|
PureBytes Links
Trading Reference Links
|
At 11:02 AM -0400 7/8/01, Bob Fulks wrote:
>It seems totally unrealistic to me since no one would scale bet sizes
>over a long period.
At 8:40 AM -0700 7/8/01, Phil Lane wrote:
>Probably missing the obvious, but why wouldn't someone scale bet size
>according to account value? Over the long or short term.
You tend to get things as in the attached GIF.
This shows, on a log scale, the account value backtesting a "decent"
system using "fixed fractional" with the bet-size set at "Optimal_F"
(the point that maximizes the "terminal wealth relative" as Ralph
Vince calls it). Not too bad looking on the surface.
The account value starts at $100,000 and ends at $50 billion. There
is this little drawdown in 7/97 where your account drops from $29
billion to $2.1 billion (losing about 93% of its value) but if we hang
in there at Optimal_F, we get it to $50 billion by the end of 1999.
No one I know would start with $100,000 and not withdraw any profits
or would stay with a system after a 93% drawdown. If backtesting is
supposed to simulate real trading, this is not realistic backtesting.
And I would have to conclude that any effort to optimize the
parameters of a system based upon any such equity curve is totally
fallacious. The performance of such a test is totally dominated by
what happens in last small period so the running TradeStation to
optimize anything would only be optimizing on the most recent market
characteristics.
Just my opinion, of course...
Bob Fulks
Attachment:
Description: "OptimalF.gif"
|