MetaTrader strategy tester produces different results with the same expert advisor

[pva]

When run three times in a row the same strategy provides different results in total net profit:
1st run: 18.56,
2nd: 12.78,
3rd: 0.72.

The difference between the first and last rounds is 25 times.

The currency pair, time period and expert advisor settings remain the same (I just clicked “Start” button continuously). The expert advisor have not gotten recompiled either.

The data was obtained from Alpari New Zealand Limited broker.

MetaTrader version was 4.00, build 225.

There were no unmatched data errors.

Please tell me what can cause such a difference in backtesting passes.

Thank you in advance.

[Paul]

Is the computer isolated from the internet if not the data might be changing between runs.

[JC]

PVA:

[...] Please tell me what can cause such a difference in backtesting passes.

There's a number of possibilities including, as Ruptor says, changes in the historic data. Other possibilities include the following:

• Your EA is very, very sensitive to changes in the spread. MT4's historic data does not include spread; the strategy tester uses whatever the latest live value is when a test is started. Therefore, the simulated spread can change between backtests, and this can potentially affect EA results.
• Ditto for tick value, if the quote currency of the symbol you are trading is different to your deposit currency. However, this is a much less common issue.
• Your EA is very, very sensitive to minor changes in price. MT4 simulates the ticks within a bar; it doesn't hold any data more granular than M1, and simulates the ticks within each minute. I assume (I may be wrong) that there's a random element to this. If your EA is very sensitive to price changes then different random simulations could potentially produce very different results.

Not an exhaustive list. Other people will doubtless have things to add.