Backtest Memory Issue? | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Tick Data Backtest Issue - Please help! 4 replies

Visual backtest result different from non-visual backtest 0 replies

A Short Memory Is A Trader's Best Friend 7 replies

2009.03.11 15:41:15Tester memory handler: tester stopped because not enough memory 19 replies

Memory leak in MT4 3 replies

Backtest Memory Issue?

  • #1
  • First Post: Dec 15, 2017 5:13am Dec 15, 2017 5:13am
I'm backtesting quite a complex strategy using trueFX tick data for multiple years on every tick using MT4. Obviously, this is a lot of number crunching.
When the backtest starts it will do the first six months or so very quickly, then gradually slow down.
If I break the 8 years up into individual years and test each year on its own, it will do each year fairly quickly so it's not a problem with volume or data. if i extend the test to two years, it does the first bit fast then gradually slows down. Testing the whole lot starts fast then ends up at a glacial pace (imagine a 10 day backtest). I've removed all printing to the log and any history loops are limited in terms of the number of trades rather than viewing the whole trade history. I've split all functions that need to be processed by each tick from the functions that can be performed each bar or each week.

any tips to try other than breaking the test up into small tests?
has anyone overcome this problem themselves and how?

regards

M
  • #2
  • Dec 15, 2017 5:05pm Dec 15, 2017 5:05pm
I faced similar problem (when I use visual mode, don't know about yours).
If my test leave a lot of objects on chart, they will slow down the computer.
I delete all of them after each period, and it will smooth like the beginning.
Hope this can help you.
  • #3
  • Dec 15, 2017 8:33pm Dec 15, 2017 8:33pm
Quoting mlewis
I'm backtesting quite a complex strategy using trueFX tick data for multiple years on every tick using MT4. Obviously, this is a lot of number crunching. When the backtest starts it will do the first six months or so very quickly, then gradually slow down. If I break the 8 years up into individual years and test each year on its own, it will do each year fairly quickly so it's not a problem with volume or data. if i extend the test to two years, it does the first bit fast then gradually slows down. Testing the whole lot starts fast then ends up...
You could convert it to MQL5 and use its tester, which is much more advanced...
  • #4
  • Dec 15, 2017 8:40pm Dec 15, 2017 8:40pm
Don't waste your time on backtests!!

It's totally bullshit data, no matter how accurate you 'Think' your data is.

When you run a backtest for a report, you use the "Strategy Tester" tool in Meta4.

"STRATEGY TESTER"

This is meant to verify triggering and signals based on your logic.

NOT MEANT as an accurate tool as to how your logic would have faired over any time period.

Don't believe me?

I've offered up $1000 to anybody who can prove that a backtest will reasonably match a forward test for any length of time dependent on their timeframe testing.

IT'S NEVER BEEN CLAIMED!!
Money Can't Buy Happiness. Poverty Can't Buy SHIT! You Choose!
  • #5
  • Dec 16, 2017 12:50am Dec 16, 2017 12:50am
Thanks for your sharing, that mean forward test is the only way to test the logics of any ea?

Quoting Magix
Don't waste your time on backtests!! It's totally bullshit data, no matter how accurate you 'Think' your data is. When you run a backtest for a report, you use the "Strategy Tester" tool in Meta4. "STRATEGY TESTER" This is meant to verify triggering and signals based on your logic. NOT MEANT as an accurate tool as to how your logic would have faired over any time period. Don't believe me? I've offered up $1000 to anybody who can prove that a backtest will reasonably match a forward test for any length of time dependent on their timeframe testing....
  • #6
  • Dec 16, 2017 1:15am Dec 16, 2017 1:15am
Back testing is good for learning.. but lacks the context of whats going on that day..
For example you might do a back test during a rate decision, a presidential election, or a day when stocks are down 400.
Without the context of the day, makes it hard to get meaningful results.
I think its best to back test certain time of the day Like London open...
The only system that will work is one designed by and for yourself.
  • #7
  • Dec 16, 2017 3:00am Dec 16, 2017 3:00am
Quoting energetic
Thanks for your sharing, that mean forward test is the only way to test the logics of any ea? {quote}
Hey Energetic!!

Correct.

Logic is tested in a backtest - Strategy. This can show you if and when triggering in and out of trades takes place and does this match with what was intended.

Performance is front test only.
Money Can't Buy Happiness. Poverty Can't Buy SHIT! You Choose!
  • #8
  • Dec 17, 2017 12:41pm Dec 17, 2017 12:41pm
Quoting extraw
I faced similar problem (when I use visual mode, don't know about yours). If my test leave a lot of objects on chart, they will slow down the computer. I delete all of them after each period, and it will smooth like the beginning. Hope this can help you.
thanks! Yeah, i too delete all objects at the open of every week and it does indeed keep visual tests from slowing. I'm not visual testing though so objects are disabled I believe.

Quoting Magix
Don't waste your time on backtests!!
I can't exactly see Renaissance Technologies risking capital without backtesting. Backtesting provides a vital function in algorithmic trading, although I agree, many amateur traders simply curve-fit to historical data. This does not make backtesting useless though, it just makes curve-fitting useless and most amateurs traders bad at avoiding it.

