How long time to valid a trading system?

Hi gentlemen.

I’m still trading in a demo account and i’m thinking about starting a real one.

My question is: how do i know if my trading system work ?

I read a lot and an often it is said that " you have to be consistent with your demo on 3 or 4 mounth ".

But there is something that i dont understand: why are you talking about time ?

If a trading system makes me trade 10 times a day , in two weeks i’ll have more experience than a guy who trades 1 times a week during 3 mounth, right ?

Wouldn’t it be better to talk about how many trades instead of how long time ?

So here is another questions: in how many trades can you do statistics that make sense ?

Thanks You :slight_smile:

If you want ‘real’ statistics on a trading system, obviously the more the better.
I’d say that to verify any system’s performance with some reliability you need to look at a [B]minimum[/B] of 1,000 trades … 5,000 would be better, 10,000 ideal.
This is why people use backtesting, because obviously forward-testing even only 1,000 trades will take pretty long.

O.

I think they use time, particularly with respect to back testing, so you can have an idea of how your system performs in various market conditions. The market is always changing so while your system may perform well during your two weeks when the market conditions suit it, two weeks is not enough time to really know how it will perform over the long term.

But the other guy who trades for 3 months will have a better idea (still not enough imo). I have been back testing my system using 2007-2011 data which I think is a reasonable time frame given the market crash of 2008, the subsequent recovery and the euro troubles so it gives me an idea of a wide range of market conditions.

its definately a combination of both, i know a cop out right. Well its a number of trades and the length of backtest data. So you need a significant number of trades i would say over 100 to get some real data. 25-30 you start to get a rough idea but the more the better. With some position trading strategies they may only get 2-5 trades a year so getting 100 or more trades in your dataset is probably going to be unlikely. Now if you are say taking 30-50 trades a day you have a large data set but you may reach 1k trades in a month so you need to do it over a long duration. I personally use 10 years if i am starting to test a brand new system. 3-5 years of the most recent data to test any modifications. I believe you should also use an ‘outside data set’ which means that you have atleast 1 year of data you did not optimize your system on, it can be taken at random. then you test your system on that outside set first. Its a lot of work but it makes sure you didnt over optimize or just curve fit to the data. meaning your system will probably not robust.

MeiHua I thought about using a 10 year time frame for my backtests but I find data from the early 2000’s to be a bit sketchy, probably because the volume of trading in those days was nowhere near to the levels we see now so you get a lot of volatility and gaps.

I like your idea of a random test year will try this out. I also use the same system on various pairs to how it performs just simply as a benchmark. It is said that a really robust system works across different pairs so even if I do not trade those pairs live I still do a back test to check. But I normally end up having to make some changes simply because of the different nature of the pairs. GBP/JPY is a lot more volatile than EUR/USD so had to make adjustments because of that.

100 isn’t nearly enough to get reliable statistical data; and 25-30 is not statistics at all but merely a snapshot.

O.

true but like i said certain systems will never be able to get 1000+ trades in a data set in some peoples lifetimes. Its not like poker where you can be certain to get 100,000 + played hands and then make a determination. You have to find something that is suitable to go on, plus back testing results and future results always have skew. What is an acceptable data set to make a judgement, and how large of a confidence factor do you been before you go on to forward testing. Also it has been shown that the capital markets have infinite volatility as 9 sigma events occur more frequently then they should. So really the statistics gets hammered on heavy tails.

Yes, true … but then one shouldn’t expect anything even halfway reliable: 25, 50 or 100 trades, simulated or real, backward or forward, will never supply reliable data.

O.

so if over 1000 back tested trades wont be halfway reliable, and 100 live trades wont supply reliable data. Whats the point in gathering data at all? Cause if that was the case how can people develop any sort of system?

If you make 2-3 bad trades in a row, loses, and also drawdraw, then its trash. It dont take trade after trade after trade, to tell you, it dont work. I know, I know, He or she or him makes 10 trades a day, wins 1, and is in profit, is differant. OR is it?

I personally, dont need months and months of gathering data to say, It dont work, or it might work. And, Is it he system, or is it You, the traders, that isnt allowing it to work… There are alot of varibles…

I’ll get flamed im sure, but if its losing, its not going to get better. Time to find something else.

Dont waist months, when its 50/50 from the start.

Just my 2 dinars,

Of course having gathered data from 100 or so trades is better than nothing, but the OP’s question was ‘How long to [B]validate[/B] a Trading System’ … validation is different from having some clues regarding probable performance, which is all you may expect from 100 or 200 trades.

That’s what software like ‘ForexTester’ and similar tools have been developed for.
It’s no problem simulating thousands of trades with these applications, no matter the timeframe.

Cheers,
O.

LOL i use forex tester 2 also. hahahaha

Hehe, ok … are you satisfied with it?

O.

Ok, so Thank you for your answerds guys :slight_smile:

Are you meaning backtest in MT4? If it so, i think its not good idea, you can set optimal settings to get profit from past charts but you just cant optimize your strategy (EA) on future non-exists charts. I usually use some tools that help me to check my strategy, most of them you can fund here - [B][U]earnforex tools[/U][/B]

It’s like falling in love, you’ll know when it’s right LOL!