Edge Test

M

Maverick1

Hi Quants,

Let's imagine you have a trading system where you enter on a break of a bar's high, conditional on a number of trading rules.

If you were to compare your trading results to random entry, would your random entry also be on a break of a random bar's high? (without trading rules of course)

What is the appropriate comparison for an edge test?

Thanks,
M1
 
Hi Achilles,

Yeah, am using the concept.

So if you were testing a situation as the above, what would you do? Would you base your random entry on the break of a random bar high?
 
I would make the entry completely random (not related to any bar break etc) and hold exits constant across both conditions tested.

For instance, the first condition tests the actual entry edge (bar break or whatever).

The second condition would have completely random entries (random time/random direction).

The key is to hold the exit logic the same across both conditions so you've isolated the entry logic only, sort of like a science experiment (independent and dependent variables).

A price action exit or a time exit (market order on daily close...market order on hourly close etc) would be sufficient.

The random entry condition is then run a hundred or whatever times. Then, the results of the original entry condition are ranked against the distribution of results for the random entry condition. If the original entry condition beats >~70% of random entry iterations (can't remember the exact number Acrary used), then the entry logic is deemed to be non-random.

I could be wrong, but that's my understanding of Acrarys edge test...
 
Interesting. I would have thought that to make the test stricter one would have to use the same type of entry, but as per your post it looks like a random long on close (in the example I gave earlier) with the same type of exit as the actual trade would work just fine for the Monte Carlo like sim. I think Acrary talked about doing the random test 1,000 times.

Also, think he used the 70th percentile as a cutoff for the total profit comparison. But your memory is pretty good... those are some 10 year old threads were talking about :-) The guy definitely had some great contributions to this site. I wonder how the story went, think he was hired as a trader or something at some point?
 
Well I cheated a bit. Collated most of his relevant material re pyramiding, risk, edge testing etc. Just rereading the notes this week. He was involved in a couple funds. Not sure where or why he finally left.

I see where you're going. Hold entry time constant and randomize entry direction. By all means test it. That condition will yield roughly half of the entries being identical to the original entry logic being tested. So I'm not sure the results would provide a good basis for comparison. I could be wrong tho.
 
Come to think of it acrary may have identified how many longs and shorts the original entry condition produced beforehand and then designed the random entry condition to reproduce the same number of longs and shorts. For example if the original entry condition produced 70 longs and 30 shorts the random entry condition was programmed to enter 70 longs and 30 shorts. I'll see I can dig it up later
 
in some of his later postings, acrary seemed to dismiss the idea of edge based trading in favour of market character based trading...might be worth looking that up, could save some time.
 
I have yet to comprehend what is the difference between acrary's edge and market character.

For me, I would compare an edge to:
1. Long only, enter at open and exit at close.
2. Short only, enter at open and exit at close.
3. Long or short (randomly), enter at open and exit at close.
 
achilles28 said:
Well I cheated a bit. Collated most of his relevant material re pyramiding, risk, edge testing etc. Just rereading the notes this week. He was involved in a couple funds. Not sure where or why he finally left.

I see where you're going. Hold entry time constant and randomize entry direction. By all means test it. That condition will yield roughly half of the entries being identical to the original entry logic being tested. So I'm not sure the results would provide a good basis for comparison. I could be wrong tho.
More...

I wouldn't hold entry time constant, (if by that you mean the time at which an entry is made, like 9.42 for example), that would random too together with a random entry. The only similar thing would be the exit rule. But now, reading the posts above, it looks like some even vary that as well. Hmmmm...
 
You must log in or register to reply here.
Back
Top