|
Wonder why most of the bankrupt know nothing
at all Investment banks that are now gone or worth pennies on the dollar, didn't
use this so called great idea.
Haven't looked. But would be
interesting to know the high of USB and what it was on the last month or
so. Just to see how great this stuff really is.
For some reason, a bull market makes many
look like financial geniuses.
----- Original Message -----
Sent: Tuesday, December 09, 2008 8:50
AM
Subject: RE: Re: Re:[RT] A note on
Forecasting / Mandelbrot
I don?t think ?Mandelbrot
conclusion has many conditions attached to it? is an accurate
conclusion.
The fractal distribution has qualities that make it almost
totally unlike the distributions used in classical statistics.
However, since it is rather obvious that billions of
dollars have been earned using these ?flawed statistics? and variants thereof
(see D. E. Shaw and Renaissance Technologies) it is also obvious that the
entire matter is amazingly complex.
Since all of these firms are secretive almost to a paranoid
degree, a determination of the actual technologies they employ will require a
major piece of detective work.
As Sergey indicates, the devil is in the
details.
Regards,
Bob Pardo
From: realtraders@yahoogroups.com
[mailto:realtraders@yahoogroups.com] On Behalf Of Stan
Rubenstein
Sent: Monday, December 08, 2008 9:33 PM
To:
realtraders@yahoogroups.com
Subject: Re: Re: Re:[RT] A
note on Forecasting
Is it possible that Mandelbrot conclusion has many
conditions attached to it not spelled
out in your comment that has a bearing on
it?
Also isn't it the "log normal" distribution that's
applicable to financial stock returns
----- Original Message
-----
Sent: Monday, December 08, 2008 10:03
PM
Subject: Re: Re: Re:[RT] A note on
Forecasting
These are too
general statements about too general subject. The most important is in
details. Here I agree with you.
And I use the
models with a lot of added conditions. The complexity of some model is not a
problem for the kind of research that I do.
As to Prediction
Company and their technologies, I have some questions for you as a
professional.
What exactly did
you use from Chaos Theory:
1) Non parametric
statistics? If so, how did you use it?
Information for
those who are not familiar with it:
Mandelbrot found
that the normal distribution is not working for financial data. It means
that we cannot apply classical statistics for financial
data.
For
example, traditional calculation of profit of
some trading system, or Black-Scholes option pricing model, are
not applicable for the real stock market.
2) R/S analysis? If
so, how did you use it?
Information for
those who are not familiar with it:
----- Original Message
-----
Sent: Monday, December 08, 2008 8:15
PM
Subject:
[Bulk] Re: Re:[RT] A note on Forecasting
All trading systems are
based on some assumption about market behavior. In simple form a trading
system says "If A occurs then B follows."An another example If an
oscillator reverses in oversold territory, buy next bar at x price.We all
know that such a system may have remarkable results over a given time
period however in the long run it will fail because it does not accurately
describe the way markets work. Markets can trend for long periods,
yielding many false signals and failed trades. However, suppose we add
another condition to the model that describes when the extended trend will
end. Now we have a model that more accurately describes how markets work
and should prove more reliable over longer periods. If we add enough
conditions to accurately describe market behavior under a number of market
scenarios, then we have a model that will have consistent returns over
time and markets. And the statistics will be a realistic expectation of
performance. This , of course, is the dream of all system
builders.
You may make the case
that markets are so dynamic that one can never detail conditions enough to
create a model that will yield reliable performance over all time frames
but I beg to differ. Both Mandelbrot and Peters agree that markets may be
predictable in the short run. I, in fact use the work of both in my market
model. And the Prediction Company certainly succeeded, did it
not?
----- Original Message
-----
Sent:
Monday, December 08, 2008 2:23 PM
Subject: Re:[RT] A note on
Forecasting
Static cycles
are not my favorite or special.
The real
question is a fundamental one: how to verify any trading strategy, based
on anything.
10 years ago I
have participated in a big research project. Its purpose was to test and
verify different trading systems based on methods of technical analysis.
Our group has found that the application of methods of classical
statistics to the stock market analysis is an extremely dangerous
thing.
Let me explain
it better on this example.
