High Frequency Trading - The Technical Analyst

Transcription

mar/apr 2006
The publication for trading and investment professionals
www.technicalanalyst.co.uk
High Frequency Trading
The way ahead?
Vodafone
Bull market tops
Relative strength
A positive outlook
for 2006
Do single stocks
peak with the Dow?
Identifying sector
outperformance
Updata for
Bloomberg
More and more Bloomberg Professional users are running
Updata Technical Analyst to unleash the power of their Bloomberg data
Some reasons why you should join them
Familiar Bloomberg codes
for fast ticker entry
Create your own portfolio
watch lists
Full Windows look and feel
with multiple desktops
Market Maps of any list
of instruments
Optimised Trailing Stops
High speed scans for
any technical criteria
Chart Bloomberg CIX Indices
High quality printing
and page setup
Hundreds of TA techniques
and tools covered
Draw or add notes, labels,
logos and images
Automatic creation of
Index Constituent Lists
For more details:
type API <GO>
on your Bloomberg
Go to 3rd Party and
download the PDF
about the Updata system
All Trade Marks are the property of their respective owners
Best Point and Figure charts
in the world
Risk Reward Ratios and price
targets for any trade
Chart your own Excel data,
Metastock and other formats
To arrange a Trial
or Demonstration
Call: 020 8874 4747
Email: ta@updata.co.uk
Download your trial at:
www.updata.co.uk/bloomberg
Schedule Automatic Chart Runs
Optimise and Test any Indicator
Report Writer for documents
and presentations
Sophisticated technical analysis
alerts with screen pop ups,
SMS messaging or email
WELCOME
Our first annual European conference took place in early February and was a great
success. The quality of speakers was unusually high and the feedback we have
received has been very encouraging. The magazine would like to thank all our
speakers, delegates and sponsors for taking part in the event.
With the sharp decline in trading commissions, high frequency trading is now taking
on a greater role as a trading strategy. We take a close look at the need for trading
simulations as a preparation for adopting an HFT strategy.
We hope you enjoy this issue of the magazine.
Matthew Clements, Editor
CONTENTS 1 > FEATURES
MAR/APR
High Frequency Trading
Is HFT the future of trading?
HFT models can identify and exploit market
anomalies in milliseconds. We look at key
issues relating to the development of
successful HFT models.
>12
Bull market tops
Do individual stocks peak at
the same time as the Dow?
>20
Paul Desmond of Lowry’s Reports looks at the
relationship between peaks in individual stocks
and highs in the Dow, and finds some surprising
results.
Interview
World renowned TA expert Martin Pring talks
about his favoured techniques, the future of
TA and his new book due out later this year.
© 2006 Clements Biss Economic Publications Limited. All rights reserved. Neither this publication nor any part
of it may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic,
mechanical, photocopying, recording or otherwise, without the prior permission of Clements Biss Economic
Publications Limited. While the publisher believes that all information contained in this publication was correct
at the time of going to press, they cannot accept liability for any errors or omissions that may appear or loss
suffered directly or indirectly by any reader as a result of any advertisement, editorial, photographs or other
material published in The Technical Analyst. No statement in this publication is to be considered as a
recommendation or solicitation to buy or sell securities or to provide investment, tax or legal advice. Readers
should be aware that this publication is not intended to replace the need to obtain professional advice in
relation to any topic discussed.
March/April 2006
> 27
>>
THE TECHNICAL ANALYST
1
www.vni.com
Advanced Analytics for Powerful Minds
The IMSL™ Numerical Libraries
The PV-WAVE® Product Family
Professional Services
The most sophisticated, flexible, scalable
and highly accessible mathematical and
statistical algorithms available for numerical
analysis today. Available in pure C, C# for
.NET, Java™, and Fortran.
Visual Data Analysis (VDA) solutions experts
trust for rapidly importing, manipulating,
analyzing and visualizing data of any size
and complexity. Includes the PV-WAVE
development environment for sophisticated
visualization, TS-WAVE™ for advanced time
series analysis and JWAVE™ for Web-based VDA.
Utilizing the best tools, highly skilled
technical experts and over three decades
of experience, Visual Numerics
collaborates with its customers to develop
optimal solutions and achieve the highest
possible return on investment.
VISUA L N UM ER I CS I N T E R NAT IO NAL , LT D .
Soane Point, 6-8 Market Place, Reading, Berkshire, RG1 2EG United Kingdom
TEL : +44 (0) 118-925-5910
EM AIL : info@vniuk.co.uk
Keltner Channels revisited
Perry Kaufman on trading systems
23
45
CONTENTS 2 > REGULARS
Editor: Matthew Clements
Managing Editor: Jim Biss
Advertising & subscriptions:
Louiza Charalambous
Marketing: Vanessa Green
Events: Adam Coole
Design & Production: Paul Simpson
The Technical Analyst is published by
Clements Biss Economic Publications Ltd
Unit 201, Panther House,
38 Mount Pleasant,
London WC1X 0AN
Tel: +44 (0)20 7833 1441
Web: www.technicalanalyst.co.uk
Email: editor@technicalanalyst.co.uk
INDUSTRY NEWS
04
MARKET VIEWS
Vodafone: A brighter outlook?
Asian & Australian equities: An Elliott Wave outlook
06
08
TECHNIQUES
High frequency trading
Relative strength: identify which sectors are outperforming
The nature of bull market tops in US stocks
A trading strategy using Keltner Channels
12
18
20
23
27
INTERVIEW
Martin Pring
SUBSCRIPTIONS
Subscription rates (6 issues)
UK: £160 per annum
Rest of world: £185 per annum
Electronic pdf: £49 per annum
For information, please contact:
subscriptions@technicalanalyst.co.uk
ADVERTISING
For information, please contact:
advertising@technicalanalyst.co.uk
PRODUCTION
Art, design and typesetting by
all-Perception Ltd.
Printed by The Friary Press
ISSN(1742-8718)
SUBJECT MATTERS
Dissecting the RSI
Research update
32
38
SOFTWARE
Poulin-Hugin: Patterns & Predictions
41
BOOK REVIEW
New Trading Systems & Methods
by Perry Kaufman
45
COMMITMENTS OF TRADERS REPORT
EVENTS
46
48
March/April 2006
3
Industry News
THOMSON ADDS DEMARK INDICATORS
Thomson Financial has recently
added a suite of 25 Tom DeMark
indicators to the charting applica-
tions on its Thomson ONE platform. DeMark Indicators are a set of
market timing tools that use quanti-
tative analysis to anticipate price
exhaustion and trend reversals across
the full range of asset classes.
The Thomson service includes a
scanning facility for US and
Canadian equities that looks for
stocks that meet buy and sell conditional requirements each night.
Thomson scans daily, weekly and
monthly historical equity charts for
completed buy and sell Setups and
Countdowns. Results are then distributed according to their exchange,
index, and/or sector. The scans
keep the results for five periods
allowing customers to review historic
signals. Thomson scans the NYSE,
NASDAQ, and TSX for the following DeMark studies: TD Setup™,
TD Sequential™, TD Aggressive
Sequential™, TD Combo™, and
TD Aggressive Combo™.
TraderMade supplies HSBC with mobile charting
HSBC has chosen Tradermade to
deliver real-time charts and quotes
via mobile phones and BlackBerrys
for its global treasury staff. TM-Cell
offers mobile access to recent price
action in a number of markets
including FX rates, precious metals,
commodities, futures and stock
indices. TraderMade say no software
needs to be downloaded onto mobile
devices and a simple bookmarked
page enables easy access.
TraderMade has also been chosen
by The ICMA Centre at The
University of Reading to supply
technical analysis products to
ICMA’s staff and students.
TraderMade will provide both
mobile and web-based access to
rates and charts for a number of
asset classes.
DrKW enhance TA research service
Dresdner Kleinwort Wasserstein
(DrKW) has updated its on-line
technical analysis service, PIA, with
the launch of a Blackberry PIA. This
allows users to receive the service
simultaneously via their PC or
Blackberry. PIA offers DrKW clients
4
such as fund managers and hedge
funds 6 market versions: equity,
emerging markets FX, fixed income,
Europe, FX and US with headline
call, targets, risk levels and sentiment
analysis. More information is available at: http://pia.drkw.com.
March/April 2006
Market Views
VODAFONE
A BRIGHTER OUTLOOK?
A
mong blue-chip tech/telecom
stocks, Vodafone was the
major casualty of the corrective phase that began in early 2000. At
that time it was the UK's largest company (by capitalisation) and by the time
the share price was topping out at 384p
it had risen by more than 200% over
the course of the previous two years.
Following that peak, however, (and the
realisation that share price valuations
were not subject to a 'new paradigm'
after all) there followed a corrective
phase that lasted almost two and half
years and led to a near 80% decline in
the value of the stock. The stock's
nadir came on July 3rd 2002 when it
closed at 80.25p, a low that was subsequently (and successfully) tested in
September of that year - price action
which confirmed that a bottom of
sorts had been created. A rally soon followed which lifted the price up to 127p
(in November 2002) and although there
have been further rallies since then, the
by Bill McNamara
overriding impression created by the
Vodafone chart over the last three years
is of a stock that is trading within a
range.
Support holds again
Since March 2003, when it bottomed
out at 101p, Vodafone shares have
oscillated between 114p and 151p - a
range of 31.5%. Over the last few
weeks, however, there has been a break
down to 109p and although from the
perspective of the daily chart this
appears to represent a break through
support, a glance at the weekly chart
(see Figure 1) creates a different
impression altogether.
This view confirms that the latest
advances from Vodafone amount to a
bounce from support, and it is interesting to note that the readings from the
14-period RSI confirm that a bottom
was being formed before the share
price actually began rising. The oscillator formed a major low in early
Figure 1.
6
March/April 2006
December following which it began to
rise, and this bullish divergence
between the price and the RSI suggested that the selling pressure was starting
to dry up even before Vodafone bottomed out a few weeks later. It is also
worth mentioning that the initial low
was tested a month later, and it is the
fact that support held at the second
time of asking that led to the beginnings of a rally.
It was serendipitous perhaps that the
lows coincided with the announcement
from the company that it was in talks to
dispose of its Japanese division to
Tokyo's Softbank, a move which
increases the chances of further divestments in the medium term. The question now is whether this latest bounce
from support will act as the springboard to further significant upside over
the course of the next few months or
whether momentum will simply fall
away again. The chart contains some
clues.
Market Views
Figure 2.
“WHILE A RETURN TO THE HEADY DAYS OF 2000 SEEMS
ALTOGETHER IMPROBABLE, THE TECHNICAL
PICTURE FOR VODAFONE DOES APPEAR
TO BE IMPROVING.”
Since 2002 the RSI has dropped
below a reading of 35% (on the weekly
chart) on just three occasions - in May
2002, July 2004 and in late 2005.
Following the drop in 2002 the
Vodafone share price rallied from 90p
to 127p in the space of six months, a
gain of 41%; in the wake of the 2004
low there was an advance from the 116p
level up to 148p, a gain of 27.6% which
took five months. In December, the
RSI dropped to 31.8%, its lowest reading in more than three and a half years,
and it would appear that such oversold
levels have a good chance of leading to
a rally of meaningful proportions.
Another factor which should be taken
into consideration is the evidence provided by Figure 2 which appears to suggest that the latest price action from the
stock amounts to a bounce from the
long-term uptrend (in which the price
advance between 1997 and 2000 comes
across as an aberration).
On this view the share price is set to
return to the (shallow) upward trajectory that has been in place over the last
ten years. This chart does not suggest
that Vodafone is set for a dramatic
revaluation in the near term, but it does
March/April 2006
seem to imply that downside is likely to
be limited.
While a return to the heady days of
2000 seem altogether improbable, the
technical picture for Vodafone does
appear to be improving and there are
certainly enough positive signals in
place to reach the conclusion that the
shares are likely to be heading higher
during 2006, and perhaps beyond.
Bill McNamara is a technical analyst at Charles Stanley stockbrokers.
7
Market Views
ASIAN & AUSTRALIAN EQUITY MARKETS
AN ELLIOTT WAVE OUTLOOK
by Wang Tao
A
sian equity markets have been
quite bullish during the past
few years, while the Australian
