. We find that agents, who use the true data-generating-process to form expectations, perform perfect forecasts. Nonetheless, the sophisticated expectations have mixed effects on the performance of the system, as forward-looking firms’ survival prove to be worst than in heuristic-driven ones, and they destabilize the system dynamics, which is more volatile and unstable.