Tester in MT4 Build 200
Bars in test 3768
Ticks modelled 2462319
Modelling quality 90.00%
Initial deposit 100000.00
Total net profit 32137.61
Gross profit 52568.34
Gross loss -20430.73
Profit factor 2.57
Expected payoff 349.32
Absolute drawdown 0.00
Maximal drawdown 4829.65 (3.53%)
Relative drawdown 3.53% (4829.65)
Total trades 92
Short positions (won %) 57 (64.91%)
Long positions (won %) 35 (71.43%)
Profit trades (% of total) 62 (67.39%)
Loss trades (% of total) 30 (32.61%)
Largest
profit trade 1710.77
loss trade -1752.32
Average
profit trade 847.88
loss trade -681.02
Maximum
consecutive wins (profit in money) 28 (21222.89)
consecutive losses (loss in money) 6 (-4662.12)
Maximal
consecutive profit (count of wins) 21222.89 (28)
consecutive loss (count of losses) -4662.12 (6)
Average
consecutive wins 7
consecutive losses 3
Bars in test 12355
Ticks modelled 3949514
Modelling quality 90.00%
Initial deposit 100000.00
Total net profit 17777.00
Gross profit 65206.00
Gross loss -47429.00
Profit factor 1.37
Expected payoff 144.53
Absolute drawdown 0.00
Maximal drawdown 20877.50 (15.20%)
Relative drawdown 15.20% (20877.50)
Total trades 123
Short positions (won %) 53 (54.72%)
Long positions (won %) 70 (62.86%)
Profit trades (% of total) 73 (59.35%)
Loss trades (% of total) 50 (40.65%)
Largest
profit trade 1996.50
loss trade -3743.50
Average
profit trade 893.23
loss trade -948.58
Maximum
consecutive wins (profit in money) 15 (16254.00)
consecutive losses (loss in money) 12 (-16440.00)
Maximal
consecutive profit (count of wins) 16254.00 (15)
consecutive loss (count of losses) -16440.00 (12)
Average
consecutive wins 4
consecutive losses 3
Again, Alpari data under build 198 is the first posting. Build 200 is the second post. I'm not too happy in the differences.
I also realize that build 200 has modified how the bars are modelled or simulated. But this is a pretty big difference. Is the Alpari data good? Or is the Build 200 data good? Or is the problem with the way 200 models the candles? Can 200 be trusted?
I"ll let the readers draw their own conclusions. My conclusion is the results of the tester, any version. Forward testing is the only valid test.
Can anyone else run similar tests to confirm my findings?
I ran two tests. One with build 198 with and Alpari Data. The other with build 200.
Dates 2005/01/03 thru 2006/09/30.
GBP/USD
Both results with 90% quality. Results are posted below. Alpari data first.Which data is having build 200? Metaquotes demo (which is ibfx)?
If yes so it should be different.
Again, Alpari data under build 198 is the first posting. Build 200 is the second post. I'm not too happy in the differences.
I also realize that build 200 has modified how the bars are modelled or simulated. But this is a pretty big difference. Is the Alpari data good? Or is the Build 200 data good? Or is the problem with the way 200 models the candles? Can 200 be trusted?
I"ll let the readers draw their own conclusions. My conclusion is the results of the tester, any version. Forward testing is the only valid test.
Can anyone else run similar tests to confirm my findings?The way data is modelled was not changed in 200. You can find the list of changes here: http://www.metaquotes.net/news
Which data is having build 200? Metaquotes demo (which is ibfx)? If yes so it should be different.
200 is not using any partucular broker's data. It uses a filtered data gathered from different banks. I believe MetaQuotes data should be very reliable. I dont think they would put a low-quality data out for their users. If you want to know more about the data, have a look at the Russian section of the forum, I believe all the info you might need/ want should be there.
The way data is modelled was not changed in 200. You can find the list of changes here: http://www.metaquotes.net/newsQUOTE]
The way data is modelled / interpolated has changed according to point 4 of metaquotes news release which states
4. Tester: Imrpoved fractal modeling of a bar. Now more smoothed patterns are used to model price movements
As to the quality of the data, it should be easy enough to scan for missing bars, or do compare for differences with previous data. Its also probably worth backtesting strategies by hand at least once, and comparing this with the output from strategy tester. You get to see if the testers working as you'd expect it to , with the added bonus of looking at some charts 
regards
zup
Metatrader backtest behavior changed
Even while still using build 198, running backtests on before perfectly optimised settings, produced garbage this morning. (several EA's tested)
After importing history data again, which did not improve the situation, I decided to upgrade to build 200, but still no improvement.
Something strange is going on here, which I cannot explain.
Anyone else who recognises the same behavior on their machine?
Even while still using build 198, running backtests on before perfectly optimised settings, produced garbage this morning. (several EA's tested)
After importing history data again, which did not improve the situation, I decided to upgrade to build 200, but still no improvement.
Something strange is going on here, which I cannot explain.
Anyone else who recognises the same behavior on their machine?What kind of behaviour? :?
As far as I see it you only mentioned "perfectly optimized settings" (how do you define "perfectly optimised"?) and the "garbage" that was produced (what exactly does it mean?).
The way data is modelled was not changed in 200. You can find the list of changes here: http://www.metaquotes.net/newsQUOTE]
The way data is modelled / interpolated has changed according to point 4 of metaquotes news release which states
4. Tester: Imrpoved fractal modeling of a bar. Now more smoothed patterns are used to model price movements
As to the quality of the data, it should be easy enough to scan for missing bars, or do compare for differences with previous data. Its also probably worth backtesting strategies by hand at least once, and comparing this with the output from strategy tester. You get to see if the testers working as you'd expect it to , with the added bonus of looking at some charts
regards
zupIf you backtest EA using alpari data (every tick) and some other data (with 90% for example) so you will have diffrent results in most of cases. Bcause of the different data. Especially for aplari/nf/fibo group and ibfx. And the results may be completely different.
I install 200 version of MetaTrader now and this version is having possibility to downlod the data for backtesting. But I am still not sure: I am downloading M1 data for my broker (which I need), or from some banks or any (which I don't need)
Just for example: if you are using North Finance so it is better backtest your EA with North Finance data. And if the settings of some EA was optimized using Alpari data and you find it profitale by backtesting so this settings will not work with IBFX broker (Will work but losing money).
What is wrong with different data? Nothing wrong but sometimes EAs are having very different settings because of different broker's data. And sometimes even different systems because of that. And I know some systems which will not work well with IBFX data and will work for Alpari data much more longer.
So, probable it is different data.
What kind of behaviour? :?
As far as I see it you only mentioned "perfectly optimized settings" (how do you define "perfectly optimised"?) and the "garbage" that was produced (what exactly does it mean?).
Same EA, same settings, same period, same whatever...
I leave it up to you to decide which is garbage and which seems perfect to me
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
I ran two tests. One with build 198 with and Alpari Data. The other with build 200.
Dates 2005/01/03 thru 2006/09/30.
GBP/USD
Both results with 90% quality. Results are posted below. Alpari data first.