Analysis of Sample Period Lead 1 Forecast Errors

Abstract

As has been mentioned earlier, BNM was fitted to 44 quarterly observations, 1960-01 through 1970-04, while ARIMA to 56 observations, 1957-01 through 1970-04. The post-sample data for both the models consisted of 16 observations, 1971-01 through 1974-04. Lead 1 forecasts were calculated from the fitted econometric as well as ARIMA models (econometric forecasts were in fact provided by the author of the BNM). Since both the econometric and ARIMA models required varying number of lagged observations to calculate forecasts, 40 econometric and, generally, 51 ARIMA forecasts for the sample period could be obtained, while for the post sample period 16 forecasts were calculated for both the models. For some of the ARIMA’s the number of within-sample forecasts was more than 51. However, to maintain uniformity, only 51 within-sample forecasts, 1958-02 through 1970-04, were considered for all the ARIMA’s. The autocorrelation and cross-correlation analysis presented later was however based only on 40 observations, covering the period for which econometric forecasts were available. The analysis of sample period lead 1 forecast errors is presented in Table 2.