My first EA, interpreting results

[phil_trade]

So you're saying choose the best pairs..

But isn't that the entire reason why steve doesn't want us to post backtest results? Because you can't assume that the way a certain pair behaved in the past will happen in the future.. ?

So you're saying choose the best pairs.. Yes

Because you can't assume that the way a certain pair behaved in the past will happen in the future..
Things are not fixed ! You have to re backtest each quarter your EA to control pairs that are ok and pairs that are not.

And with time, your strategy could evolve, so you need to re backtest again (cf Darwin )

[aSushiRoll]

To answer your question: It uses CSS.

Results different timeframes, same method ( you can see the used formula in the M30 screenshot)

For some reason TF H4 works like crap.

I will keep you updated

[aSushiRoll]

So you're saying choose the best pairs..

But isn't that the entire reason why steve doesn't want us to post backtest results? Because you can't assume that the way a certain pair behaved in the past will happen in the future.. ?

So you're saying choose the best pairs.. Yes

Because you can't assume that the way a certain pair behaved in the past will happen in the future..
Things are not fixed ! You have to re backtest each quarter your EA to control pairs that are ok and pairs that are not.

And with time, your strategy could evolve, so you need to re backtest again (cf Darwin )

Ah thanks! That sounds like a strategy I should try. Yes.. evolution

[cheva282]

Hi Sushi, good luck with your new project!! I can offer you to help you doing 99% modelling quality backtests using dukascopy tick data with spreads and comission included as well as slippage simulation, so you can get a very good BT wich of course doesnt guarantee future perfomance but will definitively be 99% accurate for past years or months you want to test it, If you need this help just PM me, Im not a coder so this is the only way I can help you.

Backtest is good ONLY for testing your EA if it does what is supposed to do.

Regards

Esteban

[bodleytunes]

Hiya,

In response to c# api's

As far as I know, Forex Strategy Builder has a C# Indicator API and allows for decent back testing with imported tick data from the likes of Dukascopy. I've had a dabble with the GUI, but not got into coding any custom stuff with C# yet.

http://forexsb.com/wiki/download

Laters,

Bod.

[SteveHopwood]

So you're saying choose the best pairs..

But isn't that the entire reason why steve doesn't want us to post backtest results? Because you can't assume that the way a certain pair behaved in the past will happen in the future.. ?

Just stopping by here tonight. No idea what this thread is about, but just a few comments about backtesting in Empty4. I really must put all this into a doc that all can use.

Never mind. Here goes.

Backtesting in Empty4 does not work because:
t

• Empty4 only stores the high/low/open/close of each candle. It uses an algorithm to guess at at the 'ticks' in between. Pretty much destroys a scalper's results immediately Good luck when the scalper profits. Bad luck when it fails. Utterly irrelevant to what happens on your live account now.
• it uses the current spread, not that actually in place at the time of the trade.
• it uses the current swap, not that actually in place at the time of the trade.
• most ea's being backtested use indicators. Indi's store only the high/low/oen/close (or do they only actually store their close?) prices. What we see on our charts is merely a historic representation of the hloc of a particular candle, and an an ea can only access the hloc that the indi stores. Doesn't matter that CCI or RSI has screamed up and down all day like a demented yoyo on Acid - if your ea works on a tick by tick basis, then all it is working with is whatever hloc price the backtester choses to present you with.

Run out of steam now - late at night and tired. I really must make up a doc about why CraT4 backtesting does not work sometime. In the meantime, only an idiot takes any notice of BT results.

[aSushiRoll]

So you're saying choose the best pairs..

But isn't that the entire reason why steve doesn't want us to post backtest results? Because you can't assume that the way a certain pair behaved in the past will happen in the future.. ?

Just stopping by here tonight. No idea what this thread is about, but just a few comments about backtesting in Empty4. I really must put all this into a doc that all can use.

Never mind. Here goes.

