This paper develops a model where agents learn about the probability of devaluations in a fixed exchange rate regim e. The true probability of devaluation is assumed to be low (or zero) b ut agents are initially unsure about the government's intentions and st art with a high prior belief. Bayesian updating dictates that private sector expectations are revised down during periods when no devaluat ion occurs, but are sharply increased when a devaluation does occur. Thi s model of learning is embedded in a model of overlapping contracts to study the process of inflation convergence after ERM entry. The main results are that agents over-estimate the authorities' desired realignment rate, so inflation will be high in anticipation; but as the authorities do not actually devalue, there is a terms of trade loss leading to recession. Before long-run equilibrium in which agents correctly estimate realignment rates can be attained, competitivenes s losses actually have to be regained by a sustained deflation. Copyright 1993 by Royal Economic Society.
