Topic: Question About Monte Carlo Testing

I understand there are different things you can randomize when performing Monte Carlo testing.  The simplest one would be to randomize the order of the transactions, looking for potential large draw downs.  In this case the balance curves would look different but, in the end, they would all come together and have the same final balance (following the cumulative rule of addition).

When I perform a Monte Carlo test and only select "Randomize History Data" then I expect all the balance curves to end with the same final value -- but they don't.  Why is that?   I guess I'm assuming "Randomize History Data" is the same as randomize transactions -- but perhaps it's not.


Re: Question About Monte Carlo Testing

"Randomize History Data" changes the prices of random bars.
The number of  the changed bars is determined by the "Count of changed bars %" setting.
It changes the O, H, L, or C price with a random value between 0 and the value set in "Data range change ATR %"

The purpose of the test is to detect potential over-optimization.

3 (edited by Irmantas 2017-12-17 19:44:37)

Re: Question About Monte Carlo Testing

Hi Sleytus,
There is no this future of mixing trades order to get max possible drawdawn, but this future would be very appreciated. Probably some strategies would be turned off by seeing more realistic drawdawns(deeper and longer), at least risk would be lowered. If you are looking for this MC test, there is some excel sheet to make this trade shuffling. … oling-you/

How mr Popov said, "Randomize History Data" is like a noise test, you add some variation to data bars to see by how much your strategy is affected by some data change. If results changes drastically to the bad, probably there is over fit, because future definitely will be not the same like backtest, and this added noise just simulated that. I made experiment some time ago with about 40 strategies on unseen data chunk, found that this and variable mixing to be effective way to filter some crap strategies. End results of filtered portfolio changed a little bit to the better.

Re: Question About Monte Carlo Testing

Ah -- I see.  Excellent answers.  Popov, Irmantas -- thank you both for making this more clear to me.

It used to be I didn't pay much attention to the Monte Carlo tests.  But as I gain more experience I've come to appreciate its value.

Irmantas -- I did try the Excel spead sheet you referred to.  But after loading my data and pressing 'Calculate' then it crashed Excel -- or Excel stopped responding.  I used about 900 data points.