Terry Fields wrote:
Roger wrote:

I think you have discovered why the Met Office have chosen a 21 point
binomial filter to smooth the curve. The more 'sophisticated' filters
ignore short term trends.

Interesting, but see below.

I would be interested to see how the polynomial filters would cope with
a subsequent downturn so how about inverting the 1950 - 2004 about 2005
(2006 = 2004 .... 2060 = 1950) and then doing several finish dates to
see when the downtown can first be clearly seen as well as what a mature
downturn would look like.

Now done that; results he

Second-order polynomial:

http://farm4.static.flickr.com/3070/...e4e85748_o.jpg

Third order:

http://farm4.static.flickr.com/3052/...dbf159c8_o.jpg

5-point moving average:

http://farm4.static.flickr.com/3041/...01d63059_o.jpg

10-point moving average:

http://farm4.static.flickr.com/3031/...072cc3b6_o.jpg

20-point moving average:

http://farm4.static.flickr.com/3109/...846c5b23_o.jpg


Several points spring to mind he

- the polynomials are, well, rubbish

- the 10- and 20-point moving averages shift everything to the
right


Of course. Phase delay..

- the 5-point moving average picks up the downturnclearly afetr 2 to
3 years, as might be expected.

I am a bit curuous as to where the data from the years 2010-2050 comes
from...
The thing that appears to come out of this exercise is that to pick up
trends at the earliest opportunity, long moving averages are not the
tool to use.

The standard advice when using these to e.g. track stock prices is to
buy/sell when the data crosses the average. This is margnally better
than tossing dice or gazing into a crystakl ball.

Hadley seems to use a (weighted) 20-pointer; I'll bet
they've done every kind of data run, in a more sophisticated manner
then we have, and have a good idea of the current position.

One of the bloggers said that the 'downturn' data doesn't appear on
the Hadley's 'Myth' page....I can't help wondering why. Perhaps, like
Micawber, they're waiting for something to turn up (or down).