Monte Carlo sim

[laparker]

Just taking the Monte Carlo sim for a spin. Can you please explain what the x axis % means. The y axis I understand, but the meaning of the x axis has got me beat.

-thanks

[NinjaTrader_Chris]

The horizontal axis of the Monte Carlo Simulation is showing the percentage of simulations that have fallen below the y axis value. For Example, if you run a Monte Carlo simulation setting the "# of Simulations" to 200 and using the Cumulative Profit Graph, The intersection of the 50% x value and the associated y value means that 100 of your simulated trading sequences will be below that cumulative profit/loss value; and oppositely the remaining 100 will have a greater cumulative profit/loss.

[laparker]

ok - got it, thanks.

I am running into a problem where every time I run a simulation, the memory usage increases until it fills up and the system and is page swapping to a standstill. Closing the MC sim makes no difference, the memory remains committed to NT.

[NinjaTrader_JoshP]

Can you please try with SampleMACrossOver as the strategy and see if you can reproduce? Thank you.

[laparker]

I'm not sure if the issue should be connected with the strategy that produced the results table. I ran your suggestion and got the same problem. I was running on x64 and I then ran your strategy on x86 and it crashed NT with what looked like an out of memory problem. Also it looks like it is only using 1 CPU core.

edit - run with 100 trades and 10,000 sims.

[NinjaTrader_Dierk]

We'll look into the memory consumption.

The 1 core usage is expected.

[darthtrader]

Not to sound like a total moron but I can't seem to find how to access the monte carlo sim...
can anyone point me in the direction?

[laparker]

Strategy analyzer backtest results page
Trades tab
Right click
select Monte Carlo

[darthtrader]

Can you please try with SampleMACrossOver as the strategy and see if you can reproduce? Thank you.

well first thanks laparker for that, knew it had to be in there someplace but was driving myself nuts looking for it after this thread.
I'm also running into the same issue using SampleMAcross..Basically, it seems to work as long as the # of simulations, # of trades can be completed before
memory is used up..I wasn't able to reproduce it on daily BAC data though until I cranked it up to 100,000 simulations with 40 trades..Not that I have use to see 4 million samples for this though. I would rather not see a max number in there though but maybe just a general guideline to how many random samples you would nee before its overkill..

Since it was suggested to try SampleMAcross...what variables is the simulation taking in from a selected strategy to randomize? At the least I would think it would need to take in average hold time..the distribution for a strategy that has an average hold time of 5 minutes isn't going to be very useful if its including random week long swing trades.

I'm also a little confused on what the difference is between:
# simulations - 100
# trades - 90

vs

# simulations - 90
# trades - 100

does it just have to do with how each trade is seeded?

[NinjaTrader_Ray]

We are looking into the memory issue. The simulation is purely random.

[darthtrader]

The simulation is purely random.

Well also no offense Ray, but that means the results of the simulation are basically useless.
I mean hey, as a trader highly interested in MC methods, I've seen how its become a "buzz word" in retail space and its just good marketting..and good marketting is good for ninja so its good for me in the long term.
I'm glad its there as opposed to not being there, but I dont know..

[NinjaTrader_Ray]

Let me clarify.

- The simulation will randomly orders the trades
- The # of simulations determines how many simulations is done
- # trades will determine the total number of trades included from your backtest in the simulation. If the value of trades is less than than the total backtested, we will randomly kick out the difference

[tradejockey]

Well also no offense Ray, but that means the results of the simulation are basically useless.
I mean hey, as a trader highly interested in MC methods, I've seen how its become a "buzz word" in retail space and its just good marketting..and good marketting is good for ninja so its good for me in the long term.
I'm glad its there as opposed to not being there, but I dont know..

well actually its useful I hope.. my take is that if you can randomly throw data at it, curve fitting will not work.. also, removing 1 to 5% outlier winners/losers will show the robustness of the script.. i.e if it is relying on top few blockbuster trades for it to be successful...

[NinjaTrader_Dierk]

Right, exactly that is the value of a Monte Carlo sim...

[LarryS]

Strategy analyzer backtest results page
Trades tab
Right click
select Monte Carlo

This seems awkward and a bit obscure.

Can this be put on a separate tab labeled something obvious like Monte Carlo?