My personal preference is to have out of sample data set be on the same symbol as the in sample data set, but with randomly chosen segments for each. That removes my concern for correlation between symbols, and places multiple instances of different market conditions into each set. Since the elements of each set are randomly selected (without replacement), a large number of runs can be made and statistics measuring the consistency of results can help detect unintended curve fitting. I am somewhat of a fanatic about methods to avoid curve fitting when developing/testing strategy ideas using back testing. I am interested in watching your results here forward for several reasons. First I wish you great success. Second, it will give me some insight into whether or not I am overly aggressive at randomizing the test data sets to avoid unintended curve fitting.
Well...this looks like it is going to be a good thread to follow...what you have offered is very interesting and supported by "Jazzguy" understanding of NN. Only one question...do you have a completely redundant system (desktop/laptop).? 2nd question... ...are you going to used 4.15 EST as EOD for all participating derivatives...? NiN
I know this system is automated but could you trade it without automation, would you recognise a potential trade without the automatic signal , just wondered?
You're simply underwriting index tail risk, so why even concern yourself with correlation? You don't have a system per se; you're offering insurance via index options. Avoid curve-fitting? Like your paper performance in 2008 in which you selectively avoided the entire year?
The system makes all of the trades automatically. All I do is turn it on in the AM. After 2:00 central it stops taking new trades and closes out any remaining open before market close.
yes 4.15 EST will be considered EOD. However, typically the system will be out of the trade by 3:45 EST. My trading machine is hosted at my office which is next door to a datacenter. I have backup power here as well as multiple fiber connections feeding the datacenter. As a fallback, every position has a worse case stop placed, and in the event I lose all connectivity I will use my phone to go flat.
Forgot to ask you about your historical data: Are you back testing with 10yrs of intraday tick data, 1min bars, or ? Also, are you trading RTH (9:30AM to 4:00PM EST) or do you include the pre-/post-market periods in your historical backtests?
Frosty, glad to see you back. Looking forward to this thread. Good luck, make the rest of us algo traders get jealous with these results
The strategy trades based on 1min bars. The strategy is restricted to opening new positions only between 9:35 to 3:00 EST