Hi Rick
As far as I can tell the EA is performing exactly the same in live mode so far as it did in demo mode. I check the trades executed (both entry and exit) each day for quality control purposes and so far I have not picked up any differences/errors.
It is probably still too early to test drawdown limits. The backtesting results have a maximum drawdown of 10% and so far the current drawdown is only 5.7%. That having occurred in a 24 hour period is high compared to backtesting but its not the highest 1 day drawdown for the model.
As far as demo testing time is concerned I felt it should be separated into 2 phases:
- Bug-fixing phase, after back-testing to pick up any issues migrating from back-testing mode to demo mode. Once I no longer picked up any bugs by making any changes to the EA I moved to phase 2.
- Execution-testing phase. This is where I simply let it run without touching it and if it completes this phase my view is its ready to go live (but at reduced initial weighting as per a previous post). In phase 2 the time I allow depends on the trade frequency of the model. For this 1st model I wanted to run it for 1 month without any changes because the model averages 40-50 trades per month (x 2 orders so that's 80-100 orders) which I believe is a sufficient number to confirm that the EA operates correctly
As far as how much drawdown will I tolerate. This is my own opinion but I have set a policy that I will allow an EA to go as much as 2x the maximum drawdown from back-testing. In this case that means 20%. I also set my phase 1 live trading strategy to run at half-weighting to effectively I will only risk a maximum 10%. If it hits that level I will stop trading it live and put it on hold till I review why that happened. Basically I designed my model and the EA so that it fits to my tolerance risk. So I guess the answer about how much optimisation/backtesting/demotesting is how long does it take to produce an EA that fits your performance criteria.
The other approach I have taken is that the EA that goes into trading initially may at some point be improved. I think you can spend forever trying to design and build an EA with a perfect equity curve - my view is to spend the necessary time to build one that meets the desired performance criteria and start making some money, and while its making some money you can always invest more time building a better one (or a different one).
Cheers, Sharks
