Learning Hyperinflations

Emprical studies of hyperinflations reveal that the rational expectations hypothesis fails to hold. To address this issue, we study a model of hyperinflation and learning in an attempt to better understand the volatility in movements of expectations, money, and prices. The findings surprisingly imply that the dynamics under neural network learning appear to support the outcome achieved under least squares learning reported in the earlier literature. Relaxing the assumption that inflationary expectations are rational, however, is essential since it improves the fit of the model to actual data from episodes of severe hyperinflation. Simulations provide ample evidence that if equilibrium in the model exists, then the inflation rate converges to the low inflation rational expectations equilibrium. This suggests a classical result: a permanent increase in the government deficit raises the stationary inflation rate (Marcet and Sargent, 1989)