stock market has been on a strong bullish trend since as far back as 1989.
Based on Elliott Wave analysis, we
believe these bullish trends will continue over the next few years.
Singapore
In September 1998, the Singapore
Straits Times Index (STI) started a
fierce bullish run from the low at
805.14, after the Asian financial crisis,
reaching 2582.94 in late 1999, a 220.8%
increase (see Figure 1). It is quite obvious that the wave pattern was impulsive
in nature and with fine tuning on a daily
chart the impulsive waves 1, 3 and 5 can
be clearly identified. We classify this
rally as giant wave (I).
When the index dropped to 1197.85,
it posed the question whether the low
at 1197.85 was the end of the ABC correction or just the completion of the C
wave (followed by D and E waves).
What is clear is that the current rally is
under the strongest wave (III) move,
the difference in the end of the correction from 2582.90 only leads to the different labeling at the start of this wave
(III).
There is a double bottom pattern
from C to E with a projected target of
2500 which is yet to be reached. Based
on the wave pattern with wave (III)
1.618 times wave (I) the projected target will be 4076.48. The historical high
was at 2582.94, which we think will be
easily broken soon.
Japan
For the Nikkei 225, the rally from 7603
in April 2003 to 11160 in September
2003 is classified as wave (1). Wave (3)
was much extended, though we are not
Figure 1. Straits Times Index
so convinced by the wave (3)-1 pattern
as it should be a more impulsive and a
clearer 5 wave mode.
It is a little bit tough to project the
wave (3) target, as wave (1) is rather
short in length. However, if we suppose wave (3) to be 2.618 times of
wave (1), we would be able to arrive at
a target of 18926.26 by adding 3557 x
2.618 to wave (2) bottom at 9614.
Currently, it makes little sense now to
assume wave (3) to be 1.618 times of
wave (1).
The alternative calculation comes
from the classical inverted head and
shoulder pattern where the head is at
7603, the neckline at 12081 with a projected target of 16559, which has been
Figure 2. Nikkei 225
March/April 2006
9
“AMONG THE
INDICES, THE
AUSTRALIAN
ALLORDINARY
INDEX IS
THE EASIEST
WAVE PATTERN
TO RECOGNIZE.”
Figure 3. Australia All Ordinary Index
reached already. We regard the current
move of the index as a wave 4 correction, which is likely to develop into a
flat pattern, but yet to be confirmed.
Australia
Among the indices, the Australian All
Ordinary Index is the easiest wave pat10
tern to recognize. The weekly wave pattern shows that wave (3) and wave (III)
were greatly extended and even now,
the 5th of wave 5 is yet to come. We
would like to target this 5th wave of
wave 5 to complete at 5073.50 before
starting the correction to 4750 or even
lower to 4550. Nevertheless, the major
March/April 2006
trend remains rather bullish with the
giant wave (III) yet to be completed by
the prior development of wave (4) and
(5).
Wang Tao is technical analyst with
Man Financial in Singapore.
Desktop Quality
in a mobile environment
Information at your fingertips
Real-Time Quotes
and Charts on your
Blackberry or
Mobile/Cell Phone
-Don’t just look at where the
market is trading now. Get
the whole picture.
-Access to high quality real-time
charts defined by us or
the user.
-No software installation
means it is quick and easy to
set up.
-Reduce the number of mobile
devices you carry.
-Accessible world-wide 24/7
8 Archers Court
48 Masons Hill
Bromley
Kent BR2 9JG
United Kingdom
Phone: +44 (0) 20 8313 0992
Fax:
+44 (0) 20 8313 0996
E-mail: sales@tradermade.com
The RIM and BlackBerry families of related marks, images and symbols are the exclusive
properties of and trademarks or registered trademarks of Research In Motion Limited used by permission.
Techniques
HIGH
FREQUENCY
TRADING:
THE WAY AHEAD?
by Lester Dye
The growth of high frequency trading (HFT) over the past three years is reflected in
the exponential increase in data message rates seen by the major electronic
exchanges. At Nasdaq's electronic platform, INET, for example, January 2006 daily
message volumes have grown to nearly 4 gigabytes, which is eight times the data
rate experienced in January 2004. In this article, we examine the implications of
detailed HFT simulation technology and advanced models for the future of trading.
12
March/April 2006
Techniques
March/April 2006
13
Techniques
H
igh Frequency Trading systems examine real-time, ticklevel, data streams for trading
opportunities which they are able to
exploit by placing and canceling orders
within milliseconds.
There are significant challenges to
high frequency modeling. The very
high data rates that must be handled
and the small latencies required for
most HFT models (see Box 1) restrict
applications to state-of-the-art servers
with high-bandwidth access. Accurate
simulation of HFT models is an intricate and expensive computational task,
since handling book data can require
processing gigabytes of data perfectly.
Doing this efficiently is difficult.
“THE KEY ADVANTAGE TO
HFT MODELING MAY BE
THE ABILITY TO
ACCURATELY SIMULATE A
MODEL PRIOR TO
COMMITTING CAPITAL.”
than a human can begin to press a key.
The key advantage to HFT modeling
may be the ability to accurately simulate
a model prior to committing capital.
This capability gives traders the ability
to select models and weight them
appropriately to manage risk. This simulation capability may be the foundation for the next generation of trading
systems.
High numbers of trades can mean
high commissions. HFT models are
often sensitive to brokerage commissions and even the seemingly small regulatory fees. On the other hand, the
models can take advantage of small
rebates available from electronic communication networks (ECNs) when
adding liquidity to the book. (It’s worth
noting that although HFT models must
be capable of trading at high frequencies, they may not necessarily do so. An
opportunistic model that trades infrequently, but is capable of trading at
high frequencies, may also be considered an HFT model).
HFT systems are sparking the rise of
a new segment in commercial software
frameworks: those designed to connect
HFT algorithm systems to a variety of
equities markets and data sources.
These frameworks, which include
Apama and Vhayu, provide connectivity solutions for large banks and brokerage firms. Other companies involved in
structure mathematical models are difficult to translate into efficient and correct algorithms where data rates vary
and where milliseconds can make the
difference between a winning trade and
a losing trade.
The advantages of HFT modeling are
compelling. The ability to take advantage of small or fleeting opportunities
in a market, and to do it reliably and
consistently and with minimum manpower, is a common goal for those
entering the field. Commercial HFT
models are robust and start and stop
automatically, removing both human
emotion and human error. HFT models can respond to complex market
conditions in milliseconds, far less time
Figure 1. A 4,000 share Intel model simulation is compared to live trading over a three week
period in which 252 trades were made.
HFT modelling
HFT model research can also be challenging. Many HFT algorithms deal
with the market micro-structure, and
many models are mathematically and
algorithmically sophisticated. Micro-
14
March/April 2006
this field concentrate on HFT modeling for specific markets. This next-generation system allows models and portfolios of predefined models to be
selected and simulated. Accurate simulation technology provides a modern
substitute for expensive live testing and
aids in cash and risk management.
These models are largely uncorrelated
to the market and are often used as a
hedge for equity portfolios, or as an
automated trading tool.
Simulation
Accurate simulation may be the most
important HFT technology. The ability to accurately test a model that makes
hundreds and thousands of trades is a
powerful tool.
Understanding a
model's behavior accurately and in different market conditions is an aid when
assembling portfolios of models that
have consistent returns. HFT simulation technology must pass a fundamental test: does it accurately match live
trading? The simulation technology
must pass this fundamental test.
Simulation of HFT models, while difficult computationally, can yield exceptionally accurate results. In Figure 1 an
Intel 4000 share HFT model made 252
trades from January 23, 2006 to
February 17, 2006. In this period the
historical net profit was $6267 and the
actual gain was $6892, for a total error
Techniques
of $625. This accuracy represents an
error of less than 0.70 per cent, or less
than $3, per trade.
In Figure 2, a Microsoft 4000 share
HFT model is compared to the simulated result over a three week period ending February 17, 2006. In this example
over 327 trades were made with an ending accuracy of $37.
To achieve this level of accuracy,
order book information must be constructed from the historical data stream
and the latencies of order placement
and partial fills must be considered. In
other words, all of the data must be
considered.
Limitations of bar data models
Many models examine data that has
been condensed into open-high-lowclose (OHLC) bars on specific time
intervals. As a result, the amount of
data
is
dramatically
reduced.
Additionally, data become much easier
to handle within algorithmic and modeling frameworks. There is a price to
pay, however, for this convenience.
OHLC data, collected over specific
time intervals, have lost some important fundamental information. The
time intervals, from which the data is
gathered to form the bar are the value
of the latency introduced in the system.
Bar intervals of 1, 5 or 15 minutes are
now latencies of information. And
opportunities that might require second or even millisecond accuracy are
now gone.
As a result, the volume and time of
the fill, as well as all book and reserve
order information, is lost. Because the
volume of the fill that comprised the
high of a specific bar is not known, the
model-independent of its sophistication-will not know whether it could
have filled 100 shares or 10,000 shares.
Models which are based on bar data
and place limit orders clearly will not
scale correctly. If the system assumes
market orders, a potentially significant
error can occur when there is slippage
from a market order. For these reasons, live models that use bar data often
have results that are difficult to reconcile with testing.
Figure 2. A 4000 share Microsoft model simulation is compared to live trading over a three week
period in which 327 trades were made.
Generally, using bar data will restrict
models to a domain that trades at low
frequencies, with market orders, and
that must achieve very large gains compared to the assumed slippage.
Additionally, the number of shares
must be small such that scaling is not an
issue, particularly if constant slippage is
assumed.
Simulation performance
Accurate simulation allows us to understand model behavior and control risk.
For example, stop losses might be
applied to determine the compromise
The latency of an order system is the amount of time it
takes for an order to be
placed. This includes the time
taken for the computer to
send an order message to the
ECN (electronic communications network) and receive a
confirmation that it was
received. For HFT systems,
latency can vary from a few
milliseconds to a few
hundred milliseconds.
Box 1.
March/April 2006
between risk and reward. Portfolios of
models can be constructed using
advanced simulation capabilities to
model risk reduction by weighting different models in a portfolio, each with
unique stops.
Simulation also provides a mechanism for determining the performance
of a model, its accuracy, draw-downs
and consistency. In Figure 3, we revisit our earlier example of a 4000 share
Intel model. In this case, you can see
there would have been periods of significant drawdown. This information
would not be available by examining
the performance of live trading over
this period.
HFT models often are not correlated
to the market or the model stock. In
Figure 4 we compare an Intel model to
a buy-and-hold strategy. It is clear that
the HFT model, over this six-month
period, provides a linear return with
lower volatility than the same capital
applied to the stock itself. This and
similar expected value models are often
used as hedging mechanisms. This particular model trades intra-day only,
thereby removing any exposure to
overnight risk. Since overnight events
are difficult to manage and liquidity
often is not available, this is particularly
important for trading professionals.
The model example in Figure 4 →
15
Techniques
traded 1781 times during this sixmonth period. This high frequency of
trading allowed it to earn $28,883 in
ECN rebates with a net profit of
$61,525 after all brokerage commissions, modeling fees and regulatory
fees.
Is HFT the future of trading?
Yes. High frequency modeling will
eventually dominate most trading. If
the current trend to fully automated
trading continues, it will only be a few
