Discussion in 'Journals' started by frostengine, Jul 2, 2011.
Another way of checking if a strategy is over optimized is to change around the bar size. This strategy typically runs on 1 minute bars, so I decided to test using various other bar lengths. Surprisingly no bar length produced bad results.. In-fact I discovered almost all other bar lengths produced even better results. Average trade expectancy kept climbing with every extra minute I added.
Using 60 minute bars, ES produced $323 per trade expectancy... attached is the equity curve.
Either I have truly discovered the "Holy Grail" of price action capable of exploiting any market or found a bug in NinjaTrader capable of seeing into the future...
Guess we will learn soon
Do you use GPUs to crunch data at all?
Please ...get away from this "Holy Grail" notion...it seems that you have designed a sound strategy that should prove to be profitable in real time.
Trading is MATTER OVER MIND mind...."Holy Grails" exist in the latter...nothing to do with trading...period.
In relation to 6E slippage...given the time of day you are trading 1 pip is sufficient....however ,after 11.30 EST the market gets faily thin and if anything you may be tempted to run a back test with 2 pips as from 11.30 EST to EOD to see that you are still in your comfort zone.
ES...1 tick will always suffice....cannot speak for CL although it too may need reviewing after the market has gone into it's midday doldrums.
Looking forward to your successful results.
For another sanity check, I had it trade using Daily bars against every stock in the S&P 500... There was only 3 stocks in the S&P 500 that it did not make money on. Trading 100 share lots, it averaged +$52 per trade
PS: PF was over 6 for this test with just north of 20,000 trades taken
There was an issue in my S&P 500 daily test. It was still using 15 tick stops which got stopped out too much. I increased that stop to take into account longer time frame and average per trade went to +$127
Goes to $172 if expanded a little further..
Interesting stuff - appreciate your candidness.
Can you expand your exit executions a bit more? Are you using a target, time stop, or are the NNs triggering your exits as well as entries?
Interesting you ask. Since my focus today has been on looking at ways at improving my exits. My exits are based off a simple indicator. However, literally moments ago I tried using my NN in reverse for exits. Essentially creating an always in the market strategy and the results exploded.
I will continue my original strategy for a while as its production ready. This new development is not... but there is a chance my strategy may end up much better than what has been shown here.
Tomorrow AM the strategy goes live on ES, CL, and 6E. I am "tempted" to run this new variant on a low risk instrument such as NQ just to get some live data on it as well.
This will be a learning week for me. This live data collected will be very value to gauge how closely results match with backtesting. Could re-inforce or cause changes in how I view these strategies and the development process around them.
Just an observation, but using longer bar lengths would imply that you are increasing your hold times as well (since your execution granularity is a function of bar duration). Have you tried just increasing your hold time and while sticking to 1min bars? I'd guess you'd see your expectation go up (as well as your volatility of returns).
But regardless, your results appear to confirm that markets are indeed fractal in nature.
Separate names with a comma.