|
PureBytes Links
Trading Reference Links
|
Just another $0.02 on continual contracts ...
I've pulled some info from Genesis on how they do continual contract
(listed below).
First, I think it's useful to ask "why so many continual contract methods"?
Each one has a purpose, and every single one has limitations. It doesn't
make sense to talk about continual contract data separate from the context
of how it is being used. As a result, testing on continual contract data
can either be a fair approximation, or totally bogus.
If you use some type of filter like a moving average, you'll need a
continual contract type that avoids creating price discontinuities. In this
case, using a back-adjusted data series (like 067) may be appropriate. If
you use price patterns or support/resistance levels, you'll probably want
something that avoids distorting actual price, (like 057). Some methods are
so specialized they require unique types of continual contracts (see Gann
contracts, below)
The type of market also impacts construction and valid application. In
commodities, the AGS, GRAINS, and some SOFTS are seasonal by definition.
The contract following harvest is tracking next year's harvest. There is
often disparity between how adjacent contracts chart prices. In these
cases, testing on continual contracts is probably OK to get a general idea,
but not for evaluating any kind of results.
I believe it is paramount to remember continuous contracts are synthetic
data. By inference, all tests performed on synthetic data produce synthetic
results. My preference is to test as close to real market conditions as
possible, so my first priority is to use actual contract data. Of course,
without special tools this can be labor intensive. A compromise is to test
on continual contract, and spot check individual contracts. I see
differences all the time.
------
|