Forecasting With The Theta Method

7Conclusions and the Way Forward

By Dimitrios D. Thomakos and Kostas I. Nikolopoulos

From its inception, in the minds of computer engineers, to its current form, as an all‐purpose, multidisciplinary, forecasting tool, the ‐method follows the time‐honored path of simplicity. Once we understood how to link the original derivations to an underlying data‐generating process, it became not only apparent why the method might have worked but also why it will continue to work in many different contexts: the keyword here is “adaptability” or “flexibility” across different types of time series. If we consider the trend‐stationary representation of the univariate ‐method, we see that with a choice (which maybe we can make more “automatic” in the future) of the trend function, we can address both stationary and nonstationary time series representations.

The results of our analyses presented in the previous chapters provide us with a wealth of practically relevant and theoretically motivated recommendations that support our abovementioned claim. Let us review some of them. First, the ‐forecasts are good for predicting direction, most times being above the benchmarks, sometimes at par, fewer times ...