I had a few requests for video solutions from section E.2.3, so I made them all. Should be ready to go as of now.

Thanks to Karin for pointing out that the formula for predictions with the Holt-Winters model and multiplicative seasonal was incorrectly dividing by the seasonal term instead of multiplying by it as it should be.

I’ve fixed this in the lesson, the lesson notes in the handouts, and on the summary sheet as well.

I found a couple of mistakes in the Negative Binomial portion of the table on the back of D.2. Those have been fixed, and the corrected pdf is up on the website now.

Be sure to download a new copy of the D.2 summary sheet.

By the way, I also added the formula for the entries of the diagonalization matrix D from the Iterative Weighted Least squares method to the summary sheet.

There was a typo in E.1.5 – the formula for the peak time in the Bass model should have had denominator p+q, not p+1 as previously stated. This is fixed in the lesson, handouts and the summary sheet.

Also on the E.1 summary sheet, there was a correction in the exponential smoothing bit. There is more smoothing when alpha is smaller, not when alpha is larger. This matches with what the lesson already said.

I added a note to E.2.3 to clarify that the acf function of an MA(q) series has actually q+1 nonzero terms, not q as previously stated. I was thinking of rho_1 through rho_k, but of course rho_0 is always one as well, so that’s one more.

I also updated the handouts and the E.2 formula sheet to reflect this change.

These two questions are not able to be answered with the given information. Computing confidence intervals for predictions in linear models involves the covariances of the coefficients, which were not available in these two problems. Therefore, both have been removed.

You don’t need to worry about missing out on anything due to this, because these were not reasonable questions, so similar questions would not appear on the exam.