How long should a back test be? Is five year enough ? on EUR JPY

Jan 24, 2013 at 11:06
3,168 Views
15 Replies
Member Since Nov 06, 2011   28 posts
Jan 24, 2013 at 11:06

Hi All,

I'm back testing an EA and for five years back it gives a very decent results on the EURJPY. (+80% per year)
But the 6th year is the past (2006) the EA losses all of the money before it can get in the profitable years (as of 2007).

Does any of you experts can tell me of the over all EUR JPY market as 'evolved' and I should not longer expect year like 2006 (at least for a couple of year) ?

Overall how reliable are back tests for simple strategies on Moving averages.

And last but not least how long should a Front test be to give valuable information? 6 months? more / less ?

Thanks for your help !!

W.
It's by making mistakes that you learn !
biosko
forex_trader_32952
Member Since Mar 30, 2011   65 posts
Jan 24, 2013 at 11:23
1 year of backtest with 99% quality data.
1 year forward test.

Very old data is not relevant as you want to trade in future not in past. Backtest is good to know if your strategy survive and if you are good MM. But it doesnt mean that it will be working in future.
Member Since Nov 06, 2011   28 posts
Jan 24, 2013 at 11:37
Hi biosko,

Thanks a lot for the quick response! You mention 99% quality data.

I use the back testing function of MT5 (with metaquotes Demo account) how do I ensure I have good quality data ?

Thanks !
It's by making mistakes that you learn !
Member Since Oct 28, 2009   1427 posts
Jan 24, 2013 at 12:47
Hi,
I would disagree with biosko's answer. It depends very much on the EA and the frequency of trading. If your EA trades three times in one year then that's not a particularly good sample to look at. If it trades 100 times a day then one year is probably overkill.

So, the more statistical data you have the more confidence you should have in your system. Where you draw the line ( 30 trades, 100 trades etc. ) is completely up to you. However the more you assess the more you can believe in the system.

Don't disregard earlier years, there is no guarantee that the markets won't 'de-evolve' and knowing that your EA at least won't wipe the account if this happens is reassuring.

With regards to 99% tick data accuracy. It matters for some strategies, generally scalpers or any EA that is interested in tight price differences. 90% is a good starting point though. If it works with 90% then test it with 99% to be sure.

Then forward trade it.

Then compare your forward trading with a backtest over the same period.

Then live trade it on a small account and compare live trading with a backtest and a forward test over the same period.

Keep doing that, it will let you know if your broker's trading conditions are deteriorating.

Best regards Steve
11:15, restate my assumptions: 1. Mathematics is the language of nature. 2. Everything around us can be represented and understood through numbers. 3. If you graph these numbers, patterns emerge. Therefore: There are patterns everywhere in nature.
Member Since Nov 06, 2011   28 posts
Jan 24, 2013 at 13:09
Hi Steve,

Thanks a lot for the additional info. EA trades 2 to 3 times per day so I guess 5 years in the past is pretty relevant?
I was just surprised to see the EA perform so well for the past five years and in 2006 loss all of the initial deposit on EURJPY.

For tick data is that a setting of the back tester of MT5?
I did the back test with 'every tick' selected but I don't see where I can adjust the tick data from 90 to 99% ?

I have indeed started a forward test on the first of Jan and it's very positive so far. Only trouble I'm having is with the myFXbook EA for reporting on the MT5 Platform (other post I added to the forum)...

Thanks for all your help! Very precious to beginners like me!

Wip



It's by making mistakes that you learn !
Member Since Oct 28, 2009   1427 posts
Jan 24, 2013 at 13:33
Hi Wip,
I don't use MT5, so I can't help with that sorry.

MT4 is still the recognised industry standard and preferred platform for most MT traders. For MT4 you have to download the data from an independent source and then load it into MT4.

If it's coming up with 90%, then my guess is it's the same as MT4 and you only have the minute data with the prices in between being interpolated by a fractal algorithm. So basically if it's important to your EA to know what the actual prices are within a minute bar then you really need to get hold of and use tick data.

Best regards Steve
11:15, restate my assumptions: 1. Mathematics is the language of nature. 2. Everything around us can be represented and understood through numbers. 3. If you graph these numbers, patterns emerge. Therefore: There are patterns everywhere in nature.
Member Since Nov 06, 2011   28 posts
Jan 24, 2013 at 13:44

Can you point to a independent source where I could download tick data?

