Practical Time Series Forecasting, Third Edition
is now in print (a Kindle version is also available
, as well as an Indian Edition
). Here is a summary of the changes:
Based on feedback from readers and instructors, the Third Edition
has two main improvements:
The first and major change is in the order of topics. The reordering includes
- reversing the order of the smoothing and regression chapters. "Smoothing Methods" (Chapter 5) now precede the two "Regression-Based Models" chapters (Chapters 6-7)
- relocating and combining the sections on autocorrelation, AR and ARIMA models, and external information into a separate new chapter (Chapter 7: "Regression-Based Methods: Capturing Autocorrelation and External Information")
- forecasting binary outcomes is now a separate chapter (Chapter 8 introduces the context of binary outcomes, performance evaluation, and logistic regression)
- neural networks are now in a separate chapter (Chapter 9)This reordering of topics is aimed at providing an easier introduction of forecasting methods which appears to be more intuitive to students. It also helps prioritize topics to be covered in a shorter course, allowing optional coverage of topics in Chapters 8-9. The restructuring also aligns this new edition with Practical Time Series Forecasting with R, Second Edition, offering instructors the flexibility to teach a mixed crowd of programmers and non-programmers.
The second update is software screenshots. Since the Second Edition, XLMiner has introduced several new versions. This edition includes screenshots of the latest version. Although it is not detrimental to learning, a better match between the current software version and the screenshots is helpful to those interested in recreating the chapter examples. Future software versions might again lead to slight discrepancies, but software differences are expected and are part of the forecasting environment.
Also new to the Third Edition
is the discussion of fixed partitioning vs. roll-forward partitioning (see Section 3.1 in Chapter 3). While roll-forward partitioning can only be performed manually with XLMiner, this concept is important and useful in many deployment scenarios.latest version. Although it is not detrimental to learning, a better match between the current software version and the screenshots is helpful to those interested in recreating the chapter examples. Future software versions might again lead to slight discrepancies, but software differences are expected and are part of the forecasting environment.
The discussion of ARIMA models now includes equations and further details on parameters and structure.