This paper compares four extrapolation methods used to forecast monthly library circulation four months ahead. For 22 of the libraries, 4 years of data were available, while for 12 of them, 7 years were available. Actual data were withheld for four months and were later used for validation. Of the four methods, two took no account of seasonality (simple averages over total sample and regression against circulation), and two did account for seasonality (Box-Jenkins and a monthly averages method). I must admit to a strong bias against the monthly averages method because, in reality, one must forecast 12 months ahead (i.e., from the previous February to make a forecast for this February). This method ignores recent data. Thus, it was surprising to find that this monthly averages method did almost as well as Box-Jenkins (MAPEs of 9.66 and 9.36, respectively, results with no statistically significant difference). The methods that ignored seasonality were worse, as expected (again, no statistically significant difference, with MAPEs of 9.94 for regression and 10.51 for the simple average).
The results are in the correct direction with respect to the rule weight the most recent data more heavily.