In the chapter on ARMA models, the example analyzed is Canadian Employment data. One of the models that is fit is an MA(4) -- see pages 164-6. When I tried to reproduce these results using software other than EVIEWS, using the data disk in the 1st edition, I couldn't. I contacted EVIEWS and they discovered a programming error in the estimation routine. They released a patch to fix EVIEWS. However, the author never re-estimated his model, and the estimates in the second edition are the same as in the first. However, my copy of the 2nd edition has no data disk! Was that thought to be an adequate solution?!
Chapter 9 ("Putting it all together") is a capstone chapter that analyzes liquor sales data using the techniques introduced in earlier chapters. After several pages (pp. 207-19) a model is selected. On pages 220-2, the residuals are examined using the Box-Ljung statistic, and deemed acceptable. However, as a careful examination of table 9.6 makes clear, the p-values for the Box-Ljung statistic were computed as if the input data were a raw series. The model generating the residuals (p. 219) had 3 autoregressive terms! This changes the d.f. in the chi-square distribution of the statistic. If you make the appropriate correction using the data in table 9.6, and compute the p-values correctly, you will see that the model residuals apparently ARE NOT white noise. One reason is a calendar effect in liquor sales: months that contain more than a usual number of Fridays and Saturdays result in more liquor sales; ones with more Sundays result in lower liquor sales. However, the author doesn't discover this, but accepts his inappropriate model on the basis of faulty distribution theory.