It's just an artifact of the noise model; a completely normal, natural, expected effect of random variation, enhanced here by
strong autocorrelation.
The brief time span of the most recent data, and
the strong autocorrelation of temperature time series, combine to make the error range considerable.
Not exact matches
First - differencing, on the other hand, introduces very
strong lag - 1
autocorrelation, which I seriously doubt you accounted for.
But as I said, even when I jack up the noise
autocorrelation it still doesn't give a
strong enough PC # 1 to come close to that of MBH98.
While
autocorrelation does suggest some justification for 74, 37y periods, it would be hard to «tweak» the
strong 60y to conform to 151y saw tooth.
I calculated the Durbin Watson statistic (DW) for
autocorrelation for the GISS time series 1979 - 2007 (using the residuals from the anomaly regression) for monthly data and determined a DW = 0.83 indicating a
strong positive
autocorrelation.
Even white noise can give that impression, and with
strong enough
autocorrelation one can get hellacious apparent «trends» that are nothing but noise.
For the same reason, tree growth and food production in one year may influence growth in the following year, and lead to a
strong serial correlation or
autocorrelation in the tree - ring record.
But if you look at the residuals and test for the presence of
autocorrelation you'll get very
strong evidence that the error term is autocorrelated.