RE: Max Drawdown: Worthwhile statistic?

["Tom" <profilic@xxxxxxxx>]

Quite simple Paul. If you have a very large data sample and there is a very
very small part of it that is different, discard this part when designing
your system.

Of course, if you encounter those conditions in the future, you will get the
drawdown but to try to put in filters purely to get rif mof the drawdown
related to a small part of the data is pure curve fitting.

Tom

-----Original Message-----
From: Paul Altman [mailto:paulha@xxxxxxxxxxxxx]
Sent: Monday, May 08, 2000 11:53 AM
To: omega-list@xxxxxxxxxx
Subject: Max Drawdown: Worthwhile statistic?


I'm looking at a promising stop&reverse system that makes approx 50 trades
a year.  I can't stop myself from monkeying with the system to try to get
the MDD down, but when I really look at the equity curve, the worst DD's
occurred over a period of a few days.  Volatile markets, an unfortunate
sequence of a few days that whipsawed me, and the MDD is trashed forever.

I'm trying to understand how to think of MDD.  Obviously, you don't want it
to be too large, even with a great system, because you'll stop trading the
system at a certain point of pain, and end up with a large loss, only to
watch the theoretical equity shoot right back up while you're on the
sidelines.  But OTOH, you don't want to throw out a great system simply
because it had an unfortunate few days a few times over many years.  This
could happen to any system, and it seems foolish to throw out a great
system that got unlucky and go with an inferior system that has better MDD
stats.

IOW, I'm struggling with the fact that the MDD is a one-time event, so it's
really apples and oranges with the rest of the system's summary measures.

Any comments on how to think more rationally about MDD would be
appreciated, and perhaps what would be a more realistic and useful measure
of MDD.  I'm familiar with a number of volatility measures including the
Ulcer Index, but intuitively it's obvious that in addition to these summary
statistics you also want to know how bad things can get when they
occasionally get really bad.  MDD, as a measure of what's really a singular
system event, seems necessary but also very misleading, particularly if
it's caused be a very fast decline.

      Paul