years before hand-trading becomes
obsolete. Traders in the future may
spend most of their time simulating
proposed models and portfolios of
models. The future of a DMA brokerage will be controlled more by the quality of its models and smallness of its
latency than by the features of its
graphical tools. Automated trading will
Figure 3. A 4000 share Intel model ran over the four month period ending February 17, 2006
is shown compared to live performance. The simulation illustrates periods of drawdown that
were not evident in the three week live trading period.
Figure 4. 10,000 share Intel model results compared to buy and hold on a six month period that
ended March 6, 2006. The model returns assume the standard intraday account margin of 4-1
and the buy and hold assumes the standard 2-1 margin required when holding the model
overnight.
16
March/April 2006
“HFT MODELS CAN
RESPOND TO COMPLEX
MARKET CONDITIONS IN
MILLISECONDS, FAR LESS
TIME THAN A HUMAN CAN
BEGIN TO PRESS A KEY.”
allow traders access to overseas markets, or twenty four hour markets.
Effectively, continuous trading will be
possible for individuals as well as institutions.
Models will monitor every tick of
every stock in your model portfolio and
eventually even your investment portfolios will be balanced automatically,
taking advantage of small price movements to buy and sell. In the case of an
investment portfolio, the trades may be
infrequent, but full book data will be
consumed by the balancing model.
An HFT model, while just an algorithmic tool, can currently be selected to
run indefinitely, just as one might select
a stock. In the future, it may be common for traders and investors to
choose models for trading just as one
might select a mutual fund. You won't
be betting on the fund manager, you
will be betting on the mathematical
abilities of the modeling team and your
ability to simulate the particular conditions to manage your risk.
Lester Dye is the founder of
Benchmark Simulation, a company
focused on HFT modeling technology. (www.benchmarksimulation.com).
Techniques
RELATIVE STRENGTH
by Julius de Kempenaer
A
technical call that the market or
a specific stock is going up or
down can be very useful and
profitable information, especially for
absolute return traders. However, for
an institutional portfolio manager
whose job it is to outperform a given
benchmark, such information is not
likely to be of much added value. One
technical tool that is available to indicate that one sector is likely to outperform a particular benchmark is Relative
Strength (RS). This is not to be confused with Welles Wilder's RSI (Relative
Strength Index).
The basic concept of Relative
Strength is pretty straightforward and
always comprises two time series that
will be compared against one another.
The comparison should always be likefor-like. For example, a price index
should be compared with another price
index instead of a total return index.
The basic formula for RS is:
RS = price item A / price item B
This RS value can then be plotted as a
series in itself, see Figure 1.
In Figure 1 the upper chart pane
shows a normal weekly price bar chart
of the Dow Jones STOXX banks index
(BB ticker SX7P Index). In the lower
chart pane the RS line of the DJ
STOXX banks index against the Dow
Jones STOXX (600) index is shown.
The interpretation of this RS-line is
simple. If the RS line moves up, the
banking sector is outperforming the
STOXX index. If the RS line is moving
down, banks are underperforming. Or
in other words, as long as the RS line is
18
Figure 1.
moving up, portfolio managers want to
be overweight banks in a European
portfolio and if the RS line is moving
down, an underweight position is more
appropriate.
Price/RS divergence
It is possible that the RS-line may on
occasion be moving contrary to the
price chart. A very good example in the
banking sector can be found in the
period from January 2001 to mid 2003.
During that period the price chart of
the banking index declined from a high
near 420 to a low near 220 at the beginning of 2003. During that same period
the RS line shows a clearly rising pattern that began back in early 2000. Like
equities in general, the stock prices of
banks were also under pressure at that
time. However, the sector got hit less
severely than the STOXX index itself,
causing the RS line to move up and the
March/April 2006
sector to outperform the benchmark.
Overweight positions in banking stocks
during that period would have created
excess returns for the manager. The
opposite situation - a rising price chart
and a declining RS line - is also possible
and well visible in Figure 1. In the period from mid 1999 to early 2000, the
banking sector index moves slightly
higher while the RS line shows a steep
decline. Although a long position in the
sector probably would have been profitable, the declining RS line clearly
shows that other sectors within the
STOXX index were doing much better
and should have been preferred over
banks during that period.
Applying moving averages to
the RS line
As we can see, the RS line moves like a
normal security. It displays defined
trends up and down and one can there-
Techniques
Figure 2.
fore determine support and resistance
levels. Trend lines can be drawn and we
can apply any technical indicator to the
line. As the RS line usually shows a
rather volatile pattern, I have applied
two moving averages in order to filter
out / display the trends in the RS graph
more clearly. The look back periods for
the moving averages are 10- and 30weeks. These parameters are non-optimized but seem to be doing a rather
good job as a filter over the period. A
lot of empirical and quantitative
research has not (yet) revealed that
other periods would lead to significantly better results. Be aware that this
approach so far is not a mechanical one
so there is always an interpretational
component to calls being made based
on the RS graphs. One thing that has
become clear over the years, however, is
that the methodology, catering for
longer-term oriented investors, is better
applied to weekly charts rather than to
daily charts which tend to have too
much noise in them.
While using this combination of
moving averages together with the RS
line, some filter rules have been
devised. They are along the lines of a
straightforward double crossover system:
If (RSMA-10 > RSMA-30) AND (RS >
RSMA-30) then OUT PERFORM
If (RSMA-10 > RSMA-30) AND (RS <
RSMA-30) then NEUTRAL
If (RSMA-10 < RSMA-30) AND (RS <
RSMA-30) then UNDER PERFORM
If (RSMA-10 < RSMA-30) AND (RS >
RSMA-30) then NEUTRAL
I assume that the outperformance
and underperformance results speak
for themselves. The neutrals are situations where, for example, the 10-week
moving average is above the 30-week
average, but where the RS line dropped
below both averages. As moving averages introduces a lag and the RS line
reacts immediately to price changes,
this can be seen as an early warning system. The same goes for the opposite
situation where the 10-week is below
the 30-week and where the RS-line rises
above both averages.
In order to better visualize the action
in the RS-graph, I started applying
coluor codes to the rules listed above.
During the outperformance periods the
bars are coloured green, during underperformance periods the bars turn red
and during the neutral periods the bars
are blue. In this way the position of the
March/April 2006
moving averages and the RS line is
immediately translated into a colour
code which makes quick interpretation
possible.
It should be stressed that this relative
approach is not suited to a mechanical
trading system for a single stock or a
single sector. Another difficulty is the
variety of restrictions with which many
fund managers have to deal. One can
imagine that a portfolio test, based on
the filters described, will have very different results for a fund manager who
has to adhere to a strict tracking error
limitation compared to a manager that
has the liberty to deviate strongly from
the benchmark. Nevertheless the
approach can be of added value for
both managers.
As stated earlier, working with moving averages introduces a lag into the
decision making process. When equity
markets and especially the TMT sectors, started to turn downward at the
end of the nineties, this caused fairly
late calls for these sectors as prices had
to travel quite a bit to the downside in
order to generate a signal. Using a technical indicator like the MACD on the
RS line can be of help in such situations. In Figure 2, the Dow Jones technology sector index is displayed with
the RS graph vs the STOXX index and
the 10-/30-week moving average with
the MACD indicator based on the RS
line of this sector.
As may be expected the RS-MACD
pinpointed the downturn of the RS line
in April 2000 while the moving averages in the RS graph only crossed in
July, some three months later. However,
the RS-MACD produced some premature buy signals during the decline of
the RS line during 2000-2001. Once
again this underscores the fact that no
single indicator is foolproof but that
added value can be retrieved from
interpreting indicators in combination
with developments in the RS graph.
Julius de Kempenaer is a technical
analyst at Kempen and Co merchant
bank in Amsterdam.
19
Techniques
The Nature of Bull Market Tops in US Stocks
by Paul F. Desmond
20
March/April 2006
Techniques
H
stock?
ow does an investor
identify the top day
for an individual
An easy answer might be that it is the
highest level reached by the Dow Jones
Industrial Average (Dow) before a
major market decline. Alternatively, it
might be that the exact top of a bull
market is the point at which the vast
majority of stocks reach their highest
price levels for many years to come.
Do the majority of stocks reach their
peak at the same time as the peak of
the Dow or do they peak in unison?
Stockmarket guru, Joseph Granville,
once surmised that one-third of stocks
reach their final bull market price peaks
in advance of the Dow, one-third reach
their highs in unison with the Dow's
peak, and one-third reach their peak
after the Dow. However, the sheer simplicity of Granville's theory suggests
that it was based more on guesswork
than on hard statistical analysis.
A Lowry Reports study found that
major market bottoms can often be
identified by evidence of panic selling
(one or more 90% downside days) in
which investors sell stocks with abandon. With the desire to sell having
been exhausted, buyers then suddenly
rush in to snap up the bargains, and
cover short positions, resulting in a
90% upside day. The combination of
panic selling across a broad spectrum
of stocks, followed quickly by broad
enthusiastic buying, produces what
might be described as a classic "V" pattern of prices at major bear market bottoms.
The nature of market tops
Bull market tops tend to develop gradually over a long period of time. The
reasons for this gradual process are
easy to understand. Just as bull markets
result from strong, persistent investor
demand, bull market tops evolve when
investors gradually stop buying. Some
investors simply run out of new money
to invest. Others begin to see individ-
ual stocks as being overvalued and
begin to hold back on new purchases.
The evolution of investor psychology
from strong buying enthusiasm for
stocks to passivity does not occur suddenly. Thus, bull market tops are commonly diffuse, possibly lulling most
investors into inaction. Perhaps it is the
slowness of the entire process that
makes it difficult to recognize a market
top.
The final days of a bull market are
substantially different than the final
ability to avoid capital losses is, in many
ways, a more important objective for
investors than making big gains. There
are several helpful tools that traders
have used for many decades to warn of
impending stock market tops, such as
the Advance-Decline Line and the
number of stocks recording new 52week highs. History shows that these
indicators often top out and begin to
contract as individual stocks fall by the
wayside months in advance of the final
Dow top. Therefore, it would not be a
“INVESTORS MUST BE ABLE TO SEE, AND HAVE TIME
TO REACT TO, THE GRADUAL DETERIORATION OF
MARKET BREADTH THAT PROCEEDS PERIODS OF
SUBSTANTIAL STOCK MARKET LOSSES.”
days of a bear market. At most bear
market lows, because fear and panic are
the dominant emotional drivers, the
vast majority of stocks tend to bottom
in unison. At most bull market tops,
where investors have been lulled into
complacency, the vast majority of
stocks seem to top out on an individual
basis. This simple study of bull market
tops should have far-reaching implications for all investors. The conventional wisdom of what a major market top
looks like must be completely revised.
Every portfolio manager must create a
new strategic plan as to how and when
to take defensive action. And, new
indicators must be devised to eliminate
the current guesswork of where individual stocks are within the primary
trend. Investors must be able to see,
and have time to react to, the gradual
deterioration of market breadth that
