The r,-;~w T.

echnical
Tr~der .
,'1 '

I'

THE NEW TECHNICAL TRAOER

WILEY FINANCE EDITIONS

FINANCIAL STATEMENT ANALYSIS

Martm S. Fridson
OYNAMIC ASSET ALLOCATION
David A. Hammer
INTERMARKET TECHNICAL ANALYSIS
John J. Murphy
INVESTING IN INTANGIBLE ASSETS
Russell L. Parr
FORECASTING FIN ANCIAL MARKETS
Tony Plummer
PORTFOUO MANAGEMENT FORMULAS
Ralph Vince
TRAOlr-.G A'IO 1 VESTING IN BONO OPTIONS
M . Anthony Wong
THE COMPLETE GUIDE TO CO'IV[RTIBLE SECURITIES
W O RLDWIOE

Laura A. Zubalake
MANAG ED FUTURES IN THE INSTITUTIONAL PORTFOLIO
Charles B. Epstein, Editor
ANALYZING ANO FORECASTING FUTURES PRICES
Anthony F. Herbst
CHAOS ANO ORDER IN THE CAPITAL MARKETS
Edg.1r E. Peters
INSIDE THE FINANCIAL FUTURCS MARKETS, 3 RD EDITION
M ark J. Powers and Mark G. Castelino
RELATIVE D IVI DE O YIELD
Anthony E. Spare
SELLlr-.G SHORT

Joseph A Walker
TREASURY OPERATIONS ANO THE FOREIGN EXCHAt-.GE
CHALLENGE
Dimitris N. Chorafas

iv

THE FOREIGN EXCHANGE ANO MONEY MARKETS CUIDE

Julian Walmsley

CORPORATE FINANCIAL RISK MANAGEMENT

Diane B. Wunnicke, David R. Wilson, Brooke Wunnicke

MONEY MANAGEMENT STRATEG IES FOR FUTURES TRADERS

Nauzer J. Balsara

THE MATHEMATICS OF MONEY MANAGEMENT

Ralph Vince

THE NEW TECHNOLOGY OF FINANCIAL MANAGEMENT

Dimitris N. Chorafas

THE NEW
TECHNICAL TRADER
Boost Your Profit by Plugging
into the Latest lndicators

THE DAY TRADER'S MANUAL

W illiam F. Eng

OPTION MARKET MAKING

Allen

J.

Tushar S. Chande and Stanley Kroll

Baird

TRADING FOR A LIVING

Dr. Alexander Elder

CORPORATE FINANCIAL DISTRESS ANO BANKRUPTCY, SECOND

EDITION
Edward l. Altman
FIXED-INCOME ARBITRAGE
M . Anthony Wong

TRADING APPLICATIONS OF JAPANESE CANDLESTICK CHARTING

Gary S. Wagner and Brad L. Matheny

FRACTAL MARKET ANALYSIS: APPLYING CHAOS THEORY TO

INVESTMENT ANO ECONOMICS
Edgar E. Peters

UNDERSTANDING SWAPS
John F. Marshall and Kenneth R. Kapner

GENETIC ALGORITHMS ANO INVESTMENT STRATEG IES

Richard

J.

Bauer, Jr.

THE NEW TECHNICAL TRADER

Tushar S. Chande and Stanley Kroll

@
JOHN WILEY & SONS, INC.
New York Chichester Brisbane Toronto Singapore

To my other cooulhors
Vid)'a, Raw, and Aroon
llus \Cl1 ts pnrued on acad--free p1ptt
Copyn&ht

c.

TC

1994 b) John Wtley 4< Sons. loe.

AU nghts ....,_<d, Pubhshcd "mul11ncously 10 CanadL

ReP<OCiucllon or tnMlluon ofany pan ofthts ..'Ort
be)ond that pcmuu<d b) Soction 107 or 108 ofthe 1976

Urul<d Sutes C'oP)n&ltt Act wuhout tbc pcm111S1on ofthc

copyn&bt o,..ntt as unlav. fuJ Rrqun1s for pcmuwoo or

fiutb<r 111forma11on >hould be tlddr<U<d 10 lhc l'mnU$10M

Dei>anmcnt. John Wtley & Sons. lnc.

This pubhc:.auon 11 c:k\i.pcd 10 prov1de accurate a.nd

aulhontata\lc 1nfonn1uon in rq.i.td 10 tbc sub}CC1 matter
cove~ h u sold Wllh the undcrst1nd1n1 that lhe
publist\cr u not en.p aed 1n ~ndcnn1 lcpl. accounun. or
olher profc.ss1on1J at:rv1cts. l flepl advu.-e or othcr c,;pc:n
ass1st.ance is rcqutttd. thc se.rv1ocs of a compctcnt
profcss1onol person should be $0Ught.

Chande, Tuhar s.. l 9S8The New Technical Trndcr : lloosc Your PTOfic by Plugging
n10 the La1e.1a Jnd1ca1ors / T ushar S. Chlndc aod Stan.ley
Kroll.
p.
cm. - (Wlley ftnancc edil101u)
Jncludes bibtioniphical rcfcrcnccs and l ude.A.
ISBN 0-471-S9780.S
l. lnvcsuncn1 analy11i.-Da1a proctS$1ng. 2. Futurcs
markcc-Da11 proee$$1n. l. Kroll, S11nlcy. JI, Ti~<.

111. Senes.
H04S29 C'4S 1994
332.6'028S--Oc20

Prinl.Cd 1n thc Un11cd Statcs of Amcnca

10 9 8 7 6

4 3 2 1

93-3663
C'IP

Preface

This is a book about new technical indicators. You will find

these indicators useful beca use, although markets ha ve
changed, technical indicators bave not. The analytical "thinkware" has lagged behind trading hardware aod software.
This is an intermedia te-leve! book in technical analysis. We
bave assumed that you have a working knowledge oftrnding,
technical analysis, computers, spreadsheeis, and technical
analysis software.
You can group the indicators in t.his book into two broad
areas: new methods of price analysis and risk control. A brief
description of the new material follows:
Linear regression analysis quantifies trends and projects
prices for developing a game plan.
VIDYA is a variable-lengtb exponcntial moving average
tbat is indexed to volatility or momentum.
Qstick is a quantitative candlestick that gi ves you a number to look at rather than a pattern to ponder.
New momentum oscillators are derivatives oftbe relative
streogth index tbat belp overcome its limitations.
Market thrust is an improvement on the Arms index.
x

PREFACE

Maximum favorable exeursion analyzcs profitability of

your losing trades; it is useful for aggressively managing
new trades.
Volatility-based stop is an advaneing stop from volatility.
Typical trade profile shows price-time evolution oftypical
a trade from your model; it is useful for open trade management.
Price targets are used to formula1e a specific trade plan..
We'IJ eplain our indicators in detail, using tutoriats and
practica! eam ples, giving speci6c rules for trading futures,
indices, stocks, or mutual funds. The last chapter shows how
to combine tbese indicators in to unique and powerful trading
systems. We hope to stimulate your efforts to adapt our new
indicators to your trading style, whieh will give you an aoalytical edge in loday's tough markets.

Acknowledgments

Modern technical analysis sofiware is at thc heart ofthis book.

We want to thank Bill Cruz of Omega Rcsearch, Miami, FL,
for giving us System Writer Plus and SuperCbarts. We liked
tbe power and simplicity of both programs. We also wan1 to
thank Allan McNichol and Equis lnternational, Sal! Lake
City, UT, for a test copy of their Technician software. The
marke t thrust material was developed from tbeir data. Many
ofthe chans were compiled bascd on data supplied by Commodity Systems, loe., of Boca Rato n, FL. We also used da1a
from TechnicaJ Tools, 1ne., of Los Altos, CA, using the.ir Continuous Contraetor sofiwarc.
No book is ever complete without multiple drafts of the
manuscripl We want to thank Nauzer Balsara for his comments on an early version of this text. We want to thank Thom
Hartle, editor of Technical Analysis of Stocks and Commodities magazine, for reviewing an early draft. Jack Hutson,
pubtisber of Stocks & Commodities, gave us permission to
use the anides upon which this book is based.
Our tbanks to Mylcs Thompson, our editor at John Wiley
& Soos, lnc., for his guidanee and support. We also want lo
thank the reviewers at John Wiley for sbaring thcir insights.

xi

Contents

1 An Abundance of lndicators

The Significant Failure Rate oflndicators,

Similarities Among Popular lndicat0rs, 3
New 1ndicators for Price Analysis, 1O
New Ideas for Risk Control, 12
Puuiug It Ali Togelher, 14
2

Linear Regression Analysis

19

How 10 Use Linear Regression Analysis, 19

Case in Point: Tbe Coffee Market and Linear Regrcssion, 25
Case in Point: lntel Corporation and Trendiness witb r2, 29
Case in Point: T-Bond Forecasts for lhe N ext Trading Oay, 34
Developing a Foreeast Oscillator, 40
Summary, 42
3 The Variable lndex Dynamic Average

49

The Simple Moving Average and lts Resp0nsveness, 50

Tracling Strategies, 54
xiii

Contents

xiv

Case in Point: VI DYA ond T-Bond Marke t Analysis, 55

Summary, 62
Tuiorial: Spreadsbcets for VIOYA, 63
4

Qslick: The Quantilalive Candleslick

THE NEW TECHNICAL TRADER

73

The Basics of Candlestick Anal)sis, 73

Qsuck: lntraday Momentum lnd1ca1or, 75
Case 1n Poant: Qsticlc and Crash of 1987, 77
Qsticlc and Momentum, 79
Quanufying Candlesuck Shadows, 82
Case in Po1nt: Analy1Jng Thrce Stocks with Qstick, 85
Summary, 90
Tutorial oo Qsllck, 90
5

New Momentum Oscillators

93

Chand c Momcntum Oscillator, 94

Tutorial: Definiog RSI, 119
Stochastic RSI Oscillator, 124
Variable Length Dynamic Momcntu m lndex, 134
Su mmary, 141
6 Markel Thrusl and Thrust Oscillalor

143

Market Thrust and Thrust Oscillator, 145

Summary, 159
7 Controlling Risk: The Key lo Profilability

161

Estimaling Risk o n New Positio ns, 163

Devclo piag n Trade Templale, 169
Aacipating Priccs for n Rislc Control Plan, 172
Praccal lssues in Rislc Comrol, 175
Controlling " ln vi1ible" Risks, 178
8 How lo Use This Book

181

A CMO-Driven VIDYA Trading Systcm, 18 1

Market Rotatioo, 189
8 ibliography
lndex

199

195

1
An Abundance of
lndicators

There is no shortage of indicators in technical analysis. We

know an indicator is a mathematical formula for analyzing
price action based on prices, or volume, or both. One popular
sofiware package has over 50 built-in indicators. The sheer
number of price-based indicators suggests the question: Are
any redundant? In fact, there are strong similarities between
price-ba.sed indicators, and using them simultaneously crea tes
redundancy.
THE SIGNIFICANT FAILURE RATE OF INOICATORS

Evea lhe bestindicatordoes not work 100 percent ofthe time;

bence, using indicators is a game of percentages. Since each
indicator has a significant failure rate, traders have developed
man y indicators to analyze prices, tbe random nature of price
cbanges being one reason why indicators fail. Therefore, traders use multiple indicators to confirm the signal of one indi.cator with another. Tbey believe that the consensus of indicators is more likely to be correct.
1

An Abundnce of lndicators

Can you rccognize the stock, commodity or index shown

in Figure 1.1 ? FoUowing well-known principies of tcchnical
analysis, note that prices topped at A, and began a downtrend
in January and February. The downtrend accelerated wben
su pport failed at the 26.00 leve!. Prices madc a rounded bottom in late March, bouncLng offB on onc final sclling climax.
Prices then rose in a tight channel to the 26.00 leve!. Prior
support became the new resistance, and prices entered a trading range. The resistance at the 26.00 Leve! is markcd C, and
is roughly at a 50 percenl retracemenl of the decline from A
to B. You can observe the hammer and doji fonnations near
the bouom, the spinning tops in tbe trading range, and the
hanging man formation near 1he top.
Don't be disappointcd ifyou didn't recognize 1he fictitious
price chart in Figure 1. 1. We developed those realistic looking

prices using a random oumber genera1or and a spreadsheel.

Nevertheless, many features sbown there are found on real
price charts, such as support and resistance, ro~nded bottoms,
and 50 percent retracements, although there 1s no reason to
find thosc features oo a chart of randomly generated prices.
Sorne elcments oftechnical analysis ma y work simply by pu re
chance; hence, their failure rate remains uncertain.
We usually do not know the true reasons for price changes.
We could specula1e that prices were overbought, or oversold,
or at resistance, or in support. lt is easy 10 develop a plausible
explanatioo for market action after the fact. But, you must
use objcctive price analysis 10 develop the consistent decision
making necessary to trade with random price movemeo ts.
You should also use strict risk control lo cope with unexpected
price movements.
The ncw material in this book should help your cfforts. Our
new indicators for price analysis and risk control will help
you overcome key limitations of existing ones. Ultima1ely,
tbey could boosl your profitability.

'5.llD

'A~~

411.llD

SIMILARIT IES AMONG POPULAR INDICATORS

'

3SOl)

52% Relracmr.1 from A-8-C

1,

'.DOl)

e
:1500)

'1
tSOCO
10.CXD

ftb

Similarlties Amons Popular lndicators

M"

a
Mo

FIGURE 1.1 A candles1ick char1 of randornly generated prlces.

