Backtesting and optimizer computing time question

[farang]

Hello,

I noticed that when testing on longer periods of time, NT takes EXPONENTIALLY longer to come back with the results.
So say, testing 1.5 years of data vs 6 months takes maybe 10x times longer, not 3 times.

All things otherwise equal: same strategy, same parameters, same machine, and on optimizer, same number&range of parameters sensitised.
The strateg(ies) I am testing (not in parallel) do not require long lookbacks, enter/exit within a few hours, and trigger regularly over the entire time frame (eg . it's not a case of having more trades/triggers at the beginning of the longer data sample).
The only thing I can think of is that it stores DataSeries, but again, the lookback for these is short.
I am running a decent machine (3.1 Ghz,8GB RAM, SSD) and not complaining about the computing time in itself, just surprised at the difference.

I am wondering whether this is expected? (and out of curiosity, why) If it is NOT expected, I'll go look for ways to "lighten up" my strategy script (suggestions welcome).

Thank you in advance

[NinjaTrader_Bertrand]

Hi farang, I would not really expect an exponential longer time here - what chart / dataseries and optimizer are you working on? Would you see the same if for example testing the same scenario and setup with the generic SampleMACrossOver strategy we ship with NT?

Also: for your DataSeries the lookback is 256 or Infinite?

Thanks,

[tradingGary]

Bertrand,

I have a somewhat similar issue, but I do have DataSeries lookback = Infinite, because some of my SMA's go a thousand bars back. (and assume that is why my backtests run longer)

Would it be possible to have lookback period = OneThousand ???

Thanks,
Gary

[NinjaTrader_Bertrand]

Hi Gary, it would be either infinite or the 256 setting, we've added a custom lookback setting here to our feedback lists though for future consideration. With the infinite setting you would have a higher memory footprint then, yes.

[farang]

Bertrand,
That's a good question- yes, the MA Sample crossover has exponential computing time as well.

That helped me identify the problem - the issue was starting the backtest at the very beginning of my data history, where I did not have a full day of ticks - somehow, it stumps NT7 for a while. Starting the backtest just one day after that solves it

Thanks!

PS: I was using a 256 lookback all along..