I think Foracy has some excellent points here. In my experience I've seen that very high profit/drawdown ratios often fall over flat as soon as you try them with out sample data. It looks too much like over-optimization or curve fitting, the results ...

I haven't run this real time so cannot comment on results. This system with optimisated parameters can generate some nice looking equity curves, but not as nice as the ones that Mikkom has been showing (I'm running single strategy / single currency ...

Sure. Without giving away too much, here is a much simplified explanation of how my system works. I'll only talk about entry criteria here. Take a group of entry criteria. For each entry criteria, give it a value (i.e. an optimisable parameter, let ...

An interesting view on this in/out sample discussion. It makes sense if you are optimising something like indicator parameters where the entry/exit criteria are essentially stable throughout the optimisation. Things get a little more complicated ...

This all makes sense to me. As part of my fitness algo "pondering" my plan was to simply make the verification part of the system. Hence, while the system does not train using the verification data, it does use it to either accept or reject a set of ...

out-sample data??? — I think this raises some interesting questions, I've been thinking a bit about this recently and the problems associated with using out sample data for verification. See if you can follow my reasoning... Continuing on from ...

That sounds sensible. I didn't realize that you were displaying a log scale on your equity curve; that makes these results even more impressive. I guess another way of looking at it is by considering the equation: Fitness = Good_Stuff / Bad_Stuff ...

Fitness Algos... — Hi Mikkom, Congratulations on generating such nice out-sample results! I would be very interested to hear you expand on your fitness algo. I've been considering a few options here; profit / drawdown is the simplest fitness ...

Agreed, the optimisation window size is paramount. Are you currently using the above technique?

It's definitely true; with 20 day sample size on H4 you only manage to capture a few market twists and turns at best, and it is really these market features that a strategy's success (or failure) can be measured. I have seen some good results when ...

I'm using 20-60 trades in a month. as a starting point. I use the H4 timeframe, I've found anything smaller than H4 and it's hard to capture the market "bigger picture" that has the best chance of persisting past the in-sample period, and ...

Thanks for the description. Do you then manually analyse the results in excel and then transfer these to your MT4 EA for live trading, or is this part also automated?

The gap between in-sample and out-sample data — Hello All, I'm currently developing my trading system that is heavily optimisation-based (i.e. it optimises on recent data to determine trade strategy & parameters), and was wondering if anyone ...

Sorry, a mistake in my last post; I mean't that the processing will all be done in my external application (not the EA) to speed it all up. I'm using dynamic optimisation but having said that it doesn't need to happen that often (daily would be ...

Yes, this is definitely another option. However, writing an optimisation engine will need to incorporate a backtester and trading logic, and if I've written that there is little advantage in having two sets of trading logic (i.e. one in the EA and ...

Thanks once again for the response. My strategies are very lean in terms of processing effort and I use relatively short optimisation periods, leading to around one optimisation pass per second (or maybe a few seconds max). However, I do a lot of ...

Thanks Adal, that sounds like a convenient way to reduce code complexity. I guess I could take it a step further by throwing away the open and close values, assuming the sl & tp settings are far enough apart such that they will essentially never be ...

Thanks, it took me a while to find it but that's where I eventually got it from! That makes perfect sense. It has the advantage that you can design your own optimisation algorithm, as the MT4 one can sometimes be limited. I hadn't thought of this, ...

Adal, Do you use the Highs and Lows during the 1 min intervals for modelling or do you just use the 1 min close? I'm interested in this as I'm also using 1 min data and I can greatly reduce memory requirements in my trade engine by holding less data ...

Thanks for the response asasa. I looked and looked for this documentation and eventually found it within the MT4 help application itself; Thanks for the tip! I noticed that the exported optimisation reports do not include the settings for the ...