Quality Quandaries: Box-Cox Transformations and Time Series Modeling-Part Ⅱ

摘要

In this column we have discussed how to forecast a seasonal time series fitted to transformed data.rnWe also showed how to interpret the forecast equation as a generalization of an exponential moving average. This provided us with an intuitive understanding of how the forecast equation works. Furthermore, we demonstrated an alternative time series model based on extrapolating the monthly linear trends using a somewhat unusual approach involving a second-order seasonal difference. Both types of models provided what appear to be quite reasonable forecasts that were better than what CP originally reported.rnThat two very different models can produce good forecasts is of course just another example illustrating that there is no such thing as the "correct" model. As George Box often has said "All models are wrong, but some are useful!" In this case we found two models that appear useful for forecasting.