Heterogeneity and Learning with Complete Markets∗

We study an endowment economy with complete markets and heterogeneous agents that do not have rational expectations, but form their beliefs using adaptive learning algorithms that may differ across individuals. We show that market completeness allows agents to smooth consumption across states of nature, but not across time, and that the initial wealth distribution is not enough to pin down the long-run equilibrium: initial differences in beliefs induce persistent consumption imbalances that are not grounded in fundamentals. In some cases these imbalances are not sustainable forever: the debt of one of the agents would grow unboundedly, and binding borrowing limits are necessary to prevent Ponzi schemes. Finally, we find that if a rational social planner attaches to the different individuals fixed Pareto weights, there exist configurations of individual expectations such that welfare is higher with financial autarky than with complete markets. The first best can be restored introducing a distortionary tax on borrowing, that transfers consumption from the more optimistic agents to the others. JEL classification: C62, D83, D84