Need help in backtesting

New comment
 
Hello everybody and happy new year.

I have a problem with backtesting. I designed an EA which was showing very good results. I noted down the results. The modelling quality was 25%. I needed to improve the modelling quality, but I coudn't, I downloaded the data, still was not able to improve the same. I also used period converter and properly converted 1 Minute data to all timeframes, I could see the .hst files visible in the history for all timeframes.

Someone suggested me to download alpari data, I did and was showing ridiculous results. the graph went down instead of up. !!!! That was not understood by me. The same ea showing different results at different times !!! I am in question what to beleive and what not to.

Now my problem is how to increase the modelling quality ???

Any help would be highly appreciated.
 
Well... I have actually the same problem. Hope somebody can help us.

Best regards and happy new year!
 
Set the start date of your test no earlier than the first full day of data included in your 1m history.
 

Set the start date of your test no earlier than the first full day of data included in your 1m history.


Hello sir,

Thank you for the reply, but it is giving the different results on different pc's the EA is same. Is backtesting that reliable ?? I am in dilemma now !! One Metatrader is from Interbank and the other pc has metatrader of Alpari. Please suggest something which will really convince me why the difference is !!

I have done following things on 1 pc:-

1. Installed Meta Trader
2. Downloaded history direct from Metatrader download button.

I have done following things on 2 pc:-
1. Installed Meta Trader Demo from Alpari
2. Downloaded Alpari Data
3. Converted to proper timeframes.

Please advise what is wrong !! The procedure given above is confirmed and done in a proper way.

Regards,
 
If you have different history data between the two PC's, the results will be different.

No two Dealers post exactly the same prices, there will be differrent defects in their histories, different holiday and weekend hours, and on and on.

Is backtesting that reliable?

There is no definitive answer to that. It is what it is, a simulation of trading actions taken based on calculations upon a non-repeating data stream.
 
If you have different history data between the two PC's, the results will be different.

No two Dealers post exactly the same prices, there will be differrent defects in their histories, different holiday and weekend hours, and on and on.

Is backtesting that reliable?

There is no definitive answer to that. It is what it is, a simulation of trading actions taken based on calculations upon a non-repeating data stream.





Dear Sir,

I can understand the above facts, they may be true, my dilemma is on which graph shall i frame my EA ? As my EA is based purely on mathamatical calculations and has got nothing to do with any of the default functions used in metatrader.

Please advise, this is my last question, I am very happy to see you help me.

Regards,
 
No idea.
 

No idea.




Sir,

Now about the modelling quality, I have set the max bars in tools, options to 9999999, but still the modelling quality is not getting increased. The data is downloaded from Metatrader, Can you give some information as to why is this happening ?

Regards,
 
Modeling quality:

I think you have set "every tick" in the tester for the "Model"

If it does not have 1 minute data for the entire period of the test to create "every tick", the modelling quality percentage is reduced.

As I said above:

"Set the start date of your test no earlier than the first full day of data included in your 1m history."
 
I feel backtesting is no use. Again this is my personal opinion. If you really want to test your EA run it for Forward testing for few weeks or months. See the results. You would be surprised how your 'BEST EA' would fail in forward testing.

So forward testing is the best test one can do on an EA.

Goodluck.
 
I feel backtesting is no use. Again this is my personal opinion. If you really want to test your EA run it for Forward testing for few weeks or months. See the results. You would be surprised how your 'BEST EA' would fail in forward testing.

So forward testing is the best test one can do on an EA.

Goodluck.




Hello Kent,

I am in this show for past several years and designed various EA's based on this garbage testing. As my vision matured I have come to following conclusions:-

1. Backtesting is useless. Even though the timeframe change and the holidays change the
logic of the bars should not change in my opinion. So there is garbage mixed up in the data.
2. As you have pointed out in your personal opinion, you are very much right that EA's should not be designed for backtesting, it should be forward tested and multiple variants should be made so you will save time.
3. And finally, do not design any EA based on the bar values or time or high or low, but design EA's on Averages or cumulatives or pivots or bar values before and not current etc. You will get assured results. If you can make EA based on mathamatical logic go for it. But when you see that your logic is perfect but the chart is going down, do not panic, put it to forward test, I am doing the same thing.

I think everyone should agree to this, lets hope so.

Thank you for your time and vision extended.

Great Regards,
New comment
Reason: