The research questions addressed by the literature on learning in macroeconomics can be classified into four categories: First, there are issues related to the convergence and stability under learning in models with unique rational expectations equilibria. Authors here are concerned mainly with the learnability of a rational expectations equilibrium, as a measure of that equilibrium's plausibility as an observed outcome in an actual economy. Second, there are issues related to convergence and stability under learning in models with multiple rational expectations equilibria. In this case, learnability serves as an equilibrium selection device, helping economists decide which equilibria are the more likely to be actually observed among the many that exist under rational expectations. A third set of issues involves the examination of transitional dynamics that accompanies the equilibrium selection process. Following some type of unexpected strcutural change or change in policy regime, for instance, economies necessarily must follow temporary transitional paths to a rational expectations equilibrium associated with the new reality. Learning is sometimes used to help model such transitional dynamics. Finally, there are issues related to the examination of learning dynamics that are intrinsically different, even asymptotically, from the dynamics of the rational expectations versions of the models. In these cases, the learning dynamics do not converge to the rational expectations fixed points, and (unexploitable) expectational errors persist indefinitely. Some authors have tried to make use of this possibility in order to build explanations of otherwise puzzling macroeconomic phenomena based on constantly changing expectations.

We study a dynamic equilibrium model in which agents have adaptive expectations and monetary authorities pursue an inflation target. We show how alternative monetary stabilization policies become more effective when fiscal constraints on deficits are implemented, although they are not binding at the equilibrium target. In particular, we show that the inflation target equilibrium can be locally, or even globally, stable for a large class of adaptive learning schemes. We also compare alternative stabilization policies in terms of their stability properties. Commonly postulated conditional Taylor-type rules tend to be dominated by other rules, such as an unconditional Friedman type.

Price fluctuations under adaptive learning in renewable resource markets such as fisheries are examined. Optimal fishery management with logistic fish population growth implies a backward-bending, discounted supply curve for bioeconomic equilibrium sustained yield. Higher discount rates bend supply backward more to generate multiple steady-state rational expectations equilibria. Under bounded rationality, adaptive learning of a linear forecasting rule generates steady-state, two-cycle as well as chaotic consistent expectations equilibria, which are self-fulfilling in sample average and autocorrelations. The possibility of “learning to believe in chaos” is robust and even enhanced by dynamic noise.

The model is a two-country overlapping generations economy with boundedly rational agents who update their decision rules using a version of the stochastic replicator dynamic. The results show that stationary rational expectations equilibria of this model are unstable under this type of evolutionary adaptation. The paper also derives a two-period-ahead forecast of the values of average fractions of savings placed in each of the two currencies. This forecast is used in decisionmaking of a rational agent who has a full knowledge of the evolutionary economy. The performance of the rational agent is compared to the performance of boundedly rational agents, based on the average utility received over time. Results show that the difference between utilities earned by rational and boundedly rational agents is small. In addition, the average utility of the best-performing boundedly rational agents is higher than the average utility of the rational agents.

Recent research has shown the importance of time horizons in models of learning in finance. The dynamics of how agents adjust to believe that the world around them is stationary may be just as crucial in the convergence to a rational expectations equilibrium as getting parameters and model specifications correct in the learning process. This paper explores the process of this evolution in learning and time horizons in a simple agent-based financial market. The results indicate that, although the simple model structure used here replicates usual rational expectations results with long-horizon agents, the route to evolving a population of both long- and short-horizon agents to long horizons alone may be difficult. Furthermore, populations with both short- and long-horizon agents increase return variability, and leave patterns in volatility and trading volume similar to actual financial markets.

The standard approach to modeling consumption/saving problems is to assume that the decisionmaker is solving a dynamic stochastic optimization problem. However, under realistic descriptions of utility and uncertainty, the optimal consumption/saving decision is so difficult that only recently have economists managed to find solutions, using numerical methods that require previously infeasible amounts of computation. Yet, empirical evidence suggests that household behavior conforms fairly well with the prescriptions of the optimal solution, raising the question of how average households can solve problems that economists, until recently, could not. This paper examines whether consumers might be able to find a reasonably good rule-of-thumb approximation to optimal behavior by trial-and-error methods, as Milton Friedman proposed long ago. We find that such individual learning methods can reliably identify reasonably good rules of thumb only if the consumer is able to spend absurdly large amounts of time searching for a good rule.

We introduce adaptive learning behavior into a general-equilibrium life-cycle economy with capital accumulation. Agents form forecasts of the rate of return to capital assets using least-squares autoregressions on past data. We show that, in contrast to the perfect-foresight dynamics, the dynamical system under learning possesses equilibria that are characterized by persistent excess volatility in returns to capital. We explore a quantitative case for theselearning equilibria. We use an evolutionary search algorithm to calibrate a version of the system under learning and show that this system can generate data that matches some features of the time-series data for U.S. stock returns and per-capita consumption. We argue that this finding provides support for the hypothesis that the observed excess volatility of asset returns can be explained by changes in investor expectations against a background of relatively small changes in fundamental factors.

A genetic algorithm (GA) is used to model learning in a financial model similar to the Grossman–Stiglitz model. Individuals need to learn how to use a signal, how to make an inference about a signal from a market-clearing price, and whether or not a signal is worth acquiring. We provide examples in which the GA does and does not converge to the rational expectations equilibrium. Similar to earlier results, the behavior depends heavily on the rate of experimentation or mutation in the GA and the size of the risky-asset supply noise in the economy.