|
PureBytes Links
Trading Reference Links
|
A Monte Carlo simulation lets you examine the dependence of a system's
performance on the sequence in which the trades are ordered.
I have found a Monte Carlo simulation useful to evaluate such things as the
distribution of drawdowns over a lookback period, or a given number of
trades. This assumes the trades are independent. So with a mechanical
system with a ten year track record of say 50 trades, I could randomly
sample the sum of ten trades and examine the resulting distribution to get a
feel for the system's annual performance.
Paul Barnes
-----Original Message-----
From: Bob Fulks [mailto:bfulks@xxxxxxxxxxxx]
Sent: Wednesday, 5 August 1998 10:48
To: cvinton@xxxxxxxxxxxxxx
Cc: L_Omega
Subject: Re: System Evaluation
Monte Carlo simulations are typically used to find the probability
distribution of some dependent variable which is a function of several
inputs, all of which have their individual probability distributions.
It is not real clear to me how you would use this technique to evaluate a
trading system. The uncertainty in a trading system's future performance is
mostly related to sequences of future prices, not to the distribution
functions of prices.
Am I missing something?
Bob
At 4:26 PM -0400 8/4/98, Cab Vinton wrote:
>For a while I've been interested in using a Monte Carlo approach with a
>random system in order to benchmark systems.
>
>I'm now beginning to question the usefulness of that approach, however.
>
>Let's say you discover after 10,000 simulations that your system is in
>the top 20 percentile (according to net profit, % winners, or whatever).
>
>But what good is this information? As far as I can tell, in using this
>approach, the better your system, the more curve-fit it is.
>
>Am I missing something here?
>
>Is there a theoretically sound way to use this methodology to judge the
>quality of a trading system?
>
|