Analysis and Forecasts of Money Multiplier Behavior 1982-4

In our last report to this committee, we noted that we had changed our fore-casting methodology, and were updating our data set each month to reflect the data that existed at that time, and then constructed truly ex-ante forecasts based on these data sets. Our analysis of the behavior of autocorrelation functions of these updated estimations indicated that the estimated residuals of some of the component models were not behaving as white noise. As a result, we have specified most of our component models (the exceptions are those for t2 and r+ l-v). The revised models, which are the basis for the forecasts below, are given in Table 1. At first glance it appears that these models are considerably different and in some cases more complicated than the previous specifications. However, if we invert the moving average portions of the new models so that they are written in the form of infinite autoregressive models, and compare these with the equivalent form of our earlier models, the differences between the old and new specifications can be seen to be relatively minor.