|
PureBytes Links
Trading Reference Links
|
R.E.Turner wrote:
>
> RT'ers
> Here's a topic to sink your teeth into. What is the current meaning behind
> the buzz word "robust"? Notice everything seems to be "robust " these
> days; that is exhibiting strength and vigor. For example a countertrend
> mechanical system, trading against the trend, which keeps shorting a
> position until it eventually turns a profit is likely to be a very" robust"
> system,,,,,,,,, and take alot of guts to use.
> Russ
I think re systems robust means the opposite of curve-fitted or
overoptimized.
A non-robust system tends to only work on the historical data that was
used to design it. A system that, in historical design, has a low
trade:rule ratio runs the risk that the rules were designed to make a
particular trade in the past work. Problem is, trades in the future may
not have the same "shape".
By contrast a robust system will have lots of trades in the test, few
rules, and the parameters in the rules (like the # days in the mov. avg.
if used) can be varied a fair amount and still produce profits. By
modifying the parameters you are indirectly testing how sensitive the
system is to the "shape" of particular trades. If it's not sensitive,
then you assume it will be profitable in trades in the future even if
they don't behave exactly like the ones in your test.
I'm certainly not a systems expert; this is just what I gather from
reading.
A topic that might be valuable for further disc. would be, just how does
one "rate" the robustness of a system. How many rules are ok? How far
should you be able to vary the parameters? When are the results of an
advertised system "too good"? Etc.
Conrad Bowers
|