Thank you. The sample sizes are certainly large enough to justify the performance numbers you posted. OTOH it looks a bit mystifying. There are only 252 trading days in a year, which translate to 504 trading days over two years. So 2394 trading signals implies either intraday trading or trading multiple instruments simultaneously. Given that most of your trades last a few days with some as long as a month, I assume you're trading multiple instruments. So far your system looks great. Good luck with your future performance.
Too good to be true??
- Thread starter Ucicelos
- Start date
Thank you. The sample sizes are certainly large enough to justify the performance numbers you posted. OTOH it looks a bit mystifying. There are only 252 trading days in a year, which translate to 504 trading days over two years. So 2394 trading signals implies either intraday trading or trading multiple instruments simultaneously. Given that most of your trades last a few days with some as long as a month, I assume you're trading multiple instruments. So far your system looks great. Good luck with your future performance.
'Sometimes you gotta say "What the Fuck", make your move. Joel, every now and then, saying "What the Fuck", brings freedom. Freedom brings opportunity, opportunity makes your future."
There are no gurus on ET. That's your first lesson. Most people here are losers.
The numbers are not too good to be true, but it may very well be that you do not have a robust system still.
Forward test it in simulator mode. If successful, trade it live with a very small amount of money risking very little money. If you start losing, stop trading.
Here's my personal opinion. The numbers are too good to be true. You've a good large sample size, based on that I would not think that the system is heavily curve fitted/optimized but I do think you've some issue in your coding logic. There's not that much edge in an automated strategy. - Again my personal opinion, may be worth less than 2 cents.
Make sure you don't "look ahead" in your backtest and also make sure you trade somewhat liquid underlyings so you have manageable slippage.
Here are some common mistakes that often mess up my system performance numbers:
If I trade futures I don't account for contract changes
If I trade stocks I don't account for dividends
If I trade stocks I use the open price - even on super liquid underlyings (> 1 mill shares/day traded) that's not working until you hit the > 50 mill shares/day category).
I get in on a level but use the wrong price (e.g. high or low of the day rather than the level price)
I mess up my logic by using ANY price input from the next day (or 5 min whatever your interval is) candle
I martingale and end up with 20 independent long or short positions that eventually start working but would give me a heart attack in real life and I'd have to risk the farm to hold them
I often explore an edge that just isn't there when I trade it for real.
-> check out http://logical-invest.com/. I love their approach and the fact that they publish each strategy but look at the strategy P&L graph where they backtest and where they start trading (live trading start). They've many systems that generate decent return in the backtest for many years and fail in real trading although they probably didn't make any backtesting mistakes. My point is, if you've a P&L graph that looks too good to be true, it probably is.
Last edited:
Thanks all, I'll take the dive and start with small trades and see how it goes. or as lawrence said..."make my move"
Thanks all, I'll take the dive and start with small trades and see how it goes. or as lawrence said..."make my move"
In simulator mode, of course. Then, small trades.
Are you really analyzing the trades and not equity? Equities generally move in correlation and if the losses happened in a row then that would be enough to wipe out your account. I'd also like to believe I could just analyze trade stats because life would be so much more easier that way but reality gets in the way.
Also, tripe check and then quadruple check for any and all kinds of peeking/using future data.
my apology, once again, if I drink don't post(but I was sure smart there for a few hours)
carry on
Thanks, I'll check the formulas to make sure that the tester is not peeking at the future data. You are right that I am looking at trade stats, the system picks from a list 1000 most liquid equities, but the system only risks 0.2% on any one equity at any point in time and max positions held over 2 year were 42, so limiting the max risk during the big move to 8.5% (assuming stop losses take me out). The data does has survivorship Bias, any other thing you think I should pay attention to?
