EA backtesting vs real life

New comment
 

So ive seen quite a few EA backtesting results on MQL5.com so far and some of the results look incredible for example look at this

However this persons performance in the championship was poor in 2012 .

 Can someone explain why there are so many good backtesting EA's but they dont necessarily work in live environments?

Ive seen so many other good backtesting results in the market as well...so im a bit confused as to the performance with live accounts.

I mean surely with back testing results like the one above, on average surely you can be a millionaire within a couple of years?

 

According to my experience - backtesting results can not be a proof that EA is profitable. Especially for now when we may see some complicated/complex EAs such as tick scalpers or EAs using close bar together with open bar (in indicators' icustom for example) and so on.

Besides, past performance is not guarantee of future performance.

Of course, there are some EAs which are performing same with backtesting. But market is changed all the time so if the strategy or EA is remaining the same one without improvements for 1 year for example so no any "good backtesting" will help with this situaion.

 
Surprising how not more people comment on this.
 
fractalfreak:
Surprising how not more people comment on this.

Probably because this question was discussed many times. I develop scalping EA and for this type of EA results in tester will very different from real account. 
However for EA that use open price method and works on H4 result will the same as tester. 

 
Some people have the skill and determination to alter the trade history to get good results on fake backtesting.  Shocked?  Don't be, money attracts the talent required.
 

Back testing will not be realistic for Scalping EAs

else back test vs live results might  be the same if not close for long term trading EAs

 

Could anybody give some PRECISE reasons why?


- tick data vs M1 data: can be solved i.e. with tickstory


- spread: can be defined


- slippage: if not manipulated, should cancel itself out


- indis: if not repainting, should show real signal


- commissions: many EAs allow to define these


So if you take a big sample of past data..........................what would be the difference be between the same EA which was running live during that time period vs the one looking back under the conditions described above?


Thx for precise answers ONLY.

 

One good reason is the delay in order execution.

You cannot get an immediate execution in the real environment like you have with the tester.

When your EA detects a condition for opening or closing an order, it needs some time to send a request to the server, which in turn requires additional time to process that order.  

This delay can take anywhere from a third of the second upwards, depending on the network latency and your broker's infrastructure. In that period, the price can be changed, and you will not get order executed at the price you requested.

On the other hand, the tester will execute the order immediately at the same price it used for the decision.

The tester will have perfect execution while delays in the live trading can lead to orders opened and closed at different times and prices.  

 
I believe there are many circumstances, obvious one is that during back testing, an EA will make a lot of Pips and enter many trades on BIG spiked candles while  on LIVE it might have happen that not even a single trade was able to go through during that spike. This is true for many scalping EAs
 
There have been many topics on the limitations of MT as a true "back testing" platform so I won't rehash them here. The general consensus amongst reliable coders is that it is great as a "logic tester" but any results as a "strategy tester" should be taken with a very large bucket of salt ;-)
 

Hi Drazen,


that would be slippage, right?

New comment
Reason: