Very clear article. So clear you can actually draw additional points from it (not called out by the author). One of the problems of using regression on time s

Very clear article. So clear you can actually draw additional points from it (not called out by the author). One of the problems of using regression on time series is the model is much better at modeling a grand mean than getting the variation right. So if you take an recent upward trend of something you are measuring and then use regression to get data for past points you don't know- you are likely to predict a fairly flat past, almost no matter what the nature of the original data. You need to (at the very least) cross validate by holding out a large contiguous interval of the past where you know the temperature and the proxies to see if the model tracks.