Topic: Ideas, maybe for FSB pro
I have been using FSB already for some time and as I said before I am very happy with it. I want to highlight some ideas to make it better (maybe for pro):
- Ability to store more strategies: I find it a pity that FSB only stores strategies better than the best strategy found at a given moment. We should be able to keep all strategies (or up to a certain number) that meet the criteria. This would be excellent so we could check how each behaves with new data, for example.
- Ability to choose other strategy sorting criteria: current options are very limited. I would like for example to sort by degree of linearity of the equity curve (this is different than imposing it as a criterium!), or by win/loss ratio (again, I know I can set this as a criterium, but I don't necessarily want to). I can even imagine a wonderful situation where, after clicking "stop" of the generator time is over, we would have different criteria with a slide bar on each and interactively restrict the resulting strategies to those meeting the a posteriori criteria. I think this would be really a very very good improvement.
- Backtesting improvements: I have data usually from 2002, more or less, but it is impossible to backtest so much data at once even on a 4h timeframe. I would like therefore to be able to choose different periods to make it faster (but to backtest at the same time these different periods!). For example, 2003-04, 2008-09, 2011-12. This way the backtesting would be much more reliable and strategies more robust.
- Choice of parameters: I don't know the details of the generator algorithm, but let me raise this point again (I've written before about it). In order to make the search more efficient, the range of parameters tested should be related to the number of digits set in the instrument parameters. For gold, for example, my broker uses 5 digits! I'm sure the program will not test the most suitable parameters in this circumstance. I believe it would not be difficult to correlate the range of parameters tested with the number of digits (for example, out of my mind: 10-100 for 2 digits, and 1000-100000 for 5 digits).