Very good thread here. This illustrates what Intradaybill kept preaching - that you can never, ever know the odds until after the trades are tallied. "Path dependent", he kept saying. This is where PHd mathematicians claim "the market shouldn't have done that" because it didn't match their model. Hint: all systems are curve fits of past data.
1. bench the system and paper trade it 2. redesign it going further back and test it on out of sample data (sometimes referred to "walk forward" testing) 3. ALL systems are curve fits.......... you can't calculate the future, ever.
Why not looking at another 4 years in the past. Or how about backtesting on another instrument/market? And was your strategy fine tuned on all the backtest data or you had some out of sample years?
sweep your account and trade it untill it blows up. Same thing at a crap game. Put your wad on the table and everytime you are up put the profit in your pocket. Keep playing until the wad is blown. Then count how much you have in your pocket, and that is how profitable your system is.
Now that you've proven the system only blows up every four years, get some investors and trade huge size for the next 3 years and 11 months.
trading is a job. just like any job, you can get laid off or fired at anytime. The purpose of trading is to pay bills, not build wealth. If you are so fortunate as to make more than you need to meet your cost of living, then invest that money like normal people do. They ought to have a website for traders and people who opened restaurants and went broke. They probably have a lot in common. I don't get it. All my resaeach showed that people were buying more and more hamburgers. I even opened up right next to McDonalds and they have one of the most successful locations in town!
at anyrate, if you are sweeping on a regular basis, it won't matter if you blow up because you will still have a stash to start another account. And if that stash has been in stocks it hasn't been doing anything lately anyway. It may not be as much as the account you just blew up, but at least you will know not to trade the same stupid system that blew up your last account.
This assumption was unfortunately way too optimistic, as you just found out. I am coming back to MonteCarlo simulations, as these are providing a reasonable tool to "objectively" define what is the worst drawdown you should be prepared for. I have one system which in backtesting didn't even reach the mean max drawdown for the corresponding number of trades. At least I knew that I should be prepared for a max drawdown way bigger than that system experienced in backtesting. Do the exercise for your system (my offer to run it for you still holds, I know you sent me a PM but for whatever reason I am not even allowed to see it, so you'll have to disclose that distribution in the thread ... it doesn't need to be complicated, essentially a number of buckets, for each bucket the average P&L per trade & the number of trades in that bucket).
So far you've indicated you haven't tested enough (4 years isn't even often a full business cycle) and that you're over-leveraged. These are valuable lessons to learn (ideally beforehand, but it happens), and that is OK. The real question is, out of a sample of say 1,000 trades, is what is happening right now statistically significant? Have you run a monte-carlo to see what the odds are of your current draw-down given your win ratio/profit factor (which is always an assumption for future use as others here have pointed out!). What happens with high win-rate, low-margin systems, is that when they break, they really break dramatically, because that win-rate lures users into riding the edge of acceptable leverage, since the high "apparent" win-rate leads them to very liberal assumptions about potential draw-down.
I tested some systems with equities going back to the mid 90's. Guess what? They didn't do shit when I forward tested them. Did you forward test this? Also, don't forget, your application you were using was most likely giving you preferential fills on stops and limits ( if used ). FYI - designing one and JUST ONE of these systems takes years of research and hard work. Your sample is much too small for starters. Just some helpful advice. My best advice to you is to spend more time researching and try a variety of backtesting engines. You will find out eventually that you will get many different fills and conclusions amongst them all. Which one is right? Probably none of them. GL