A method I use which works well: Suppose you have a parameter "X" Run the test and graph the results vs. "X" There should be range of "X" over which the system is profitable I Choose "X" in the middle of the range. Now get a little funky: if you have two parameters, this makes a 3-D graph (profit vs. X,Y), look for a nice broad mountaintop; If you have three parameters, you have a 4-D graph, good luck with that one...
But when does a person cross the line into curve-fitting? I've manipulated variables that have resulted in improving the yield when backtested, and I've managed to convince (or fool) myself as to the rationale for using them. But seriously, I think as long as there is some correlation with the variable to results (e.g., holding all other variables constant, as you increase/decrease one variable there is a subsequent increase/decrease in the backtested returns), there is a strong argument that incorporating and tweaking the variable is justified. Obviously if you incorporate a variable that causes a jump in returns at a particular value range but moves arbitrarily beyond that, then that would be a case of overfitting. Having said that, it appears that roughly a third of respondents actually think there are "consistently" profitable systems out there, while another 45% think that it is possible to come up with system that is profitable at least in the short term. For my part, I think it's possible to come up with a consistently profitable strategy as long as it takes advantage of three constants in the market -- 1) There are always newcomers to the market and who can be exploited (this is more so during extended bull runs) 2) The market has a short term memory 3) Despite the current bear market and predominance of short sellers, overall the market is based on investors' optimism for the future, and there are always assets out there that can benefit from that optimism.
There are plenty of good mechanical systems, however most of them are long term. They run off daily, weekly charts. Most people just don't have the deep pockets or the patience to stick with them. This high frequency trading or day trading if you will, is largely a creation of the prop firms to extracts commissions out of desperate, poorly capitalized traders. No offense to those of you who day trade, but imho the odds will one day catch up with you. Ed Seykota once said he has never come across a short term system ( < 30 min) that is profitable in the long run.
To heck with Ed Seykota....I've traded exclusively short term in one area for 7 years, only had a handful of losing days, NEVER a unprofitable week, NEVER an unprofitable month. About 100,000 trades. Average stats: 76% Win, 11% Lose, 13% even. The ways that I've seen virtually all traders I know approach the market are wrong.....the way a $100 billion mutual fund manager could think he can beat index returns consistently---it'll never happen. Think about what others do not think about. Try what others have not tried. Best of all keep it simple. PS: Pulled $12k today.
That's it? You've been trading for 7 years with that win/loss ratio and only made 12K today? Shit, one guy I know made 134K today and has a much worse win/loss ratio, plus has been trading for only 3 years. Let's just say you matched the indices today and made 4% on your portfolio, that's 12k on 300K. I knew Ed Seykota, and you ain't no Ed Seykota.
Those are sick stats, man, if you're legit. The kind of results that encourage me to plod onward. I saw a previous post of yours regarding your $2k-3k/day in commissions. What do you trade?
"Never a unprofitable week" Anyone buying this? Assume 12 weeks for vacation and miscellaneous. 40 weeks X 7 years = 280 consecutive profitable week. I believe it, when I see it.
Oh, I see. So, you found the Holy Grail of Trading. hehe. I think those stats are possible but highly unlikely to be continued if you SCALE it up. Things become harder to get in and out once you get to 8 and 9 figures asset. If you are good you can double $1M easily or even $10M. At $100M it gets a bit tougher. At $1B, it's almost a miracle. At $10B to $100B, you are God. hehe. This is over one year time frame. Duh, if you had bought the market in 1982 and sold on March 2000, then you could easily more than double, triple, 10x a $1B. But that's over a very long time period. So compounded annual return will still be around 20-25%. As Bill Eckhardt said, "The % of winners is the least important statistics of the strategy and perhaps even inversely related to performances." But it's human nature to go for high % methods. good luck dude! Glad to hear you had a good day today. Keep it up!