|
PureBytes Links
Trading Reference Links
|
Some performace measures to optimize here:
http://www.andreassteiner.net/performanceanalysis/index.html
http://iasg.pertrac2000.com/PerTrac2000Stataistics.htm
I use sortino ratio.
Dário
-----Original Message-----
From: Gooly [mailto:gooly@xxxxxx]
Sent: quarta-feira, 12 de Novembro de 2003 15:06
To: Omega forum posting
Subject: Equity curves
Hi List,
there were some posts about the evaluation of equity curves and one metioned
Chande with his relation of Final Equity divided by the Error of the linear
regression. Another post mentioned the sharp ratio, which is s.th. simular
devided by the non-risk-interest-rate.
I may be wrong and would be glad if s.o. points to my failor, but I think that
both aren't the best methods if you want to impove a system by optimizing
parameters.
In the specific case of optimizing ne system within one or more markets the
input of the (allways same!!) non-risk-interest-rate in the evaluation of the
outcome does not give you any additional information, it just changes the
number. Only if you compare different types of investment this sharp ratio
makes sense - no?
And Chande relation - if used to optimize - does not leed you to the best
system, I think. Here an example:
lets assume this equity curve: (1 trd = 1000 and 4 trds = 0, ..)
1000,1000,1000,1000,1000,2000,2000,2000,2000,2000,
3000,3000,3000,3000,3000,4000,4000,4000,4000,4000,
5000,5000,5000,5000,5000,6000,6000,6000,6000,6000,7000));
This leeds to this from the linear regression:
'Final Equity: 7000 Eqt/Err: 2406 slope: 197.58 intercept: 451.61
Error: 290.99 No.Trades: 31'
Chande's relation is 2406
now this one: (1 trd = 5000, 1 trd = -1000, 3 trd = 0, 1 trd = 6000, ...)
5000,4000,4000,4000,4000,10000,9000,9000,9000,9000,
15000,14000,14000,14000,14000,20000,19000,19000,19000,19000,
25000,24000,24000,24000,24000,30000,29000,29000,29000,29000,35000));
This much better result leeds (from lienar regression) to
Final Equity: 35000 Eqt/Err: 1956 slope: 987.9 intercept: 1483.87
Error: 1789.08 No.Trades: 31'
Chande's relation here is 1956, much less so you would reject this version.
I think if you optimize a system, then you try to find those parameters that
finds better the correct enter- and exits times and - I think - it happens
that what is shown in the above two equity curves, some trades are much
better, some a little worse and you try to get form curve-1 to curve-2. But
Chande "won't let you do that".
Additionally to that I myself found that not the systems with a maximized
Eqt/Err-relation performed better in different markets but those that where
optimized by maximizing the final Equity did.
Refutations appreciated, ...
carl
|