|
PureBytes Links
Trading Reference Links
|
Well stated Gary, and I don't disagree that if your systems are primarily
reliant on moving averages or other indicators, that the potential error lag
could screw up the results for X bars instead of just one trade...
howsomever...
If one's logic is intolerant of gaps, then what's the difference between 4
additional gaps caused by rollover, vs. all the other gaps in a trading
year? Seems to me if the logic can't handle gaps then you're in for a rough
ride anyway. And working so hard to eliminate these 4 particular gaps/year
may not solve the "robustness" question entirely.
I have embraced gaps plenty of times in my logic, and so as you put it, the
extra few tend to get buried in the noise of a sufficiently large backtest.
In one case, it turned out to be far more critical to avoid one major
holiday overnight position in one year, than all the rollover gaps put
together over 4 years.
And obviously intraday only strategies are immune to this.
Of course, if one were running a hedge fund, then one's outlook could change
in terms of what the procedure should be at rollover time. I'm more speaking
to the individual here.
Best regards,
Gene Pope
----- Original Message -----
From: "Gary Fritz" <fritz@xxxxxxxx>
To: "Gene Pope" <gene@xxxxxxxxxxxxx>
Cc: <omega-list@xxxxxxxxxx>
Sent: Saturday, April 13, 2002 12:35 PM
Subject: Re: Why Continous contract software
> > In other words, *whatever* method one uses in live trading at
> > rollover is the same method that should be coded into TS. So if you
> > close out the day before rollover day, then your backtest code
> > should reflect that.
>
> For my style of trading anyway, the actual impact of rolling from one
> contract to another is practically in the noise. It's zero or one
> additional trade / slip / commission per quarter. Often it's only 0-
> 1 per year in my tests, since most of my systems are not always in
> the market. Counting it or not counting it would have almost no
> impact on the final results. I ignore it.
>
> Having smooth data (no jumps at rollover) throughout the test,
> however, is critical. If you ignore that, the results will be badly
> skewed and won't bear much resemblance to realtime trading.
>
> Gary
>
>
|