Prediction(!!!) of future temperature using "cycles" + trend

This is an update of an earlier post and is just a "fun" thing

Get HADCRUT3V global temp record
Average over 6 months to remove some of the "noise"
Create a series of narrow band filters on the resultant temperature plot.
Tune each filter manually to isolate peaks in the output.
These SHOULD show peaks wherever there is a signal of, say, Scarfetta's 60 years.
Take the output of each narrow band filter and generate a cosine wave that is as near as possible the same amplitude and phase as the filtered signal.
Do this a number of times isolating each frequency.
Add together the generated cosines. Multiply the result by a factor (approx 3 in the plot below). If there is a suitable long period - low frequency - signal isolated the resultant should match the original signal. IT DID NOT so a trend was added.
y = 2.44231E-07x^3 - 1.36387E-03x^2 + 2.53884E+00x - 1.57576E+03 (not good as it deviates before 1850.
This is what I got :


Note
No 60 year signal
No massive TSI signal (there is some!)
The significant signals are all around 2 to 6 years
The plot shows prediction for the next few years!!!!

Get HADCRUT from CRU website
Get Excel from microsoft
Get bandpass filter from
http://www.web-reg.de/index.html
in general set the bandwidth to months/150 (e.g. period start 21.19 end 21.29 months ie. months/200 in this case)