Cagan argued in 1956 that during a `hyperinflation' defined as inflation in excess of 50% per month expectations of inflation will dominate other factors in the demand for real money balances, such as interest rates and the level of transactions, in a special case of Friedman's general theory of money demand. The stability of such a relationship during a hyperinflation is a most stringent test of the theory of money demand, and many researchers have re-examined Cagan's model and emprical work various alternatives to the assumption of `adaptive expectations' used by Cagan.

In Discussion Paper No. 473, Research Fellow Mark Taylor uses new cointegration techniques to test the hyperinflation model and to obtain highly efficient estimates of its parameters, subject only to the very weak assumption that agents' forecasting errors are stationary. In Taylor's model, real money balances depend on expected inflation, and during a hyperinflation both series are non-stationary processes, but their percentage changes approximate to stationarity. There must therefore exist some stationary linear combination of real balances and inflation. If this is correct, a cointegrating parameter between current real money balances and current inflation will be a highly efficient estimate of the semi-elasticity of real money demand with respect to expected inflation, which will not be specific to the process of expectations formation. If no such cointegration can be found, the model is rejected.

Taylor applies these techniques to data on the hyperinflations in Austria (1921-2), Germany (1920-3), Greece (1943-4), Hungary (1922-4), Poland (1922-4) and Russia (1921-4). He finds that both real money balances and inflation were first-difference stationary during the Austrian, Polish, Russian and Hungarian hyperinflations. The results provide some weak support for Cagan's hyperinflation model for Hungary, Poland and Russia, but they reject the hypothesis that the authorities expanded the money supply to maximize the inflation tax revenue. Taylor uses these results to test the hyperinflation model under rational expectations, and the data reject this hypothesis for Russia and for Hungary but can not reject it at the 5% significance level for Poland.

The real money balances and inflation series for Germany needed to be differenced twice, however, to induce stationarity. Respecifying the model to include expected exchange rate depreciation, which allows the substitution of domestic money not only by real assets, but also by foreign nominal assets, appears to indicate that the latter were the closer substitute for domestic money during this period.

The Hyperinflation Model of Money Demand Revisited
Mark P Taylor

Discussion Paper No. 473, October 1990 (IM)