|
PureBytes Links
Trading Reference Links
|
To: "Neil Harrington" <njhprovo@xxxxxxxxxx>,"'Mark Brown'"
<markbrown@xxxxxxxxxxxxx>,<omega-list@xxxxxxxxxx>,"'Bob Fulks'"
<bfulks@xxxxxxxxxxxx>
Subject: Re: Trend or No Trend, Gambler Indicators
Cc: <flag@xxxxxxxxxxxx>
At 11:24 AM 10/13/98 -0600, Neil Harrington wrote:
<snip>
>
>7. What about improved gambler indicators, like Mark Jurik's JMA, the Visine
>of MAs that "gets the lag out"? Does this make it a non-gambler indicator?
>
>Thanks for any thoughts.
>
>Neil
>
Neil:
To be precise, Jurik's JMA, of which I am a big fan, is a mathematical
function. When used by itself, it smoothes (or tames) data with minimal
lag. Hopefully, the information that is destroyed in the process (usually
called noise) is not needed, so that the desireable information is emphasized.
JMA can be combined with another slower JMA to form a moving average
indicator. It has now become a gambler function, because moving average
crossover by definition lags the true action. Try to daytrade with a moving
average crossover: it will put you on the high velocity part of the curve
everytime (or, based on your parameters, whipsaw you to death).
So I see JMA as an outstanding tool. As with any tool, you can use it to do
very ordinary things with it.
There was some earlier discussion on the list about Momentum and
Acceleration being leading indicators. This isn't true. However, I have no
doubt that a skilled discretionary trader can use these indicators with
success. Why? Because the skillful trader is filtering the trades using
some <probably> unknown, unconcious mechanism.
In a previous private post, I discussed my view of a time series. You can
model a time series as some type of autoregressive process that also has
independent noise added to it. I believe that a price series acts as a
state machine. It can be in a fairly predictable state where the
autoregressive contribution is high. At these times, indicators like
Momentum may work (for a while). When the state changes so that the
autoregressive contribution is low, that breaks down.
I think the successful discretionary traders know when those state
transitions occur, which allows them to filter otherwise ordinary
indicators. Some of the information is available to floor traders, as they
watch the action live. (I think a squawk box is a poor substitute for this.)
I like to think that there is a less subjective path to intelligent
trading. I have seen some evidence that there is intermarket information
that sheds light on the problem. To my surprise, I'm also finding that
there is information in the price series itself that is contributory.
However, the tools I'm using to explore this area are neural nets and fuzzy
inference systems. In essence, I'm trying to construct a "perfect"
artificial discretionary trader. And I'm not doing it the "traditional" way
by just throwing huge quantites of data at a neural net and hoping it will
converge to a usable solution.
This is intended as food for thought only.
Allan
"Everything ... is true, except the parts I made up. And they might not be
the parts you think." - Michael Flynn
|