|
PureBytes Links
Trading Reference Links
|
<x-html><!DOCTYPE HTML PUBLIC "-//W3C//DTD W3 HTML//EN">
<HTML>
<HEAD>
<META content=text/html;charset=iso-8859-1 http-equiv=Content-Type><!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<META content='"MSHTML 4.72.3110.7"' name=GENERATOR>
</HEAD>
<BODY bgColor=#ffffff>
<DIV><FONT color=#000000 size=2>ARIMA stands for Auto-Regressive Integrated
Moving Average. It is a standard statistical tool for analysizing and
predicting time series. Unlike spectal methods which transform the time
series into the frequency domain (e.g., Fourier Analysis and Maximum
Entropy), ARIMA stays in the time domain. It predicts into the
future by summing past values plus a random shock (the AR part) with a series of
random shocks plus the mean (the MA part). The AR component is analgous to
the series' memory of itself, while the MA part smooths the uncertainties of the
external world.</FONT></DIV>
<DIV><FONT color=#000000 size=2></FONT> </DIV>
<DIV><FONT color=#000000 size=2>ARIMA can give good results but it is not a
black box--the analyst must guide the model. First, t</FONT><FONT
color=#000000 size=2>he series being predicted must be "stationary"
which means that the statistics don't change over time. A non-stationary
series, which market prices most certainly are, can often be transformed into a
stationary one by removing the trend and seasonal components. The most
common way to remove a trend is to take differences of the values, e.g., Pn+1 -
Pn. Any transforms to make the series stationary must be reversed when
making predictions. A difference is reversed by an integration, hence the
"I" in the acronym.</FONT></DIV>
<DIV><FONT color=#000000 size=2></FONT> </DIV>
<DIV><FONT size=2>After making the series stationary, the analyst must choose
which terms to include.in the model. For example, the analyst might choose
simple linear regression, i.e. Pn+1 = A*Pn + Zn, where A is a parameter and Zn
is a random shock (at time n). This is where the "art" comes
in. While a computer can estimate the parameters, the structure of the
model must be specified by the analyst. The linear regression model just
specified would be written in short hand as ARIMA(1,0,0), that is, an ARIMA
model with 1 AR term, no differencing, and no moving average terms.</FONT></DIV>
<DIV><FONT size=2></FONT> </DIV>
<DIV><FONT size=2>ARIMA is used extensively by the US government to project
economic and other data into the future. The government even provides (for
free) the software it uses, ARIMA11. </FONT></DIV>
<DIV><FONT size=2></FONT> </DIV>
<DIV><FONT color=#000000 size=2>Some References:</FONT></DIV>
<DIV><FONT size=2>Timothy Masters: "Neural, Novel, & Hybrid Algorithms
for Time Series Prediction", published by Wiley. Besides ARIMA, the
software includes Fourier and Maximum Entropy Spectral analysis, neural nets,
plus all the transforms and filters you need in an integrated package.
There's a little bit of math, and the UI is explained by the commands it
generates, but it's simpler than most introductory books on time series
analysis. He doesn't explain how to apply the analysis to trading, but
some of the included examples are bond prices. </FONT></DIV>
<DIV><FONT size=2></FONT> </DIV>
<DIV><FONT size=2>For applications to trading see chapter 2 in Perry Kaufman's
"Trading Systems and Methods", also published by Wiley. His
explaination skips estimating the parameters (he assumes you've got a program to
do this), but he gives an excellent overview of how the analysist would fit the
model. He gives 3 basic ways to trade it: (1) If ARIMA forecasts an
uptrend and if prices fall below the forecast, go long (expecting lower risk and
more profit), (2) If forecast is for higher prices, hold long, otherwise go
short, (3) use confidence bands to determine overbought/oversold levels (most
programs provide confidence bands). Enter long when prices penetrate
lowest 95% confidence bands, close out when prices penetrate 50% level.
