You seem to have multiple down days in a row. Have you checked to see if the daily results are correlated? ie If yesterday you made money are you more likely to make money today or not. Maybe only trade when you would have made money yesterday. See what that does to results/equity curve. Just an idea to try to reduce drawdown peroids.
Bot made +$50..... backtesting after the day ended revealed the bot should have made +$100.... what happened is i exited a trade prematurely because I got scared going into the Fed meeting.. and figured it was best to close out the current trade... I have got to learn DO NOT LOOK at the computer the whole day while the bot is trading
Did you reserve a portion of your backtest data for out of sample testing? How many hardcoded numerical parameters does your system contain? Do we know for sure that your system is not overfitted on your backtest data? This point was raised in a post a couple of pages back, but I don't remember seeing it addressed; apologies if it was.
Todays fed move was pretty tame. I would not want an ats running during some of those moves. I dont see how any system could cope with extreme swings.
waxwing, Let me briefly describe how this system came about. First several months ago I was developing an automated system for trading the ES. I started a post here about it, you can probably find it in the Automated trading section somewhere. I did not like the draw downs as they were even more severe than what I currently face on the ER2. However I took that same basic strategy and modified it slightly to arrive at the strategy I now trade. All the modifications I did was only on 1 quarter of data not the full 1.5 years worth of data. Thinking about this modified strategy I am now running backtests on the ES using both the originally developed ES strategy and the modified for ER2 strategy... Some time later tonight I will provide some backtesting charts of both the modified version that I fit for ER2 and the original ES version all on the ES to see how they stack up on the ES... This should also show if and how robust the strategy is by showing how it does on a different instrument. Plus I will show 5 years worth of backtesting on the ES.... I also at some point need to backtest my original ES strategy on the captured ER2 data and see how the original performs to make sure the changed I made for the ER2 is a better strategy or not than how it was originally months ago....
you would be saying the exact opposite if the fed meeting blew up your account. better safe than sorry. i understand interfering with the bot invalidates the results obtained by backtesting. but I think its counter productive to automatically assume any input you have to give to your bot will have a negative effect, even if the total profit at the end of the day is reduced somewhat. I often exit positions early when I feel its right, if I end up with less profit b/c of it, I just look at it as paying for piece of mind. my 2c on the subject.
Thanks. I don't understand all the details, but if you're confident you haven't overfitted and you've used a good chunk of out of sample then there's nothing to worry about there.
out of sample data? I dont believe this system is adaptive or incorporates any kind of ML. out sample data does'nt not apply to this situation if i understand what you mean by it. with ML systems you would have a training set and a sample set. with a static system you can assign all your data to the sample set since training is not needed. correct me if i'm wrong.
I couldn't disagree more with walter. In my opinion one of the nastiest habits of all is early profit taking. If you think that big news events introduce unacceptable volatility, then introduce a systematic rule to get out (or not get in) at those times. Unfortunately you have to find a way to backtest that, which I think is tricky. Much better is to just leave it alone and stop tinkering! I'm currently running mechanical (but not automated) systems and I have exactly the same issues. I was developing this same habit - pulling out because I didn't want the thing to turn on me. Now I'm being ruthless with myself not to allow this.
I don't know what you mean by ML (your reference to training data reminds me of neural nets, AI?), but if you backtest on all your historical data and then make changes to your parameters, you are in severe danger of optimizing your system on the historical data, giving the possibility of great results on the backtest which will not be matched at all in live trading.