You can fmd brief descriptions of different indicators in various books or software roanuals, but you wi ll rarely see th e
derivation of these indicators or an analysis that describes
wbat is new or different about an indica1or. We think i1's
reasonablc to ask: What do the Lndicators mean? Hcre we will
examine several popular Lodicators based oo price and show
how similar they are.
Tbese popular indicators measure price momentum in one
form or an other: directional movcment system, momenturo,
relati ve strcngth index (RSI), stochastic oscillator and commodity channel index (CCI). The William's %Risa complement of thc stochastic oscillator; hencc, it has the same Loformation as 1hc stochastic osciUator. The price-rate-of-

An Abundance of lndicators

~~~~~~~~~~~~~.:..;;;..:.;;:.:::.:::.:.:..:::.::....::::...:..:::::.::::~

changc is, by definition, a momentum calculation, and we

will not discuss it separately. The price oscillator, though
smoothed, is also similar to momentum.

Charting the Similarities

You may wish to verify visually thc similarities among lhese
indicators using the price action of Philip Morris (MO) stock.
In Figure 1.2 you can see the 14-day momentum, relative
strcngth index, and stochastic oscillator for MO from Octobcr, .1992 through July, 1993. The three iodicators look alike,
part1cularly oear significan! turning points. Note that momentum has an unbounded scale. On the olher hand, both
RSI and the stochastic oscillator vary from O to 1OO.

Slmilarititt Among Popular lndicators

Figure 1.3 shows the similarity bctwecn the 14-day RSI and
the 14-day plus d.irectional index indicator for MO. Observe
how both indicators peak and bottom together. Figure 1.4
with
compares the 14-<lay commodity channel index
the 14-day stochastic oscillator. T hese two indicators are also
similar in appearance, and their turning points occur at about
the same time.
The CCI is compared with a price oscillator in Figure 1.5.
Tbe price oscillator is !he ditTereoce bctween today's close
aod a 14-day simple moving average of tbe close. The CCI
behaves like this oscillator, lhough it bas ditTerent scaling
factor; both track momentum (see Figure 1.6).
There are other examples ofusing the diJference in moving
averages to measure momentum. For example, notice how
the 14-<lay momentum to the popular moving average con-

cccn

.v...v.
~~,,........~./\.,.....

IDO

2.00
10 m
00

(.::::.;:.:.=-=:.=..::.:::.....-=-~__..:.~~~~~~~~~~~~1-00

00
1-~.:....-=-=:....:..~__:~~..x....:.....:....-l1.:-~-=-1..o<===L-~---l ~m
N

93

FIGU RE 1.2 A comparison o the 14 day momentum, relative

strenglh lndex, and stochastic oscillator for Philip Morris stock.

93

AGURE 1.3 A comparison of the 14.day relative strength index and

the plus directional movement (DX +) for Philip M orris stock.

An Abundnce of lndicators

Similuities Among Popular lndktors

OOl
SOOl

PRIC~

OSCUAi;oA C.A~RAGE(C.!'l

OOl
OOl

l!il OO

moo
SOLIO~: ce~ OASHEO ~ PRICE OSCl.lATOR

o
FIGURE 1.4 A comparison of the 14-day commodity channel index
to the 14-day slow stochastic oscillator for the same data as in Figures
t.2 and t.3.

vergence-divergence (MACO ) ind.icator bottomed together in

Figure 1.7. Tbe MACD is the difference between 12-day and
26-day exponential moving averages of the daily close. As
momentum declines, the diffcrence between two exponenal
averages declines also.
Tbese figures show that indicators deri ved from prices are
similar. Tbe differences in smoothing determine whether they
lead or lag behind other indicators. There also are subtle differences in the definition of these indicators. For example,
not every indicator uses the daily high, low, and close in its
calculations.

93

..

ACURE 1.s A comparison of 14-day commodoty channel index to a

price oscillator for Philip Morris stock.

Correlation Among lndicators

A quantitative way to show the similaritics between indicators is to calculate the stascal correlations between them.
To do this we use linear regression analysis on indicator pairs.
The simil~rity is quanfied using the coefficient of determination (denoted by r1). If two indicators move together, they
are perfectly correlated, and the value of r1 will be LO. 1)ie
value of r' will be O if they move randomly compared with
one another. The higher the value of r2 , the less random the
relaonship between the two indicators.
. .
The statistical correlations between these mdicators was
calculated usLng the Commodity Systems, lnc. (CSI) 39 Perpetual contract for deutsche mark futu~es over a rece~t 1~Q..
day period. We have listed the coefficient of detenrunauon

An Abundnce ol lndicators

New lndlciltors lor Price Analysis

ro aoo

DAS>Eb LJ<E Pllice OSCI.AlllR

SOLIO LtolE. t ..O..Y "40~NTIJM

93

FIGU RE 1.6 A comparison o lhe 14- day momen1um and the price
oscillator in Figure 1.5 far Philtp M orris stock.

FIGURE 1.7 A comparison of the 14-day price momentum to the moving average convergence-dovergence indicator or Philip M orrs s10<.k.

(r') values in Table l. J. All the indicators are correlated, thus

the r2 values approach 1; thcrefore, using these indicators
simultaneously does not provide additional information. The
high correlation is not surpri sing, since thc indicators are
based on daily prices. Subtle differences in their definition
and scaling affect the mcasured correlation.
The purpose of this discussion is to point out the unavoidable similaries between price-based indicators. Each indicator pro vides a different pcrspective on price action. Your
trading style and analytical approach will determine whether
you use one indicator or another. However, using all these
indicators together does not yield additional information.

Table 1. 1 Statislical Correlation Us ng 14 -Oay Calculations for

Deulsche Mark CSI #39 Perpe1ual Contract
lndicato r Pair
Momentum vs. RSI
M omentum vs. Stochastcs
Momentum vs. CMO
RSI vs. S1ochast1cs
RSI vs. Plus D1rec:.1ional lndex
CMO vs. ADX
CMO = Chandc Momentum Oscillator
ADX - Wilder 's Average Directoonal lndex
RSI - Wilden Relative Strength lndex

CSI = Commodoty Systems. lnc.

r - 1, perfect conelatoon
ri - O, no correl.a11on

r'
0.93
0.78
0.93
0.77
0.78
0.82

10

An Abundance of lndicators

NEW INOICATORS FOR PRI CE ANALYSIS

Wh y tben, are new metbods of price analysis necessary? You

~e~ them to give you an edge in tbe markets. Obviously,
mdicators that overcome specific limitations of existing ones
would be highly desirable. In fact, insights from new indicators may be gained without a strong correlation wilh existing indicators. Tbese new indicators faJJ into two groups:
price-based indicators and risk control tools. Most of the indicators can be used on any market, whetber futures commodities, stocks, indices, or mutual funds, altbough' onemarkct thrust-is specificaJJy designed to analyu tbe stock
market. The indicators addrcss man y problems, including improved p_au~m recognition, variable iodicator lengtb, aod
pnce prOJCCUOn. Thus, thesc nexible indicators will fulfilJ a
variety of needs.

L.inear Regression Analysis

Wouldn't you trade bctter if yo u could peek into tbe future?

Unlike mosi technical indicators, linear regression analysis
can develop price forccasts for the next trading day. Tbe forecasts are not meant to provide the precise bigb or low of the
next day; rathcr, they providc sorne guidance for developing
a spe~ilic plan for trading. Tbis method also allows you to
quant1fy the strenglh of the linear trend in the data and is
belpful in following thc long-term trend.
'

Variable Length Moving Averages

You kn?w that moving avcrages use a fixed number of days

of preVJous data. T his is a significant limitarion, since the
most profitable leogth for the average changes without notice.
Wouldn't il be helpful to use a moving average that adjusts

New lndlcators for Price Analysis

11

its " lcngth" automatically, based on price a.ction? Tbe v_ariable length dynamic index (VIDYA) does JUSt that. h is a
modified exponential moving average tbat adapts to market
volatility, incrcasiog its lengtb wh~n prices trad~ in a narrow
rangc, and shortening it wben pnces move rap1dly. ~YA
slows down when prices are quiet, and speeds up when pnces
make their move. You can adjust the resp0nsiveness, or dynamic range, to suit your trading stylc. Tbe dynamic range
is thc range of effective lengtbs tbat VlDYA can use, say, from
J to 30 days. VlOYA is therefore a lcxible moving avera.ge,
which is a significan! improvement on fixed-lengtb movmg
averages.

Qstick: Quantitative Candlestick

Japanese analysts believe that tbe candlestick method of price

pattcrn recognition has prcdictive value, helping the trader
react faster than when using a moving average. However,
intcrprcting patterns is still a subjective process. Qstick extracts the essc ncc of tbe candlestick approacb by taking a
moving average of tbe daJly dilfereoce between the clo~i ng
and the open ing price. Qstick, tbe quantified candlestJck,
givcs you a numbcr to evaluatc, ratber than a p~ttern to ponder, rcducing thc subjectivity in using candlest1.cks.

Momentum Oscillators

The Chande momentum oscillator (CMO) is a pure momentum oscillator. The stochRSl combines tbe p0werfuJ ideas of
relativc strength and a rangc location oscillator. They will
often show pricc extremes that the relative strengtb index
(RSI) will not, hclping to overcome its limitations.
Thc CMO can show oet momenturn at a glance, and can
be combined with VlDYA to forro a dynamic average keyed

12

An Abundance of lndicators

to market momcntum. The stochRSI quickJy shows price extremes and momentum swing failures because it reaches new
lows and higbs faster tban the RSI itself. You thus can combine the two popular ways to use RSJ into a single indicator.
We complete our oscillator group by extending the idea
bebind VIDYA to RSI, defining tbe dynamic momentum index. DM 1 also adjusts its own length using market volatility;
tbus, you do not have to specify lhe number of days in the
calculations. DMI often leads RSI into overbought or oversold territory by many days, a usefuJ featllJ'C most traders
couJd exploit.

Stock Market Thrust a nd Thrust Oscillato r

FinaJJy, we present a new way to anaJyze stock markct advance decline data. Up thrust (down thrust) is the product of
the number of advancing (declining) issues and the up volume
(down volume). Market thrust is the daily difference between
up tbrust and down thrust, and can be cumulated or smoothed
without distortions, a significant improvement on tbe trader's
index or TRl.N. TRJN has an unbounded scale on days with
large volume in declining issues. We ovcrcome this limitation
by defining a thrust oscillator tbat provides a bounded range
for relative volume nows on up and down days. The thrust
oscillator may be visualized as a volume oscillator or an advance/decline oscillator, usefuJ because major market bottoms tend to occur when the 21-day smootbed thrust oscillator is at or below - 0.30.
NEW IDEAS FO R RISK CO NTROL
Along with the new indicators for price anaJysis, we aJso introduce the following ideas for risk control.

Maximum Favorable Excursion

The maximum favorable excursion (MFE) anaJyzes losing
trades from a trading system to hclp manage open trades more
aggressively during the first fcw days of lhe trade. For example, whcn the maximum profit is less than sorne amount,
the stops should be closer than wben the profit is greater than
sorne other amount.

Typica l Trade Profile

Another ncw way to manage an open trade is tbe idea of
typical trade profile, wbicb shows the evolution of prices in
time for trades from a given model. Hcre we analyze daily
equity ofall the tradcs from a model to derive the "signature"
of tradc equity changcs in time. This price-time profile can
be uscd to close out trades that do not follow tbe usual patb.
Thus, you can ca1cb losers early and casb out of fast-rising
winners using thcse profiles. They provide an objective way
to manage the open trades.

Contingency Planning
Contingency planning sbould also be part of a risk control
strategy. We look at the practicaJ issue of developing a gamc
plan, suggcsting a way to projcct tbe possible price range for
the next day using absolute momentum. These numbers then
providc targets for a "what if?" simulation, a proactive approach 10 trade planning. Tbis will help you preplace your
orders and trade more mechanicaJJy. In our discussion we
also touch u pon practicaJ issues of trailing stops, trading tactics, portfolio selection, and asset allocation. Sorne of these
ideas may alrcady be familiar to you.

14

An Abundante ol lndicators

PUTTING IT ALL TOGETHER

s. -

Thcre are many new ideas in this book that you can integrate
into your trading s tyle. The ideas are flexible and powerful,
so you can easily adapt them to your analytical approach and
planning prooess. The more ideas you can integrate, the more
you can boost your trading pro6tability.

Tutorial: Reasons for Similarities Among lndicators

This tutorial will clarify the reasons for thc similarities among
the indicators shown in Figures 1.2 through 1.7. We begin by
defining momentum as thc differeoce betwecn today's close
{day O) and the close x days ago.
momentum -

Co - C,.

(1. 1}

For our purposes we will use x - 14 days. Momentum can

be positive or negative, so wc also define the absolute value
of momcntum.
jmoment11mj -

ICo - c,1

(1.2)

A common practice is to scparate momentum into days

when priccs close up and days when prices close down.
up-day momcntum - Co - e, if Co > e,
- O otherwise
down-day momcntum - e, -Co if Co < e,
- O othcrwise

(l. 3)

This dcfinition givcs positivc numbers for both up-day and

down-day momentum. We can now sum up-day momentum
and down-day momeotum over 14 days.

15

Pulting 11 Ali Togcther

S4

14-day sum ofup-day moroeotum

14-day sum of dowo-day momcntum

(1.4)

We now write RSI using these definitions that ignore the

smoothing scheme in RSI calculations for the sake of simpliciry.
momeotum ~ (S. - S.)
RSI - 100 (S,J(S. + S.))

( 1.5)

This shows that momeotum and RSI are closely related

bccause thcy botb involve the term
Thus, tbe high correlation is 10 be cxpected. The unsmoothed stocbastic oscillator is defined usiog the close, bighest high (HH), and lowest
low (LL) ovcr a 14-day period. It shows where the close is
within its rangc over 14 days.

s..

.
(Co - LL,.)
s1ochas1tc - (HH,. _ LL,.)

(I.6)

The el ose tcnds to be near the higb or Jow of the day wben
markets makc new highs or lows. You can check many price
cbarts to vcrify this observation. Hence, highest bigh can be
replaccd by the bighest close (HC). This is usually a good
approxirriation over tbc calculation period. Similarly, the Iowest low is rcplaced by the lowest close (LC).

.
(Co - LC,.)
stochasuc - (HC,. _ LC,.)

( l. 7)

This now takes on thc appearance of a momentum calculation, where the number of days can vary between today
and x days. Assume that the lowest close was 14 days ago.
The numerator would then be (Co - C,.), a momenlum calcuJation. Thus, we expect to see a broad similarity between

16

An Abundance of lndiC<ltors

stochastic oscillator aod momentum. There will be sorne lags

beca use of lhe smoothing scheme in stochasc calculations.
When a market is moviog strongly up or down, the price
raoge tends 10 lie primarily beyond lhe high or low of the
previous day. Suppose the market makes oew bighs for the
move 14 days in a row. Assume also 1ha1 each day the market
closes at the high. Here the uosmoolhed RSJ - 100 since s.
= O. The plus directiooal movement (DX +) also would be
100 (before smoothiog) since the largest pan of each day's
action (directional movcment) was equal to tbe daily true
range. The bread similarity between RSI and DX + follows
from their de6nitions (see Figure 1.3). Smoothing schemes
and actual de6nitions account for lhe differeoces.
The commodity chaonel index starts by de6ning M, the
mean price of each day, as the average ofthe daily high, low,
and close. Next, a 14 day moving average of tbe mean price
(M...) is used to calculate a devialion D.
D=M - M ,

(1. 8)

A scaling factor is developcd by taking a 14-day moving

average of the absolute deviatioo lD,.). Tbe 14-day CCI is
lheo the ratio o f D and its scaling factor.

CCI

D / (0.0l 5 DJ,.)

(1.9)

Note that the nume rator D determines the sigo aod change
in CCI. We can replace the mean p rice by the daily close C.
This provides an cxcellent approximation more thao 95 perceot of thc time. Hence, the deviation D now becomes a
momentum-like calculation, where we take the difference between today's close and its 14-day simple moving average
(C..):

CCI - (Co - C..)/ (0.015 * kC - C...,h.)

( l.10)

The CCI then reduces 10 a price oscillator, with one-day

Putling lt All Together

17

and 14-day moving averages (see Figure 1.5). Price oscillators

also measure momentum, as you can see in Figures 1.6 and
l.7. There are differeoces io scaling caused by the denominator of each iodicator. Tbere are also differences in the
smoothing process within each indicator. However, we can
expect broad similarities, particularly at key tuming points.
This also explatns the sim1lari1y between momentum and tbe
moving average coovergence-divcrgence (MACO) indicator,
since MACD is a price oscillator using exponential moving
averages.

Tuto rial: Gene ra ting Ra ndo m Prices

We used an Exccl (vcrsion 3.0) spreadsheet with a random
oumber generator to build a price pattem. For our purpose,
we assume that the random number generator is perfect. The
following tutorial illustrates the process of generatiog ten days
of data. You can follow the same process to generate more
days as nceded.
Wc first gcncrate 10 random numbers that are either + J
or - l to signi fy an up day or down day. We use lhe following
rule:
(il\ rand()

>

rand(), 1, - 1)).

The built-in Excel function rand() rctums a random number between O and 1 inclusi ve. This rule generates two diffcrcnt random numbers; if the 6rst number is greater than
thc second, today's price is greater than yesterday's price
