Customer Reviews

SAS for Forecasting Time Series, Second Edition

5 star:		(1)
4 star:		(1)
3 star:		(1)
2 star:		(1)
1 star:		(1)

 

 

The SAS Institute's Books by Users series contains many excellent manuals. The ones by Paul Allison (on survival analysis and on logistic regression) and by Stokes, Davis and Koch (categorical data analysis) are particularly well-written and illuminating. Unfortunately, Brocklebank and Dickey's contribution on time series analysis falls far short of the mean.
The problem is not the statistical content, which is quite reliable, but rather than explanatory style. Chapters are disorganized, with many ideas introduced before being explained. Furthermore, the authors have adopted an unfortunate habit of constantly referring to "you" (i.e., the reader). "You" will do this. "You" will decide to do that. In many case, it was far from clear why such decisions would be made.
The most serious problem, though, is the treatment of SAS code. This is supposed to be a book about ideas AND about syntax. But code is repeatedly presented with any kind of line-by-line explantion. Readers ("you" again) are left to wonder what the various elements of code mean, and how they control the computations done.
I was very disappointed with this book. Unfortunately, the only alternative is to use the SAS documentation, and that's not really a very good alternative.

While the publishers describe SAS for Forecasting Time Series as a manual, the authors have provided more than SAS statements and the resulting outputs. Theoretical explanations, equations, and matrix algebra forms of equations fill the book. This superb manual is the product of the Research and Development Director of Analytic Solutions at SAS and of the Professor of Statistics who was the co-inventor of the Dickey-Fuller test. In addition to the coverage of the essential univariate and multivariate time series analysis topics (e.g., ARIMA models), the authors included entire chapters or large portions of chapters on: Cointegration, State Space Modeling, Spectral Analysis, and Data Mining.
My only disappointment with this manual was the lack of an entire chapter on forecast accuracy. Four pages of references did not include a single reference to articles about forecasting competitions. The authors could have: (1) held back recent data in their examples (2) made forecasts with their best models (3) explained how to identify significant changes over time in error terms, standard errors, and in correlations (4) explained when and how to re-calculate model parameters (5) discussed the choice of unbiased forecast accuracy measures for comparing forecasts from ARIMA and regression models.

.. the benchmark being SAS/ETS documentation. Let me recommend printing out selected chapters of SAS/ETS User's Guide - available online - describing the procedures that you (may) need, such as MODEL, FORECAST, ARIMA, VARMAX, UTC and STATESPACE.

I had hoped that it would turn out to be at least a good book (if not excellent) by seeing such big names in the list of authors but am terribly disappointed! The table which normally could have been printed in a fraction of page has been printed on a complete page. Even if that was not enough, 5-6 pages continuously filled with 5-6 tables (all displaying the same meaning) can be commonly seen throughout the book. There is a terrible amount of repetition in printed matter also.
It seems (though might be unintentionally) that a lot of stress has been given to enhance the page count of the book while giving almost no consideration to the quality of the material.
In addition, a lot of important matter (for example non-linear time series models) have just not been covered.

If you're interested in advanced methods of forecasting time series data using SAS then this is the book to have. It is loaded with examples and interpretation of output as well as a nice concise explanation of theory. Everything you would expect from such renowned authors.