Has anybody tried locking the strategy log folder to prevent logs getting really big and slowing things down?
  • #9
  • Dec 17, 2017 3:38pm Dec 17, 2017 3:38pm
Quoting Magix
Don't believe me? I've offered up $1000 to anybody who can prove that a backtest will reasonably match a forward test for any length of time dependent on their timeframe testing. IT'S NEVER BEEN CLAIMED!!
ummm is this still up for grabs? I'll take a swipe at that. You select the sample period for the backtest (make it statistically significant) and I'll post the backtest results and set a demo up and run it in real time from my server for the same duration of time. And to make it fair, I'll put the $1000 into a live account running the same algo and give it back to you when the live account hits 100% return.
  • #10
  • Dec 18, 2017 5:55pm Dec 18, 2017 5:55pm
Quoting mlewis
{quote} ummm is this still up for grabs? I'll take a swipe at that. You select the sample period for the backtest (make it statistically significant) and I'll post the backtest results and set a demo up and run it in real time from my server for the same duration of time. And to make it fair, I'll put the $1000 into a live account running the same algo and give it back to you when the live account hits 100% return.

Do I get a copy of the Source?
Money Can't Buy Happiness. Poverty Can't Buy SHIT! You Choose!
  • #11
  • Dec 19, 2017 6:01pm Dec 19, 2017 6:01pm
Quoting Magix
Do I get a copy of the Source?
no source I'm afraid.
  • #12
  • Dec 20, 2017 5:06am Dec 20, 2017 5:06am
Quoting mlewis
{quote} no source I'm afraid.
Then, no bananas.

Easy enough to code trades into EA to match forward tested results.

All the best to you this season!

Money Can't Buy Happiness. Poverty Can't Buy SHIT! You Choose!
  • #13
  • Dec 20, 2017 6:38am Dec 20, 2017 6:38am
Quoting Magix
Then, no bananas. Easy enough to code trades into EA to match forward tested results. All the best to you this season!
Lol. No problem I had a feeling there would be a snag.

I think you might be confused though. If I publish a backtest of historical data, then publish a link to a verified MyFxBook account at the same time (before the live forward test starts), I would need a time machine to falsify making them identical by hardcoding trades into the EA. Hell, you could even copy trade the forward test to your own demo if you wanted to make sure it was real.

If I run a forward test, I can easily falsify the backtest results to match the forward test, but that's not what we're chatting about here. if you know a way to do it the other way around then PM me and let's put something together!! I'll code it for sure! now that would be a merry Xmas!!

...but anyway, it doesn't' matter now. I wish you a great holiday season too!

M
  • #14
  • Dec 20, 2017 6:41am Dec 20, 2017 6:41am
Quoting mlewis
{quote} Lol. No problem I had a feeling there would be a snag. I think you might be confused though. If I publish a backtest of historical data, then publish a link to a verified MyFxBook account at the same time (before the live forward test starts), I would need a time machine to falsify making them identical by hardcoding trades into the EA. Hell, you could even copy trade the forward test to your own demo if you wanted to make sure it was real. If I run a forward test, I can easily falsify the backtest results to match the forward test,...
lol, so essentially he'll go ahead and take your EA off your hands now, and if it's good enough then maybe he'll pay you a grand for it.
  • #15
  • Dec 20, 2017 6:48am Dec 20, 2017 6:48am
Quoting Nicholishen
{quote} lol, so essentially he'll go ahead and take your EA off your hands now, and if it's good enough then maybe he'll pay you a grand for it.
I've probably given away better...

Link

Link

Link

Money Can't Buy Happiness. Poverty Can't Buy SHIT! You Choose!
  • #16
  • Dec 20, 2017 7:48am Dec 20, 2017 7:48am
wait... your backtest looks quite a lot like your live, forward (demo) test. in fact, the strategy will always look like that except for black swan events that risk running out of margin on the very rare occasions that you don't retrace enough to close the basket, which would then blow the account. Did you not just prove yourself wrong with your own EA? all you would have to do is turn the leverage down far enough so that the max drawdown for the past ten years if no more than 15% and you be reasonably confident that the system will survive almost anything AND that the backtests will look very similar to the live trading.

I guess you gotta pay yourself a grand now
  • #17
  • Dec 20, 2017 9:14am Dec 20, 2017 9:14am
Quoting mlewis
wait... your backtest looks quite a lot like your live, forward (demo) test.

They are 3 separate live running accounts.

I don't post backtests for the same reason I originally told you not to bother with them.

They are totally fucking worthless.

If you'd like, I will load these into their own TE's.
Money Can't Buy Happiness. Poverty Can't Buy SHIT! You Choose!
  • #18
  • Dec 20, 2017 9:42am Dec 20, 2017 9:42am
Quoting Magix
{quote} I've probably given away better... Link Link Link
...ex4's aren't source
  • #19
  • Dec 20, 2017 10:59am Dec 20, 2017 10:59am
Quoting Magix
If you'd like, I will load these into their own TE's.
please, don't bother I really don't care about this anymore. it's obvious you were just blowing hot air with that $1000 thing. Good luck with the volatility.

can we go back to talking about the MT4 back-tester memory thing again, please?
  • #20
  • Dec 20, 2017 11:34am Dec 20, 2017 11:34am
Quoting mlewis
{quote} please, don't bother I really don't care about this anymore. it's obvious you were just blowing hot air with that $1000 thing. Good luck with the volatility. can we go back to talking about the MT4 back-tester memory thing again, please?

Sigh...

If you had anything, you would have posted it already.

Since you don't I figured I'd save you a little time and not bother with bullshit backtests.

Best of luck with your backtests.

Quoting Nicholishen
{quote} ...ex4's aren't source
No Shit, Sherlock.
Money Can't Buy Happiness. Poverty Can't Buy SHIT! You Choose!
0 traders viewing now