(+ 1). O therwise, today's price is lower than yesterday's (- 1).
To start the series, we assume that the close of tbe 6rst day
is 40.00. This eould be lhe price of a stock or the price of a
commodity in cents per pound or cents per gallon. We also
assume tbat today's high or low c.an be no more than 2 cents
above or below the close. We thus generate a random number

18

An Abundance of lndicators

between O and 2 and add it to the closc to get the higb. Similarly, we generate anotber random number between O and 2
and subtract it from the close to get the low. Tbc rules are as
follows:
today's higb - today's close + rand( ) * 2
today's low - today's close - rand() 2
t.oday's close - yesterday's close + rand( )*2*( + 1 or -1)
Tbe Excel function rand( )*2 generales a random number
berween O and 2 inclusive. For day 2, we generate a random
number betweeo O aod 2, muJply it by either + 1 or - 1 as
determined before, and add it to the previous close. Next, we
find two more random numbers to compute the bigb and low.
We continue the process for ten periods. You can add
rand( )*3 to increase tbe range of price action.

2
Linear Regression
Analysis

Linear regression analysis is a weU known metbod of data

analysis. However, not ali tccbnical analysts use it routincly.
Thcrefore, we will approach tbis material from a user's point
of vicw, rather than attempting a mathematically rigorous
application of the linear regression process. The technical analyst must cope with unccnainties using every indicator that
is available, and any impcrfcctions in the application of tbis
method m ust be weigbed against the usefulness of the resultan t information. You'll sec that there are several ways to
use the resullS that will add value to your analysis.

HOW TO USE LI NEA R REGRESSIO N ANALYSIS

You can visually determine from a bar chan if prices are
trending, and use the mathematics of the linear regression
method to lit lhe "best" line to a series of prices. This method
uses the formula of least squares to find the best fitting line.
This calculation gjves us the slope and intercept of the "bestfit" line, as well as tbe strength of the linear trend. In tbis
19

20

Lineu Regresson Analysis

chapter, you'll discover how to trade with the slope and trend
strength.
Using !he equatioo of the best-fit line, you can estimate
values of the possible prices for the oext trading day. The idea
is not to predict the precise high or low for the next day,
although you could occasionalJy come quite close; it is instead
to have price targets that can be used to develop a game plan
for trading. You can tbeo trade objectively within the heat of
banle.

The Linear Regression Method

The linear regressioo method solves the following equation:
y -

X+ C.

(2. 1)

Here x is the indepeodent variable, y is the dependent variable, m is the slope, aod C is a constant intercept. You can
imagine a plot along the x and y axes of the two variables.
This equation describes their relationship in a quantitative
form.
The output of the regression calculations gives values for
m and C. We also get !he coefficient of determina ti o o, denoted
by r2 Refer to the tutorial at the end of this chapter to sce
why r2 measures the relative trend strength.
We like to use five days of closing data for short-term trad
ing using regression aoaJysis. You may wish, however, to experiment with the number of days used in the calcuJations,
as well as try the daily high and low prices for mak:ing fore
casts.

Samp/e Calculations
Let's look at a sample calcuJation using the closing pric.es of
a recent gold futures contract Assuming time as the inde-

21

H ow To Use Linear Regresslon Anal)'11s

pendent variable and price as the dependent variable, entcr

thc valucs in a spreadsbeet. {Refer to Table 2. l during thc
following discussion.)
In Table 2.1, we want to fit a straight line to five days of
daily closes of the gold contrae!. Hence, under x, the inde
pendent variable (column A), is simply the days from 1
through 5. The depcndent variable shown as y (column B),
is the daily close of the gold futuros contract
We will square each of the daily variable values and write
them in the next two coturnos (C and O). Thus, the line for
day 5 shows that the square of 5 is 25 (5 X 5) and the square
of 388.20 is 150,699.20. The last column, E, is the product
of each pair of dependent and independent variables.
The value for day 5 is 1,941 (S X 388.2). The sum of eacb
of the five daily values is calcuJated in ils respective column.
In Table 2. 1, we define three other terms used in the calTable 2.1

B
y

x-squared

D
y-squared

E
xy

1.00
2.00
3.00
4.00
5.00

378.10
376.10
379.50
379.20
388.20

1.00
4.00
9.00
16.00
25.00

142959.61
141451.21
144020.25
143792.64
150699.24

378.10
752.20
1138.50
1516.80
1941 .00

15.00

1901.10

55.00

722922.95

5726.60

A
X

Sum
n q, q, q, -

Linear Regression Calculations Using Daily Clase of

August, 1993 Gold Contrae!

number ol cJay~ In 1he cakulations - 5

sn -

!5726.6 - !05 1901. 1

23.3
(55
({ 15 IS)/ 5)) - 10
!722922.95 - <1 1901 1 190 1. n Sll - 86.7 1

slope -

q, / q, 0. 1 (5726 6

(3 - 1901 1)) - 2.33

m1e"ep1 - ((0.2 1901 1) - (J 2.JJ)) - 373.23

coefc1en1 ol de1erm1na1ion (r') - (q, q,) / (q, q,) - 0.626
day 6 forecas1 dose - ((6 2.H) -'- 373.23) - 38 7.21
aciual close
)86.70

22

Linear Regression Analysis

culations, q,, Qz, and q,, which are computed using rhe totals
in the various columns and the number of data points. You
can find the slope and intercept of the 5-day regression from
the following equations:
slope(5 day) = 0.1 *(sum of producl - 3*sum of y values),
(2.2)
intercept = 0.2*sum of y values - 3*slope.

In Table 2. 1 the sum of the product is in column E and

the sum of y values is in column B. We can also lind the slope
using q' and qz. We get the forecast by substituting the slope
and the intercept into the regression equation and using the
value 6 for x, the independent variable. The coefficient of
determination, r2 , was 0.626, showinga statistically signilicant
trend. Note that the 5-day trend was up and the best-lii line
was rising $2.33 per day over the calculation period.

23

How To Use Linear Regression Analysis

390
388
386
384
382
380
378
376
374
372
370

Linear regression "best Ht" line

O"

... .. .

go

....
...~--"

.. .:.:--~-----"'
God futuras dallv close

o.,..
FIGURE 2.1 An example of linear regression calculations using the
daily dose of the August, 1993 Comex Gold futures contract. The
solid line j oins the daily closing prices.

Graphs of the Output

Forecast

In Figure 2. 1, note how the dotted best-lit line smooths out

the variations in daily data, showing the upward trend in

prices. In Figure 2.2 we step the regression line one day forward to find the regression forecasi. In our example, prices
retreated to close sligh tly below the day-6 forecast: at 386.70
instead of 387.21. The forecast is simply a " point estima te"
ofthe daily close. We can use variation in the data to calculate
a range ofvalues for the closing price. Then actual close could
le within a band above and below tbe poinl estimate. Neverlheless, the point estmate is usualJy sufficient to develop a
game plan for irading.

....

390
Unea1 regrNsion

\...

385

380

....

~best fit" lln~

.... ...

.. ~

'

375

Gold furures daitv c&ose

3
Oaya

FI GURE 2.2 The day-6 closing price forecast using the " best-fit" linear regression line.

Outputs: Slope and r'

Tbe fi.rst output of a linear regression anal ysis is the slope of

the trend line. The slope is positive when prices are rising

24

linear Regression Anafysis

and negative wben tbey are fa!Jing. Tbe slope measures tbe
cxpected change in price per uni1 of time, wbich, wbeo coovcrted into dollars per cootract or share, iodicales iftbe m arkct is making big or small moves. The slope cooverted into
dollars is a useful filler for trading or ignoring markets.
Linear regression aoalysis also gj ves us 1he strenglh of tbe
linear relationsbip, denoted by r2, the coefficieol of determination. lt is a quick measure of 1he trendiness in the data,
ranging from O to l. Ifthere is no trend, that is, random price
ac tion , the r2 value will be close 10 O. A perfect linear trend
givcs a value of l. Figure 2.3 shows two simulated sets of
data: a random set and another with an exact linear trend.
Thc r2 val ue is close to O for th e random set, and is 1.0 for
the cxact linear treod.

Using r
Thc primary use of r2 is as a confinning indicator. It is a
lagging indicator that shows the s1rengtb of tbe trend. The
critica! val ue ofr2 depends on the number of days in tbe data:
whcn the r2 value is greater than the critica! value, a statis-

12

10

,,. ...... ......

Cot1ffieltni of ~tttrmlna tlon 1 .0

......

---

......

_...
'

',

' '

2
o ~~~~~~~~~~~~~~~~~~~~~--'

1>..,.

FIGU RE 2.3 Simulated data showing lrending and trendless data.

Tht Coffee Marktt and Untar Rtgr~uion

25

tica!Jy significant trend exis1s. For a 10-, 20-, 30- or 50-day

regression, the critica! values you can use are 0.40, 0.20, 0.13,
and 0.08 respectively. (You can find details of tbese calculations in the tutorial section on linear regression later in thi s
cbap1er.) The slope ofthe regrcssion will tell you the direction
of 1be trend. You can then put on posi1ions witb the crend if
you wisb.

Using Both Slop e and r

You should use the slope a nd thc r2 valucs togetber. For cxamplc, a strong trend with a sma!J slope may not interese the
short-tcnn trader. A moderate vaJue of tbe slope, with only
a wcak trend in the data, may be a warning that the trend is
chaogjng. Hence, the slope and the strength of thc regression
are valuable inputs in to a trading model High values of the
sJope occur when thc markct is trcnding strongly. As the trend
weakens, both the slope and r2 wiJJ bead toward O.
lo ordcr to trade with the sigoificant trend, simpJy check
if the slope is positivc or negati ve. You could open a long
position when the slope firsl becomes sigoificantly positive.
Conversely, you couJd open a short position when the slope
is significantly negative. You also wiJJ then see the slope
make a peak (positive or negative) and tum 1oward O. At this
point, you may choose 10 take profits, tigbten your stops, or
take an antitTend position. The market may take a long and
slow trip back to O, in whi ch case, an antitrend position would
be uoprofitable. Hence, therc is a variety of trend-following
a nd antitrending strategies possible using linear regrcssion
analysis.

CASE IN POI NT: THE COFFEE MARKET

ANO LI NEAR REGRESSION

We can illustrate these ideas using the Commodity Systcms,

!ne., (CSI) Perpetua! contract for the coffee futures marke t.

26

Linear Regression Analyss

Figure 2.4 gi ves an overview of the #39 Perpctual contract

daily close for coffee. Coffee bottomed in August and September, 1992, and rose steadily into the middle ofDecernber,
1992. Tbe coffee market was cboppy as it formed a top, lben
collapsed in late January. Co1fee rebounded strongly from an
oversold condition, fell again, and then entered a brief sideways period.
We begin by arbitrarily choosing 15 days as the lime period
for linear regression calculations. Figure 2.5 shows the slope
of the coffee data from 08/ 03/ 92 to 03/ 23/ 93. Overlaid on
the slope is the coeflicient of determination, r 2 You will notice
that r2 approaches O as the slope changes sigo. Notice lbat
high values of slope coexist wit.h high values ofr2: this is often
true for trending markets. When the trend is changing, tbe
slope is also changi.ng signs, and r2 approaches O.
Since there are trends witbin trends, tbe lengtb of your
regression calculations will inJ1 uence wbat yo u see. The si o pes
for lbe coffee contrae! over 7-day and J5--0ay intervals are
sbown in Figure 2.6. The volatility of tbe colfee market can
be seen by the variations in the slope of the 7-day regression.

90.000 . . . . - - - - - - - - - - - - - - - - - - - - - - - - .

The Cofee Market and linear Regression

27

1 6-0t't' SIOl)4

1.5 '-----~-------------------'
08/03/92 0 8131/92 09129/92 10127192 11124/92 12/24/92 0 1/25/93 0 2123/93 0 3123/93

FIGURE 2.5 The 15-day slope and r' from a linear regression analyss
of the CSI data in Figure 2.4 for coffee.

2-r-----:-- - ----------.----.-----.
7-daJ slope

1.5

'{

0.5 .

80.000
0 .5
70 .000
60.000 .
50.000 .
40.000 ...__ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _.._

FIGURE 2.4 The daily close of the Commodity Systems, lnc. #39
Perpe1ual contract for the coffee futures prices.

.,

. , .s

1S-ctay slope

2 .__ _ _ _ _ _ _ _ _ _ _ _ _ ___,_ _ _ _ ___,

08/0319 2 0 813, 192 09/29192 10 127192 , 1/24/92 12/24/92 0 1125/93 02/23193 03/23/93

FIGURE 2.6 A comparison of the 15-day and 7-day slopes for the
coffee data in Figure 2.4.

28

linear Regression Analysis

The 15-day regression slope is less volatile, showing lhat the

regression process smooths thc data considerably as the length
of the regression increases. Thc shorter regression changed
di rections and sigos more often, producing more trades. However, the amplitude of the move, and hence the profils, were
not the same for each change in direction. Thus, your choice
ofthe length ofthe regression will influence both your trading
frequency and profits. In volatile markcts, a shorter length is
more effecti ve.
ln ou.r coffee market example, onc simple trading strategy
could be based on tbe observation that high values of slope
coexist with bigb values of r1. You could go long when the
slope is positive aod r2 is above 0.20, showiog a statistically
significant tren<i You could close thc long whcn the slope was
no longer statistically signjficant (r1 bclow 0.20). This approacb would take you long at the end of September, and
close you out in the dip in mid-Novembcr. You would go
long agaio nine days later, and close out the long at the end
of December. Then, you would go short in the first week of
Jaouary, and close out the short in mid-February. Thus, you
could catch the major moves with this strategy.
Figure 2.6 also shows how a countcrtrend strategy may be
carried out. Usiog the 7-day slopc, values above + l and below - 1 signa! market extremes. You should be prepared to
take an opposite position as the slopc cxcceds these limits.
Then, you could wait for the slopc to turn down or up before
placing your trades.
You can experiment witb regressions of different lengths.
You also should test the models ovcr four to five years of
data to obtain reliable estmates of overbought and oversold
regions for a countertrend strategy. Thcrc is no assurance lhat
future prices will follow precisely the same pattern. Hence,
you may wish to use two or more regrcssioo leogths in your
analysis.

lnltl Corporation and Trendinen wilh r'

29

CASE IN PO INT: INTEL CORPORATIO N

ANO TRENDINESS WITH r1
The linear rcgression r1 is an altemati ve to other measures
oftrendiness such as the average directional index (ADX) or
the vertical homontal filler (V HF). Let's compare these indicators using Metastock software vcrsion 3.5.
We will use daily data for lntcl Corporation (lNTC) from
01/01/91 to 12/3 1/92 t~ make our comparison. This time
span includes periods when lntcl showed many different price
patterns: a broad trading band, and strong up- and downtrends. Here are the Mctastock formulas for smoothed r>and
VHF. (Picase refer to J. Welles Wilder's book in the Bibliography for a detailed discussion of ADX.)
smoolhed r2 ~ mov(pwr(corr(cum(l ),c,14,0),2)*100,14,s)
smoothed VHF - IOO*(mov(vh.ftc,14),14.s))
(2.3)
The ADX, smoothed VHF and smoothed r2 are in broad
agreement. ln Figure 2.7 we show a smoothed r' lhat is a 14day simple moviog average of 14-day r1. Low values imply
an imminent change in trend; high va.lues show price extremes. The low values in 1991 led to tradeable trends in
In tel. Observe how the smoothcd ri peaked i.n O1/92 before
thc actual high in February. This indicator sbowed a falling
trend (loss of momentum) evcn as prices moved upslowly in
October, 1992.
In Figure 2.8 we have a smoothed VHF, lhat is, a 14-day
simple moving average of a 14-clay VHF. We multiplied the
smoothed VHF by 100 to produce numbcrs on the same scale
(O to 100) as the other two indicators. Observe the broad
similarities to the smoothed r1 shown in Figure 2. 7. The
smoothed r2 declines to lower values than the smoothed VHF.
Then, in Figure 2.9, wc have the 14-clay ADX showing thc
trend. Observe the similarities to Figures 2. 7 and 2.8. Thc
ADX stayed flat for most of this period except far the yearend rallies. Notice how the ADX declined in October-

: \: ' ' f': - ~ : /l . . : - ; :J\

' ~ : f;\: fr : : / ; ~\rt ;'(\'. ..N:'Vr \.,
1

, -/'J _

. ....

! ~ iT

cv.

(\14

- '<t -. . ....' ~.-~

11,f'+ 1,:t'/'....'"''
...y1.
; V-''.

.. , " .. r--- ,........

1 -

>

,.

_J
: .

, (

,. ':
~ ~

1\ ..,.

- . \...) /JtJ . ~
- '
'\ . '.
+,. :
! ... .,;,,_,,:.,, ..;. . . ti ... ' . .

.l.. -

....
~
..
"'
,.
'J,.. .."'
....

'.

., ,

.. .':;
;::;
0
:.
:
;
i,,,r'
. ! - :;
' : . f

1.

'

:~." ~
1 1
~ r";~~.,.,~:"

,. .. , ,.

ri

'91

11

t '1~ ; ,._. ,

l'I

l'I

.i ;

. : :

"

~
;
= .. . . ~ ~ . :
,

'
l'I

"

l'I

,.

,,_

.."'

l'I

..,.
....
....

.)

FIGURE 2.7 A 14-day sirnple rnoving Average of lhe 14-day r2 used

as a measure of trendiness in the price o f lntel stock.

lntel Corporation and Trendiness with r1

..
....

..
'
'

-~

.,

E L
. - ~ ...
;)O

20

'

'

10

!OC

'

7 11

... '

+- "

_,.,

"

'

.1
'

,.

'

'

..

'

'

... . . . ' ... L..

..

31

,,

'

..,... .

+ '!

............. '
.i

.,

~
r
~I\.'. . ' - . . ~ .

: . ~\/~ : . ::::.: : ~~;::

~

")!:!

'

'1

1'11'1"

:u"\1x- '

JI

1'1

SO

'

''

'

w0

' 92 f

..,_, '"'

nl\.ltJ

''

l\SO

NO

FIGURE 2.9 The 14-day average directional index used as a rneasure

of trendiness in lntel stock .

to

l ,.

eo
,..,

'

'

, r
....
.,... .,.

..

.,

: ..:.::.L...~j::::-:::.;.-~:.i :1:: _:.:~

9'tl

.. .

.. .

"t .... ...

f ~

:: .: : : .
..

.. :

40

. 'l{.

'

'

t 1

l'I

'

'

1/': ....
~~~'(!
...