proceeds periods of substantial stock
market losses.
Many investors have experienced the
frustration of making big stock market
gains in a bull market, only to watch the
gains turn into big losses during the
subsequent bear market. Thus, the
March/April 2006
surprise to find that all stocks do not
reach their peaks simultaneously or in
unison with the Dow. But, it is the
degree and the intensity of the divergences of individual stocks from the
Dow that had never been measured
before.
The Crash of 1929
On September 3rd 1929, the New York
Stock Exchange (NYSE) saw the Dow
reach its high prior to the 1929 Crash.
In simply looking at the trading data
from that day, at the volume of trading
and at the highest prices for each stock,
it becomes apparent that some stocks
had traded at prices below their 1929
highs. Other stocks were considerably
below their yearly high. That seemed
strange for a day on which the Dow
was at the absolute highest point in history and at a level that would not be
seen again for the next 20 years. Upon
closer examination, it was difficult to
find stocks that were at their highs on
that fateful day.
Intuitively, something seemed to be
very wrong. On a day when common
sense would dictate that most →
21
Techniques
% of stocks at or
% of stock
% of stock
less than 2% 20% off or more 30% off or more
from high
of new highs
from high
B ull market
top day
% stocks at new
highs
09/03/1929
2.30%
15.62%
31.84%
18.77%
03/10/1937
6.05%
21.34%
5.94%
1.06%
05/29/1946
8.59%
30.44%
6.30%
0.86%
04/06/1956
5.32%
23.36%
1.92%
0.42%
01/05/1960
1.60%
5.83%
23.25%
7.67%
12/13/1961
3.56%
11.83%
25.29%
11.60%
02/09/1966
9.66%
19.04%
9.52%
2.68%
12/03/1968
9.43%
20.12%
9.51%
2.36%
01/11/1973
5.30%
11.82%
34.22%
20.51%
09 /21/1976
10.97%
22.88%
21.65%
10.09%
04/27/1981
7.09%
15.18%
28.01%
9.39%
08/25/1987
6.23%
15.23%
17.37%
7.44%
07/16/1990
5.35%
18.11%
37.31%
22.74%
01/14/2000
3.54%
6.31%
55.33%
32.45%
AVERAGE
5.98%
16.88%
21.97%
10.54%
Table 1. Examination of trading at 14 peaks in the Dow
stocks should have closed at their alltime highs, it was determined that very
few stocks had closed at, or even near,
their 1929 highs. Indeed, many stocks
were down from their year highs by
20% or more. Thus began a detailed
examination of the trading of
September 3, 1929. The results were
most surprising (see first line of Table
1).
On the day on which the Dow
reached its absolute high for the 1920s
bull market, the percentage of stocks
making new 1929 highs on September
3rd was only 2.3%, or 19 out of a total
of 826 stocks that were traded on the
NYSE that day. Equally surprising, only
15.62% of all issues traded on the
NYSE were either at, or within 2% of
their 1929 highs. In other words, about
84% of all stocks had topped out and
had begun to decline at some time prior
to September 3rd. In fact, on that day
31.84% of the stocks traded had
already declined by 20% from their
22
1929 highs. Thus it became apparent
that the absolute top for the vast majority of stocks had probably occurred
months before September 3rd.
Nevertheless, there had been no single,
outstanding day of rally prior to
September 3rd that investors could
identify as the ideal point at which to
shift portfolios to a more defensive
composition.
The pressing question was whether
the 1929 case was an anomaly or
whether similar conditions would be
found at other important bull market
tops throughout history. Therefore, we
expanded our study to include each of
the fourteen major bull market tops,
based on the Dow, from 1929 through
2000. Our basic assumption was that
most stocks reached their highest prices
in unison with the Dow. However, our
study for each stock traded, comparing
their bull market highs to their closing
prices on the peak days of the Dow,
showed an unexpected picture (see
March/April 2006
Table 1).
Conclusion
These findings defy the conventional
wisdom about the nature of stock market tops. In each case, 11% or less of
stocks (average 5.98%) made new highs
along with the new high in the Dow.
Further, in 9 of the 14 cases covered in
this study, a significant number of
NYSE-listed stocks (average 21.97%)
had already dropped in price by 20% or
more before the Dow had reached its
bull market peak.
The primary conclusion to be drawn
from these fourteen cases is that the
vast majority of NYSE-listed stocks
reached their bull market highs well
before the peak of the Dow. If a portfolio manager had been able to sell out
on the absolute top day of the Dow in
each of the fourteen cases studied here,
the portfolios would have already lost
value in most cases. Investors who may
have thought themselves lucky enough
to sell all of their stocks on the exact
top day of the Dow could have actually suffered significant losses. The similarity of the statistics in these fourteen
cases suggests a pattern of deterioration at major market tops that investors
cannot afford to ignore.
Our study also appears to show that
the Dow is a less than ideal proxy for
the broad list of stocks. For example,
in the case of 1929, none of the 30
component stocks were making new
highs along with the Dow on
September 3. This is due to a large
extent on the reporting of closing
numbers for the Dow on a theoretical
basis. The study also suggests that, even
at that early time in the history of the
30-stock Dow, the price weighting of
the components was producing an
undue influence on the movements of
the Dow.
Paul F. Desmond is president of
Lowry’s Reports, Inc.
(www.lowrysreports.com).
Techniques
A TRADING STRATEGY USING
KELTNER CHANNELS by Jason Leavitt
K
eltner Channels are a volatility-based technical indicator
developed by Chester Keltner,
a grain trader in the US in the 1930s,
who described the bands in his 1960
book, 'How to Make Money in Stocks'.
The channels are most often compared
to Bollinger Bands because they are
both price envelope type indicators, but
Bollinger Bands have always been used
to a far greater extent by traders. This
article revisits Keltner Channels and
looks at the basic trading strategies
offered by the indicator.
The basics
A Keltner Channel consists of a centerline and an upper and lower band with
prices having the greatest probability of
falling within the boundaries of the
outer bands. If prices move outside the
bands, then a trading opportunity
exists.
US trader, Linda Bradford Raschke,
modified the original Keltner Channel
parameters in the early 1980's and it is
her version that is used in today's charting packages. The center line consists
of an exponential moving average
(EMA) with the upper and lower band
plotted as a multiple of the Average
True Range (ATR) from the centre line.
Raschke recommends using a 20-period EMA while plotting the bands
2.5ATR (10) from the centre line. That
means the 10-period ATR is calculated
and the upper and lower bands are plotted 2.5 times this ATR to form the
channel.
Today's charting software permits
users to specify all variables involved
(the EMA, the number of days used to
calculate the ATR, and the multiple the
bands are displaced from the centre
line) and traders can use any time frame
desired (daily, 60-min., 5-min. etc).
Figure 1 shows a simple example of a
Keltner Channel applied to a Nasdaq
chart. The parameters selected (20, 2.5,
10) are the 20-day EMA and a 2.5, 10day ATR. In this example, the 10-day
ATR is 0.56, so the bands are placed
1.40 (2.5 * 0.56) above and below the
centre line.
Many books describe Keltner
Channels as measuring volatility but
this is overly simplistic. The centre line
indicates the trend while the outer
bands are determined by the ATR
which is a measure of the price's intraday range. Bollinger Bands, on the
other hand, are plotted 2 standard deviations from its centre line and are based
on closing prices only. They are a measure of the dispersion of data points
regardless of how big or small the
intraday range is. Keltner Channels
measure intraday volatility while
Bollinger Bands measure volatility →
Figure 1.
March/April 2006
23
Techniques
Figure 2.
of closing prices.
Trading strategies
Charles Keltner used his indicator to
identify the beginning of a trend. He
believed the upper and lower boundaries defined normal price fluctuations
while a close outside a band signaled a
change in character. Because of this,
Keltner said buy when price exceeds
the upper band and sell when it exceeds
the lower band i.e. penetration of a
band implied momentum was assumed
to continue.
Our research shows that simply following this rule is not a profitable trading strategy. However, the most useful
statement that can be made regarding
the outer bands is the following: The
outer bands represent a line in the sand
and there is a high probability something significant occurs there. There is
some debate as to exactly how Keltners
should be traded. Some prefer
Keltner's original breakout strategy
while others do the exact opposite. I
will present both sides of the story and
offer my own opinion.
Simple strategies
First and foremost, I do not believe
Keltners should be used as a primary
24
indicator. Users simply don't get the
divergences attained with other indicators such as the MACD or RSI, and
they don't get explosive moves after a
narrowing of the bands as seen with
Bollinger Bands. There is also no indication of volume which is important to
confirm price moves. As such, Keltners
should be used as a secondary indicator. Secondly, Keltners are best used as
a trend indicator. This should be obvious because the centre line is the 20period EMA.
1. 1) If the channel is flat, the issue is
range-bound, so selling resistance
and buying support would be the
call until proven wrong. This plan
provides many solid risk/reward
trades
2. If the channel trends up, buying
dips (often to the 20-day EMA) and
playing breakouts is the way to go
3. If the channel trends down, shorting is the preferred strategy
4. Don't trade against a trend, so use
Keltners as a trend indicator
A more complex strategy
Bollinger Bands are the true volatility
bands in that they narrow and widen
March/April 2006
with price activity, so penetration of a
band offers a momentum play. But
Keltner Channels don't have the same
characteristics. The bands are a constant distance from the centre line over
the entire chart. I do not find momentum breakout scenarios playing out
with enough regularity to be trusted. A
quick glance through several hundred
charts reveals a great tendency to reenter the bands after penetration, and a
reversal often ensues.
If prices trade above the upper band,
or even close above the band, but then
re-enter and close within the bands, a
short signal is offered. Your stop would
be above the recently made reaction
high and your targets would be 1) the
centre line and 2) the lower band. The
opposite is true for a move below the
lower band and subsequent move back
within the channel (see Figure 2.) This
strategy is very much for more aggressive traders because you're trading
against the trend.
Users are encouraged to experiment
with the Keltner parameters to find
what works for best for his or her own
trading style.
Jason Leavitt is head trader at
Leavitt Brothers.
Asia and Japan Hedge Fund Directory
2005
Qsjodjqbm!Tqpotps;!
Tfdpoebsz!Tqpotps;!
Interview
THE TECHNICAL ANALYST TALKS TO...
Martin J. Pring entered the financial markets
in 1969 and has since become one of the
most respected and internationally-renowned
technical analysts in the industry. He founded Pring Research in 1981 and began providing analysis for financial institutions and
individual investors around the world. Since
1984, he has published the "Intermarket
Review" a monthly market letter offering a
long-term synopsis of the world's major
financial markets. He is also chairman of
Pring Turner Capital Group, a money management firm.
Martin is the author of several important
books on technical analysis, including the
classic Technical Analysis Explained, now in
its fourth edition, and Investment
Psychology Explained. His latest book,
Active Asset Allocation Around the Business
Cycle, will be released by McGraw Hill in
the spring of 2006.
TA: What is the investment remit for Pring Turner Capital
Group and to what extent are you involved in its investment decisions?
MP: I provide the investment strategy input for PTG and
my two partners execute the conclusions plus add their own
ideas into clients' portfolios. Typically this involves identifying where we are in the business cycle and what assets are
appropriate for that particular stage. We are very risk averse
and regard preservation of capital in real terms as our number one objective. When our business cycle and technical
work suggests that the risks are low relative to the potential
reward, that's when we are more aggressive. As a result we
March/April 2006
underperformed most asset managers at the end of the tech
bubble, but early on in the current bull market many of our
clients were at new all-time highs compared to the S&P
which was still down quite a bit from its all-time high.
TA: What are your preferred analytical techniques and
strategies?
