Now, I know that so far in this book, we've approached automatic model selection techniques with a fair bit of healthy skepticism. In light of that, you might ask why I would spend time extolling the virtues of ets here all of a sudden.
While it's true that there's danger in automatic model selection – and we'll see an example of that later in this section – I'm more open to automated techniques for forecasting than for other domains. This is for two main reasons.
For one, in many of the sectors that employ forecasting, there is often a need to forecast estimations for a ton of different series, often, and at a high frequency basis. Think of an airline that has to set prices for flight tickets for hundreds of flights and update those prices on a daily (or even hourly) basis, based on projected consumer demand. Because of the intractability of hand-tuning models for each and every one of these series, many current approaches use in-house hard-coded algorithms for...