"'

"

"'

- .. .

r;,..: >;,u:;

! "... .. i ... ,.. .. '- -. . ' ' '. ~

--

- '1~
~

,;.; .

,,
. ..

..

i-

-- . . . - .. -

. <
-~ ,,,,,

, ., . , i ;

:,. : :1:

... ~.

! ..

"V., ftl""
o

,.

o . ., r

' '

'

"

"'

"

'

'

.1

,.

''

"

FIGURE 2.8 The 14 -day simple rnoving average o f the 14-day vertical
hori zontal filter used as a measure o f trendiness in lntel stock.
30

Decerober, 1992, even as prices moved slowly higher. The

ADX had smaller swings than eilher the srnoolhed r' or the
smoothed VHF.
A more direct comparison between the smoothed r2 and
the smoothed VHF is shown in Figure 2.1 O; a comparison
berweeo smoothed r2 and ADX is shown in Figure 2.11. Notice that the cwo indicators move in unison at key tuming
points.
Ali three measures oftrendiness roove upas trends emerge.
Hence, they move up even when prices are falli.ng. They also
usually decline during tTendJess periods. The smoothed r' and
smoothed VHF move in unison at key tuming points. But,
the srooothed r' is more sensitive than the ADX, oft.en staying
at higber levels even as the ADX declines. This sensitivity
demonstra tes the con.trast, since a declining ADX implies no
trend, wbereas high values of smoothed r' shows a persisten.t
tren d.

.'

-......

o..,... u - ..,_,......,

~,..,,

, _ _ l- 1 -.o.. 1no . .......,.

"-

of

,_...., .-. .

I~..

tol\lf 11- i.-.;1.., . , . . , . _,....,

,,,.,.....,~

,.

,,..
::'.

....
..
........
..
'

"'

'"

1:

.'''

: ,!

20

'

,.

..

u
.

1 '

...:

'

s1

t1

.s

"'

FIGURE 2.10

....
....,.
..,...
..
........

"'

l'I

o n ,. "

"

"

'

.J

"

li

The superimposed smoothed r2 and smoothed VHF for

lntel.

..
....,.
,.
....
......"
......
..
~

folid , . .,..

~--.ci...,

__ ............

O
- u,,.,
- '""'
I~ ...l ... lo -o"lf'>O

..
:

""

. '.
'.

,.:

'
'
'

r:

:
1

~; :

"

'

' ....

'

' J'
..

v
'

i(

"
r ;

:: '. : :
: '' i :

....'.
'

'

-t i

ti

11

ti

"'

.o!

,_.,

,.

11

"

ti

.J

/1

..'

.."'

....
..
20

'

'{

...."'
..

" ,."'

! '

~~

.:.~~~..

-f'~

...."?,

,'!

""'"'""",."e1

FIGURE 2.11

The superimposed smoothed r' and average directional

index for lntel.
32

lntel Corporation and Trendiness with r'

33

A Detailed Comparison: r2 , ADX, and VHF

In 199 l , lntel was in a tracling range between 38 and 60. The

trading range narrowcd to approxiniately 40-45 during most
of October through Deccmber of 1991. lt broke out of this
range in the rally sparked by interest rate cuts on December
20 that year.
The l4-day ADX stayed llat below 25 from April through
September (see Figure 2. 9). It rallied briel y that month during
a short sell-off, and quieted down again during the tight trading range. lt rose with the rally frorn December, l 99 l through
February, l 992. It then declined and stayed range-bound for
the rest of 1992, even as Intel moved steadily upward, only
responcling to the breakout above 70 in December.
Thus, the ADX rises when there are strong movcs in the
stock. It often declines during a tradeable downward move
in priccs, and it does not react well when prices rise unevenly.
For example, the 14-day ADX declined in October and No
vember, 1992, while prices rose choppily from 62 to 70.
The smoothed r' is more responsive to price changes, as
you can see in Figure 2. l I. Note how it moved up faster aod
soooer than the equivalent ADX in November aod December, 1991, and again in mid-October, 1992. H peaked simultaneously with the ADX, pegging the imminenl trend change
in June, 1993, as Intel neared 55, and sold off sharply into
July. lt was a tradeable short from 55 lo about 42. From this
you can see, the smoothed r2 has inleresting properties as a
measure of trcndiness.
Adam White, a contributing editor to the Technical Traders
Bulletin, proposed the vertical horizontal fter (VHF) to mea
sure trendiness. VHF compares the range between the high
and the low to the average absolute daily momentum. White
does not explicitly discuss using in a smoothed form. VHF
is usually uscd with 28 days in the calculations. Howcver, we
find it more interesng to use a sborter time period witb
smoothing to isolate the underlying trend.
Tbe smoothed 14-day VHF is similar to thc smoothed r1

T-Bond

(see Figure 2. 1O). Both can go up e ven as prices move down.

Smce you can use the slope for direction, thc combination of
slope and smoothed r 2 provides both the direction and
strength of Lhc trend. The smoothed r1 shows an imminent
trend change by falling below O. l. This happens eveo wben
the smoothed VHF remains at high values, giving the
smoothed r2 an edge.

A Common Weakness

Ali three measures of trendiness show a common weakness.

After prices make a quick move (up or down) and then briefly
reverse direction, the trendiness measures begin to decline,
and will kcep fa!Jing (suggesting no trend) e ven if prices la ter
resume their initiaJ directioo. T bus, tbey often show a loss
of momentum rather than an actual change in tbe major
treod. The main reason for tbis weakness is that we are using
a fixed number of days for these calculations. We get a diffcrent view of tbe underlying trend as we change tbe period
of the calculations. So, tbere is sorne ambiguity about determining trendiness, since tbe number of days in thc caJculatioos influences the resuJts.
CASE IN POINT: T-BOND FO RECA STS
FO R THE NEXT TRA DING D AY

We can use Linear regression anaJysis to develop a regression

forccast for the next trading day. First, we dcvelop linear
regressions separately for the bigh, low, and close. Then, we
substitute thc values of slope and interccpt into the linear
regression equation and caJculate the oew vaJuc for the next
tradiog day. Coosequently, we can forccast tbe bigh, low, and
close. Statisticians call tbis approacb thc point forecast. We
aJso can develop a confidcnce interval for the forccast bigh,
Iow, and close. This intervaJ is the rangc within wbicb we can

For~cuts

for the Next Trading Oay

35

expcct the acruaJ values to faJJ with, say, 95 percent confidence.

Strategies
We don't cxpcct to forecast precisely the actual bigh, low, or
close. The purpose of the forecasts is to develop a range of
cxpccted prices for the next trading day, and then to use the
range to chalk out a trading plan for specific contingencies.
You can plan a variety of actions with the forecast, such as
sctting stops, initiating new positions, or closing old ones.
You can estmate your risk and reward, and write orders to
takc profits, cut losses, or selecti vely add or reduce existing
positions.

A T-Bond Spreadsheet

We will use the TREND function from Microsoft Excel 10

illustrate tbe use offorecasts. Table 2.2 shows the daily bigb,
low, and close ofthe September, 1993 Treasury Bond contract
from 01/27/1993 through 02/17/1993. The rows are ourobcrcd 1 to 16 starting at tbe top, and the columns are Iabeled
A through G starting at the lcft; the label DATE occurs in. tbe
ccll designated A l.
We used the built-in TRENO fuoction in Excel to caJculate
the 5-day forecast. Please notice that we insened the forecast
one day ahead of the caJcuJations. Thus, tbe actual trading
range and the forecast trading range are on tbe same line. We
wrote the TREND function as follows to calcula te the forecast
high in cell E7:
forecast high in cell E7
- TREND(b2:b6,,{6},TRUE) (2.4)

Linear Regression Analysis

36

Table 2.2 A 5-day Forecast Developed with an Excel Spreadsheet

+- 5-Day Forecast ->
Date
High
Low
Close
930127
930128
930129
930201
930202
930203
930204
930205
930208
930209
930210
930211
930212
930216
930217

104.5
104.969
105.094
104 .938
104.719
104.813
105.531
105.906
105 844
105.688
105.563
105.5
105.844
106.156
105.875

104.156
104.344
104.438
104.375
104.375
104.406
105. 188
105
105.563
105.219
104.563
104.563
105.031
105.219
105.344

104.313
104.5
104.719
104.938
104.469
104.781
105.344
105.875
105.563
105.438
104.625
105.344
105.5
105.375
105.844

HIGH

LOW

CLOSE

104.966
104.7
105.244
106.006
106.366
106.175
105.659
105.372
105.631
106.116

104.478
104.406
105.01 6
105.287
105.797
105.675
104.797
104.419
104.472
105.059

104.813
104.775
105.178
105.906
106.191
105.859
104.806
104.769
105.228
105.481

Tbis defines the first five values ofthe bigh in cells B2 through
B6 as the o oes to be used for the forecast. By not speci fyiog
the range of x values, we are using thc dcfault valucs 1, 2, 3,
4, 5 for the fivc days. This is exactly what we did in Table
2.1 earlier. The !6} givcs us tbe forecast for the oext tradiog
day. The last i1em, TRUE, says thal lhe inlercept should be
calculated as usual. This ensures that the best-fit line will not
go througb the origi.n.
Note that we always use the days 1, 2, 3, 4, 5 as tbe independent x variables for each successive forccast. Wc
forecast the high for day 6 usiog data from days 1 through 5;
for day 7, we use data from days 2 through 7, droppiog the
data from day l. Accordingly, we assume that the market's
" memory" cxtends five days back. We will show bow to extend the calculatioo to cell ES usiog the TRENO function:
forecast high in cell E8
- TREND(b3:b7,,{6},TRUE) (2.5)

T Bond Forecasts for the

N~xt

37

Troding Day

Forecasts for the low and tbc close use exactly the same
calculatioo me thod. lf you do not have a TREND function
in your spreadsheet, you can use the calculation scheme
showo in Table 2. 1. For completeness, we will show bow we
wrote tbe TRENO function for forecasting thc low and the
close. ln effect, you can use 1he COPY function in the spreadsheet 10 copy the formula from cell E7 into the neighboring
columns, cells F7 and G7. Similarly, you can then copy the
formula from these cells into ali subsequent rows to calculate
tbe rcmaioing forecas1s.
forecast low in cell F7 - TREND(c2:c6,,{6},TRUE}
forecast close in ccll G7 - TREND(d2:d6,,{6],TR UE}
(2.6)

Graphs for T-Bo nd Forecasts

The daily range and 5-day forecast for tbe September 1993
Treasury Bond contract is shown in Figure 2.1 2. This market
was mak:ing swing moves in February and March, 1993. You

103
102L-~~~~~~~~~~~~~~~~~~~--'
$
: ~ ~
~
~ b
X ~ o
~ ~
e
~ ~

~ ~

~ ~

~ ~ 9

~ ~ ;

~ ~ ~ ~ ~ ~ ~

~ ~ ~ ~ 9

~ ~

~ ~ ~

ACURE 2.12 The da1ly pnce range of the Seplember, 1993 Treasury
Bond futures contract and a S-day forecast of the daily close.

38

Unear R~grcssion Analysis

can see that a close above or below the forecast tipped off the
trend for the next severa! days. For example, a olose above
the forecast on February 11 Jed into a tradeable 8-<iay rally.
The market corrected briefly by going sideways for three days.
1t resumed anothcr smaU four-day swing, lcading to a shortterm top oear 111-16. The next series of swing moves were
also tradeable on a short-term basis.
Figure 2.13 shows the same T-Bond 09/93 contract in the
months of April, May, and June, 1993, indicating the envelopcs formed by the 5-day forecast of the daily bighs and lows.
T he envclopes tend to narrow beforc signiftcant moves. A
daily close outside the envelope a lso provided clues offuture
market directioo. For example, note the close above the higb
forecast in early April, May, and Junc that led to higber prices.
A close below tbe envelope also preceded down moves in
mid-April and mid-May. The envelope oflen acted as a point
of resistaoce and support as the markct searched for direction,
aod couJd be used to plan tradiog strategy.
The T-Bond market trended steadily upward in June 1993.
1t was oot making large swing moves. Figure 2.14 shows the

T Bond Forecasts fo r the Next Trdlng Dy

39

115
cr'

11
113
112

~-ol\

, 11

" J1
'

{'"' i'

j~

t~

f'f r

'

.)
1 ~~
-"'

\
Otitv rano

A CU RE 2.14 The daily price range of the September, t 993 Treasury

Bond futures contract and a 5-day forecast of the daily close.

daily forecast close and price range. Here, too, a close above
the forecast helped gauge market directioo. Nevertheless, the
markct often traded in a narrow raoge in this month; therefore, oot every close below the forecast led to down moves.