MP: I start by trying to identify the duration and magnitude
of the primary trend. This is obvious for any investor but
even traders with a 2-4-week time horizon need to undergo
the same exercise because if a whipsaw breakout is going to
develop it will invariably happen in a contrary direction to
27
Interview
the main trend.
The tools I use depend on the market I am following.
Inevitably I will use my long-term KST in conjunction with
a 12- month or 65-week EMA, provided they have offered
reliable signals in the past. I also look at intermarket relationships, such as stocks vs commodities, commodities vs
bonds etc.
TA: How much value do you place on rule-based techniques such as DeMark?
MP: I look at technical indicators as offering evidence in
the weight of the evidence approach. The more that are
pointing in a certain direction the greater the probability the
trend will reverse. I look upon any approach as another
indicator in the technical arsenal whether it be DeMark,
Elliott, Gann ect ..
TA: Do you think the standard parameters for studies such
as Bollinger Bands and RSI, e.g. SD2/20day & 14day RSI
70/30, are still valid and equally applicable to every market?
MP: Since prices are determined by psychology and human
nature is constant the simple answer is yes. However, since
certain securities are more volatile than others it's usually a
good idea to establish different parameters for them. In
other words the technical principles are the same for a utility stock as it is for gold shares. However, since gold shares
have a high beta they require different overbought/oversold
parameters for, say a ROC. With the RSI this is not so, but
parameters for it need to vary with the time span. For 14days its 70/30 for 9 its 80/20 for 65 its 35/65 and so forth.
Bottom line, use whatever parameters work for you, and for
heaven's sake do not be shy about experimenting.
TA: Perry Kaufman has spoken convincingly on the astrology of the markets, for example the correlation between
phases of the moon and US stock market highs/lows, and
the work of Bill Meridian. How much value do you attach
to such observations?
MP: I believe that prices are determined by psychology and
that changes in the electro magnetic currents in the brain
affect psychology. I suspect that gravitational pull from the
planets affects these personal magnetic currents and I have
noticed that occasionally markets reach major turning
points at the time of eclipses, so I feel there is something
there. However the whole thing is too complex, even for
computers to decipher because there are so many possible
combinations. I looked at this many years ago and discarded
it. However, anyone who thinks there is something in it
should use astrology as one indicator in the weight of the
evidence approach. To use it exclusively is to invite disaster.
TA: Looking at the broad picture, how do you think TA has
March/April 2006
evolved since you first entered the market?
MP: First, we have many more, but not necessarily better,
indicators. Second, TA is being applied to all markets. In the
old days it was just equities. Most important of all, time
frames have shrunk considerably with the advent of on line
trading. This has not meant that people are making more
money but it has increased the use of TA because you have
to trade off the charts if you are an intraday trader, the fundamentals do not change that quickly.
TA: With the plethora of indicators, studies & oscillators
available, how should a trader decide which ones are best to
use? Do some perform better than others?
MP: Read books for guidance and then rely on your own
experimentation for execution. You are the guy with the
money on the line so you need to be confident about the
indicators you use. They may not be the ones that I use but
if they work for you that's half the battle.
TA: Do you think TA is becoming more quantitative in
nature and moving away from the subjective interpretation
of chart patterns/trendlines?
MP: To a certain degree as it has become more sophisticated but that is more a function of the merge between charting and quant analysis that anything inherent in TA.
TA: The use of program trading & algorithmic execution
strategies has increased significantly in the stock markets
(the NYSE's weekly Program Trading Statistics typically
report that between 55 and 60% of all volume is executed
through program trading). What do you consider to be the
impact of program trading on technical analysis patterns
and techniques?
MP: It's all about psychology and human nature more or
less remains constant, so do charts. People will continue to
make the same mistakes and the charts will continue to
trace out the same messages whether program trading is
there or not.
TA: Many traders report that traditional TA-strategies such
as trend following in the FX markets are less profitable than
they once were. Do you think the most established TA
techniques will remain useful or do you subscribe to the
view that market anomalies are constantly opening and closing, and techniques/strategies must constantly adapt to
exploit them?
MP: No, all markets alternate between trending and trading
range. Compare the DJIA in 1966-1983 where it was in a
trading range, to the trendy 1980's and 1990's, or gold
between 1976-1980 and the trading range between 1980 and
29
Interview
2004. Markets will do what they have to fool the majority.
The only exception develops when new instruments or
securities are listed where for a time there may be a great
hedge strategy, but sooner or later someone will find out
and the opportunity is no longer there.
rotation and how this can all be applied with the use of
Exchange Traded Funds. You may ask why the title is not
"Follow the Money Investing". The answer is I did not
think of it until the title was in the publishers catalogue,
when it was obviously too late to do anything about it.
TA: What role do you think technical analysis has to play in
the development of mechanical trading systems?
TA: You are giving a two-day introductory TA training seminar in Mumbai, India in May. What role do you think the
emerging markets such as India and China will play in the
development of trading techniques and TA in particular?
MP: I wrote a book on mechanical trading systems called
"Breaking the Black Box" and I found out that only ones
that worked consistently were based on long-term time
frames and intermarket relationships. Some work very well,
but who has the patience to wait 6-months or a year for the
next signal?
“THE BUSINESS CYCLE
GOES THROUGH A SET
CHRONOLOGICAL SEQUENCE
OF EVENTS… THE MONEY THE
FED INJECTS DURING A
RECESSION FIRST FINDS ITS
WAY TO BONDS, THEN STOCKS
AND FINALLY COMMODITIES.”
TA: Looking ahead, you have a new book coming out in
Spring 2006 - Active Asset Allocation Around the Business
Cycle (McGraw Hill). Can you explain some more about
what this will tell us?
MP: I wanted to call this book "Follow the Money
Investing" because the business cycle goes through a set
chronological sequence of events. Bonds, stocks and commodities do the same thing as the money moves from one
part of the system to another. In other words the money
the Fed injects during a recession first finds its way to
bonds, then stocks and finally commodities. This sets up
business cycle seasons each of which is favourable for a
particular asset class. I call these the six stages because there
are three markets, bonds stocks and commodities and each
has two turning points, a top and a bottom. In most cycles
they repeat in a set chronological sequence. The book
explains what these stages are, how they can be identified
and what asset mixes are best for each stage. As you can
imagine we talk a lot about intermarket relationships, sector
30
MP: Incidentally, one of my sessions in Mumbai will cover
the six stage approach form my latest book. But to answer
your question, perhaps some arbitrage opportunities will
develop but basically the answer to your question is none
whatsoever except that they may make other markets more
liquid and liquid markets tend to be easier to apply technical
analysis.
TA: There's been much talk about a sharp fall in the US
stock markets by the end of this year. Do you go along with
this view?
MP: A prerequisite for a market top is an extended trend of
rising rates. We have certainly seen that at the short end but
not the long end in the US. There is a great correlation
between housing starts and long-term bond yields. Right
now yields are flat and January's housing number was a
cyclical high. In order to make the bear case for equities I
think you need to anticipate a weak economy. Housing
starts are the most leading indicator you can get and if they
are not showing weakness it's difficult to see the market
crashing. From a psychological point of view if everybody
is calling for a drop late this year, and I am not sure that is
the prevailing sentiment, we probably have seen a rally in
the interim that will convince these same people that it will
not happen. When they are all looking up that's the time to
look down.
TA: What one thing, above all else, have you learnt from
your 27 years in the markets?
MP: It's all about psychology.. It's all about psychology…
It's all about psychology… It's all about psychology… I
think you get the point?!
Martin Pring will be giving a two-day introductory TA
training seminar in Mumbai, India at the end of May.
Visit www.technicalanalyst.co.uk/training for further
details.
March/April 2006
presents
The Technical Analyst Conference
Mumbai, India 2006
Taj Lands End Hotel
Wednesday 31st May, 2006
Profitable trading strategies for the financial markets
A one day conference for India’s trading and investment
community featuring some of the world’s leading experts in technical
analysis, trading techniques and behavioural finance. The conference will
present the best strategies to enhance both short and long term
trading returns across all markets.
• Trading and analytical software demonstrations
• Exhibition areas
• Practical masterclass sessions
• Key speaker presentations from internationally renowned analysts,
traders and investment managers, including:
Jefferey Kennedy
Elliott Wave
International
Robin Griffiths
Rathbones
Martin Pring
Author of Technical
Analysis Explained
For further information and to register:
Go to www.ta-conferences.com or
Call +44 (0)20 7833 1441
Sponsored by the UK’s Society of Technical Anlaysts
Subject Matters
DISSECTING THE RSI
by Giorgos Siligardos
I
n his 1978 book "New Concepts in
Technical Trading Systems" J.
Welles Wilder Jr. presented several
methods and trading techniques,
including one of the most widely used
technical oscillators today: The Relative
Strength Index (RSI).
This article shows from a visual point
of view what the RSI actually measures
and demonstrates how the RSI offers a
strictly mathematical way to express
many price movements and patterns. It
offers a useful way to re-visit the RSI
and to understand how we can expect
this indicator to behave under different
market conditions.
The RSI formula
The RSI evaluates the percentage of
the up-moves relative to all price moves
(up and down) for a particular period
(see Box 1). Wilde's original formula
used an exponential moving average to
average out the up and down moves,
but - for the sake of simplicity - let us
first start with an understanding of the
RSI using a simple moving average.
Figure 1 illustrates what the simpleRSI
measures with regard to a 7-period simple smoothing.
The original RSI
Let's now come back to Wilder's RSI
which uses exponential smoothing. The
exponential smoothing gives different
weights to each price movement. The
weights lie between 0 and 1 and vary
gradually from the recent data to the
past data by a geometric progression so
that the more recent a price move is the
more weight it is given. From a visual
point of view this is equivalent to contracting the segments of the line chart
32
The RSI Formula
The RSI evaluates the percentage of the up-moves relative to all moves. Using
a simplified version of Wilde's original RSI formula, the RSI of k periods is
defined as:
where AUC(k) is the (2k-1) exponential moving average of upward price
moves, UC, and ADC(k) is the (2k-1) exponential moving average of downward price moves, DC. UC and DC are always zero or positive numbers and
are defined as follows:
UC: If the closing price is less than the closing price of the day/period
before, then UC = 0. If the closing price is greater than the closing price of
the day/period before, then UC = C - C-1
DC: If the closing price is less than the closing price of the day/period
before, then DC = C-1 - C. If closing price is greater than the closing price of
the day/period before, then DC = 0.
It is clear that UC is designed to track only the distance covered by the
upward movements of the Closing Price while DC is designed to track only
the distance covered by the downward movements of the Closing Price. It is
also easy to see that in all cases:
SimpleRSI
The idea behind the RSI can be shown more clearly if a simple smoothing
