Introduction

Economists widely agree that inflation expectations play a crucial role in hyperinflationary episodes. There is little agreement, however, on how these expectations are formed, and two competing approaches to the question of expectation formation have emerged. On one hand, the rational expectations approach posits that agents base their expectations on full knowledge of the structure and equilibrium of the economy. On the other, the passive adaptive learning approach endows agents with simple learning rules that map past inflation observations to future forecasts, with the form of the mapping incorporating little direct information about the structure of the economy.

Each of these approaches faces difficulties from both conceptual and empirical standpoints. The rational expectations approach (for example, Sargent and Wallace, 1987) gives no account of how agents come to acquire rational expectations, and in particular there is no specification of what agents would believe were they to observe a history that violated rational expectations. For models with multiple rational expectations solutions, the approach faces the problem of selecting among the (possibly infinite) set of equilibrium paths. Empirically, it is difficult to reconcile rational expectations paths in the paradigm hyperinflation model with the historical record, which shows hyperinflations to have involved increasing and accelerating inflation rates, terminating in major fiscal restructurings or ‘currency collapse’. Finally, there are problems in reconciling the rational expectations approach with experimental results.