|
PureBytes Links
Trading Reference Links
|
Leslie's point about compounding and specific historical trade sequences
distorting results seems to be well taken, and Monte Carlo simulation does
sound like a good way to avoid such problems.
Mark, some basket models, such as Aberration (which I know you have
criticized) are very simple, use the same parameters for all securities,
were released a long time ago, and produce positive returns over time if one
can tolerate significant drawdowns. In several of these respects, they
resemble Oddball.
As for walk forward testing, it does seem possible to abuse it: if you take
100 models that work well on in sample data, and cherry pick 10 that also
work well on out of sample data, does that really mean that those 10 will
necessarily do better than the other 90 in real trading?
One of the appeals of a basket system that uses the same parameters on every
market is that if it works in back testing, it seems reasonable to expect it
to work going forward if the character of the markets does not drastically
change. Another appeal is that the equity curve smooths out when you trade
multiple securities. Which does not mean that it makes sense to include
securities with net losses.
I found a presentation on the Internet by Keith Fitchen, vendor of
Aberration, with some interesting information. He demonstrated that
parameter optimization on a per-security basis did not trade well on unseen
data, but selecting which markets to trade on the basis of historical
profitability did carry over pretty well into the future.
By the way, the results David recently posted for his weekly-bar SAR model
look quite good. Can anyone tell from examining or testing his code whether
there are subtle issues distorting the results? One thing that I can see is
that TradeStation is probably assuming that the exact stop level prices are
traded, not taking gaps into account; I'm not sure whether resulting
distortions would be as likely to be profitable or unprofitable, or whether
they would be biased to the profitable side.
-----Original Message-----
From: Mark Brown [mailto:markbrown@xxxxxxxxxxxxx]
Sent: Sunday, April 13, 2003 10:46 AM
To: omega-list@xxxxxxxxxx
Subject: Re[2]: Available Portfolio testing programs for TS2000i
Hello Leslie,
LW> History does not re-occur in the same sequence in the future as it
LW> occurred in the past. Thus, correlation, between markets change
LW> constantly.
that goodness "someone" was able to finally put into words the exact
reason these basket models being sold are mostly well thought out
mathematical moving target (meaning changing portfolio and position
size) scams.
btw i feel the same about forward walk, myself.
--
Best regards,
Mark mailto:markbrown@xxxxxxxxxxxxx
|