BeckTest EA Perfect, Real different.

New comment
 
Hi, this is my first contribution to this forum.

I back test this EA using pair G/U in TF 4H and the model is Control Points (base on the nearest.....).
The Result from 10.000 become 301.000.
I Real test since 2 days ago, I think the back is different from real.

my question is how to make this ea work the same as it works in back test.
I mean if back test using Control points (based on the nearest less timeframe with fractal interpolation of 12 control points) so this ea should use that to.


I'm new in programming, and new in trading too.

Here is the back test.
 
Test your EA in the model is Every Tick. Also read the article Strategy Tester: Modes of Modeling during Testing
 
I have found the back tester is often misleading in it's results, whichever mode is used. It must be realised that even "Every Tick" is not actually *every tick*, it is an approximation of some sort, I think based on sugnificant moves.

Someone suggested to me that using the history centre allows for more accurate testing, but the back tester is very useful during development to make sure that the code works as expected.
 
Correctly written expert works as expexted whichever mode is used. I promise you.
 
Hi Rosh,

With respect, I have written perfectly good experts that can be run in one location using Control Points, but regularly and repeatably give completely different results, using the same timeframes, properties, etc.

Same code, same platform, same settings.

Different results.

In my efforts to get the same results, we even ended up changing the local time of one of the computers (which were in physically different timezones), and setting the timezones of both systems to GMT.

Occasionally it has happened using Every Tick as well.

Perhaps this is expected to you, but it is certainly not to myself.
 
Thank's for the reply, I stil try this ea and will post the result later


Rosh wrote:
Test your EA in the model is Every Tick. Also read the article Strategy Tester: Modes of Modeling during Testing
 
If your EA must perfect trade actions (open/close/modify) on every tick, you must back test it (your EA) in mode Every Ticks.
In my opinion, this is obvious. For maximum reliability of the results it is desirable to have data on the smaller period (period 1 minutes).
 
When running the tester with recalculate on, it compiles the fxt file - which it will use for the backtest - using real time data. That is, if you run a recalculated data backtest with different real-time spread, the whole result will be different.
 
Backtesting a the later builds of metatrader has to be done in disconnected mode to be consistant, otherwise metatrader will download live data and try to 'blend' it into one file giving the different backtest results which everybody loves so much. So use alpari data in disconnected mode and run the 'every tick' model, but make sure you only trade at the start of each bar to avoid opening trades on the 'interpolated' data. Before you load the alpari data make sure you delete all prev. .fxt and .hst files. Even then, if you are doing small timeframe trading, your results will vary heavly on the data, it's best in this case to match data with the broker, problem is the data you download from most metatrader brokers is c**p! Catch 22!

It's all so simple!
Why metaquotes thinks it's a good idea to allow people to write experts which will trade off interpolated data is beyond me, how is this in any way helpful?
 
My problem is, I cannot be bothered to create an EA since it will not be accurate. I am led to believe that by creating an EA, it will never be accurate and so if you have time on your hand, you can forward test an EA. This to me im afraid is not worth it since EA needs to have worked from old data to new.

So Craig, is there a way of really having an EA that has accurate backtested trades?

So if I follow what you have said, it will work?

Like this link for instance: www.metatrader.info/node/67 If I Follow these steps i will be ok?
 
kingfx:
My problem is, I cannot be bothered to create an EA since it will not be accurate. I am led to believe that by creating an EA, it will never be accurate and so if you have time on your hand, you can forward test an EA. This to me im afraid is not worth it since EA needs to have worked from old data to new.

So Craig, is there a way of really having an EA that has accurate backtested trades?

So if I follow what you have said, it will work?

Like this link for instance: www.metatrader.info/node/67 If I Follow these steps i will be ok?
This link has nothing to do with data accuracy, only basic setup of the platform. Your work with the backtester will be a bit easier if you follow those steps, but not more accurate, since the tick data, that is used for backtesting will be simulated, based on the same M1 data, which is kind of default. You will need more accurate data (like historical ticks) to get more accurate results.
Note: M1 data is compressed data with a loss format, which means that unzipping is impossible, you cannot get the original price movement from an M1 bar, only 1 certain, and two relative points.

If you are new to MT4, and using M1 data is new for you, then the answer may be yes, it will seem more accurate to you, but it won't be THE accuracy you would expect, knowing that you likely will make money only on some percent of probability on your side, that is some cents of expectancy on your side. I mean, that your being profitable or not could depend on some trades, or some pips. I would be the most happy, if it wouldn't be this way, so someone please confute it! :D

You can read more about this issue here:
'Made $ 258 663.50 from $3000 using Problematic Backtesting by just buying at open (Fractal ZigZag)'
'Indicative vs "Real" quote data...'

About forward testing: IMHO, you need certain backtest, since with certain strategies may have problems that turn out after a year. You will need another year to test the solution after this (I've tried it myself. :) ), and so on.
Strategies that rely on expectancy (I guess most of the traditional ones) need a certain number of trades to be able to make the law of big numbers work, which again can take years.
12
New comment