Backtesting in Empty4 does not work because:
t

• Empty4 only stores the high/low/open/close of each candle. It uses an algorithm to guess at at the 'ticks' in between. Pretty much destroys a scalper's results immediately Good luck when the scalper profits. Bad luck when it fails. Utterly irrelevant to what happens on your live account now.
• it uses the current spread, not that actually in place at the time of the trade.
• it uses the current swap, not that actually in place at the time of the trade.
• most ea's being backtested use indicators. Indi's store only the high/low/oen/close (or do they only actually store their close?) prices. What we see on our charts is merely a historic representation of the hloc of a particular candle, and an an ea can only access the hloc that the indi stores. Doesn't matter that CCI or RSI has screamed up and down all day like a demented yoyo on Acid - if your ea works on a tick by tick basis, then all it is working with is whatever hloc price the backtester choses to present you with.

Run out of steam now - late at night and tired. I really must make up a doc about why CraT4 backtesting does not work sometime. In the meantime, only an idiot takes any notice of BT results.

Steve thanks for your answer. I know you're getting sick and tired talking about this subject.

In that case the results I posted are valid. My indi only uses the previous bars' close.

[phil_trade]

I dare a crime of lese-majesty, sorry Steve

M1 OHLC data could be ok if you backtest DAILY or more strategy ( may be H4 and H1) AND if and only if you respect a good test process :

1 - Choose backtest data from your LIVE broker if possible and check your data -> no hole, no curiosity...

2 - Backtest from largest period A

3 - Forward Test from a new quarter period B

4 - re backtest this new period B to check if your backtest result match your forward result

5 - Pray so that Steve is not too angry after that

cheers

Philippe

[garyfritz]

Phil is right about trading a high-TF strategy on a low-TF chart. The general rule is, if you can ever have more than one stop inside a single candle, all bets are off. As Steve said, Empty4 doesn't remember what happened inside the candle, so you can't tell which stop got hit first. But if your strategy basically runs on an H4 timeframe, so the stops are far apart from each other, then testing on an M1 or possibly even M5 or M15 timeframe should work fairly well. Maybe. Sometimes.

Remember though that Empty4 EAs can use any TF they fancy -- not just the TF of the chart. So you really don't know what the EA is doing inside unless you wrote it or studied it carefully.

If you're running a high-TF strategy that averages 10's or 100's of pips per trade, then MAYBE Empty4 can test it somewhat accurately. The small errors of spread, swap, etc will be minor issues compared to your trade size so those errors can probably be ignored. If Empty4 says the EA is profitable, there's a reasonable chance it might actually BE profitable. (But again that assumes your testing TF is much smaller than your stops, so you never have more than one stop in a single candle.)

But for small-trade scalpers, like Steve mentioned, Empty4 is a terrible testing platform. The errors and approximations that Empty4 makes will overwhelm the tiny profits that your strategy makes, and your strategy results will be "in the noise" instead of the errors. If a pip or three or 10 (per trade) makes the difference between profit and loss for your strategy, don't believe anything Empty4 tells you, and don't waste your time backtesting.

[Jeuro]

Hey guys,


How should I proceed from here?
1) I could just create an EA from this, which runs ALL symbols therefor averaging out the positives with the negatives, I should get a pretty straigth line up (I'll add calculation & chart tomorrow, I'm tired as it's 04:36 AM now)
2) I could see on which pairs it performs best and have the EA run on those. Except.. a backtest is in no way an indicator of how it will perform in the future on those pairs.... correct?
Sushi

Number 1 is the best first step on how to proceed.
We could code the whole thing to trigger in Jforex, avoiding the data/interpolation issues, plus
will work in all pairs simultaneously given you a very exact past max dd.

Regarding number 2.. you are correct, a BT, in most cases, is no way indication of future performance.
But in this case , if by running all the pairs or at least all the majors with "1 " setting, and producing a net profit every year tested, would be a very good indication that the strategy would work in the future (because of the diversification).

Worth the try.

J.