method is used instead of the original exponential smoothing. Thus, the
Relative Strength Index of 3 periods using the simple smoothing is therefore:
in such a way that the recent segments
are contracted less than the past ones.
Wilder's RSI then expresses the percentage of the up-moves relative to all
moves (much like the simpleRSI above)
using the contracted segments instead
of the actual segments. Figure 2 provides an illustration of this.
Interpretation
The more time consuming and severe
the corrections of a bullish trend are,
the lower the percentage of the upMarch/April 2006
moves to all moves is and the lower the
RSI is. On the other hand, the more
time consuming and stronger the corrections of a bearish trend are, the
higher the percentage of the up-moves
to all moves is and the higher the RSI is.
It should be clear now that when CP is
the value of the Closing Price of a bar,
the value of the RSI for that bar
depends on the path the daily Closing
Price traced to reach CP in respect to
corrections with more weight given to
the recent price movement.
Subject Matters
Very high RSI values mean that
recently the percent of the distance
covered by the Closing Price during its
up-moves relative to all moves is very
high. Simply stated, the Closing-Price
moved upward with very few and weak
downward corrections or there were
past corrections but the most recent
upward movements were much more
powerful. On the other hand, very low
RSI values mean that recently the
down-moves relative to all moves is
very low. In other words, the ClosingPrice moved downward with very few
and weak upward corrections or there
were past corrections but the most
recent downward movements were
much more powerful.
Based on the analysis above, the
motive behind the categorization of
the RSI as an overbought/oversold
indicator is obvious. A situation where
there are forceful directional price
movements accompanied by few (if
any) and weak corrections is what creates a tensed situation and pushes the
RSI to its extremes. Since the price of
the vast majority of trading vehicles
moves in zigzags, such a tensed situation is often interpreted as a forewarning of a countertrend appeasement
and that is why extreme values of the
RSI are commonly interpreted as a
countertrend signal.
An example
Let's now take a look at a Figure 3 - a
daily chart for ADCT (ADC
TELECOMMUN NE). Below the
price line there is a sub chart with
Wilder's RSI(14) plotted (the 14-period
RSI is the most commonly used RSI
setting). Although the RSI uses exponential smoothing (which means that
all past values are taken into account),
very old RSI values take extremely low
weights and do not contribute much to
very recent RSI values. For example,
99% of the value of a 27-period exponential average of an indicator depends
upon the past 62 values of that indicator.
With this in mind, we will take a look
Figure 1. SimpleRSI The black thick segmented line on the left represents a closing-price line
constructed using 7 hypothetical closing-prices. The line has 3 up-moves (coloured blue) and 3
down-moves (coloured red). A 7-period simpleRSI separates the up-moves from the downmoves and then evaluates the percentage of the up-moves relative to all moves. In this case,
the simpleRSI value would be approximately 65.
at three values of the RSI(14) in conjunction with the price plot contained
in a 62-day window (Note that Wilder's
14-period RSI uses a 27-period exponential smoothing). Point 1 in the plot
of the RSI(14) is at the 46.44 level. The
46.44 value was determined 99% from
the 62 daily closing prices encompassed
by the green brackets but with more
weight given to the prices lying in the
right hand of the brackets. Point 2 is at
the 78.42 level. The 78.42 value was
Figure 2. Wilder's RSI The black thick segmented line in the left represents a hypothetical
closing-price. Wilder's RSI exponentially distorts (contracts) the price moves by contracting the
old moves more than the recent ones. Then it separates the distorted up-moves (coloured
blue) from the distorted down-moves (coloured red) and evaluates the percentage of the distorted up-moves relative to all distorted moves. In this case, the RSI value would be approximately 65.
March/April 2006
33
Subject Matters
of the blue brackets was enough to
push the RSI value at point 3 down to
41.62.
Figure 3. ADCT. Each one of the green, red and blue brackets encompasses price movement
lasting 62 days. The green, red and blue downward arrows show the price points which correspond to points 1, 2, and 3 of the RSI(14) respectively.
Figure 4. A frequently seen feature - a continuation pattern? Values of the RSI(14) at points 2,
4 and 3 depend 99% upon the price data covered by the red, magenta and blue curves respectively. While the price moved sideways from point A to B, the RSI(14) formed a choppy
downward formation from the high at point 2.
determined 99% from the 62 daily closing prices encompassed by the red
brackets but with more weight given to
the prices lying in the right hand of the
brackets. Finally, Point 3 is at the 41.62
level and this value was determined
99% from the 62 daily closing prices
encompassed by the blue brackets but
with more weight given to the prices
lying in the right hand of the brackets.
It is clear that the price movement covered by the red brackets has more "bull-
ish" characteristics than the price
movements covered by both the green
brackets and the blue brackets. Also,
the reason why point 3 is lower than
point 1 is that Wilder's RSI uses exponential smoothing thus giving more
weight to recent price moves. Though
the price decline in the left hand of the
green bracket is severe it does not contribute much to the RSI value at point
1. On the other hand, the almost flat
oscillating price movement to the right
March/April 2006
Interpreting an RSI feature
Since we now know how the values of
the RSI are constructed and what they
represent, we are able to interpret specific movements of the RSI in conjunction with the corresponding movement
of the price. As an example we will
study a case where the RSI moves in
one direction but the price consolidates
and fails to move in the same direction.
Figure 4 shows the daily chart of the
ADCT again along with a subchart of
Wilder's RSI(14). We will explain the
transition of the RSI(14) from point 2
to point 3 so another point (point 4)
has been added between points 2 and 3.
The value of point 4 is based 99%
upon the 62 price data prior to it which
are covered by the magenta curve. The
red and blue curves determine 62-days
price movements too. Points A and B
of the price plot correspond to points
2 and 3 respectively. Notice that as the
price data from point A to point B
unfolds, the 62-day time span curves
cover less and less bullish price movements. The magenta curve encompasses less bullish price movements than
the red curve and more bullish price
movements than the blue curve. The
terminology "bullish price movements"
means either many or large up-moves
accompanied by few and small downmoves or too many small up-moves
and almost absent down-moves. This
transition from a strong bullish situation to a lesser bullish one is what
makes the RSI drop from the high level
of point 2 to the level of point 3.
So what did the RSI indicate by a
choppy fall from point 2 to point 3
accompanied by a price consolidation?
It indicated that some kind of continuation pattern may have formed
between points A and B. One of course
can say this is obvious - the price plot
itself is more-or-less horizontal
between points A and B - but the RSI
gives us a way of quantifying the mag-
35
Subject Matters
Figure 5. Trend Reversal The turquoise eclipses show a case where the RSI(14) moved down
sharply while the closing price withstood the RSI(14) for a significant time. Nevertheless, the
downward movement of the RSI(14) was strong and long enough to finally yank the price
down. The red eclipses show a case where the RSI(14) suddenly sinks sharply from a high level
accompanied by a severe price decline. This usually indicates that the price has more to the
downside especially when the RSI(14) fall goes below the 35 level.
Figure 6. Divergence In this daily chart of Alcoa Inc the cyan filled rectangles show a divergence pattern where the RSI(14) declines while price keeps moving upward. Divergences after
prolonged up trends are prone to produce reversals or at least serious corrections. The red
filled rectangles show a case where the price consolidates while the RSI(14) declines from a
high level, indicating a continuation of the main trend.
nitude of the correction to the previous
bullish trend. Indeed, many technical
analysts interpret a consolidating price
with a choppy fall in the RSI as a bullish continuation signal in itself.
Quantification is what gives the technical trader the opportunity to compare
situations and plan rigid strategies. In
our example the fact that point 3 is at
the 41.62 level gives us the opportunity
36
to compare this situation to other similar situations, which may help us form
an expectation of what may happen
next.
Completions
Not all cases where the RSI drops from
high to low levels indicate that a continuation pattern is forming. Such drops
may occur during the initial phase of a
March/April 2006
downtrend as shown in Figure 5 (see
the red eclipses). In cases like this, the
fall of the RSI will be accompanied by
a significant price decline and it will be
stronger and straighter than the choppy
one which appeared in the previous
example (Figure 4).
Also, the rare occasions when the RSI
drops from high to low levels but the
price keeps moving upward (a phenomenon known as divergence) is different
to cases where the price consolidates.
Divergence is more prone to reverse
the trend rather than continue it especially if it takes place after a prolonged
bullish trend. As an example, in Figure
6 the daily chart of Alcoa Inc is shown
along with a sub chart of RSI(14). The
cyan filled rectangles show an example
of divergence where the RSI(14)
declines but price keeps moving
upward in an oscillating fashion. The
results of the divergence are obvious.
For comparison purposes, the red filled
rectangles show the case where the
price consolidates while the RSI(14)
declines from a high level.
Chart analysis and the RSI
Although some may use the RSI to
indicate overbought/oversold levels in
isolation, this indicator is far more
powerful when used in conjunction
with chart analysis. In fact, interpretation of the RSI / price relationship e.g. divergences and continuation signals - potentially offers many trading
opportunities
for
the
trader.
Fundamental to any such analysis, however, is an in-depth understanding of
how the RSI behaves in different market conditions, which has been the aim
of this article. Only in this way will its
many "weird" habits be revealed and
demystified.
Giorgos Siligardos is a scientific fellow
in the Department of Finance &
Insurance at the Technological
Educational Institute of Crete. His
websites are: www.tem.uoc.gr/~siligard (academic) www.daedalussoft.com
(commercial).
The Technical Analyst is pleased to offer a range of training courses
for institutional traders and investment managers. These one day courses
are designed for more experienced professionals who wish to
expand on their existing knowledge.
Subject Matters
RESEARCH UPDATE:
ON PICKING STOCKS AND FOOTBALL
PLAYERS by Ben Marshall
TA & Small Cap Stocks
In an unpublished working paper entitled "Simple Technical Trading
Strategies: Returns, Risk and Size"
Satyajit Chandrashekar investigates the
profitability of simple technical trading
strategies across firms of varying market capitalisations.
Most empirical technical analysis
studies use stock market index data
which is based on the largest stocks in
the market. However, the theoretical
paper of Blume, Easley, and O'Hara
(1994) finds that technical analysis may
have most value on small stocks
because there is generally greater uncertainty about the prospects of these
stocks and because these stocks are
38
more likely to be affected by private
(inside) information.
Chandrashekar considers the value of
a range of simple moving average rule
across different market capitalisations
using the ten CRSP decile indices from
1963-2002. He finds strong support
for the proposition that technical analysis is more profitable among smaller
stocks. More specifically, moving average strategies earn excess returns of
1.7% per month on average on small
stock indices, but fail to earn returns in
excess of a buy-and-hold strategy for
large stock indices.
Chandrashekar's result holds after
being subjected to numerous robustness checks. It is possible that aggre-
March/April 2006
gate risk factors are an explanation for
the results. The author examined by
adjusting all the results for factors that
have been proposed as aggregate risk
factors such as market, size, book-tomarket, momentum, and liquidity risk.
None of these factors were found to
explain the result.
Chandrashekar also examines
whether time-varying risk premia is
