|
PureBytes Links
Trading Reference Links
|
At 8:05 PM -0400 4/13/98, steinbr@xxxxxxxxxxxx wrote:
>In all deference to Bob's opinion which I do respect.... my experience has
>been quite the opposite over long periods of backtesting.
>
>I have seen $300,000 profitable systems over a 10 year span turn into $50k
>losers just by moving the beginning dates by 10 days. I believe this means
>that the system is subject to initial conditions. The most naive systems
>fall prey to this when trades are interdependent. For instance, if the
>system ignores all entry signals and takes only exit signals once a
>position is taken. If that initial trade lasts for 180 days then you have
>effectively ignored the system for 180 days. Now if you move your beginning
>point 10 days and that trade is never taken because another trade is on,
>well then all things change and the system results over a series of
>beginning points never really converges. It is sort of a chaotic system.
I think we agree.
Obviously, if one trade is most of your profit and you lose that trade, it
makes a big difference. But anybody who uses the results of such
backtesting as an indication of future potential is nuts.
My point was that with 10000 bars and any reasonable system that produces
lots of trades (with no one contributing a lot of the profit), losing a few
bars should not have a big effect. If it does. I would be very suspicious
of the system.
Bob Fulks
|