Forecasting using automatic identification procedures: A comparative analysis

The goal of automatic forecasting procedures is to select models which are as accurate as those produced by competent modelers. A major step in this process is deciding upon the order of differencing. Two standard tests, based upon linear regression and autoregression respectively, are compared with a new procedure based upon the use of Walsh functions. All three are found to be effective in deciding upon the order of differencing required. Forecasts generated by automatic methods, without and with allowance for possible interventions, are then contrasted with those generated by experts. The goal of modeling competency is seen to be achieved by the proposed methods.