Understanding Monte Carlo

[lookOutBelow]

Hello.

I've searched here in the forums and elsewhere on the net to try and determine what exactly NT is telling me with the Monte Carlo results. Even after reading the treads and various Google results, I'm still not clear at what I'm looking at and how the output can help me refine my strategies.

Can anyone shed some light for me? I've attached an example image from one of my strategies. The intersection of the plot and the 50% lines up with what the back testing results (roughly $58,300) are telling me over my 6mnth period. But, I'm not really sure what to make of the rest of the plot.

Does it mean that I have a 80% chance of landing in the $54,000 to $61,000 range given some randomness? I must be missing the point completely. How can I use this information to either improve the results or know more about my strategies weaknesses or strengths?

[NinjaTrader_Brett]

Hello,

Thank you for your forum post.

Please use the following guide for this as this is a very good written page on this on the help guide that should explain what monte carlo is and also how to read the chart.

http://www.ninjatrader.com/support/helpGuides/nt7/running_a_monte_carlo_simulati.htm

Let me know if I can be of further assistance.

[lookOutBelow]

Hi Brett.

Yeah, I actually read that earlier and was still not clear. However, after posting this thread, I found some good links on the internet that explained things much more clearly.

Here is a good read for anyone that has questions about Monte Carlo (He has a free pdf book available for download):

http://www.tradelabstrategies.com/

One thing that confused me about NT Monte Carlo is the Percentage at the bottom of the graph. Shouldn't that be labeled as percentile not percentage? Because, after all, the proper way to read that graph would be.. At 80 percentile, the simulation is telling you'll make X amount of money or less and 20% of the time you'll be making X amount or more.


Thanks,
Rick

[NinjaTrader_Brett]

Hello,

Yes they could be interpreted both ways. Means the same thing essentially.

Let me know if I can be of further assistance.
.

[Owlman]

I don't really get it either...

I've been looking at Monte Carlo simulations too, and I'm not sure if I'm reading it right. I think I get how they are generated, but not how to interpret the results. Using LookOutBelow's example, is it fair to say that he has a 90% chance of making more than about $33k (i.e. a 10% chance of making less than that) and a 2% chance of making $100k? And the 50/50 shot is at about $58k?

Also, I've noticed if I increase the number of simulations I get multiple lines on the graph. What do the multiple lines mean?

Thanks,
Ian

[NinjaTrader_Austin]

Ian, yes, you've interpreted the graph correctly.

The multiple lines are each different simulation.

[Owlman]

Thanks. I'm now using the Monte Carlo simulations to help refine my strategies, very useful!

Ian

[dave1992]

There's something really simple I'm missing about this.

As I understand it in Ninja, each MC run randomizes the order of trades, and by default, uses all the trades in each run.

So what changes the cumulative profit each time then? If I add up a load of trade P&L values I get the same cumulative total at the end regardless of in which order I add them up.

thanks
Dave

[NinjaTrader_JoshP]

Dave,

Monte Carlo grabs your trades and randomizes them. It then does this multiple times to generate multiple equity curves. Then it generates statistics off of the multiple equity curves. What is shown is the statistics of those numerous equity curves. As trades are randomized you do not necessarily end up with the same equity curve especially when you run the risk of being wiped out.

[dave1992]

Dave,

Monte Carlo grabs your trades and randomizes them. It then does this multiple times to generate multiple equity curves. Then it generates statistics off of the multiple equity curves.

What statistics?

What is shown is the statistics of those numerous equity curves. As trades are randomized you do not necessarily end up with the same equity curve especially when you run the risk of being wiped out.

Yes, but each equity curve starts and finishes at the same place, so how are some of the simulations able to end up at different cumulative profit - only talking about positive profits here - ignore wipe out for now.

[NinjaTrader_JoshP]

Dave,

The whole graph is the statistic. You choose what kind of statistic you want the graph to be showing you.

Monte Carlo grabs trades by random to add into its test. This is not to say it can only grab the trade once. The same trade theoretically can be grabbed twice, three times, etc. thus resulting in different probabilities and equity curves.

[dave1992]

Dave,

The whole graph is the statistic. You choose what kind of statistic you want the graph to be showing you.

Monte Carlo grabs trades by random to add into its test. This is not to say it can only grab the trade once. The same trade theoretically can be grabbed twice, three times, etc. thus resulting in different probabilities and equity curves.

So what's the algorithm for selection then? Select first trade at random from full set. Select second trade at random from full set, etc...

If so, I'm struggling to see any benefit of those statistics.

[NinjaTrader_JoshP]

Hi dave1992,

Yes, it is grabbed from the full set each time. This can result in some trades being grabbed multiple times and some not being grabbed at all.

Basically the benefit of this approach is that, with enough iterations, you are able to generate a "normal distribution". This distribution can then be used to give you an idea of ranges that your strategy could run into. The key here is to be running enough iterations. A single random result by itself is not meaningful, but when you generate lots of iterations you start to get a fuller picture as to possibilities of the strategy.

The reason Monte Carlo draws from the full set is to try to remove curve fitting issues and to try and better simulate a wider range of possibilities than a limited backtest could provide.