Page 1 of 1

Determining Fixed Fraction Percent using Monte Carlo

Posted: Sun Aug 01, 2010 2:11 pm
by sr100m
I am thinking about the process of evaluating a system - I don't have TB yet - I would be interested in whether it can do the following.

So here is a scenario with a question.

I design a system trading a number of instruments, each traded with a couple of methods one LTF, one countertrend. It has a positive expectancy.

I would like to trade it using a fixed fraction position size % that limits the drawdown to 30% and a MAR above 1.

For further confidence I would like to use monte carlo analysis and select the fixed fraction percent where 95% of drawdowns are under 30%.

Is it possible to ask TB to run this kind of combined stepped test and monte carlo analysis or would I have to run a monte carlo on each of the different fixed fraction steps and record the results of each manually?

Or is this overkill?

Posted: Sun Aug 01, 2010 3:32 pm
by sluggo
Max Drawdown is a smooth, nonlinear function of the fixed fractional risk per trade. See the red curve in (this) post for an example.

Precisely BECAUSE it is smooth, simple search strategies will quickly home in on a desired risk number which produces a given drawdown.

As an example, I pulled out an old trading system and ran Blox Monte Carlo analysis on it, with a HUGE portfolio, at seven different values of fixed fractional betsize. Then I plotted the results. Because the plot is smooth, it is not frightening to visually interpolate the data and say that fixed fractional risk of 0.190% produces a MC max drawdown of 30%, for this system, with this portfolio, over this span of historical prices, using this set of assumptions about slippage and commissions, etc

Now that I know the FF risk (0.190%) which satisfies a certain Monte Carlo Max Drawdown constraint (30%), for this system and this portfolio etc, I can begin to look at other performance stats when operating at this particular FF risk.

Posted: Mon Aug 02, 2010 4:55 am
by sr100m
Thank you sluggo.

88 seconds of my time seems a reasonable price to pay to generate the curve manually.