|
PureBytes Links
Trading Reference Links
|
Another point on this topic: it can be very helpful to examine the equity
curve for places where the system does poorly and then look at the market
behavior during those periods and try to develop filters that prevent the
system from trading when the market is exhibiting behavior that the system
can't handle.
Kent
----- Original Message -----
From: "Paul M. Zislis" <pzislis@xxxxxxx>
To: <omega-list@xxxxxxxxxx>; "LPrice" <lprice1023@xxxxxxxxxxxxx>
Sent: Friday, February 28, 2003 1:12 PM
Subject: Re[2]: The Due Effect......
Replying to your message of Friday, February 28, 2003, 12:53:13 PM,
My feeling is that the equity curve is telling us only second-hand
that our system is not working as well as it did in the past. Rather
than focusing on the equity curve itself, I think it is more
profitable to focus on what it is about the market and system being
traded that leads to the poorer results; i.e., identify additional
tests or filters on trading based on patterns, your favorite
indicators, etc. Of course, you need to be sure you aren't curve
fitting. I agree that self adapting systems that make trade decisions
based on changing market characteristics is a better way to go than
basing trading decisions on the equity curve.
L> This thread has stimulated a lot of thinking for me. A few years ago I
L> experimented with various moving averages of the equity
L> curve and also looked into analyzing the equity curve as a data stream
L> in order to come up with some advance idea as to the
L> performance of the system. However, I have concluded that the most
L> important issue to understand is why systems stop
L> working. Markets constantly change and it becomes clear that the reason
L> a system fails to perform during any given time
L> period is because the decision rules of the system no longer fit the
L> market conditions at the time. Of course this seems
L> elementary, yet it leads to the whole question of self adaptivity and
L> optimization. In analyzing many different compressions of
L> the same security for the same period, ranging from tick to daily by 1
L> minute increments, it is possible to determine some
L> relationships of noise to identifiable trending behavior as that ratio
L> moves through the various compressions. On a simple level,
L> it is clear that the smaller the compression, the more noise, however it
L> does not follow that every compression yields a logical
L> progression of that ratio. It is also instructive to analyze different
L> time frames in each of the compressions to understand how
L> this ratio evolves.
L> We know that markets go through periods of congestion and periods of
L> direction. We also understand that there are
L> relationships between securities, and various indices. Systems work
L> when whatever paradigm of market conditions that were
L> designed into the system exist. When the market has evolved into an new
L> circumstance, the system stops working and we
L> begin lose money. Confidence in the system is really based on the idea
L> the we believe that the market will return to the former
L> iteration that produced results in the past. Depending on the level of
L> refinement of the systems decision rules, that may or not
L> happen. For example, we all know that over enough time and in the right
L> compressions, moving average systems can produce
L> good results. But we also know that these systems go through periods of
L> deep draw downs. Perhaps the reason the simple
L> systems produce better results over time, is because they contain a more
L> common market condition and the system is designed
L> to, by it's very nature, ignore more of the intermediate market
L> perturbations.
L> What is clear is we can organize the data that is created by a security
L> in many different ways and that the act of organizing the
L> data is significant to the ultimate potential to be systematized.
L> Perhaps we bring order to the chaos of a data stream by that
L> very act. We can also change the ratio of noise to trend for any given
L> time period by altering the compression, however what
L> we cannot do is predict that ratio in the future. It constantly
L> changes. This is the challenge when we try to create evolutionary
L> systems that can adapt to market conditions.
L> Thanks to all who have contributed to this thread,
L> Lawrence Price
|