Ok, so I've started my new system. On the backtest it came up with AR 202% PF 2.17 SR 1.45 and UI 5.8. This looks good but I think I may have over oprimized it just a little. I think it should still make a profit.

I have some more questions. I haven't managed to work out how to get the script to retrain itself periodically so what should I do, just manually retrain it once a month? Should I retrain more or less often?

When you are trading a system and you press the train button a new instance of zorro opens up. When you press train on that second version does that rewrite the optimisation file and the version that is trading live reads the new settings in the file?

Also I have a question about multiple instances of zorro on the same computer. I know fxcm don't allow more than one access of the API per computer. Does this still apply if you are trading on a different account? My ultimate aim is to have 5 or 6 different systems running and keep the good ones and ditch the bad ones. I don't really want to have 5 or 6 different computers to run all my systems. At least with metatrader you can have as many instances as you like on the same pc.

Retraining must be explicitly allowed in the script; you can see an example in Workshop 5. The retrain period is normally identical to the WFO test period. The trading Zorro instance automatically reloads the parameter and factor files whenever they are edited or retrained.

For the multiple instances issue of the FXCM order2go interface, the account number does not matter. We have indeed 10 laptops here nailed to a board for running several different systems simultaneously. However, the multiple instances problem is apparently related with the FXCM relogin function. When you set g->nWeekend to 2 so that Zorro logs off every weekend, it seems that multiple instances can run on the same PC.

So I can't just hit the train button every couple of weeks and get the same result?

Also when it comes to optimising would it be better to find parameters that give a reasonable return for all pairs traded rather than to use different parameters for each pair?

I have not so much trading experience to give a general answer - it also depends a lot on the strategy.

Training too often has the disadvantage that the parameters change too often, and that can have an adverse effect when series suddenly change and open trades are continued with changed parameters. You can emulate this by using a very short WFO test period - then you can see how this affects your system.

As to the second question, different pairs normally require different strategy parameters. I'd say that it is indeed an indicator of robustness of a strategy when it returns profits in a WFO test with parameters generated for all pairs together. But I have not tried this so far.

I feel inclined to try and use the same parameters on each pair. I have retested my system with sliders to change the parameters instead of optimising them and tested the portfolio.

I can't get a sharpe ratio or ulcer index that I like but it shows a return of 250% a year. Realistically I don't think that it will return that much but I just have a feeling that the parameters used and the system itself will be more robust. Even if it doesn't make 250% I think it has a good chance of making between 20 and 100% which is a return I would be very happy with.