Forecasting the Real Interest Rate

A method first employed by Mishkin to forecast the ex ante real interest rate involves regressing the ex post real rate on various explanatory variables. The fitted values from this regression are taken to be the ex ante real interest rate series. We argue that this method does not reproduce how forecasters actually generate ex ante real rates because the method generates in-sample predictions and uses information unavailable to forecasters at the time of the forecast. Our paper develops several regression-based forecasting models to generate alternative ex ante real rates using only information available at the time of the forecast. Unlike Mishkin, we update our model with new information as time passes and generate out-of-sample forecasts. We show that the parameters of the model change over time, yielding results different from Mishkin. While Mishkin's technique is found to yield better forecast diagnostic statistics, we contend that our models more closely approximate how forecasters operate. If so, then our ex ante interest rates will be closer to the true, unobservable ex ante rates. Finally, as an application of our technique, we consider the debate between the real business cycle theorists and the monetarist school on the effects of money on output. Recent studies of this question have argued that money growth may be a poor proxy of monetary policy. Accordingly, we use two other measures of monetary policy more consistent with the actual operating procedures of the Federal Reserve. Our results lend some support to the monetary models.