"A conventional assumption in dynamic models is that agents form their expectations in a very sophisticated manner. In particular, that they have Rational Expectations (RE). We develop some tools to relax this assumption while retaining fully optimal behaviour by agents. We study implications for asset pricing and macro policy.We assume that agents have a consistent set of beliefs that is close, but not equal, to RE. Agents are ""Internally Rational"", that is, they behave rationally given their system of beliefs. Thus, it is conceptually a small deviation from RE. It provides microfoundations for models of adaptive learning, since the learning algorithm is determined by agents’ optimal behaviour. In previous work we have shown that this framework can match stock price and housing price fluctuations, and that policy implications are quite different.In this project we intend to: i) develop further the foundations of internally rational (IR) learning, ii) apply this to explain observed asset price price behavior, such as stock prices, bond prices, inflation, commodity derivatives, and exchange rates, iii) extend the IR framework to the case when agents entertain various models, iv) optimal policy under IR learning and under private information when some hidden shocks are not revealed ex-post. Along the way we will address policy issues such as: effects of creating derivative markets, sovereign spread as a signal of sovereign default risk, tests of fiscal sustainability, fiscal policy when agents learn, monetary policy (more specifically, QE measures and interest rate policy), and the role of credibility in macro policy."