Thanks for all the info !
It's by making mistakes that you learn !
Member Since Oct 28, 2009   1427 posts
Jan 24, 2013 at 13:59
Start here

https://eareview.net/tick-data

Best regards Steve
11:15, restate my assumptions: 1. Mathematics is the language of nature. 2. Everything around us can be represented and understood through numbers. 3. If you graph these numbers, patterns emerge. Therefore: There are patterns everywhere in nature.
biosko
forex_trader_32952
Member Since Mar 30, 2011   65 posts
Jan 24, 2013 at 14:14
Well, just try it and run your system.
After 2 years in business with live trading you will see that old data is not so relevant.

About 90% vs 99% data. With 90% data you can miss some big short spikes. Maybe its not relevant for trader who place 1 trade per week and hold it for month. But its relevant for intraday trader as your TP and SL will be tighter.
Member Since Nov 06, 2011   28 posts
Jan 24, 2013 at 14:59
Noted thanks a lot guys! I'll do some more tests :D
It's by making mistakes that you learn !
Member Since Apr 14, 2011   72 posts
Jan 24, 2013 at 17:34
Old data is not relevant at all. It only aids in giving people false confidence and you are really inadvertently curve fitting your settings to past data. Obviously, horrible performance on historical data sucks but problem is people get overly confident on good performance on historical data. And you only end up curve fitting your system to past data. That is basically pointless. You cannot trash system sellers and EA vendors for curve fitting while actively curve fitting yourself.
Never ever ever give up!!!
Member Since Oct 28, 2009   1427 posts
Jan 25, 2013 at 09:49 (edited Jan 25, 2013 at 09:53)
Totally disagree with you here. If you optimise an EA over say three months then those three months could be totally unrepresentative of the larger sample.

I'm going to take a popular EA here to demonstrate.

If I analyse the live trades of Forex Growth Bot from 1st October 2012 to 1st January 2013 - a three month period involving a total of 115 trades you would think. Okay that's a reasonable period and a fair amount of trades to take a look at.

End result -5.95%

If I analyse the live trades of Forex Growth Bot from November 9th 2010 to February 9th 2011 - a three month period involving a total of 91 trades you might think. Okay, that's too long ago it's not going to tell me what the market is like today. And it doesn't, we get a whopping 538% ( although if I remember correctly it was trading high risk initially, something a lot of EA vendors do to make initial growth on a system ).

So, which is correct? Neither. What is correct is the 2 + years of live trading of FGB that shows a fairly straight equity curve rising. Only a reasonable statistical sample shows us this.

You can backtest FGB to 1999 and while the first three years the equity curve is practically flat it then goes on to describe a reasonably stable equity curve. This gives people the confidence to say 'Hey there is a good chance this EA will NEVER blow my account as it's traded over so many different market conditions. Come what may I think this EA will handle it'

Curve fitting is when somebody has the short sightedness to believe that three months of performance or whatever on current market conditions is what they should be trading on.
Best regards Steve
11:15, restate my assumptions: 1. Mathematics is the language of nature. 2. Everything around us can be represented and understood through numbers. 3. If you graph these numbers, patterns emerge. Therefore: There are patterns everywhere in nature.
biosko
forex_trader_32952
Member Since Mar 30, 2011   65 posts
Jan 25, 2013 at 14:09
It doesn't really matter what was in past. Only future matters.
Don't rely on bot. Use your brain. Modify your strategy with market. Use proper MM.
Member Since Oct 28, 2009   1427 posts
Jan 25, 2013 at 14:25
If what was in the past doesn't matter then by extension all indicators are pointless as they are based on past data. If all indicators are pointless then how do you trade? Please share.
11:15, restate my assumptions: 1. Mathematics is the language of nature. 2. Everything around us can be represented and understood through numbers. 3. If you graph these numbers, patterns emerge. Therefore: There are patterns everywhere in nature.
biosko
forex_trader_32952
Member Since Mar 30, 2011   65 posts
Jan 25, 2013 at 15:55
Fundamentals, sentimentals and proper MM.
Member Since Oct 28, 2009   1427 posts
Jan 25, 2013 at 16:44
We're talking about optimising an EA here, 99% of which are I would imagine are based on technical trading rather than fundamental trading. While I appreciate that as a fundamental trader history is not of too great a significance, for technical trading it's all that you can base systems upon.
Best regards Steve
11:15, restate my assumptions: 1. Mathematics is the language of nature. 2. Everything around us can be represented and understood through numbers. 3. If you graph these numbers, patterns emerge. Therefore: There are patterns everywhere in nature.
Sign In / Sign Up to comment
You must be connected to Myfxbook in order to leave a comment
*Commercial use and spam will not be tolerated, and may result in account termination.
Tip: Posting an image/youtube url will automatically embed it in your post!
Tip: Type the @ sign to auto complete a username participating in this discussion.