Page 1 of 1

The Rumpled one camp

Posted: Tue Jun 20, 2006 5:26 am
by JPT
OK, I have this great entry for any of the 1M emin's , ER2, NQ,ES,YM and some of the Forex 15M that I have looked at. I have been trading the ER2, one contract only and usually make 240 to 300 per day. I have a few losing days, but generally am making this with only two to three trades per day.

Now here is the problem. When I write my rules as a strategy, and put in the cost of the trades + slippage, I get a great equity curve directly down.

In real life, I am making a trade and then placing my order to cover 100 to 120 dollars away. So far since 5/23, I am getting this point 81.2% of the time.

When I optimize the stop loss it comes out to 240 so that is what I use in my discretionary testing.

What bothers me is that this system does not walk forward as a strategy but actually it does in real life.

Since it is counter trend and tries to catch a falling knife, I am using the MACD to offer some form of confirmation, at least wait till the slope of the MACD goes flat.

I also am not taking those buys or sells if the bars are way below or above the Kelterner channels.

I guess, I am frustrated that I have done all of this mathematical work before and found Grail robustness systems of 70%+ and now I can't get this one to work in theory but in practice it is almost a living.

I am dramatically afraid that the first time I add money management to the entries, I will get a large draw down and wipe me out.

Am I being too paranoid or should I just go sit on the beach with TRO.

If I could have another trend change filter (any suggestions would be welcome) then maybe I would try to test it again under Grail.

After reading Michal's theory on costs, maybe I am on the right path but it drives me crazy that I have a 70%+ program but when testing it has some very large loses. When I add a SL to it, it helps but just takes down the % also.

Any suggestions on what direction I should go to provide myself iwth the mathematical or statistical rigor so I can start adding to my positions and branching out to other instruments.

I need to automate this since one can't follow the charts all of the time and you do miss trades.

any thoughts out there.

I hope I made myself clear on this.

Regards

Posted: Tue Jun 20, 2006 4:40 pm
by eudamonia
JPT,

Let's try a couple of hypotheticals to get to the bottom of the problem. Say in real time you've been making money with this strategy for about a month. And then you test the strategy in backtesting for that same month and it turns out to be a loser. If this is the case then clearly the problem is with the assumptions in the strategy backtesting. Go back and look at all of the entries and exits for both your real time results and backtested results. Pinpoint the error in the backtesting and make adjustments to the code.

Second example is that you've been trading this system successfully for about a month. Your backtesting is exactly the same for the past month (i.e. results of real-time and backtesting sync up perfectly). But your worried because 4 months ago the system lost money and kept losing money historically. Solution - go back in time on your charts and manually apply all of your rules. If it is still a loser then I would consider using the Grail to determine if parameters need to be occasionally reoptimized. If the Grail doesn't have an adequate solution I would demo trade the system for awhile. If it continues to work in real-time going forward and the Grail shows it as a failure then you know that your assumptions in the Grail WFO are wrong. Reconfigure, reoptimize and see if that takes care of the problem.

TRO has pointed out numerous times that bumble bees can't fly according to the laws of physics (at least on paper). In my opinion this is not a good reason to discount science, but rather to make sure your assumptions and formulas are correct.

The Grail in particular is highly sensitive to your assumptions. Michal has pointed out that the Grail is not the "Grail". You've got to use your best judgement and common sense as well.

Good luck on your strategy. Currently, I am working on a similar problem (ie. a strategy that has walked-forward in real time with real money for over a year successfully) and I'm trying to get it to work on paper. Knowing this strategy works in real-time gives me the confidence to spend whatever hours are necessary tweaking code and the Grail to get my answers to match up. You may need to engage in a similar process.

Based on what you describe above, your largest source of likely error comes from those "extra" discretionary methods like not taking a trade too far below the Keltner Channel. Make sure you code those assumptions in if you want them included. Good luck.

Edward

Posted: Tue Jun 20, 2006 9:33 pm
by JPT
Edward:

Thanks for your support. I guess I will start the detail work. I think the differences are that I take some discretion and eliminate some of the trades that are going to be obvious losers. I have to code this into strategy to better test this.

I discovered this by going thru your first suggestion today.

Thanks again.

I guess I have not done a great job in coding all of the rules.

Regards

your development

Posted: Thu Nov 30, 2006 7:46 am
by traxino
Hi,

I became a board member a few days ago.
What is the stage of your work.
I have a couple of indicators for different tasks and in
general I´m interested in exchanging ideas and experiences
creating a stable strategy. I work with the GO from TradeSmart.

Regards

Peter