i have 27000 bars of test data which is 1 min bars . a strategy makes 1.5 pf in around 200 trades . drawdown is not too bad and profit is good. it includes some parameters which was trained on this data. when tried on a out of sample data of 50000 one min bars consisting of the same equity it makes 1.37 pf. do you think that is a viable strategy ? strategy is live for 3 days and it is not bad.
Is 1.5 calculated on a per trade basis or on some given period of time (i.e. daily performance)? If it's a per trade basis on 1min bars after comms/slippage, that's pretty good imo. Seems like 200 trades on 27000 bars isn't a particularly high amount, and perhaps more importantly 27000 bars doesn't cover a long period of time, especially not for something traded 24hrs. I'd be cautious about assuming this will continue to work for an extended period of time. m2c. Still, if you're trading it live and making money, then that's really all that matters.
1.5 PF is calculated on the whole 27000 bar time . this is a daily traded instrument . system closes positions at the end of the day. the test sample contains trend up-down and choppy times. although it is a trend following kind system it shows more profit on choppy days. i know 3 months is a short time to test and 5 months for out of sample data. But that is the all data i have. i am using small part of the portfolio so that i can assume that is a walk forward testing. i use it live 3 days and today is the forth. it is behaving as expected from the test data. the test data shows quite a steady profit . i am ready to accept it will not be very long lived but which system is? i remember to read something about any PF lower than 2 is rubbish. that is why i am worried.
IMHO, it can work just fine if you monitor it and in parallel (daily) test to see if all the trades behaved as they should... as to the PF... I have a short strategy that generates PF3 and a long that generates F1.6... they are both executed at the same time, so they balance each other... basically, make sure you have a decent distribution of long/short in your portfolio... and monitor it.. and if it works for you, it works for you... that's it...
Just purchase the historical data you need to do a proper job. Using 8 months is just not enough. These guys have very decent historical data: http://www.tickdata.com/
roger that all your rigorous testing, walkforward, limited or no curve fit means nothing if your underlying data covers a too short period in time. I consider anything building a strategy less than 5 years as risky. 8+ years is actually better. FWIW, I have 3+ years of system development experience - just wrote about my personal experience so that you can decide how much weight you may want to put on my opinion.
Depending on the previous post, too, as a strategy developer for 10 years, there must be at least 1 year of minute data to use for the in-sample parametrization and 2 years invariably to draw any conclusions about long term profitability. You might optimize on the most recent 1 year, then on the 2 year, otherwise more data the better, especially with 1 minute data need a lot more than 2 years, 3-5 and more the better. If it were tick data 2 years would be fine, and 3-5 would be a golden backtest probably. Your pf has no "consistency" if the short and longs are not substantially similar to each other in W:L. Find some more data.
My opinion only please respect; one min data = noise pf 1.5 in 200 trades = noise live for 3 days = noise/luck Take the same strategy and test it on another equity in the same sector. What do you get?
It is an opinion but not logical. Live over 60 days no more noise luck, agree there, then 200 trades is not attributable to luck. 1 minute data is noise but the strategy can still work. Mostly need more time to shake out. You can apply individually optimized strategies to other symbols but it has to be parametrized for that market meaning the strategy must be fitted to the market for its size and not simply changing data to other symbols. It can be the same strat just for that market with newly optimized parameters from the same theory. If you meant to do this anyway fine but I don't see any explanation about the need for that process.
PF isn't an issue, 1.5 is perfectly fine. But 200 trades and 8 months isn't enough. Just monitor it and collect more data or preferably buy more - you need at least 4-5 years worth.