Sunday, May 20, 2007

Testing - Composite Cross Over

Many times I like to add some complexity to my simpler filters to see what would happen if...

Today I was looking at the cross over filter v6. That looks like this -
show stocks where close is between 15 and 35
and average volume(90) > 500000
and open < ema(21)
and open < ema(8)
and open < ema(4)
and close > ema(21)
and close > ema(4)
and close > ema(8)
and close > open
and close 3 days ago < ema(21)
and close 5 days ago < close 3 days ago
and close 5 days ago < ema(90)

And I added a line to the filter which contained a constraint for the linear regression slope - that looked like this

and linear regression slope(45) < 0

After testing the 45 day version I tried the linear regression slope less than 0 for a 15 day period and a 30 day period. And then I changed the less than 0 to greater than 0 -
and linear regression slope(45) > 0

And tested every version in that configuration and here are the results

What this shows is that the original cross over v6 has a good win percentage, a good short term (4 day) ROI and a good 30 day net change. It also has a good number of selections. None of the other filters quite matched or exceeded these results in total. But there are some things to learn from this.

If you want a filter that doesn't pick many stocks but those that it picks have a high probability of winning with a great reward to risk factor and a super ROI then you would use the linear regression slope for 30 days greater than 0.

On the other hand while the filters where the linear regression slope was less than 0 did not match the original, they weren't too awfully bad either. That suggests that maybe combining the LRS with the cross over isn't such a good idea but maybe the LRS less than 0 has some promise either on its own or combined with another good filter.

Something to think about - something to look into.