Kaufman gives several variations of the above. Personally, I don't like
(1), it expects that a market which diverges from the forecast will correct back
to it. Also, he assumes for (1) and (2) that the trend is up and hedges
(3) with entering the long only if the trend is up.</FONT></DIV>
<DIV> </DIV>
<DIV><FONT color=#000000 size=2>-- Jeff.</FONT></DIV>
<DIV> </DIV>
<BLOCKQUOTE
style="BORDER-LEFT: #000000 solid 2px; MARGIN-LEFT: 5px; PADDING-LEFT: 5px">
<DIV><FONT face=Arial size=2><B>-----Original Message-----</B><BR><B>From:
</B>gary bodnar <<A
href="mailto:gbodnar@xxxxxxxxxxxxxxxx";>gbodnar@xxxxxxxxxxxxxxxx</A>><BR><B>To:
</B>RealTraders Discussion Group <<A
href="mailto:realtraders@xxxxxxxxxxxxxx";>realtraders@xxxxxxxxxxxxxx</A>><BR><B>Date:
</B>Sunday, April 18, 1999 10:39 AM<BR><B>Subject: </B>ARIMA Box-Jenkins
Analysis on YenSpot<BR><BR></DIV></FONT>Folks <BR>Does anyone know anything
about ARIMA Box-Jenkins analysis? <BR>The analysis will give out forecasts
among other things.. <BR>Today, the idiotbox gave out forecast on Yen and it
should go to 117.77-120.05. The upperbound should be @117.92-122.08. The
lowerbound is @117.61-118.02. This is not a trading suggestion. It is what I
got from this idiotbox program..
<P> <A
href="http://www.microtec.net/~ichiadmi/frameut.html#Begin";>Arimex
Box-Jenkins Time Series analysis for Excel</A>
<P>Gary </P></BLOCKQUOTE></BODY></HTML>
</x-html>From ???@??? Tue Apr 20 06:10:51 1999
Received: from list.listserver.com (198.68.191.15)
by mail05.rapidsite.net (RS ver 1.0.2) with SMTP id 11566
for <neal@xxxxxxxxxxxxx>; Tue, 20 Apr 1999 04:54:47 -0400 (EDT)
Received: from localhost (localhost [127.0.0.1])
by accessone.com (8.8.5/8.8.5/PIH) with SMTP id BAA11357;
Tue, 20 Apr 1999 01:50:44 -0700 (PDT)
Received: from pop1.tm.net.my (pop.tm.net.my [202.188.95.1])
by accessone.com (8.8.5/8.8.5/PIH) with ESMTP id BAA10994
for <realtraders@xxxxxxxxxxxxxx>; Tue, 20 Apr 1999 01:48:36 -0700 (PDT)
Received: from vincentl ([203.56.11.165]) by pop1.tm.net.my
(InterMail v03.02.05 118 121 101) with SMTP
id <19990420084801.EDHT276@xxxxxxxx>;
Tue, 20 Apr 1999 16:48:01 +0800
Message-Id: <3.0.32.19990420185514.011550f8@xxxxxxxxxxxxx>
Date: Tue, 20 Apr 1999 18:55:24 +1000
Reply-To: fibocal@xxxxxxxxx
Sender: owner-realtraders@xxxxxxxxxxxxxx
From: Peter Lim <fibocal@xxxxxxxxx>
To: RealTraders Discussion Group <realtraders@xxxxxxxxxxxxxx>
Subject: Re: Probability applications to stock prices
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
X-To: gil ward <gw1933@xxxxxxxxxxx>
X-Cc: RealTraders Discussion Group <realtraders@xxxxxxxxxxxxxx>
X-Sender: fibocal@xxxxxxxxxxxxx
X-Mailer: Windows Eudora Pro Version 3.0 (32)
X-Listprocessor-Version: 8.1 -- ListProcessor(tm) by CREN
X-Loop-Detect: 1
X-UIDL: 76a8c316206a3ccff0d644f3fa37cd17.10
Hi Gil & Lutz,
The url is:
http://www.ChoiceDaytrades.com
Have a look. Post me on developments.
Best regards,
Peter
=========================
At 10:48 19/04/99 -0400, you wrote:
>Peter, know where to find out about RPM software?
>Gil
>
>Peter Lim wrote:
>
>> Hi Rich,
>>
>> There are probability tables done for that purposes. Dr JT Jackson has an
>> add on software to Fibonacci Trader that generates the proabaility values
>> for prices to reach certain targets. You can get to read about how these
>> are computed and the actual tables themselves from the book, "Detecting
>> High Profit Day Trades in the Futures Markets-- using zone Pattern
>> Probability Analysis". His methodology does not include usage for the
>> futures index.
>>
>> There is a software called RPM (Reliable Pattern Matching) that generates
>> reports on the probabilities of targets being achieved on any particular
>> day once the first open day's trade price is know, if I am not mistaken.
>> That's for the S&P futures.
>>
>> Hope it helps.
>>
>> Peter
>>
>> ============================
>> At 10:12 19/04/99 -0500, you wrote:
>> >Hi everyone,
>> >
>> > Does anybody know the formula to compute the probability that a stock
will
>> >hit a given price at a given time? ie: given a volatility of x and
current
>> >stock price of p, what is the probability that the stock will move to
price
>> >q in three days?
>> >
>> > rich
>> >
>> >
>
>
|