|
PureBytes Links
Trading Reference Links
|
> At 11:09 AM -0400 4/14/99, stuart wrote:
>
> >Thank you for your reply. You are correct that there is no easy way to
> >study all of the indicators available commercially. I have rewritten my
> >inquiry to the group to see if there can be a generalized way that the
> >users can list their approval or lack of to the various vendors. For
> >instance, if someone has purchased indicators from Mr. Jurik and they like
> >them, they could just say that Jurik's indicators have added value to
> >their trading decisions. If someone has researched many of the
> >commercially available indicators, and a thorough response would be too
> >demanding to be typed to the list, I would be glad to call them at my
> >expense if they would have the time to respond.
>
>
> There has been much discussion on all the lists in the past about the value of indicators. It is a favorite subject of mine so I will try to respond and throw in a bit of additional "wisdom" I have accumulated. This has turned out to be more like an "essay" than a "reply", for which I apologize, but as others have noted, your question is not simple to answer.
>
<BIG snip>
>
> Bob Fulks
>
Bob -
I wasn't going to dive into this thread, but I am promising to have
something to offer "soon". Don't know when you were at MIT, but
we may have learned our digital filtering from the same profs ... but
I went elsewhere to graduate ...
I have several years of indicators --- hundreds of them --- calculated
at 15-minute intervals for the DOW, S&P-100 and NASDAQ-100 components.
Until 6 weeks ago I spent my days (and pieces of nights) running this
system and doing approx. 15% per month based on some novel math that
I ran alongside the indicators. Looking at hundreds of indicators CAN'T
work, which is why I developed a new math to supplant the indicators ---
THAT'S what works. Not having enough capital to make a living at
this, I've put it aside until I can re-engineer it into a stand-alone
system.
Anyhow, to cut to the point at hand ... each indicator has an impulse
response when related to market moves. Some are as quick as 45-minute
predictors, others several hours. Some highlight individual stocks, others
simply don't. I have on my to-do list the simple "system identification"
task ... identify the impulse response of each indicator AND the "risk"
or variance of each impulse response. THEN I'll know which indicators
work, whether they work as co-varying or contra-varying with price moves,
and what time frames (note the plural) they work over.
I got the MatLab System Identification Toolbox to do this. It didn't
work and the vendor said "tough luck, it will be fixed in the next
release". That entails not only buying a new copy (because they
upgraded MatLab itself and raised the price to the ceiling!) but buying
a new copy of MatLab. Having published a paper using this same
technique in 1970, I'm about to dust off the old code (Fortan on
a GE-4060) and do it all over again.
IF my system beats any indicator, I'll publish the indicator impulse
responses on the Web. By the way, I discovered that AIQ's indicators
had bizarre behavior (blow up after a 72-day lag, i.e.) by using just
these methods back in '96.
Now the ROOT of the problem is simple: the market is not linear
AND the indicators are mostly non-linear themselves. I don't believe
that a strong time correlation between an indicator and the market
price moves will really work, for the non-linearity will lead to
spurious "peaks" in the impulse responses. Only averaging over a
LOT of data (which is just what I have) will begin to bring out the
"true" dynamics.
Sorry to offer this as a promisary note, but I'm engaged in some
patent (non-software/non-market) activities for a few more weeks.
Cheers,
Rob Lake
Environmental Modeling Inc.
rbl@xxxxxxxxxxx
|