I realize this thread has gone stale, however I've been working on developing my idea to live framework, and the three webinars and Taking a Trading System Live thread that kevinkdog has put together are heads and shoulders above the other posts I've seen on how to develop a robust system to test your system--and then sound ideas about issues surrounding taking the system live.
Here are the relevant links:
1. Webinar and Thread: Developing a Rule Based Trading Algorithm (
2. Webinar and Thread: Designing an Algo Trading System From A to Z - code included (
3. Webinar and Thread: Five Ways to Incorrectly Build a System (
4. Thread: Taking a Trading System Live (
Following Kevin's systematic approach to system development has definitely put me on what I perceive to be the right track for me. I have a long way to go.
In addition to Kevin's approach, the other issues that I have come up against are:
1. Data: Insuring that you have the best historical data that you can get your hands on to test against. If you can't check this box, do not bother with going to box 2.
2. Testing Software: After you are satisfied with your data, then one must have an in depth understanding of the software testing tools. For me, and most nexusfi.com (formerly BMT) users, that is NT. Therefore, one must understand how every line in the Strategy Analyzer works, why it is there and how changing any one parameter will impact the quality of the test. If you can't check this box, do not bother going to box 3.
3. Optimizer Types: A corollary to the testing software is having an in depth understanding of how the testing software optimizes your test results. For NT7, this means understanding what the optimization assumptions are for the "Optimize on..." parameter within the Strategy Analyzer. If I had to guess, most folks likely just select "Max. Net Profit" and then run the test, however understanding custom fitness types is pretty important. Again, if you can't check this box, do not bother going to box 4.
4. Strategy Coding: Next up is strategy coding, which is an ever evolving art and science as you progress up the curve. First, steal good code--always. Doing so will help you get up the curve quicker. Second, learn how to add a secondary data series to manage back test order exits. If you can't check this box...you get the picture.
After you have gained a modicum of confidence in boxes 1 - 4 and then reviewed Kevin's algo development approach to build your own algo development approach, then you are probably in a good position to go live.
Clearly there are other boxes that need to be checked to--especially trading psychology. But the above should help you on the mechanical algo front.
Hopefully my comments will spur some of the more experienced algo developers to chime in, but we'll see. These are just my thoughts on what I've learned over the past year.