115
11 4

01l1v rnoe

Contingency Planning with the Forecast "Template"

S~;,v forecast of hio.h

113
112
111
110

109
1oe
107
108

=
ACURE 2.13 The daily price range of the Seplember, 1993 Treasury
Bond futures contract and a 5day forecast envelope.

Thc daily forecast is usually a good gucss for the next tradi ng
day, and can be uscd as a tcmplate. When the market is particularly strong or weak, it doses above tlle forecast higb or
below the forecast low. The most important benefit of using
the forecast template is Lhat you will ba ve speciJie oumbers
for planniog actions if lhe market trades at or beyond a particular price.
The types of plans you can develop are varied. One simple
strategy is to buy or sen on the close. Let"s assume that the
el ose is goi ng to be above thc forecast for that day. You could
then buy on the close, expccting higher prices ahead.
Aoother approach is to use the daily forecast higb and low

linear Regression Analysis

40

as reerence poinls. You could try to shon near the highs of

the day i you eel a shon-term top or trading range is about
to occur. Conversely, you could try to buy near the projected
low o the day if you expect a shon-term bottom or trading
range. To trade witb tbe trend, you could buy above tbe projected high and sell below the low.
You also can use the daily farecast high and low to set stops.
lfyou are long, you may wisb to cash out below the forecast
low. You can use the forecast far with-the-trend or againstthc-trend trades. Your approach will dcpend on your t.rading
horizon and markct strategy. Tbus, thc forccast is uscful in
developing a garue plan far trading.
We ha ve analyzed tbe regression farecasts far most major
futures markets by counting the number o times the market
closed abovc the forecast high or bclow the farecast low from
01/01/87 to 03/01/93. On average, tbe market closed beyond
the forecast 66 percent of the time. For example, over 1552
trading days, the Treasury Bond market closed above the
forecast high on 516 days and below the farecast low on 484
days. A close beyond the forecasts often provides a tradeable
signa! of shon-term market strenglh or weakness.
DEVELOPI NG A FORECAST OSCILLATOR

To dcvclop an oscillator based on th e forecast, you simply

subtract thc forecast from the closc, and cxprcss it as a percentage o the clase. This forecast oscilla tor will be positive
i the close is above tbe forecas1. You can combine i1 witb a
simple moving average to get an early signa! o a potential
trend changc. You also can use it as an overbought/oversold
indicator.
In Table 2.3, we show tbe calculations using tbe same data
for the 5-day forecas1 from Table 2.2. Wc define tbe forecast
osciJlator %F as foUows:
%F -

l~(C -

Cr)/C

(2.7)

41

Oeveloping Forecast Otcillator

Table 2.3

Forecast Osc1llator Calculat1ons

Date

Close

Forecast
Close

930127
930128
930129
930201
930202
930203
930204
930205
930206
930209
930210
930211
930212
930216
930217

104.3125
104.5
104.7186
104.9375
104.4&68
104.7613
105.3436
105.675
105.5&25
105.4375
104.625
105.3438
105.5
105.375
105.6437

104.81255
104.7750&
105.17818
105.90&28
10&.19061
105.85935
104.60623
104 76879
105 22815
105.46126

Forecast
Oscillator of
Close

- 1
16

22
-12
-25
- 40
17
23
4
11

Herc, C is thc daily closc and Cr is today's forecast close from

thc previous five daily clases. Multiplication by 100 could be
changcd to multiplication by any suitable scaling constan!.
'In thc calculations abovc we ha ve con verted tbe %Fin to ticks
of l/32nds of a point, since that is the smaUest incremenl in
which the T-Bond contract is traded. For example, for 02/
03/93, the difference between tbe actual close and tbe.forecast
was -0.03 125. Multiplying this by 32 convens tbe differeoce
into - 1/32 or - 1 tick. We convened the difference into an
integer far convenience.
Figure 2.15 shows the forecast oscillator of tbe close and
its 3-<lay simple moving average. T he osciUator shows the
difference between tbe actual close and the forecast close expressed in l/ 32nds or ticks. A close below the forecast close
gives negative values, predicting lower prices ahead.. The
crossover of tbe oscillator and its moving average gave early

linear Regression Analysis

42
40
30

Summary

43

Clearly, the linear regression method is a flexible analysis too!

that can be easily adapted to your trading style.

20

10

o
10
20

Tuto rial: Linear Regressio n Analysis

The linear regression method tries to lit an equation of the

form:

30

(2.8)

FIGURE 2.15 A forecast oscillator for the Septem ber. 1993 Treasury
Bond futures conlract.

warnings, which are useful for planning a trading strategy.

You can sce the actual market action of the September, 1993
T-Bond contract in Figure 2.14. For example, note the crossover in Figure 2.15 on 06/03/93 and 06/ 17/93. You can see
a decisive failure on 06/14/93 as %F closed well below its
trailing moving average. Despite that, the actual close was
practically the same as the forecast.
In practice, you could also develop a forecast oscillator for
the high and the low. This can provide more information oo
potential trend changes.
SUMMARY

The linear regression method provides tbe slope and r2, the
coefficient of determination. You can measure trendiness witb
r2 since the slope can be used for following price trends. Tbe
slope can also be used to identify overbougbt or oversold
condit.ions and to develop forecasts for the next trading day
tbat can be used to develop a game plan for your trading.

to the data. In tbis equatioo, x is tbe independent variable,

y is the dependent variable, m is the slope of the liae and e
is tbe y-axis intercept; in etfect, e is a constant. Assume that
by visual inspection, we can deduce approximate value.s of
tbe slope and intercept. We will call tbese approximate values
MO and CO. Once we k:now MO and CO, we can calco.late
values ofy,. At each point, we willhave an error term, defioed

as:
e; = y - Y;

y - (MO

* X; + CO)

(2.9)

Here, the subscript (.) denotes di.fferent values of the independent variable x. The least squares method minimizes
thc sum of the squares of thc error terms. The error terrn is
thc vertical deviation from the line to the actual data at that
value ofx; the deviation can be positive or negative. By squaring the deviation, we get only positive numbers. The linear
regression finds the particular values of slope and intercept
tbat min.imize the sum of the squared deviations. Note that
tbe first useful piece of information from linear regression is
the slope. Once we know the slope and the intercept, we can
draw or calculate the best-fit line for different values of x.
We will now calculate two deviations. The fi.rst deviation
is the vertical distaoce between the fined line and a given
data point:

44
Uneor Regression Analysis
..;,_~~~~~~~~~~~~~..;;;;

e, - Y - Yr

Here we bave used the subscript r to show the error from the
best-fit line. lfwe squared cach dcviation e, and added them,
we wo uld bave the SSE or Sum ofSquares from fitting errors.
Thus:
SSE - sum (e,)2

(2.11)

where we sum over ali data points.

Next, we calculate tbe average of ali values of y, tbe dependent variable. We can now measure the deviation ofeach
value of y from y1, tbe average value of all y values, or:

Y. - (y - yJ

(2.12)

We can now define SST or sum of squares total, wbicb is

the sum of all the tenns in Equation 2.11 . SST is thc sum of
squared deviations of eacb y from the average value of the
dependent variable. We define SST as:
SST - s um

r2

(2.10)

(y~)2

(2.13)

We expect SSE to be less than SST, sincc the method of

least squares minimizes the deviations from thc fitted line.
lntuitively, we can expcct SSE to be O if ali thc points fall
perfectly on a straight line. Every point that movcs away from
the best-fit straight linc will increase SSE.
In tbe worst case, the fittcd line is no better than a horizontal line drawn at the average value ofy, and the deviations
will be the same as the deviation from the average value of
y. Hence, SSE will equal SST. We can now define a quantity
to measure tbc strength ofthe linear relationship. This quantity is r2, tbc coefficient of determination defined as:

(2. 14)

(SST - SSE) / SST

lo the best case, with ali points on a straight line, SSE

O, and r2 - t; ithe worst case, SSE - SST and r2 = O. The

coefficient of determination is the ratio of tbe variation explained by tbe line to the total variation in y. By definition,
we know that when r2 is O, tbe slope of tbe regression line is
also O. We can now say that r 2 is a measure of the relative
s trength of the linear relationship, thus of the trendiness in
the data.
We can draw statistical inferences from the linear regression
with a 90 or 95 percent conlidencc leve l. For example, we can
test the hypothesis tbat the slope of the linear regression is
O. lfthe slope is statistically significant, we will reject tbe null
bypothesis that tbe slope is O. The usual process for testing
a bypothesis is to set up an analysis of variance (ANOVA)
table to calculate the F-statistic. The ANOVA table is shown
in Table 2.4.
Here, n is tbe number of days in tbe linear regression. For
us, tbe value of o was 14 days, so the error term will ha ve
12 degrees of freedom. We havc airead y defined SSE and SST
in our calculations of r2 With a confidence level of, say, 90
percent, we look up thc va luc of the F-ratio wilh 90 percent
Table 2.4 ANOVA Table for linear Regression
Source of
Variation

Oegrees o
Freedom

Regress1on
model

Mean Square

F-ratio

MSR SSR/ 1

MSR/ MSE

SSR

Error
(residual)

(n -

Total

(n - 1)

Sum of
Squares

2)

SSE

MSE - SSE/
(n -

number o days 1n regression

SST

2)

46

Line-ar Rtgression Anilysis

confidence leve!, and 1 and 12 degrees of frecdom. (You can

find the F-ratio tables in most int.roductory books on statistics.)
No te two importan! values of F-ratio: 3.18 and 4.75. Wilh
a 90 pcrcent confidence leve), the F-ratio must exceed 3. 18
to yicld a statistically sjgnificant sJope. Similarly, the F-ratio
must excced 4.75 for the slope 10 be statistically significant
with 95 perccnt confidence level. Both values are for a 14day regression.
We wilJ rewrite the F-ratio in terms of ri for ease of computation, and use the relationship SST - SSE + SSR for
simplifying the equation below. We also will use the deJioition
of r2.
F-ratio -

MSR/ MSE
(n - 2) (SSR/ SSE)
(n - 2) (SST - SSE)/ SSE
(n - 2) r, ( 1/ (SSE/ SST))

You can Jind values of the F-ratio from an introductory

book on statistics to calcula te the critica! values for any regression you wish. We tabulated at lengtb tbe critica! values for
linear rcgressions of 10 to 50 days in length in Table 2.6. The
confidence level was 95 percent for all calculations in that
table. Tbe slope is statistically significan! wben the coefficicnt
of determination of tbe regression is greater than the criticaJ
value.
Table 2.6 shows lhat as thc number of days in the linear
regression increases, the critica! value of r2 decreases. Tbus,
it is easier to spot trends as the lengtb of tbe regression incrcases. When tbe calculated value of r 2 is less than its critica!
value, tben tbe slopc is not statisticaUy dilferent from O. We
Table 2.5 Critica! Values of r'

(2.15)

- (n - 2) (r2/(I - r 2))

This equation says that, as r2 increases, the F-ratio increases,

and we are more likely to havc a significant slope. Remember,
the important values of F-ratio are 3. l 8 and 4. 75 for a 14day linear regression. We can substitutc thcsc values into the
F-ratio equation and Jind tbe critica! va lucs of r' that yield
statistically sigojficant slopes. Wc can thcrcforc express r' in
1crms of the F-ratio:
G - F/(n -2)
r 2 - G/(l+G)

47

Summary

(2. 16)

Here, F is the F-ratio for a given confi dence level, with 1

and (n
2) degrees of freedom. From the equation for G ,
the criticaJ values are 0.21 and 0.29 (rounded). These correspond to a conJidencc level of90 and 95 pcrcent (see Table
2.5). Thus, when the r2 value excceds 0.29, the slope is statistically sigoificant.

Confidence Level

Cnllc.11 Value of r'

90 percent with 1 and 12 DOF

95 percent with 1 and 12 DOF

0.2095 (0.21 rounded)

0.2836 (0.29 rounded)

Slope of 14day regression is sta11st1cally siernficant 1f critica! value of r'

exceeded.

is

Table 2.6 Crilical Values of r'

n (days)

DOF =
n - 2

F-ratio
0.95, l , n - 2

G - F/
(n - 2)

r' Critica! =
G/(1 + G)

10
20
30
50

8
18
28
48

5.32
4.41
4.20
4.04

0.665
0.245
0.150
0.084

0.40
0.20
0.13
0.08

Slope of n day regression is staustially s1gnfocanl 1f criuul value,. exceeded

Conf1dence level = 95 percent.
n Number of days in the regress1on
OOF - Oegrees of freedom
Values of r are rounded.

48

Lnar Regression A.nalysis

can then say Lhat thc data have no statistically significan!

tren d.
You will find that high values of r2 often occur wilh hlgh
~alues o'. slo~. The value of r2 is closc to O wben lhe slope
is changrng s1gns. Consequently, you should consider both
the slope and r2 in your analysis.

3
The Variable lndex
Dynamic Average

Change is the only constan! in life, and nowhere is it more

evdent than on thc trading floor. Our financia! and futures
markets seem to change every minute of every day. Tbese
markets are dynamic because traders constantly adjust to
ehanging perccptions and parti.cipants. Yet, in spite of this
dynamism, wc usually use static indcators in our analyss,
and thesc statc ndicators don't change the time period of
analysis. For cxample, many traders use a 9-day or 14-day
pcriod to calculatc their tecboical ndcators such as RSI , thc
relati ve strength ndex.
Dynamc ndicators, in contrast, vary the time period used
in analyzng market actoo. In ths chapter we will show you
how to use thc variable index dynarnc average (VIDYA), a
dynamc exponential moving average thal adjusts its effective
lenglh usng a market variable. For instance. we will index
VIDYA to thc standard deviation of closing prices, as weU as
10 a momcntum oscillator, and to the coefficient of determioation, r2 The responsiveness and dyoamic range of VIDYA changes with the indexing approach, and you can adapt
it to suit your needs.
49

SO _ _ __ _ _ _ _ ___T::.:h:.;:e~V~~
ri:.:.:.b~
le~l~
n::.
d::.~D~y~
n~
m
~i~
c~
A~
v~~
g~
e

Thc Slmple Moving Average and lis Re:!SP:.::::.:

"S:::.iv:.;:e::.:
n:.;:S::.
S _ _ _ _ _ _ _s_1

THE SIMPLE MOVING AVERAGE

ANO ITS RESPONSIVENESS

Table 3.1 Comparing Respons1veness of 5 Day and 10-Day SMAs

A simple moving average (SMA} uses a fixed number of days

to calculate a new value. For instance, when we use the most
reccnt 10 days' dala to calculate a 10-day simple moving
average, we disregard market volatility. Unfortunately, the
gap between the current price and thc moving average usually
increascs when the market moves quickly. In percenlage
tenns, thc price gap could easily be l- 20 percenl from the
moving average.
Let' s considcr two simple moving a veragcs of che daily
closing pricc; let the sborter moving average use five days and
the longer average use ten days. lf the market makes a quick
move, the shorter moving average wilJ respond carlier than
the longer moving average. In tbe 5-day SMA calculations,
cach oflhe daily closes, including tbe la test clase, has a weigbt
of20 percent. Thc 10-day SMA assigns a weight of 10 percent
to the latest daily close. The shorter average responds more
quickly than the longcr one because it uses a grcatcr proportion of new data, and usually has a smaller price gap between
the latest price and the moving average.
Thcse calculations are s.i mulated in Table 3.1. The columns
show how the 20 percent slice of new data makcs the 5-day
SMA more responsive to price changes than the 10-day SMA.
For cxample, in the last row, the clase is 14. 75 percent abovc
thc 5-day SMA. The same clase is 20.48 percent above the
10-day SMA. Tbus, the more rcsponsivc average has narrowed the gap between itself and the latest clase.
This table illustrates the key idea behind a variable length
moving average: We can make the average responsive to data
by reducing its length when volatility incrcases, or increasing
it wben volatility decreases. By malcing these changes automatically, if possible, the average could adapt its length to
market actio n.

