مقاله سال ۲۰۱۶

Hill, Zhang, and Burch (2015) proposed a measure called a ‘‘forecastability quotient’’ that estimates ‘‘. . . the economic advantage of using a time phased reorder point system (TPOP) with time series forecasting rather than a simple reorder point system in an independent demand inventory management context’’. In plain language, their version of the TPOP uses the forecasts from a time series method to determine when to reorder stock, while their simple reorder point system uses a moving average forecast (with the length or type of moving average unspecified).

Hill et al. define the forecastability quotient (Q) as the standard deviation of the time series divided by the standard deviation of the forecast errors. If this ratio is more than one, the authors conclude that forecasting should be done using a time series method. If the ratio is less than one, the authors reject all time series methods and conclude that forecasting should be done using a moving average. Although research has long since established that a method’s ex ante fit often has little to do with its ex post accuracy, Hill et al. claim that Q in the holdout sample can be predicted as a function of Q in the fitting sample, together with the parameters in Holt’s additive trend method of exponential smoothing (Holt, 2004). We have a number of corrections and objections to Hill et al.’s analysis and conclusions, but our most important point is that Holt’s method is the wrong one for the data.