driving the results. It is possible that
the moving average rules generate buy
signals in more risky times for the small
stock indices so the additional returns
earned are simply compensation for the
extra risk incurred. Chandrashekar
finds this is not the case.
The impact of nonsynchronous trad-
Subject Matters
“INDIVIDUALS DO NOT ACT
RATIONALLY, EVEN WHEN LARGE
AMOUNTS OF MONEY
ARE AT STAKE.”
ing is also investigated. It could be that
the stronger predictability across smaller stocks is due to them being traded
infrequently and this generates a spurious positive correlation in the small
stock index returns. Chandrashekar
accounts for this by introducing a lag of
n days after receiving a buy or a sell signal before it is acted on. He finds that
the results still hold after correcting for
nonsynchrounous trading of up to ten
days.
Does money concentrate the
mind?
Cade Massey and Richard Thaler study
the concepts of rational expectations
and market efficiency in a working
paper entitled "The Loser's Curse:
Overconfidence vs. Market Efficiency
in the National Football League Draft."
Rational expectations and market efficiency are two of the most important
pillars of modern finance theory.
Individuals are expected to make unbiased predictions about the future and
markets are assumed to aggregate individual expectations into unbiased estimates
of
fundamental
value.
Acceptance of these two theories
implies the rejection of the proposition
that technical analysis has value.
On the other side of the other argument, research from the field of behavioural finance proposes that investors
make systematic errors and these errors
lead to asset prices diverging from their
fundamental values. Within this framework it is quite possible that technical
analysis does have value.
Tests of rational expectations tend to
focus on laboratory experiments rather
than more realistic settings where there
are large amounts of money at stake,
while tests of market efficiency are
inhibited by the inability to accurately
assess market value.
Massey and Thaler study both these
concepts in a very interesting setting the annual draft of the National
Football League (NFL). In the draft
teams get the right to choose new players in an order based on the team's performance in the prior season (the team
that finished last gets the first draft
pick). In this system a player picked
early in the draft is expected to perform
better than a player picked late in the
draft. The draft system also makes
allowance for teams to trade their right
to pick in a certain position in the draft.
Massey and Thaler use the value
attached to draft pick trades to determine the market value of draft picks.
This is compared to the surplus value
(to the team) of the players chosen with
the draft picks. Surplus value is defined
as the player's performance value - estimated from the labour market for NFL
veterans - less his compensation.
Massey and Thaler find strong evidence that teams overestimate their
ability to discriminate between stars
and flops. They show that early picks
are dramatically overvalued. The surplus value of picks increases during the
first round of the draft. The players
selected with the final pick in the first
March/April 2006
round produce, on average, more surpluses to their team than the first pick
even though they only cost one quarter
the price. Massey and Thaler also note
that this inefficiency in the draft market
is not eliminated by market forces
because the existence of a few smart
teams cannot correct the mis-pricing
through arbitrage. After all, successful
teams do not get given the early picks in
the draft so they are not given the
opportunity to trade them away.
In summary, Massey and Thaler provide strong evidence that individuals do
not act rationally, even when large
amounts of money are at stake.
Ben Marshall is a Senior Lecturer in
the Department of Finance,
Banking and Property, Massey
University,
New
Zealand.
(B.Marshall@Massey.ac.nz). His
research interests include investigating the profitability of technical
analysis techniques, with a focus on
the application of rigorous statistical methodologies.
References
Blume, L., Easley, D., & O'Hara, M. (1994).
Market statistics and technical analysis - the role of
volume. Journal of Finance, 49(1), 153-183.
Chanrashekar, S. (2005). Simple technical trading strategies: Returns, Risk and Size. Working
Paper.
Massey, C & Thaler, R. (2005). The loser's curse:
Overconfidence vs. market efficiency in the National
Football League draft. Working Paper.
39
Software
POULIN-HUGIN
PATTERNS & PREDICTIONS
Poulin-Hugin, an international software firm that specializes in multi-factor analysis and
Bayesian software, addresses the key research and analysis related challenges that face aggressive fund managers, and proposes a new means of addressing them through their "Patterns
and Predictions" software suite.
In developing a unique 'edge', there are
at least six primary research and analysis related challenges that face the forward looking fund manager.
•
Uncovering market inefficiencies
and the windows of opportunity to
exploit them.
•
Identifying, tracking and managing
investment risks.
•
Building and operating a proprietary, multi-factor analytic decision
making model for the firm.
•
The ability to easily test security or
market related assumptions before
making them a part of the firm's
investment
process.
•
Establishing a structure for consistent implementation of the research
and analysis process by the money
manager.
•
Demonstrating to prospective
investors the transparency of the
firm's investment process.
Historically, the larger money management firms have had an easier time
meeting these challenges with in-house
quants teams. Now, however, with the
new Poulin-Hugin Patterns and
Predictions software, even small money
management firms, which make up the
majority of hedge fund managers, have
the potential to more affordably and
effectively address these challenges and
benefit from enhancing their proprietary research capabilities.
Poulin-Hugin's software enables a
fund manager to develop their own
custom model and run analysis using
any variables/factors they wish (e.g.,
stocks, indices, other factors such as
econometric). The money manager can
import historical data and run real time
data feeds (e.g., from P/Es to moving
averages, etc.). In constructing their
model the money manager can easily
aggregate a series of factors, additionally employing personal expertise to rank
and/or give weight to variables in their
model. Of greatest note, model construction with Poulin-Hugin Patterns
and Predictions software enables the
Figure 1.
March/April 2006
41
Software
hedge fund manager who is not an
expert in data modelling to easily visualize the construction and interrelation
of the multiple factors comprising their
trading model.
For example, in predicting the daily
closing price of a commodity (one of
the model templates included with
Patterns and Predictions Professional
software) you can see how our technology may be used to predict the daily
frozen concentrated orange juice
(FCOJ) contract close level, based on
average daily temperatures in more
than 150 different US cities. The application is designed to identify patterns in
the correlation between average daily
temperatures that impact on the level
of the futures contract's closing price
(e.g., the FCOJ price).
The knowledge base in this model
template can be used to make inferences about FCOJ pricing. (That is, the
FCOJ model generates a trading signal
for the level on the FCOJ price for the
next day given certain temperature
ranges. The baseline accuracy is a 60%
probability of being correct on any
given day.). The system can further perform various types of analysis such as
"value of information analysis" (determining which city or group of city temperatures is the most informative when
trying to predict the FCOJ contract)
and scenario based sensitivity analysis
(how a change in one city temperature
affects all others).
In Figure 1, each oval represents a
variable or factor, illustrating progressive complexity. The factor with label
"FCOJ" represents the level of the
FCOJ price, while each of the other
factors represents the average daily
temperature in a certain city.
Further analysis of this model illustrates another powerful feature of
Poulin-Hugin Bayesian 'Hierarchy
modeling'. In this approach, 'new' or
'dark' factors are discovered. In each
case the new factors represent a relationship between the factors below it.
Therefore, we 'discover' the new factors COMODTY, NFACTOR, and
TFACTOR (see Figure 2). These relationships may or may not be accounted
for by publicly available information or
even by generally accepted economic
theory, but appear nevertheless.
•
COMODTY = A 'Commodity' factor that accounts for a variety of
non-FCOJ farm raised commodity
prices. Any number of commodities can be entered to test the relevance of this custom indicator.
•
NFACTOR = A 'New' undefined
factor that "discovered" the relationship between Weather and
Commodity data. This factor
accounts for the unknown similarities in TTfarm raised commodity
prices, including FCOJ.
•
TFACTOR = A trading factor that
accounts for BOTH market volatility and econometric factors. We
would then assume that this factor
was in fact representative of other
active trading also responding to
these sub-factors. It's called TFACTOR because the non-linear reltionship is probably due to behavior
of other automated trading programs that we can't quantify.
FCOJ Model Summary
The FCOJ/Weather model predicts the
FCOJ historical price with a predictable
accuracy. By combining the use of
Value of Information analysis and
Sensitivity 'What-If' analysis, we can
focus in greater detail on the parts of
the model that are more predictive,
thereby improving accuracy. This technique can be used to isolate parts of a
data feed that are worthy of more
attention, and thereby further increasing the accuracy of our models.
Finally, by using hierarchical analysis
we can quantify unknown factors influencing events, perhaps even the trading
patterns of another investment entity.
Through a combination of this software and our professional services
group (a resource for helping hedge
fund managers develop or refine models for their own confidential investment process), any firm can create similar original analysis.
However, those familiar with
advanced modelling might still ask why
they should go with a Bayesian algorithms solution at all. A Bayesian-based
investment analytics approach has
advantages over other algorithmic
methods such as Markov Chain Monte
Carlo, Neural Network and Rule Based
problem solving. Table 1 sets out the
dramatically different capabilities that a
Bayesian system makes available to
financial analysis.
Handling of Uncertainty
Often the connections between different factors - reflected by the rules
defined by user assumptions - are not
absolutely certain. Bayesian-based analytics excel at handling uncertainty.
Understanding of Assumptions
An analyst/expert can understand what
elements correlate with what other elements; something you can't accomplish
Figure 2.
42
March/April 2006
Software
Neural Network
MCMC
Bayesian
Rule Based
Handling of Uncertainty
–
Heterogeneous Modeling
–
–
Provable Probabilities
–
–
–
Understanding of Assumptions
–
–
Analyst/Expert Modification
–
–
Contextual Nodes
–
–
Test Data Independent
–
–
(Markov Chain Monte Carlo)
Table 1.
using a neural network or strict Monte
Carlo.
Analyst/Expert Modification
Probabilities can be assessed using a
combination of theoretical insight,
empiric studies independent of the
constructed system, training and various more or less subjective estimates
such as well known economic
factors/rules.
Provable Probabilities
It can be proved that the method calculates the new probabilities correctly
(e.g., based on the axioms of the classical probability theory).
Contextual Nodes
Neural Network Perceptrones in the
hidden layers only have a meaning in
the context of the functionality of the
model's network. (A neural network
consists of several layers of nodes. All
nodes in a layer are in principle connected to all nodes in the layer just
below. A node along with the in-going
edges belonging to it is called a perceptrone.)
Heterogeneous Modeling
Multiple data types can be combined in
a model.
Test Data Independent
While purely frequentist approaches
(MCMC) require test data to define a
pattern, the Bayesian approach can
measure the data and then calculate the
statistical relevance to the value
observed.
Poulin-Hugin - leaders in the field of
Bayesian-based multi-factor analysis
March/April 2006
modeling software - is an international
software firm whose business divisions
include the U.S. headquartered Poulin
Holdings LLC and Denmark headquartered Hugin Expert A/S. The company's software has been adopted by leading R&D departments in 25 countries
and is used where reasoning under
uncertainty is required. Bayesian-based
multi-factor analysis is particularly useful in areas such as decision analysis,