Close
100
125
110
90

108
11 9
135
110
100
125
140

20%of
Close

10% of
Close

5 day SMA

10-day SMA

20.00
25.00
22.00
18.00
21.60
23 .80
27.00
22.00
20.00
25.00
28.00

10.00
12.50
11.00
9.00
10.80
11 .90
13.50
11 .00
10.00
12.50
14.00

106.60
11 0.40
1 12.40
11 2.40
114.40
117.80
122.00

112.20
116.20

The Sday SM/\ gives an equal weoght of 20 percenl to S day's dala.

The 10-day SMA gives an equal weoghl of 10 perct'nl to 10 day's dala.
The S-day SMA responds faster beuuse ol lakes a larger bole oul of new data.

A Volatility lndex

We need a volatility index to tell us when the pricc action is

heating up or cooling down. Volatilily can be measured as
the s1andard deviation in the closing prices over the past x
days.
In arder to form the index, wc also need a reference value
of the standard deviation over x days. !h.e re.fcrence _value
will tell us if the observed standard deVlat1on 1s too higb or
too low. Then, we can define a volatility index that refers
current volatility to historical volatility as:
k - sigma(x-days) / sigma(reference).

(3. 1)

He re, sigma(x-days} is tbe stand~rd de~tio? of closing

prices over x days. sigma(reference} 1s the histoncal value of
the standard deviation over x days. For example, you could
use a 20-day moving average of sigma as the reference value.

52

The Vuiable lndb Oynmlc Averge

Tbis would theo become the historical refereocc value of

sigma. You also could use the average value of sigma overa
time period of 5 times tha1 used to calcula1c the standard
deviation.
A third approach is to find lhe maximum and minimum
value of thc standard deviation over x days. You can theo
choose . arbu:rary value for the reference that is 1/ 4 or 1/5
the maximum value.
Our oext problem is to choose a moving average. We will
work with an exponential moving average. Thc mathematics
of exponcntial averages requires that its index be a constan l.
Ho~ever, we wi!I arbitrarily dcviatefrom this logic by mak:ing
thc index a v~nable. You can visualize this as calcu1ating a
new ~xponenllal average each day. Rather than using a fixed
~ct1on to uJ?<iate the new value, we will use a changing frac.
uon. Ou:r senes of numbers will then represcnt values of exponential moving averages with variable length.
Wc.can now write ~ eq uat.i on for VIDYA using the usual
equat10n for cxponenttal moving averages:
VIDY A - alpha

* k * ~ + (1-alpha * k) C,.

53

The Simple Moving Average nd lls llesponsiveness

number ofdays in an expooeotial moving average. The index,

alpha, of the average is giveo in the following equation:
alpha - (2/(N

(3.3)

1)).

We can sol ve for N in tenns of alpha:

N - (2 - alpha)/alpha.
Wc can inserta known value of alpba to sol ve for thc num
ber of days in the moving average. We will do just that, by
building a table to show how the leogth of YIDYA changes
as the volatility index changes (Table 3.2).
This table shows thal, as the market enters a quiet phase
and k < 1, the length of VIDYA increascs. Conversely, wben
thc market isactiveand k > 1, the Leogth ofVIDYA decreases.
VLDYA will simulate a 9-day exponential moving average
whcn Lhe product of k and alpba is 0.20. As the product of

(3.2)

In this. equation, today's close and yesterday's close ha ve the

~ubscnpts Oand 1; we have previously defined the volatility
index, k. Thc constant, alpha, determines thc effective lcngth
of the exponeotial moviog average we waot to modulatc.
When k - 1, we have an exponeotial average determined by
alpha. When k > 1, we talce a larger bite of the new data aod
the effective length of the average decreascs. When k < 1,
we take a smaller bite out of the ncw data, and the effective
leogth of the average then increases. In choosing values for
alpba and k, we must ensure that the term ( 1 - alpha k)
oever becomes negativo.
Dynamic Range of VIDYA

We can now estima te the equivaleot length of an expoocntial

moviog average from the index. For examplc, Jet N be the

Table 3.2

Oynamic Range and Effective Length of VIOYA

No. Oays in
Vola11fity
(lndex. k)

Alpha (lndex
for 9-day EMA)

k'Alpha (1ndex
for VIDYA)

Average
(effective lenglh)

0.20
0.40
0.60
0.80
1.00
1.20
1.40
1.60
1.80
2.00

0.20
0.20
0.20
0.20
0.20
0.20
0.20
0.20
0.20
0.20

0.04
0.08
0.12
0.16
0.20
0.24
0.28
0.32
0.36
0.40

49.00
24.00
15.67
11.50
9.00
7.33
6.14
S.25
4.56
4.00

A 9-day f'xponential movmg average (EMA) has alpha

0.20.

The efiec11ve length of a VIOYA (2 - k' alpha)/(k Jlpha).

The dynamic range is the range of effective lengths 49
4 - 45.

54

The Variable lndex Dynamic Average

k alpha vares tenfold from 0.04 to 0.4 due to rising volatilhy, the effective lcngth of VIDYA decreases from 49 to 4
days, a wide dynamic range. An exponential moving average
with a length of 49 days is much slower than one with a Jength
of 4 days. Therefore, VIDYA will move faster than the equivalent exponental moving average i.n response to rising market
volatili ty.

TRADING STRATEGIES
Ali your usual strategies with moving averages can be used
with VLDYA. For example, you can use a shon and a long
VIDYA and use a moving average crossover decision model.
Or, you can take the diffcrence between the two avcrages and
trade that witb a third moving average, whicb is the MACD
or moving average convergence-divergence approach discussed in Chapters 1 and 2. You can use moving averages
and trading bands or use VlDYA to set trailing stops. Since
VIDYA adjusts its length automatically in response to market
volatility, it will generally remain closer to prices than the
equivalen! exponential moving average. Hence, its responsivcness can give you more timely trades, which is the main
advantage ofusing VIDYA.
You can use volatility bands around VIDYA to make trading decisions and crcate a breakout system, trading prices in
thc direction of their brealwut outside the bands. Your next
signal to close the trade or reverse is when prices close inside
the bands for the first time. You also can use bands shifted
above and below VlDYA by a fixed percentage, sucb as 1
percenL Tbe upper and lower bands act as points of suppon
and rcsistance. Wben prices close near tbe uppcr or lower
band, you would look for markei reversals. As usual, when
the market is trending strongly, it can close outside the bands.
This is a signa! that the trend wilJ persist. You wou ld then
wait far the prices to close inside the bands befare taking
countertrend positions.

VIDYA and T-Bond~M

:::::rc:k:;:el:_:A:'..'.n::a'.!.ly_::sis:__ _ _ _ _ _ __ __ _
SS

CASE IN POINT: VIDYA ANO T-BOND

MARKET ANALYSIS
Figure 3.1 shows the US Bond 06(93 con~ct along with
VlDYA and an equivalen! exponenttal mov1ng average. You
can easily see that VLDYA is more r~sponsive than the e~
ponential moving average. The elfecuve length of VIDYA is
shown in Figure 3.2 for the data in Figure 3. 1. The units of
the y-axis are the number of days in the equiv3:1~nt .expooenal moving average. You will see that, as volatiluy mercases,
the effective length of VI DYA decreases. For cxample, observe
the decrease in length in February, 1993 as the market moved
up quickly. You can see that VlDYA autom~!ically a~justs its
Jength to daily changes in market volauhty. Nonce how
VIDYA moved up during 02/93 in Figure 3.1.
You also can use VlDYA to set unique trailing stops on a
closing basis. For example, if you were long, you would close
your long tomorrow ifthe market closed below VIDYA today
and was going to close lower tomorrow. Or, you can go l?ng
wben prices close above VIDYA, and go sbon when pnces
close below VlDYA. As an aggressive trader you could anticpate a close above or below VIDYA 10 initiate your trade.
Thls would lct you put on a position a day earler tban discussed above. VIDYA lets you set unique stops that cannot
be gunned casily.

Shifted Trading Bands

Figure 3.3 shows an example ofv_IDYA and 1 i:ic:rcent trading
bands. You could experiment with bands of different separations if you wish. The lower band, drawn 1 percent below
VJDYA acted as support for closing priccs during the correction 'rrom record h.ighs in Marcb. Notice that the initial
low on 03/26/93 at the band led to the lower Iow also al th.e
Jower band. You could have gone long the T-Bond when 11

'1lt 11'.6

"

VIOVA nd T-8-0nd M.>rktt Anl~sis

57

91>0[6
tlt()(6
lll'I~&

LOtU6

SOt0[6
Qt0[6
llf :'<&
9WK6
tZfOf 6
Uf!Y:~

<
o

>-

Q()[()(6

>

..

;orors
Z0.11(6

!!
>

....

.5

: Pll0C6
UZ()(6

a. '6

:j

91lOC6
'll0C6
60ll)(6
<;ol0f6
rozor6
10lOC6
Qll (6
9ZIOC6
Zll0f6
!Vl()(f

...i
j
(,)

>

o"'

..,u
~

~
~6
ti 10[6
ll 10[6
ij\1 ,{6
-llll Jl6
P010[6
Ol:lllt

.8

"'
;;)

56

-- -

"'o

...o

FIGURE 3.2

The effeclive length (111 days) ol VIDYA shown in Figure

3.1 .

: 9ZZO6

>

-...

QlfO\ 6
91(0C6
Zlf0\6
O'I '(6

ll3
11 2
1 ll

"
109

108

FIGURE 3.3 The )une, 1993, T -Bond futurc>s contracl, smoothed wilh
a variable lndex dynamic averJge .

S8

The Vuiblr lndrx Oynmic Avrrgr

bounced orr the lower band, with a close stop for proper risk
control. During the upward move in ApriJ, 1993, prces ctosed
outside the upper 1 percen1 band, showing a strong upward
move. As expected, prces continued tbeir climb for four more
days before moving sideways.

OlHlr6
91Hlr6
rl70f6
- Ztt0f6
LOr0(6
~1'0(6

l01(){6
(){f(){6
9lrof6
lf(){6
Zlf0f6
g1f(){6
9lrof6

e?

Volatility-Based Trading Bands

Figure 3.4 shows volatility bands formed using the dynamic

range values derived for VIDYA calculauons. These bands
can be traded as a breakout sys1cm, since a close above the
uppcr band or below thc lower band often presents tradeable
opponunities. Details of the calculaons are in the tutora!
at the end of the chaptcr.
Another use oftbe bands is to define trcndiness. The market
is trending up if thc close is outside thc upper band; conversely, the market is trending down if the close is below the
lower band. You also could use the direction ofthc bands to
judge the trcncl A close within thc bands prcdicts a trendless
or consolidation phasc.

lf(){6

01r01:6
!IOfOf6
KK'Of6

,,..
-g

-..

9Zl0f6
~ZZOf6

lll()(6
Qll(){6
9ll(){6
1ll0f6
6Ql(){6

.9

:f

g)l(){6

i..

lndexi ng to Momentum or r2
Ano1her useful feature of VIDYA is that it can be indcxed to
any dimcnsionless market variable that vares from O to l.
Thus, you don't need the standard deviation to calculate
VIDYA. For example, you could index VIDYA to thc coefficient of determination, r', since r2 measures trendiness, is
dimcnsionless, and vares between O and l. You could also
indcx VIDYA to a momentum oscillator that varies bctween
O and 1, such as the rclative strength index (RSI). We like
using the absolute value of the 9-<!ay Chande momentum
oscillator (CMO) as the momentum index. The CMO, a variant of the RSI, is described in Chapter 5.

l0f()(6

1>

fOZOf6
10l0r6
9l1(){6
9ZUJl"6
lll0f6
Oll0f6
91l0f6
711(){6
ll l0f6
90l0f6
901(){6
r010f6
orz1z&

-3

..

.e

~
<
>o

l.!

>

...~

l
"-

-N

-o

.,o

o
S9

60

The Vnible lndex Oynnnic Aerge

Figure 3.5 shows the June 1993 T-Bond contract in February-Marcb, 1993, as it becarne the actively traded coo tract.
The VJOYA, based on absolute CMO, was seositive to market
action during rapid moves. Simultaneously, it flattened out
during sideways periods.
In Figure 3.6, one VlDYA is calculated using standard deviation and another uses a 9-day absolute CMO. Tbe scaling
multiplier for tbe CM O was 0.50. T his meaos tbe 9-day CMO
varied between Oand 0.50 for these calculations. (Tbe details
of thc calculations are in tbe tutora!.) VIDYA based on a
scaled, absolute CMO was more sensitive for the Juoe, 1993
Treasury Bond contract. lt accelerated more quickly than the
VlDYA based on standard dcviation. Note how the CMObased VIDYA flattened out in December, 1992 and accelerated in January, 1993. The two metbods ofcalculation change
tbe way VJDYA responds. Note that you can also cbange Lhe
sensitivity by altering the scaling variables.
In Figure 3.7, we look at the same contract in tbe same
time period, but calcula te VlDYA using r2. Here too, we scaled
Lhe r2 with a multiplier of 0.50 so that its values range from

61

VIOYA nd T lond Mrkel Anl~sis

..---------------~-=
~~

108

Tl!o<Ml~doN

105
104

103

/'

102

l
101

l\2

OtIV doH. TBond. 06/93

ccntrOC1

111

110
109
112

101

110

ot 9-<NV hlll t~ of ~
,..,....., o.s r - " '

1oa t:~=~~~___,~~-=-=~::-;:~;-;;;:,
~
~~O-N~~-~~sa~~~~a~

109 .

~~!!~~!~~~~~~~~~~~~~~~~~.~~~

108

107

VIOY~ u&ino coethcint of dtttfMln,ation 111QuJted-I

108

111

VtOYA UWIQ !rday ~baolvll CMO

FIGURE 3.S VIOYA indexed to market momentum using absolute

value of the 9-day Chande momentum osc lllator.

FIGURE 3.7 A variable index dynamic average that

':"s.

mde:ed to
ng 0 5 times the coefficient of determ1nauon (r ) from
.
mark et act1on us1

a 9-day hnear regression analysis of the da1ly e1ose.

62

112

The Variable lndex Oynamic Average

SUMMARY

Dallv cloH, T Bond, Cle/83

contttct

"1
110
109
108

107

.,....,,.._

Tutorial: SprNdshee!!
ts~fo::,:r_V:..:1:::
0.::
YA.:.__ _ _ _ _ _ _ __ _ _ _6
_3

__ .. _

VJOVA Ullif"O cotfftbllm of dtt~ Ir ~~

- - 0 6 6 ..-

FIGURf 3.8 A VIOYA for the June 1993 T-Bond da1ly dose similar
to the one in Figure 3.7. W e used 0.65 times 1he r' value from lhe 9.
day linear regression to calculate VI OYA.

