|
PureBytes Links
Trading Reference Links
|
--- In amibroker@xxxxxxxxxxxxxxx, "brian_z111" <brian_z111@xxx> wrote:
> Re testing on several stocks.
>
> If the system is 'good' on one symbol, (the sample size is valid) and
> it is also good on a second symbol (also with a valid sample size) is
> that any different from performing an IS and an OOS test?
It's perhaps a matter of terminology, but if one were to say "you don't have a system to even test until you've locked down the parameters", I think such a statement would be easily understood (even if someone doesn't normally define 'system' that way).
In terms of the IS/OOS sequence, one can then say that the "system" comes into existence after the IS procedure.
The system may now be tested:
OOS (time-wise) on the same symbol
IS (time-wise) on a different symbol(s)
OOS (time-wise)on a different symbol(s)
all of which provide data that was unused in the original IS procedure.
These tests answer 3 different questions regarding how particular markets over particular time periods behaved wrt. to success/failure of this system in the past (we still have no data for the future).
1. Did the modelable market behavior persist into some later period for the original symbol?
2. Was <the same> modelable market behavior present in a different symbol(s) concurrently?
3. Was <the same> modelable market behavior present in a different symbol(s) in a later period?
(I presume OOS to be "later" data in all cases.)
Every "Yes" answer is considered good rather than bad, but no amount of "Yes" answers can tell us if the market behavior for the future for any given symbol will continue to cooperate.
This all applies to a single, discovered "system", not the sequential process of system discovery/rediscovery itself.
However, what might reasonably be said in relation to "Yes" answers here is that they can be a good indication that the system is actually capturing market dynamics (during all the test periods), rather than being simply a mathematical curve fit of the original IS period (as might always be constructed by optimizing an arbitrary number of irrelevant polynomial functions).
More formally, one would have to analyze the probability that the various test datas matched the putatively modeled dynamics by chance. I doubt that most traders would proceed to this point ... ![:D]()
__._,_.___
Please note that this group is for discussion between users only.
To get support from AmiBroker please send an e-mail directly to
SUPPORT {at} amibroker.com
For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG:
http://www.amibroker.com/devlog/
For other support material please check also:
http://www.amibroker.com/support.html
![]()
__,_._,___
|