Methodology

# Signal Processing Structure Overview

There have been some interesting questions in the comments to my latest article. That is nice to see, explaining the application is not easier than creating it.

Historical prices get downloaded for the desired portfolio of stocks. Each security gets it’s own worksheet. The SPY Signal graphic I publish is an example. All the other securities look the same for the first seven columns.

If Vanilla analysis is required, indicators for all the sheets will be calculated, but Specter is the thing we are analyzing so only indicators for SPY are calculated. That results in the SPY Signals.

This looks exactly like the other securities except for xEM and xNN. xNN is experimental at the moment.

I used to go directly from this to producing a sheet I call the Single Garbage Can. The logic to do that is complex, so I developed a sheet called MapSignal, originally as a means of proving that the original coding was correct.

MapSignal winds up looking something like this:

This thing drives me nuts to look at because the xCC for 8/12 is the end of day signal for 8/11. The xFC signal that is active today, will appear on the 8/13 line tonight or tomorrow once the returns for today are known. The xCC for 8/12 is built from the third character of the xEM column on SPY Signals. The hex signal for 8/12 is actually xCD, xC for 8/11 and xD for 8/10. xD gets converted to xC because of the rules.

The column LF is the Neo-Classical Specter/FF5 signal. Note xFC on 8/6, etc. is F and the others are L.

TF1 – TF4 are the various trend following signals. Y means trend following is active, N means it isn’t. 13em in this scheme is 3. 1 = 3em, 2 = 7em, 3 = 13em, 4 = 16em. I’ve been looking at 16 instead of 21 lately.

If trend following for TF4 is active, the Y in TF4 for 7/28 would change the F in LF2 to L.

Essentially any other strategy instructions could be stored in any TFx type column, and be processed with exactly the same code except the column numbers can be variable.

Once the instructions for the date range are set, the single garbage can is created.

A huge advantage of pivot tables is that the developer doesn’t have to develop indexing logic, so everything can be dumped into a single sheet. Job 1 for the developer is to make things as easy as possible for himself.

This structure can handle most of the analysis I’ve done so far. The aesthetics are not up to my usual standards, but this is a relatively advanced prototype that is about due to be restructured.

Note that a glance at FCC2 shows some 6% natural log gains for xFC where 13em was flat. In hindsight, we would rather be long those days. It’s important to remain objective about strengths and weaknesses.