Box and Jenkins: Developments Post-1970

Abstract

Box and Jenkins’ book had an immediate impact on the practice of time series analysis and it was not long before Chatfield and Prothero (1973) presented an application of their seasonal modelling procedure to short-term sales forecasting, concluding that‘with our particular set of data, the Box-Jenkins procedure has not proved satisfactory. … Clearly one cannot generalize from one set of data, but discussion with other statisticians who have tried the Box-Jenkins approach has strengthened our view that the procedure is likely to be less useful in seasonal sales forecasting than in other time-series applications particularly when the data exhibit high multiplicative seasonal variation’ (ibid., page 313). They suspected that, in part, this could be due to their choice of logarithms to transform the series prior to their ARIMA analysis, and this was also the view of several of the academic discussants of the paper, although others, typically from the perspective of industrial and sales forecasting, concurred with Chatfield and Prothero that a major drawback of the Box-Jenkins approach was that ‘the subjective assessment involved in choosing a model means that considerable experience is required in interpreting sample correlation functions. In addition the procedure is expensive in terms of both computing and staff time’ (ibid., page 313).