These cxamples show wbat occurred when w~ .u~ three dif

ferent approaches to calculate VlDYA: a volatJl1ty mdex bascd
on lhe standard deviation of closing prices, a momentum
oscillator and the streogth of lhe linear trend (r') from linear
regressio~ analysis. VIDYA automatically adapted t~ mar~et
action and selectcd an effective length bascd on the mdexmg
method. We also saw tbat t.he scaling multiplier for each
melhod can be changed to increase lhe dynarojc range of
VIDYA and malee i1 more scnsitive to price changes.
Not only can you trade witb VIDYA using lh~ usual mo~g
average trading s1ra1egies, but you can also use 1110 set uruque
trailng stops or develop volatility bands for a breakout ~
ing systcm. Finally, you can use shifted bands 10 find res~st
ancc a.nd support. ln short, VlDYA is a highly flexible movmg
avernge that you can casily adapt to suit your trading style.

O to 0.50. You can compare Figures 3.6 and 3.7 to see how
changing the indexing variable changed lhe response of VJ.
DYA. Whcn the linear rcgression analysis showed a strong
trend, lhe VIDYA based on r2 accelcrated rapidly. H moved
more quickly Iban lhe VLDYA based on the absolute CMO
for example, co mpare thc lwo types of V!DYA in m id-to-lat~
February.
Not only can you changc lhe iodcxing variable, bul you
can also change the scali og variable. In Figure 3.8 you can
see the effect of using a new multiplicr of 0.65 for r2. The
effect of tbe cha ngo in multiplier from 0.50 to 0.65 did not
malee much difference when thc market was trendlcss. Once
lhe linear regression detectcd a trend on 02/ 19/93, lhe VJDYA
with the higher multipUer moved more quickly. Notice how
~e VIDYA llattcncd out as r2 decreascd during consolida
uons. You can use a VIOYA bascd on r2 to set aggressive
trailing stops that are unique and not easily gunned

TUTORIAL: SPREADSHEETS FOR VIDYA

Once you go through thc steps of sct~ing ~P VID-YA, you can
use i 1 for man y of your own strategies. Smce ll is a movmg
average, all the usual moving average strategies can be applied. For example, you could use the crossover of two moving averages 10 generatc trading signals, or you could use tbe
moving average to set traiJing stops.

Calculating VIDYA Using Standard Deviation

First, we wanl to use a market volatility index to assess price
action. To illustratc we will use data for lhe Treasury Bond
March, 1993 contra~t with the ICklay standard deviation of
closing prices 10 define volatili1y. The 5Cklay average of the
10-day standard deviation was 0.64. The range of values for
tbe 10-day standard deviation was from a low of0.13 10 a

The Varable lndex Oynamlc Average

Table 3.4 VIOYA Calcula11ons

high of 1.47. Here is the range ofvalues oflhe volatility index

(wilh rounding):
k
k

=
=

0. 13/0.64 - 0.21
l.47/0.64 - 2.29

(3.4)

Note that we calculated the volatiUty index k by using a

reference value of 0.64. We obtained the refercnce value by
averaging 50 days of data on the 10-day standard deviation.
You could also have chosen an arbitrary reference value. For
example, the range ofvalues for the 10-day standard deviation
of the close was 0.13 to l.47. lf you had arbitrarily chosen a
rcfcrence value of 1, the range of k would be from 0.13 to
2.29. The key is LO ensure that the tenn (1 - k*alpha) always
rcmains positive. Hencc, if you were using a value of 0. 1 for
alpha, then the quantity ( 1- k*alpha) would be positve for
a reference value of l. You can test a range ofvalucs Lo choose
a responsiveness of VlDYA you like.
We wil1 use a l9-day exponential moving average, with an
alpha of 0.1, given by the ratio of 2/(19+ 1) or 2/20. The
dynamism of thls VIDYA can be calculated from the range
ofvalu.es in Table 3.3. Thus, we wil1 effectively vary the number of days in the exponen tia] average ali the way from about
8 to about 94. The equivalent exponential moving average
would aJways use 19 days.
Table 3.4 illustrates the stan of calculations in a spreadsheet
format. We first calculatcd the 10-day standard dcviation of
the close (0.96), and thcn divided it by 0.64 to find the k value
Table J.J

The Oynamism o Vidya

Alpha

k Alpha

0 .21

0.1
0.1
0.1

.021
0.1
0.229

2.29

No. of Days
(roundcd)

94
19

65

Tutora!: Spreadsheeb for VIOYA

Date

Close

921127
921130
921201
921202
921203
921204
921207
921208
921209
921210
921211
921214
9212 15
921216
921217

102.00
101.97
102. 19
102.19
102.31
103.16
103.97
104.19
104.19
104.25
104.13
103.88
103.91
104.09
104. 13

10-day
StdDev

0.96
0.94
0.65
0.74
0.58
0.30

k-c/0.64

1.50
1.47
1.33
1.16
0.91
0.47

VIDYA

104.19
104.20
104.19
104.15
104.12
104.12
104.12

EXPMA

104.19
104.19
104.19
104.16
104.13
104.13
104.13

of 1.50. We used the previous day's (92/ 12/09) close to fin~

the first valuc of VTDYA. The first value for 92/1~/10 IS
104.20, calculated by using
VTDYA ~ (0.1 *1.5)*104.25 + (J-0.l*l.5)*104.19
= 104.20.

(3.5)

Tbe corresponding exponential avcraged has an alpha of

O. l. h is shown in the last column. We connue the calcu
lations for all later days. At the end of February, 1993, there
was a sharp rally in the bond market in reaction to Presiden!
cti nton's budget proposals. The volatility increased and
VfDYA responded more quickly than the equivalent c;xponential movi.ng a vcrage. Table 3.5 shows these calculations.
As the volatility increased from 93/02/22 to 93/02/26, thc
values of k increascd rapidly too. This increase made VTDYA
take a larger chunk out ofthe new data. However, the equiv-

The Variable lndex Oynamic Averge

66

Table 3.5 VIDYA Calculat1ons

Date

Close

930201
930202
930203
930204
930205
930208
930209
930210
930211
930212
930216
9302 17
930218
9302 19
930222
930223
930224
930225
930226

107.41
106.94
107.25
107.81
108.34
108.03
107.91
107.09
107.81
107.97
107.84
108.31
109.34
109.56
110.3 1
111.72
111 .09
110.97
110.88

Calculating Volatility Bands wi th VIDYA

10-day
StdDev

c/0.64
'k'

VIOYA

EXPMA

0.59
0.55
0.46
0.44
0.48
0.53

0.93
0.86
0.72
0.69
0.75
0.83
0.79
0.73
0.69
0.68
0.67
0.59
0.85
1.09
1.47
2.11
2.30
2.19
2.08

l 06.1 s
1OG.22
106.29
106.40
106.54
106.67
106 76
106.79
106.86
106.93
107.00
107.07
107.27
107.52
107.93
108.73
109.27
109.64
109.90

106.02
106.11
106.23
106.38
106.58
106.73
106.84
106.87
106.96
107.06
107.14
107.26
107.47
107.68
107.94
108.32
108.60
108.83
109.04

o.so

0.47
0.44
0.43
0.43
0.38
0.54
0.70
0.94
1.35
1.47
1.40
1.33

67

Tutorial: SpreadsheeU or VlOYA

alent exponential moving average continued to use 19 days

in its calculations.
You also can see lhe responsiveness in this table. From 93/
02/ 19 to 93/92/24, VIDYA iocreased by l. 75, when the exponeotial moving average increased by just 0.92. VIDYA
wcnt from 107.52 to 109.27, while the exponential moving
average went from 107.68 to 108.60. Thesecalculations show
that VIDYA follows prices more closely than thc equivalent
exponential moving average, a feature traders can exploit to
their advantage using a broad variety of strategics.

To develop volatility bands using VIDYA (see Table 3.6), we

calculate the VJDYA as shown above, and then add the volatility-based componcnts to form the bands. We first calculate
the 10-day standard deviation in closing price. We then fiad
the index k by dividing the standard deviation value by 0.64.
This gi ves us the volatility-based term. Next, we must multiply these terms by the index value of the equivalen! ex~
nential moving average. We can then add any amount ofthis
term to VIDYA to form the upper band.
Using a multiplier of 2, we add and subtract 2 times the
volatility term from VIDYA. For examplc, for 92/12/10, we

Date
921127
92 1130
921201
921202
921203
921204
921207
921208
921209
921210
921211
921214
921215
921216
921217
921218
921221

Table 3.6 VlOYA Volatility Band Calculations

Lower
Upper
Band
Band
StdDev/
VIDYA
VIDYA
+
0.64
10-day
2'.l'k
2.1
k
k
VIDYA
Close StdDev
- 'k'
102.00
101.97
102.1 9
102. 19
102.3 1
103.16
103.97
104. 19
104. 19
104.25
104.13
103.88
103.91
104.09
104.13
104.28
104.88

0.96
0.94
0.85
0.74
0.58
0.30
0.13
0.26

1.50
1.47
1.33
1.1
0.91
0.4 7
0.21
0.41

104. 19
104.20
104.19
104.15
104.12
104.12
104. 12
104.12
104.15

104.SO
104.48
104.41
104.35
104.30
104.21
104. 16
104.23

103.90
103.89
103.88
103.88
103.93
104.02
104.08
104.07

68

The Variable lndex Oynamic Average

do the ollowing calculations, noting tbat 0.1 1s the index of

the equivalent exponential moving average:
upper band lower band -

VIDYA + 2 * 0.1 * k
104.20 + 2*0.1*1.50 - 104.50,
VIDYA
2 0. 1 * k
104.20 - 2 .. 0.1 1.50 - 103.90.

(3.6)

Tbe resultan! volatility bands can be used as a breakout

system. A close above the upper baod is a buy signa! for
tomorrow, and a close today below tbe lower band is a sell
signa! for tomorrow. Once you are long, a close below the
upper band is a signal to close the trade. You should be prepared to go long if priccs close abovc the upper band again.
A clase inside the lowcr band is a signal to clase open short
tradcs.
You could develop other trading strategies using bands,
sucb as antitrend trades. You could use VlDYA 10 set stops
oo a closing basis, or you could use a price crossover to buy
or seU. This tutora! should gi ve you a fecl far the power and
lexibility ofusing VIDYA in your trading strategy.

Tutora!: Spreadsheets for VIOYA

CMO - (S.
S.)/(S. + S.),
absCMO - <:MOj.

(3.7)

Tbe CMO is the net momentum as a fraction ofthe absolute

momentum over nine days, and itcan be positive or negative.
The absCMO is the absolute CMO, that is, the value ofCMO
without regard 10 its sigo. CMO vares between -1 and + 1,
while absCMO vares betweeo O and l. We can now define
VIDYA using absCMO.
Here we have used a scaling muluplier of 0.5 to scalc
absCMO, but you can experiment with. other v:U~es ~ you
wish. The benefit of using a simple scaling muluplier 1s that
we get a linear conversion from absCMO _10 V~DYA. lf you
squared the CMO to elinnate the negauve sigo,_ th_eo thc
conversion becomes nonlinear. The higher the muluplier, thc
more responsive the average will be. Hence:
VIDYA - (0.5 absCMO)*C,,
+ (1 - 0.5 absCMO) * VlDYA,

(3.8a)

where today's close is C,, and VIDYA, is ycsterday's value of

VJDYA. Here wc demonstrate our calculations by using data
for 02/12/93 shown in Table 3.7:

Calculating VIDYA Using CMO

We can use any market-related index that vares from O to J
to calculate VIDYA: far example, we can use the Chande
momentum oscillator (CMO) 10 index VJDYA.
We'll use the daily closiog data for the June, 1993 Treasury
Bond futures contrae!. Here, Mtm Up is the momentum wben
today's clase is above yesterday's clase. 1f today's elose is
belowyesterday'sclosc, Mtm Up is O. TheMtm Dn is positive
if today's clase is below yesterday's close, otbcrwise it is O.
In the following equation, we first define the 9-day sum of
the up day (S.) and dowo day momentum (S.) as:

69

s. s. (S. - S.) -

+ S.)
CMO
absCMO
VlDYA,
k
VIDYA
VTDYA

(S0

0.3125+0.5625+0.5313 +0.7 l88 + 0.1563

2.2814,
0.4688+0.3125+0.125+0.8126 - 1.7189,
0.5625,
4.0003,
0.5625/4.003 = 0.140614,
(3.8b)
0.140614,
106.5625,
0.5*0.140614 = 0.070307,
0.070307 106.7188 + (1-0.070307)* 106.5625,
106.573489 - 106.5735.

70

The Vuible lndex Dynamic Average

Table 3.7 Calculatmg VIDYA with CMO

Date

Close

930201
930202
930203
930204
930205
930208
930209
930210
930211
930212
930216
930217
930218
930219
930222
930223
930224
930225
930226

106.1563
105.6875
106
106.5625
107.0938
106.7813
106.6563
105.8437
106.5625
106.7188
106.5938
107.0625
108.0625
108.28 12
109.0625
110.4686
109.8437
109.7188
109.625

9-day Abs
CMO

Tutorial: Spreadsheets for VIDYA

71

H ere, we calculated the r1 using the INDEX and LINEST

funcons in Excel 3.0.

Mtm Up

Mt m Dn

0.4688

0.3125
0.5625
0.5313

o
o
o

INDEX(LINEST(b3:b l l ,$c$2:$c$10,TRUE,TRUE),3)
(3.9a)

o
o

0.3125
0.125
0.8 126

We scaled this value by 0.65 or 0.50, and used il to calculate

V JDYA.
Ncxt, let's review calculations for 02/ 12/93 below, using
data frorn Table 3.8:

r' -

0.7188
0.1563

o
o

0.4687
1
0.2 187
0.7813
1.4063

o
o
o
o
o

0.6251
0.1249
0.0938

o
o

VIDYA

0.125

0.1 40614
0.24786
0.278674
0.352925
0.301554
0.517702
0.670306
0.727246
0.643316
0.599983

106.5625
106.5735
106.576
106.6438
106.6941
107. 1033
107.6104
108.5684
109.0321
109.253
109.3646

Ca lculatin g VIDYA with r1

We have seen ho~ the coefficient of determinaon, r', is a
measure of trendmess that you can calculate from linear
regression analysis. Sioce r' vares between O and 1 and is
always positivc, it is a natural candidate for an indexing variable for VlDYA.
We'U use thc Exoel 3.0 spreadsbeet for calculations, whicb
features a built- in function, L INEST, that we can use along
with the built-in function, INDEX, to calculate r2. For more
details on r 2, it's best 10 review the secon on linear regression
analysis. Wc' U use ni ne days for thc lcnglh of tbe regression
calculations.

kl - 0.65 . 0.1292494 - 0.08401211,

V IDYA - 0.08401211 106.71 88
+ {1-0.08401211)* 106.5625,
- 106.5756.

(3.9b)

Table 3.8 Calculating VIDYA w1th r'

Date

Close

930201
930202
930203
930204
930205
930208
930209
9302 10
930211
930212
930216
930217
930218

106.1563
105.6875
106.0
106.5625
107.0938
106.7813
106.6563
105.8437
106.5625
106.7188
106.5938
107.0625
108.0625

R-Squared

VIDYA
0.65

VIDYA

106.5625
106.5756
106.576
106.5767
106.7593

106.5625
106.5726
106.573
106.5735
106.7142

o.so

1
2
3
4

5
5
7
8
9
0.1292494
0.0336062
0.0022159
0.1889946

72

The Vu iable lndl!l Oynamic Average

Th~se cal~tion.s show that you can adapt VJDYA to market

acuo? . usrng a dimensionless market-related index such as
vol~tility, .momentum oscillators, or r'. You can adjust the
scal1~g to mercase or decrease the sensitivity of this adaptive
moVJng average.

