|
PureBytes Links
Trading Reference Links
|
Alberto Torchio wrote:
>
> Ken,
>
> > Are you saying that if all testing is done on back adjusted continous
> > contracts, when trading the real deal you should then only take entry /
> > exit signals from back adjusted CC's? This is preferable to taking
> > signals from individual contracts, since individual contracts were not
> > used in the back testing?
>
> The rationale behind backward adjusting continous contracts is that there should then be
> no difference between single contracts and the continous. Or at least no major
> difference, providing you consider commissions and slippage for roll-over when
> backtesting with continous.
> I hope that if I'm wrong someone explains (again) the whole matter.
>
> Alberto Torchio
> Torino, Italy
I must first admit to being a newbie but these are my thoughts:
I believe cont. ctcts. properly take into account the $ P/L you
will experience in a trade that you actually carried into a later
contract (after adj for comm/slip.), *for a given date span*. But I
believe there would be some differences in certain cases in the exact
positions or dates held, between the two methods.
Suppose you are trading a momentum system. There will be cases
where the former contract will have a steeper slope in the price than
the following contract for the same dates. So there might be a buy
signal in the former contract and not the latter. The cont. contract
will get you in just as you would have if you were watching the earlier
contract. What about at rollover? You currently have a position in the
former. If you were just switching "real" contracts, when you graphed
the latter contract, you would say, "hey i don't have a signal here" and
maybe not roll. With the cc you would keep the position altho likely
your indicators would indicate a weakening trend. Depending how you
systems functions you might or might not get out.
There could be a similar situation where you would get a signal in
the cc after rollover that you would not get in the individual ones.
Suppose you buy if volatility decreases. Many times, I think, the
former contract has somewhat higher volatility than later ones. So in a
borderline case, your system might get excited just after the rollover
thinking that vol. had narrowed, while actually part of that decrease
was the contract shift. It might put you in a position that you would
not otherwise have. This would be a "systematic" enough error that i
think i would want to trade off of the cc's for a system like this if
that is how i tested it. (by "systematic" i mean tending to be in the
same direction - here more likely to take a pos. in the cc than the
individual ones). It appear in this case that testing and trading the
same way is important and that using the individual contracts might be
better.
An example where cc's might be better: suppose you trade a very
longterm relative momentum (price change / daily range). You would not
get accurate readings in the early parts of certain contracts that have
little activity. (I'm thinking of something like 5 yr notes that seems
to have negligible ranges before it becomes the front month- if i
remember right). Perhaps here testing and trading cc's would be best.
For some systems there may be differences in cc vs. individual
contracts, but you may conclude that the differences are random. In
this case i myself would at least consider trading on the individual
contracts even if i had tested on the cc's, in the hopes of identifying
the bullishness or bearishness of that particular contract month in this
particular situation. But i may need to rethink that.
I think the conclusion is that the different behavior will be limited
to the initial days after the roll, equal to the length of time your
indicators look back, but that such differences could be more common
than expected. It seems that some thought about the particular
system/indicators one is using might suggest a preference to one method
or the other. Of importance is whether these differences are random,
or are biased towards taking trades. If biased by the choice of data
series, it seems wisest to then trade with the same data series; but if
it's believed that the differences are random, one may prefer to test
with cc's for convenience and trade the "real thing".
Bottom line is that testing is approximate. Not only is the future
variable but there are probably all kinds of approximations in the
testing only some of which we can be aware of.
Conrad
Bowers
|