Monte Carlo Simulation of trades in backtesting a la Van Tharp

aarbee said:

...For example, what is the highest DD at 1, 10, 50, 90 percentile level etc...

Click to expand...

Which levels (1, 10, 50, 90, etc) do you apply in your decison making, and how?

bingk6 said:

... The ideal situation is for the high and low bounds to be quite close together across a large number of simulation runs, and they are all sufficiently profitable, then you can have added confidence in your system because it has shown its ability to perform regardless of the actual trades taken...

Click to expand...

I am trying to understand what sort of "confidence intervals" (or percentiles, or number of standard deviations, for example) others use for evaluating whether or not insample/out of sample data and MC sim results are "in agreement" or not.

So, what do you mean above by "quite close together"?

For anyone interested, here's an interesting paper on use of MC sims and its use in system design/analysis...

http://www.tradingblox.com/Files/MC_resampling_Nbars.pdf


For anyone who gets the chance to read through, do you have any feel for how general are the following conclusions from the above paper, or any other comments?

a) "Thus it is recommended to use ... 10 million resampled daily returns [for convergence]." (page 15)
... so if your actual curve has, say, 150 points then 67,000 MC simulations would be needed to give confidence of convergence?

b) "... as the portflio size is reduced, serial correlation in the equity curve is also reduced." (page 20)
... so, from the perspective of how serial correlation of returns affects MC results if not taken into account, the effect is less in any case when dealing with a single instrument rather than with a portfolio of instruments? -> MC sims for estimating Max DD etc are better on single instruments than on baskets? .... hmmmm but are instruments like ES or FDAX more like single instruments or portfolios from the serial correlations perspective???