Thanks for the feedback.
Of course, it will be fully tested for reliability in both a demo and live trading account using personal funds in the coming weeks before accepting any major amount of capital.
A more specific question: What measurements or statistics will help an individual or institution based on the back tested system to make a solid case for, say, $20,000 to run with after initial testing?
I personally use Profit Factor, weekly and monthly returns along with visual inspection to make sure the trades all enter and exit as expected.
What other measurements will help make a solid case?
As far as the strategy itself:
The big insight (for me anyway) was to diversify a portfolio of models per each instrument. That smooths out the equity curve very nicely.
So the system operates a variety of models including trend, channeling, and counter-trend models simultaneously per instrument and on multiple instruments.
Right now it uses Forex pairs.
It has all the necessities like adaptive bad tick filtering, auto restart, etc.
It was tested on 5 years of tick data using common forex pairs.
While I traded other mechanical systems live, this particular one has not gone live yet. The integration with MBTrading broker API is under way.
Still, since I trade both discretionary and automated for years, I know this model is the real deal.
The plan will be:
When the switch flips to live in a couple weeks, it will trade a number of trades on a demo account for testing.
Then it will trade using my personal money through a fair number of trades to work out any kinks with the order processing on the live system.
So it will still be a month or more from being fully live yet.
About the software itself: It's built using C# on Windows and was done entirely in-house except for the open source graphing component.
This system has taken a ferocious ( by that I mean absolutely dreadrul) amount of hard work and pain to implement. That's because it included:
1. Solving tick data performance issues to rapidly backtest. 2. Building object oriented framework for speedily making adjustment. 3. Adding total flexibility for multi time frames, per model, per indicator, per anything. 4. Automated adaptive tick filtering for bad ticks in real time. 5. Automated regressions testing the code using NUnit. 6. Creating the graph display for visual inspection of trades. 7. Designing smart genetic algorithm for quick solutions to multi-variate models. 8. Making flexible plug and play framework for connecting and multiplexing separate models onto each instruction. 9. Implementing trade statistics calculations for trade, daily, weekly, monthly, all with winner and losers separately, etc.
Last but by far not least was actually building, testing, scrapping, fixing, redoing, starting over, fine tuning, optimizing, avoiding curve fit and generally figuring out how to make the the models themselves make MONEY!
So now it has a very clean performing trend model for the back bone. It uses multi-timeframes for indicators, entry and exit.
That trend component feeds in to the various channel and counter trend models to trigger them.
Obviously, there's a limit to how much detail about the models themselves can be divulged here.
Please advise on my questions.
Sincerely,
Wayne
