|
PureBytes Links
Trading Reference Links
|
Schedlbauer, Joseph E. <JoeS@xxxxxxxxxxx> wrote:
>
> Would it not be true that the MC simulation, due to the high number
>of trade sequences it generates, would come up with many sequences of
>trades which contain unrealistically long sequences of losers, and
>therefore overstate the system's max drawdown?
There would be unrealisticaly long sequences of losers. There would
also be unrealistically long sequences of winners. There would also be
unrealistically evenly distributed sequences of winners and losers.
There would be every conceivable sequence of trades if the MC were
allowed to iterate for long enough. Would this overstate the max dd?
Moot point - if you believe that one sequence of trades is as likely as
any other sequence of trades then you can use MC to establish the
probability of a dd of a certain magnitude. In this view, the sequence
of trades that resulted from the application of a trading system to a
data series is just one discrete sequence of trades, no more
significant than any other.
It's interesting to note that after a sufficient number of trials a MC
will cover every conceivable arrangement of winners and losers,
including all winning trades consecutively, followed by all losing
trades consecutively and vice versa - the mother of all dd's! Not that
this bizarre possibility detracts no more or less than any other
sequence from the usefulness of MC.
> Wouldn't it be better to look at the system's
> losing percentage and average losing trade and use statistics to
>estimate the probability of a particular level of drawdown?
What "statistics"? In the process of averaging, you are losing
information.
IMHO cumulative frequency distribution analysis is the proper method of
establishing the probability of a draw of a certain size.
Elliot
>
>
> -----Original Message-----
> From: alex@xxxxxxxxxxxxxx [mailto:alex@xxxxxxxxxxxxxx]
> Sent: Monday, July 15, 2002 12:18 PM
> To: elliotdeane@xxxxxxxxx
> Cc: omega-list@xxxxxxxxxx
> Subject: Re: Limited life span of mechanical systems?
>
>
> Elliot wrote:
> >I would be interested to know what you consider the
> >"Proper" method for employing MC analysis.
>
> You didn't address this to me, but I tried to answer it yesterday.
> I'll try again. The answer has nothing to do with equity curve
> filtering; that's a whole 'nuther topic, one which I don't think is
> all that useful to pursue.
>
> >In my view, any analysis that ignores the time-serial
> >nature of trading system returns is of limited
> >utility.
>
> Limited utility is better than none. As Dennis said, it's another
> arrow in your quiver.
>
> In my view, Monte Carlo simulation is useful for getting an idea of
> the best and worst your strategy can throw at you, given that you
> have a statistically significant collection of trades that represent
> the performance of your strategy.
>
> I use it for evaluating position sizing strategies. What happens
> when I risk 4.5% of my equity on every trade? The original sequence
> of trades says I'll earn 30%/year on my equity, with an 18% max
> drawdown. That bit of information is well and good, but it doesn't
> give me the whole story.
>
> So I do a MC simulation. (I'm making these numbers up.) Now I see
> that for several hundred different sequences of the same trades, the
> average earnings is 32%/year with an average of a 29% drawdown seen.
> The smallest max drawdown any trial had was 17%, the largest was
> 65%, with a standard deviation of 8%. I defined "ruin" as losing
> 60%, and 5% of the simulations ended up in ruin.
>
> This analysis tells me whether the original sequence of trades is
> representative of what I can expect if a similar set of trades
> having a similar expectation occurred in another sequence.
>
> What I do now is adjust the position sizing strategy to maximize my
> MC return/drawdown ratio while not allowing the average max drawdown
> to exceed some threshold of pain like 25%.
>
> Once I do that, then I go back to the original sequence with the new
> optimum position sizing strategy, and apply it.
>
> Often, I find if I do this analysis on the original sequence ONLY,
> it turns out not to be conservative enough in a MC simulation. The
> MC simulation gives me confidence in the strategy's performance no
> matter what order the trades come in. You just don't have that
> confidence analyzing the original sequence only.
>
> --
> ,|___ Alex Matulich -- alex@xxxxxxxxxxxxxx
> // +__> Director of Research and Development
> // \
> // __) Unicorn Research Corporation -- http://unicorn.us.com
>
|