decision support, prediction, and risk
management.
For further information about
Poulin-Hugin's
Patterns
&
Predictions contact Chris Poulin.
Email: chris@poulinhugin.com.
Phone: +1 617 755 9049. Web:
www.poulinhugin.com
43
Berkeley
LtdAnnounce
announcethe
theLaunch
launchofof
Berkeley Futures
Futures Ltd
• Automated trading
• Trading signals from indicators
• Strategy backtesting
• Strategy optimisation
• Inter-product spreading
• Strategy scripting tool
• Strategy simulation tool
• Position management on signals
• Advanced charting
• Chart trade indicators
• Chart order entry
• Fills and orders visible in charts
• Bracket orders
• Trendline stops
• Ladder order entry
• Order management
• Paper trading
Berkeley Futures Limited has been offering dealing services in Derivatives to
institutions and individuals since 1986.
We deal in Futures, Options, CFDs, Bullion, Forex and Equities for Individuals,
Corporates, Hedge Funds, Introducing Brokers and SIPPs.
Jackson House, 18 Savile Row, London W1S 3PW
For more information on Berkeley IQ-Trader or the services that Berkeley Futures Ltd offer please contact
Marc Quinn on +44 (0)207 758 4777 or by email at mquinn@bfl.co.uk or see our website, www.bfl.co.uk
Berkeley Futures Ltd is authorised and regulated by the Financial Services Authority. Please note that dealing in equities, futures, options, foreign
exchange and CFD’s are all areas of investment in which it is possible to lose money. The risks attached to dealing in off-exchange products such
as foreign exchange and CFD’s differ from those attached to trading in on-exchange products. If you trade in any geared/contingent liability
product it is possible to lose in excess of the funds you may have put in as your initial deposit. Investing in any of the products mentioned may not
be suitable for you and if you are in any doubt you should consult your financial adviser.
Book Review
NEW TRADING SYSTEMS AND METHODS
P
New Trading Systems and
Methods
By Perry J. Kaufman
John Wiley and Sons
1174 pages, £72.25
ISBN 0-471-26847-X
New Trading Systems and
Methods can be purchased from
the Technical Analysis bookshop.
To order please call 01730 233870
and quote "The Technical Analyst
Magazine".
erry Kaufman's book on trading systems is an essential publication for
all traders and investment managers interested in automated and
mechanical trading. Mechanical trading strategies are generally much
more quantitative in nature and, as such, have led to the subject of technical
analysis being expanded into new areas and away from traditional techniques
such as patterns and trendlines. Kaufman's style is succinct and professional
and his highly readable book takes over a thousand pages to outline the various
techniques that can be applied to a mechanical trading strategy. What is more,
he covers various techniques that are seldom mentioned in the plethora of TA
books that regularly appear. These include Fourier and spectral analysis for
cycle identification, regression analysis and as a method for forecasting markets,
system testing and volatility. Furthermore, the author's coverage of traditional
techniques such as point-and-figure, swing trading, Elliott Wave and oscillators
is original and progressive, rather than just being merely a rehash of other
works.
Many of the strategies are presented with TradeStation program codes which
will probably be of most value to the US based user. However, many of the
more statistical techniques can use Excel to generate results which means much
analysis can be carried out without having to rely on specialised software. An
example of a very simple programmable, and easily testable, strategy described
in the book is the N-day breakout. This popular trend following technique says:
BUY when today's high move above the high of the past N days and SELL
when today's low moves below the low of the past N days. Of course, the success of this system depends on the 'correct' choice of N (the number of days,
weeks, months etc…). This will depend on the market being traded, the volatility of prices and the trader's preference for risk. This strategy works best in
volatile markets and is particularly effective in trading the Nasdaq 100.
Backtesting results using N=1 day to N=100 days between 1998 and 2003
showed N=6-8 days and N=14-100 yielded profitable trades.
A sharp reduction in commission rates on trading futures and stocks has lent
added impetus to the growth of mechanical trading. Trades can now be executed more quickly but importantly, more frequently, in order to take advantage of
market volatility and changes in the direction of prices. This is where mechanical trading comes into its own. Detecting short-term trading opportunities
requires techniques that are more quantitative in nature (breakout rules, for
example) which are ideally suited to software that makes binary decisions where
no subjectivity is required (e.g. interpreting head-and-shoulders patterns).
Kaufman's latest publication is probably the best book on technical trading
strategies and techniques available currently available. It also clearly illustrates
the more quantitative direction that technical analysis is taking as an important
component of automated and programmable trading systems. Unlike many
authors, he resists the temptation to make unrealistic promises as to the profitability of the various strategies (especially to those who may be new to trading), stressing time and again that, ultimately, the secret of automated trading
success lies wholly with the trader. A must read.
March/April 2006
45
Commitments of Traders Report
COMMITMENTS OF TRADERS REPORT
8 March 2005 - 7 March 2006
Futures only (open interest) commercial and non-commercial net positions
10-year US Treasury
Source: CBOT
250000
1400000
5-year US Treasury
Source: CBOT
1200000
50000
Non commercial (LHS)
Commercial
Commercial
200000
1200000
0
1000000
150000
1000000
-50000
100000
800000
800000
50000
-100000
0
600000
600000
-150000
-50000
400000
400000
-100000
-200000
200000
-150000
200000
0
-250000
-200000
-250000
08/03/2005
-200000
31/05/2005
23/08/2005
15/11/2005
Dow Jones Industrial Average
20000
07/02/2006
Source: CBOT
35000
Commercial
0
-300000
08/03/2005
31/05/2005
23/08/2005
15/11/2005
Swiss franc
07/02/2006
Source: CME
10000
90000
Commercial
30000
80000
0
15000
25000
70000
-10000
60000
20000
-20000
10000
50000
15000
5000
-30000
40000
-40000
30000
10000
5000
20000
-50000
0
0
10000
-5000
-60000
0
-5000
-10000
08/03/2005
31/05/2005
23/08/2005
15/11/2005
Pound sterling
50000
-10000
-70000
-15000
-80000
08/03/2005
07/02/2006
Source: CME
-20000
31/05/2005
23/08/2005
15/11/2005
Yen
90000
Commercial
-10000
07/02/2006
Source: CME
0
200000
Commercial
80000
40000
180000
-10000
70000
30000
160000
-20000
60000
140000
20000
-30000
50000
120000
10000
40000
-40000
100000
0
30000
80000
-50000
-10000
20000
60000
-60000
-20000
10000
-30000
-40000
08/03/2005
46
40000
-70000
0
-10000
31/05/2005
23/08/2005
15/11/2005
07/02/2006
20000
-80000
08/03/2005
March/April 2006
0
31/05/2005
23/08/2005
15/11/2005
07/02/2006
Commitments of Traders Report
Euro
Source: CME
140000
50000
Commercial
120000
Review of the Commitments of Traders
Report released on March 10th, 2006
40000
100000
30000
80000
60000
20000
40000
10000
20000
0
0
-20000
-10000
-40000
-20000
-60000
-30000
08/03/2005
-80000
31/05/2005
23/08/2005
15/11/2005
Nasdaq
07/02/2006
Source: CME
15000
100000
Commercial
10000
80000
5000
60000
0
40000
-5000
20000
-10000
0
-15000
-20000
-20000
08/03/2005
-40000
31/05/2005
23/08/2005
15/11/2005
Gold
07/02/2006
Source: CEI
The COT Report gives an overall view of the
interrelationship of the markets that just doesn't
come from anywhere else. The current report
suggests a coming weakening of the dollar
(large commercial net buy hedging in the Swiss
franc and Yen), a coming top in US interest
rates (large commercial net buy hedging in US
bond futures), and a fall in the US stock market
(large commercial net sell hedging in the US
stock index futures). I'm looking for an increase
in long-term rates to trigger a stock market
decline that is then fuelled by a flight out of the
dollar. The stock market decline will create
excess liquidity that turns to bonds in a run for
safety, and further weakens the dollar as money
goes home.
Back in November I was looking for a rising US
stock market, falling longer term interest rates, a
weakening dollar and another up leg in commodity prices led by the grains. Then in the
January issue I said those extreme net commercial large positions in the currencies, interest
rates and grains had moved back to neutral.
That suggested we could see retesting of the
lows in these markets but that US stock index
futures were suggesting we were nearing a top
in the stock market.
150000
200000
Commercial
180000
100000
160000
50000
140000
120000
0
Now, rising US interest rates can cause an
aggressive decline in the stock market that
could lead the currency and interest rate markets up a path of a run to safety and some of
that money could run to hard assets setting a
top in the commodity markets and grains.
100000
-50000
80000
60000
-100000
40000
-150000
George Slezak
www.commitmentsoftraders.com
20000
0
08/03/2005
-200000
31/05/2005
23/08/2005
15/11/2005
07/02/2006
March/April 2006
47
Events
EVENTS 2006
The Technical Analyst Conference
Mumbai, India 2006
Taj Lands End Hotel – 31 May 2006
Effective trading strategies
for the financial markets
Date
Event
Venue
May
31
India Conference
2006
Mumbai,
India
August
South Africa
Conference 2006
Johannesburg,
South Africa
September
Middle East
Conference 2006
Dubai,
UEA
October
04/05
Mechanical Trading
Conference 2006
London,
UK
November
The Technical Analyst
Awards 2006
London,
UK
email: events@technicalanalyst.co.uk
telephone:+44 (0) 207 833 1441
website: www.ta-conferences.com
48
March/April 2006
GET QUALIFIED IN TECHNICAL ANALYSIS
The Society of Technical Analysts (STA) represents and accredits
professional and private Technical Analysts operating in the UK
The next STA Diploma
exam date is
20th April, 2006
Originally established in the 1960s, the STA provides its members:
• Education
Monthly lectures and regular teaching courses in technical analysis
• Research
The STA Journal publishes research papers on TA techniques and approaches
• Meetings
Provide members the opportunity to discuss technical approaches and markets
• Representation The STA lobbies on behalf of analysts with data vendors, exchanges and regulators.
The STA represents the UK at the International Federation of Technical Analysts (IFTA)
• Accreditation
The STA Diploma Exam is internationally recognised as a professional level qualification
in Technical Analysis
Come and meet us at our stand at the TA Magazine, European Conference in London on 8th and 9th of February
For more information on how to join and what is involved in passing
the STA Diploma exam, visit our website at: www.sta-uk.org or call
us on +44 7000 710207
E L E C T R O N I C
M E TA L S
OPTIONS
INTEREST RATES
in ‘06
AGRICULTURAL
EQUITIES
METALS
MARKET DATA
Make your next transaction on the CBOT® Metals Complex the premier Gold and Silver trading platform. In one click, you
will discover the cost saving features and enjoy the benefits
our 100% electronic platform delivers. Check it out for yourself.
See it… Click it… Trade It - the future of precious metals.
Contract Features
22-Hour Trading Day
Complete Transparency
Straight-thru-processing
Established Electronic Market
Embedded System Stops
Lower Transaction Fees
Real-time Trade Matching
Trade Certainty
CBOT e-Gold CBOT e-Silver
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
CBOT
Last:
Change:
Gold
BUY ORDERS
QTY
PRICE
10
13
14
116
105
101
15
6
25
3
528.7
+2.4
SELL ORDERS
PRICE
QTY
10
121
119
104
33
15
6
25
1
1
528.7
528.6
528.5
528.4
528.3
528.2
528.1
527.9
527.7
527.5
528.9
529.0
529.1
529.2
529.3
529.4
529.6
529.7
529.8
530.0
For Illustrative Purposes Only
G e t y o u r M a r k e t D a t a s t r a i g h t f r o m t h e s o u r c e . V i s i t : w w w. c b o t . c o m / m a r k e t d a t a
To View Our FREE Live Books Visit: www.cbot.com/metals
The information herein is taken from sources believed to be reliable. However, it is intended
for purposes of information and education only and is not guaranteed by the Chicago Board
of Trade as to accuracy, completeness, nor any trading result, and does not constitute trading
advice or constitute a solicitation of the purchase or sale of any futures or options. The Rules
and Regulations of the Chicago Board of Trade should be consulted as the authoritative
source on all current contract specifications and regulations.
©2006 Board of Trade of the City of Chicago, Inc.
All Rights Reserved
www.cbot.com

High Frequency Trading - The Technical Analyst

Transcription

Similar documents

Wwwwfesgsgdgdggdx By Lin Yingjun The US

Trading floor

Linvest Chart Chat (112) Weekly Update Feb 12

Small Business Development Fund

Upredictable Markets a Certainty

The PDF

PPT - Service Leadership Education

NEVER do these 10 things + 3 Trading Stocks

0809 MARKETPLACE

Classified Advertising