WWL - Is there another way

Posted December 8th, 2007 by knigit

I have been wondering about the What Worked Lately approach to screen selection and I am hoping to test the following:

Find the stocks over the past X periods that produced the highest return with the lowest variance in screen picks. That is, find good screens that made their gains from multiple stocks - not just one stock. My reasoning is that I think such screens were more than lucky.

The only way I can think of to test this is to run each screen for its 1st, 2nd, 3rd, 4th...nth rank and the post-process the data.

Does anyone have another approach?

Thanks

knigit

CAGR ›

concept: Lowest variance in screen picks

On December 18th, 2007 mariopapas said:

Hi knigit,
My name is mariopapas ;
What do you mean by :: running several runs of the same screening program with ranks 1st, 2nd, 3rd, 4th, etc... ?

re: concept: Lowest variance in screen picks

On December 21st, 2007 knigit said:

Hello mariopapas-

Here is an example of what I am doing to get the results for this calculation:

Run 193404 shows the returns for the #1 pick for the screen GSX.
Run 193403 shows the returns for the #2 pick, and so on.

This method requires a new run on the backtester for each rank. I was hoping to minimize the load on the backtester, but I think this will work out OK.

Thanks for the reply,

knigit

I will assume you are

On January 2nd, 2008 Keelix said:

I will assume you are looking for some kind of measure of variance of returns for picks in a given period? E.g. if you are running a top 5 screen, you want to measure the statistics for individual returns when selling and somehow generate an aggregate number from this, possibly mean/median/min/max/sd? If we can agree upon measures which makes sense might try to put something like that into the backtester.

Reply

On January 2nd, 2008 knigit said:

Keelix-

Thank you for the response.

Yes, your description is precisely what I am after. I was planning on calculating the sd once I gathered all the data into Excel. Another metric may be better - I do not know - but sd should allow me to achieve the goal.

If it would be possible to implement, I would greatly appreciate it.

Thoughts?