Is back-testing EA's fair enough?

Neelesh

Recruit
Messages
4
Please guide me. I don't know whether to believe this EA or not. Attached the back-testing pdf report.

I have performed a back-testing one of the EA's and was evaluating in the following manner.

1. Started the EA on a demo account on June 6th 2010. Yes very recently.
2. On 15th June, 2010, I started the back-testing from 14th April, 2010 to till date.

I assumed that the trades the EA has performed on the demo account should match up while in the back-testing. The first trade in the demo account was on 8th June. I was surprised to see little differentiation at the open price, timings between the demo live and back-testing. All of sudden they started matching up, the timing and the price. But one of the trade that never occurred on the demo-live was present in the back-testing. The last trade. I don't know the reason, but the back-testing just stopped here. So is it fair enough to evaluate a EA just based on its back-testing. I am attaching the back-testing report, which displays the full settings. Sorry, I need to remove the EA name as I got it from one of my friend.
 

Attachments

  • 14Apr_14June.pdf
    53.7 KB · Views: 45
Evaluation

Hello,

You should get better backtesting datas, and test your EA for a longer period to really see what would be the result.

Among 2079 bars in test, you've got 655 charts errors, and no information about the modeling quality. So I won't pay much attention to this graph.

Here is a great article about how you can get great modeling quality and datas for a long period:
http://tinyurl.com/29xzcsd


Then you'll be able to test the EA for 2 year period ;)
 
Backtesting usually won't perfectly match a forward test during the same time period for a couple of reasons. First, most backtest datasets don't have every tic. Depending on how sensitive the indicators the EA is using, that could seriously alter entries and exits. Second, backtests don't have slippage on opening and closing.

As Gargamel said, get some more data first. If your backtest lines up 90% with forward testing, then it could be a semi-reliable way to help you fine tune the EA.
 
Thanks guys for the replies. I will try to modify the things and will test again with the EA. The EA am testing is not developed by me but my friend has bought it and he is not using it.
 
I don't believe backtests. First of all tester doesn't count commissions.
In live trading tehere is a lot of things that can bring down the EA's profitability. For example slippage, requote, connection errors, etc.
And the most important is you two month test is nothing.
I've made an EA which makes 1000% profit in 6 months in backtest.
But in real trading it is not profitable because of the slippages, and requotes.
I suggest use forward test.
 
No matter what your conclusions are about an EA/Strategy, you need to forward test it.

Put it in a micro account, and set it to 1 micro.

Some strategies I developed for example, caught breakouts. Looked great in backtesting, but when it was triggering the entry, the market was screaming in my direction. In live trading I could not catch the entries.

Just one example.

So no matter what, trust nothing until it has:

(1). Proven itself on a demo account for a significant amount of closed trades.

(2). Proven itself in a live micro account with one micro for a significant amount of trades.

(3). Slowly, gently begin trusting it more.

-Trader 5of7 of TheCollectiveFX-
 
The main purpose of backtesting is to do a general fine tuning and see if the thing is worth the effort of forward testing (starting on demo).
 
I don't believe backtests. First of all tester doesn't count commissions.
In live trading tehere is a lot of things that can bring down the EA's profitability. For example slippage, requote, connection errors, etc.
And the most important is you two month test is nothing.
I've made an EA which makes 1000% profit in 6 months in backtest.
But in real trading it is not profitable because of the slippages, and requotes.
I suggest use forward test.

Just to be a little more precise: bactesting calculates commissions if You run the backtest with an active account which has commissions. At least it does for me. Also, backtesting includes swap fees too, however, it will not be explicitly present in the statement (it's just included int the profit of every affected trade).
By the way, backtesting cannot simulate many other things already mentioned. It just gives You a general idea of how the trading strategy would look like. But make sure that you have the most perfect data for testing (90% modelling quality, 0 chart errors).

Regards,
MikKi
 
Back
Top