Let say we have
found a system that provides 70 winning signals from 100. The
university's course of statistics says that this fact is not occasional
with the probability of 99.5% (Chi Square=20x20/50=400/50=8
=> P=99.5%) It means that we can assume that there is a high
possibility that this system will work well in the future as it
does in the present. And somebody may decide that the Holy Grail is
found finally. But - it is not true. Statistics of the real stock market
and the market's logic are different from this one. If the
system works good enough for 100 current examples, it does not mean
that it will work the same for other
samples.
Jim, I want to
emphasize that I do not name here the models that we used for the
research. My group tried different things: TA indicators, risk/money
management, arbitrage systems, then different math models (like
Spectrum, autoregression), astro cycles as well. This problem still
presents for all of them.
I believe that
this problem is described well in these
books:
1) "The (Mis)behavior of Markets" of Benoit
Mandelbrot;
and 2) "Chaos and Order in the Capital Markets: A New View
of Cycles, Prices, and Market Volatility" of Edgar E. Peters.
In financial analysis, we have
to work with big data samples.
PS. Jim, it
seems to me that you are mixing two different things: fixed (or static)
cycles and dominant cycles. As an example, I would not believe if
somebody states that the 20-days cycle is found that has worked for 20
years. From another side, if somebody states that withing the last 100
days the 20-days cycle has been found, it is quite
possible.
Next 100 days
there might be some other cycle (27-days, for example). It is closer to
MESA and wavelet analysis, not to normal fixed cycles
analysis.
----- Original
Message -----
Sent:
Monday, December 08, 2008 4:27 PM
Subject: [Bulk] Re: [Bulk] [RT] A note on
Forecasting
The inability of a
methodology to return reliable and consistent performance is an
indication that the underlying hypothesis is flawed. For example,
methods based on static cycles or projections based on static cycles
will have inconsistent performance over different stretches of time
because static cycles are not fundamentally correct model of market
activity.
There are
characteristics of market movement and trader psychology that do not
change over time and methods based on these will exhibit consistent
performance. be it 100 or 700 samples.
----- Original
Message -----
Sent: Monday, December 08, 2008 11:52
AM
Subject: Re: [Bulk] [RT] A note on
Forecasting
Actually,
the question about financial statistics is a tricky one. The
important things there are not only win/loss ratios, the intervals
where these ratios are calculated should be considered as well. I
have had many cases when a trading strategy worked very well for a
half a year. And then it died
forever.
As an
example, see this intermediate backtesting result for huge
intraday data:
The system
provided 65% good signals (469 win./ 247 los.) during some perios
(several months).
After that
53% only, and then 59%.
100 trades
is not enough to get the reliable statistics (we use at least 500
trades, in this example 700
trades).
One of this
forum's participants is Robert Pardo, he can comment this better
than me.
----- Original
Message -----
Sent: Monday, December 08, 2008 12:31
PM
Subject: [Bulk] [RT] A note on
Forecasting
My pivot trading
methodology depends on anticipating and trading as close to the
pivot points as possible. My
argument is that trades near the pivot points are the lowest risk
and highest reward points to trade. I operate my trading as a
business - I buy inventory and sell to capture a minimum
profit margin. I have spent most of my trading career studying the
characteristics of markets at turning points (pivots) and
constructing trading tools to anticipate and trade near those
points. These tools deliver consistent reliability of profitable
trades between 70% and 80%.
I document my
trading concepts by forward testing, not computer generated back
testing. In other words I trade the tools in real time and record
the results. For example, my latest application to the ESZ08 has
generated about 78% profitable trades on a five minute chart over
the past 6 weeks.
One of the issues
I have with the people that post forecast on this list is that
they do not provide reliability measures of their techniques.
Failed forecasts are rarely addressed and specific application
details are not provided. Consequently I usually delete them
without consideration - after all - a stopped clock is right twice
a day.
So I recommend
that anyone who posts a forecast provide the statistics
documenting the same performance of technique over at least 100
applications. For example my techniques are good within one bar of
the forecast 70% to 80% of the time depending on market. With that
information, readers can better judge the value of the
post.
Jim
White
Pivot Research & Trading
Co.
PivotTrader.com
__._,_.___
![]()
__,_._,___
|
