We investigated the required transformations for variance
stabilization and decided to take logarithms in the case of Boots,
Booties and Flats data (for more details see Cryer and Chan [26]).
Fig. 2 shows the sample ACF and the sample PACF for the five retail
series after transforming. It can be seen that in general the sample
ACFs decay very slowly at regular lags and at multiples of seasonal
period 12 and the sample PACFs have a large spike at lag 1 and cut
off to zero after lag 2 or 3, suggesting that seasonal and/or ordinary
differencing might be necessary