|
PureBytes Links
Trading Reference Links
|
Mark makes some good points. I think the most important one is....
> But you know yourself when you have applied a system to the
> next trading contract out "what happens" your trades are no longer where
> they were on the old contract, except for the most recent maybe and thats
> maybe.
Depending on the type of system, that can happen with any kind of any
kind of continuous data and is one of the biggest gotchas. If your
signals change, you are using the wrong data.
We are really talking about 2 things here. Mark is saying that he gets
the best signals from perpetual data and I believe him. The other side
of the coin is looking at the *results* of a backtest on perpetual data.
The signals might be perfect but, depending on the system, the p/l
numbers might be distorted.
Example buy and hold the S&P futures starting in 1982. Every quarter
roll to the front contract. I don't have them up on a chart but, as I
recall, perpetual data will show that you made several hundred big
points more than spread adjusted data. The spread adjusted data is much
closer to what really would have happened. Every quarter you would have
the "expense" of selling the cheap expiring contract and buying the more
expensive front month.
Another example....
myvar = factor/close;
This one would work better with perpetual data because the closing price
is much nearer to the actual historical front month.
myvar = factor/(close - close[20]);
This one might work better with spread adjusted data because (c - c[20])
is exactly the same as the historical front month (except at rollover
time). Perpetual data will always be slightly wrong for this calc. Maybe
that matters and maybe it doesn't.
Bottom line... understand your system (which Mark does very well) and
use the best data for your particular system.
--
Dennis
|