Re: Measuring System Streakiness

["Jim Piper" <jimpiper100@xxxxxxxxxxx>]

 Thanks for the idea. Unfortunately I got a correlation of almost exactly
 zero! (-0.03).

 The equity curve I have is almost a straight line with little drawdown
 anywhere . Against this, however, is a fairly modest performance (profit
 factor 1.6)

 I would like to develop a position sizing method to suit these
 characteristics.
 I hate to use the M word but a Martingale approach (with a limit on maximum
 position size) works very well with these historical trades in the order
 they're given.
 Even randomly re-ordering the trades with the same martingale system works
 well most of the time, but occasionally the drawdowns look ugly.

 In general, and for obvious reasons, I am wary of the Martingale approach,
 so I'd like to know that the reverting tendency of these trades has some
 statistical relevance.

 Jim


 ----- Original Message -----
 From: "Schindler Trading" <schindlertrading@xxxxxxxxxxx>
 To: "Jim Piper" <jimpiper100@xxxxxxxxxxx>; <omega-list@xxxxxxxxxx>
 Sent: Tuesday, May 14, 2002 5:53 PM
 Subject: Re: Measuring System Streakiness


 > One suggestion is to look at the auto-correlation -- the correlation of
 > subsequent trades.  For a "streaky" system, the auto-correlation will be
 > positive, for a random system the auto-correlation will be zero.
 >
 > If you have Excel, just take your list of profits and losses, then copy
 and
 > paste it to the right and up one cell, then use the CORREL() function (if
 I
 > recall correctly) on the table.
 >
 > My system has a modest negative auto correlation -- a big win, for
 example,
 > tends to be followed by a loss.  It isn't very prominent, though, and not
 a
 > strong enough signal to cause me not to take the trade following a big
 win.
 >
 > Let me know what you find.
 >
 > Aaron Schindler
 >
 >
 > ----- Original Message -----
 > From: "Jim Piper" <jimpiper100@xxxxxxxxxxx>
 > To: <omega-list@xxxxxxxxxx>
 > Sent: Tuesday, May 14, 2002 10:04 AM
 > Subject: Measuring System Streakiness
 >
 >
 > > Anyone know of a statistical technique to measure the 'streakiness' of
a
 > > system?
 > >
 > > I would like to find out whether or not my system exhibits greater
 > > streakiness than the same system with a randomly generated order of
 > trades.
 > >
 > > While I could calculate mean streak lengths and standard deviations for
 > the
 > > real system, for example, I'm not sure how to go about doing the same
 > thing
 > > for a randomly generated model, or even if it's possible to do so.
 > >
 > > Any help appreciated
 > >
 > > Jim
 > >
 > >
 > >
 >