4
QSTICK: The
Quantitative Candlestick

Originating in Ja pan, the candlestick metbod of price analysis

is mainly a pattern-recognition technique. Hence, it has predictivc power, unlikc moving avcrages or oscillators. The
drawback to using this approach is that pattern interpreta1ion
is very subjec1ive, and so this method remains qualitative in
nature.
Building on thc basics of candlcstick analysis, our new indicator, Qstick, cnables you to quan tify candlcsck analysis
and improve thc interpretaon ofcandlesck pattems. Qstick
is a moving average of the difference betwcen the daily closc
and the open-a difference that is at the hcart of candlestick
analysis.

THE BASICS OF CANDLESTICK ANALYSIS

We can draw a candlestick using the same price data found
on a bar chart. Tbus, a candlcstick uses the open, high, low
and closing prices for the time duraon under study. This
73

74

QSTICK: TI\;, Quantittive C.ndle<tick

lcts you combine candlcsticks with trend lncs, moving averages, trading bands and so fonb.
lo candlestick analysis, each day's pricc action appears lk:e
a candle. The body of thc candle is drawn as a rectangular
range betwcen the open and close, and it is black (or filled)
if the close is below the open. If the close is above the open,
then the body is white (or cmpcy). "Down days are dark" is
an easy way to remember this scheme. Thin Lincsjoining the
high and low are called the upper and lower shadows, or tails.
Figure 4. 1 shows a bar chan and the equivalen! candlestick
chan. The data are for thc July, 1993 Couon #2 futures contract. While the two chans use the same data, tbe candlestick
chan is more striking in revealing bullish aod bearish days.
Most analysts look at 1wo imponant factors in cvaluating
candlesticks. The first clcment is the appearance of thc new

Qstick: lntr<lday Momentus lndicator

75

candlestick by 11self. You can look at the range betweeo the

open and the close and its pro~nion to the shad~ws. But,
it is the "big picture" fonned usmg man.y candlesucks. that
is more important than a single candlesuck. Often, a smgle
candlestick is more significan! when we view it within the
broad technical picture.
This Icads to the second and more important facet of candlestick pattem evaluation: the pauem formed by two or
more candlesticks. The range between thc open and the close
shows the power of bu lis or bears. A market has great power
or convictioo when the open and closc are far aparL Converscly a market lacks power or conviction when the open
and cl~se are not far apart. The length of the shadows is also
imponant. Powerful markets have short shadows, whereas
markets tha t are wcakeni ng or changing oftcn show long shadows. Successi ve candlesticks forrn pattcrns tbat may suggcst
a strengthening or wcakcning of tbe market.

QSTICK: INTRADAY MOMENTUM INOICATOR

00
00
00

We have noted that thc range between the open and the close
is the most imponant e lcment of candlestick analysis. Tbe
diffcrence close minus open, is a mcasurc of intrad ay (or
intraperio'd) momcntum. We have quantified this rangc to
devclop a trend-following indicator. This indicator, called
Qstick, can be dcfincd as the moving average of the intraday
momentum.
We will use eight days of data for calculating the simple
moving average. You can experiment witb o tber time periods,
from minutes to montbs.

00

m
Jun

Jul

FIGU RE 4.1 A bar chJrt and candleslick churt for the Cotton #2
futures conlract using the same mformation for ead1 day

Qstick - Average((C - 0),8) (SupcrCharts formula)

- Mov((C- 0),8,S) (Metastock formula)
(4.1)

In this equation, o is the opening price and e is the closing

price. Tbe daily range between the open and the close can be

76

QSl ICK: The Quantitative Candlestick

posjt:ive or negative. Qstick will be positive wben the market

closes above its opening price.
Note that there is sorne variability in the reported data on
opening and closing prices. For example, tbe opening price
couJd be Lhe price of the first trade, or it could be tbe first
trade reportcd, which may not be first trade executed in the
pits. Or we could use Lhc average price of aJJ trades in the
first minute o trading. The same is true for closing prices.
ln. our examples we will calcula te Qsck witb a simple
!Uovmg averac. (You can use an exponential moving average
ifyou prefer.) For example, for sbort-term analysis of a futures
market, we prefer a 5-day Qsck. Jf you are using weekJy
chans, you can use a 30-week Qstick for tbc long-term trcnd;
or you can use a 5-week Qstick for intermediate-term analysis.
On a daily chart, a 20-day Qstick could be used for int.ermediate tenn analysis.

Qstick and Crash of 1987

77

mentum as a signa! ofhigher prices, anda decrease in intraday

momentum as a signa! of lower prioes. A sell signal occurs
wben the shorter average crosses bclow the longer one.
A longer-term, trend-foUowing approacb with Qsck is LO
use the zero crossing of tbe Qsck values. You will observe
that a change in sign often accompanies a change in market
trend. Wben Qsck is negativc and then becomes positive,
you can go long tbe next day. When Qsck is positive and
becomes negative, you can go sbort tbe next day or clase out
your long posit.ion. Aggressive traders could anticpate tbe
zero crossover and put the trade on a day carlier.
CASE IN POINT: QSTICK ANO CRASH O F 1987

Figure 4.2 shows Qslick data for the Standard & Poor's-500
Index during the three montbs prior to tbe crash of 1987. We

Qstick Trading Strategies

You can plot Qstick over a candlestick cbart to spot diverge!1ces. There .is ?ften a divergence between Qstick and tbe
prices beforc s1gnificant tops and bottoms. A di vergencc lasttng severa! days is the signa! of an imminent trend change
because of a change in intraday momentum. T his occurs bccause intraday momentum oftcn changes bcfore interday momentum.
We can smooth Qstick itselfwith another moving average
and check to see ifQstick is above or below it. When markets
make giant price moves, a slackening of the momentum can
be seen when Qstick crosses below (or above) its trailing average.
You can also use the Qstick idea witb a combination of
moving averages to generate signals. Wben using a short and
. long moving average of the intraday momentum, a buy
s1gnal occurs when the shorter average crosses abo ve tbe
longer one. Hence, we intcrpret an increase in intraday mo-

llOOOO

m .ooo
310000

300000
01t1tk -026 018 000
100
100

FIGURE 4.2 An 8-day Qstock plotted under the S&P-500 index in the
period pnor to the October, 1987 crash.

QSTICK: The Quanttotve Candlestick

Qstick and Mo:::m::;::.:nl:;:u::,:m_ _ _ _ _ __ _ __ __ _ _ __79

calculated Qstick witb an 8-day period and also have shown

an 8-day simple moviog average of Qstick values. We found
divergences between tbe index and the Qstick indicator at t.be
August top. These divergcnces also occurred al tbe next lower
hig.b in early October. Wbeo tbe Qstick crossed under its
moving average, a downward price move foUowed.
AJtboug.b the Qstick analysis could not have foreseen the
magnitade of the crash, it did signa! a downtrend when it
closed below its moving average. This happened at the 320
leve! on the index. Note how the Qstick was accelerating away
from its own moving average leading into the crash.
This analysis does not dcpend on using eig.bt days of data
10 calculate Qstick. You could vary the number of days and
get the same results. In Figure 4.3 we show the same data as
in Figure 4.2, but with a 16-<lay Qstick. The leogth of the

smoothing average is eight days in both figures. Note the

divergence between priccs and Qstick in August: Qstick
peaked beforc prices made a sccondary, lower hig.b in early
October.

78

QSTICK ANO MOMENTUM

Because Qstick is dcfined using a momcntam calculation,
therc is a broad similarity between Qstick and momentum.
Neverthelcss, the two can diverge s1goificantly based on market action. We illustrate this feature usiog the Septcmber, 1993
Colfee futures contracl in Figure 4.4. We show tbe 8-day
Qstick and momcntum superimposcd on one another. Qsck

CQfFEE!&!IJ.O 1 01/lY.13 C-7JEO

o-nes IP7356 lI0.35 \/W1J821

3'11.cm
335 cm
330.cm
325.cm
m cm
316.0::0
310 cm

co

305 cm
300 cm
Ostlck 1 51 -O 3l OCO

1.50

CO
-OCO

50

FIGURE 4.3 A 16-day Qslick and 1ts 8-day simple moving average,
shown with the Standard & POO(sSOO index m 1987.

.. oo

FIGURE 4.4 A companson of the 8-day Qstock and 8-day momentum

for the September, 1993 coffee contract. The heavy line is the Qstick
and the thm line is the momentum.

l_lo_ __ _ _ __ _ _ _~QSTICK: The Quantitati~ C.ondlestick

Qsticlc and Momentum

an~ momc~1um djvcrge and converge depending on markel

acuon; spec1fically, they tend 10 converge when marke1s make
trencling moves, aod diverge during sideways pcriods. Qstick
often moves faster and farthcr than momentum.
Wben the open is near the prior day's closc, then Qstick
and momentum converge. Ma1hematically, the dilfcrence be1ween today's close (C) and the open (O), C-0, is roughly
equal to e-e,, when yesterday's close. e,, and today's open
O are not far apart. Thls occurs during quiet pcriods in the
mru:ket. Whcn the open is far away from the prior close, 1hen
Qsuck and momentum diverge. This djvergence usually occurs in trcnding markets dri vcn by extemal evcnts.

You no doubl recognize the similarity 10 thc RSI. You can

now gauge ifwhlte or black candles dominated recent market
ac1ion. IMl values will be greater than 70 if white candles
predomina te. Conversely, !MI values will be bclow 20 ifblack
bodies predominate. Wh cn the market is in a trading range,
IM1 values wilJ be in the neutral range of 40 to 60.
As a marke1 1ries 10 bouom after a seU off, there are gradually more candles with whlte bodies, even 1hough priccs
remain in a narrow range. We can detcct tbis shifi using the
IMI, because its values will increase 1owards 70. Sirnilarly,
as a market begins to 1op, thcre wilJ be more black candlcs,
causing IMl values to decline 1oward s 20. Thereforc, lMI can
be uscd to quantify sub1le shifts in intraday momentum tbat
usually precede changes in interday momcntum. Thus, the
IM1 is another way to quantify candlesticks.
Noticc, for examplc, 1he !MI for the deutsche mark March,
1993 contract in Figure 4.5. The interna! momen1um indcx
peakcd and botwmed shortly before closing priccs.
In short, the lMI is useful because 1be intraday momentum
differs from interday momentum (Qstick) at kcy turning
points. As a rcsult, we can use the lMI and Qsck togethcr
for a better understanding of price action.

lntraday Momentum lndex

We notcd that lhe dilferencc bctween the closc and the open

is a1 the hcart of candlcstick analysis and quantificd this difference using a sim ple moving average to get Qstick. Further,
~t ~ould be useful to have an oscillator 10 pinpoint extremes
to mtraday momeotum. We could develop such an oscillator
the ~traday momentum index {IM!), by adapting the com~
putauonal approach used in relative strength index (RSI) calculat10ns.
Follow!ng the ~SI approach, we separa te thc intraday momentum 1.nto bu llish days and bearish days, gctting positive
numbers in .both columns. Wc can then define an intraday
momeotum mdex (IMI) over, say, 14 days using the cquation
below:
ISu = Sum of Intraday Momentum UP (C > 0),
!So ~ Sum of ln1raday Momcntum Down (C < O),

lMI _
!Su

- (!Su + IS0 ) !OO.

(4.2)

81

:::

90.00

8000

0.625
0 .620 .

Dailydose

70.00

6000

0 .615

::::
l
0.600

- 5000

::.:1

OM OAILY CLOSE THIS SCALE

o595

0.590

0585

1--~-- -~--~

11 /> 3~2

,.,,
11'25192

12; 0Ml2

12111192

11

2000
'00

_ __,_ 000
12'31192 01 13i93

01'25~

FIGURE 4.5 The interna! momenium index for deutsche mark data

8
_ 2_

_ _ _ _ _ _ _ _ _...::>.
Q.STICK: The Quantitative Candlestick

Quantifying Candlestick Shadows

83

QUANTIFYING CANDLESTICK SHAOOWS

Justas we have quan1ified the candleslick body by defining
Qstick and IMI, we can quantify the sbadows of the candlestick patterns by separa ting the upper shadow and lower
sbadow for ca.ch day. Tbis will bigbligh1 days witb unusually
long upper or lower shadows, a useful fealu.re, since tbese are
thougbt lo occur near tuming points. You can define tbe shadows as:
upper sbadow lower sbadow -

higb
higb
open
close

close (if close >

open (if close <
low (if close >
low (if close <

open),
open),
open),
open).

(4.3)

You can take a simple or exponential moving average of

tbe daily upper and lower shadows 10 show tbe recent trend.
CandJestick analysts have observed that, when a market is
definite about its direction, it shows short shadows. Hence,
when thc moving average of thc shadows is declining, we can
be confidcnt about market directio n. However, when a market
is unsurc about its direction, it has long shadows. We can
observe this indecision asan incrcase in tbe moving averages.
To sp01 markct indecision, follow the daily shadows and their
moving averages.
Candlestick analys1s ha ve categorized the formations caJJed
hammer or hanging man by their long sbadows. These are
reversa! patterns, and it is the length of the sbadows that is
more important than the color of the body. Subsequently, a
plot ofthe upper and lowcr shadows wiJJ quickly identify days
with unusually long shadows. You can tben compare tbe
lengths of the shadows to those at previous tuming points.
Thc Treasury Bond markct oflen shows long sbadows. Figure 4.6 illustrates the candlestick cbart for tbc 09/ 93 T-Bond
contract along with the measured upper and lower tails plotted below. We bave circled tbe interest:ing arcas ofthe pauern.
Note thc long upper and lower shadows showing indecision

FIG URE 4.6 Candlestick analysis of the September, 1993, Treasury

Bond futures contract, and a graph of the upper and Jower shadows.

(or nearby rcsistance and support) i.n March. These shadows

wcre much longer than usual, as shown by the lower grapbs,
warning of an intermedia te top.
Also note the upper shadows showing local resistance in
carly April in this figure. When the marke! pushed througb
this resistancc, it moved quickly to new hgbs. Once agarn,
there was rcsistaoce near highs of March, as shown by the
long upper shadows. The sideways movement produced large
shadows, showing indecision. Similarly, the base fonncd. to
mid-to-late May also produced large lower shadows, showtng
nearby support.
The currency markets also show long shadows, as you can
scc in Figure 4.7 of the 09/ 93 deutsche mark contrae!. Long
Jower shadows show local support, wbereas upper shadows
show local resistance. Cboppy sideways trading (such as late

QSTICK: The Quantlltive Cand lestick

85

Analyzing Thru Stock! with Qstick

CASE IN PO INT: ANA LYZING THREE STOCKS

WITH QSTICK
In thls section you'll see applications of the Qstick idea to
longer-term analysis of stock prices. These examples show
how quantified ca ndlesticks can improve your analysis of
stock price action. Figure 4.8 shows a weekly chart of Amgen
(AMGN) covering the period June, 1992 through June, 1993.
The figure includes a 10-week simple movi ng average of thc
week.ly close and a 5-week Qstick smoothed with a 5-weck
simple rooving average.
Amgeo rallicd from mid-1991, peaking at 78V in early January, 1992. The broad top in Qstick in Scptcmber, 1992 led
to a sideways roove into November. Then, Qstick bonomed
and rose sharply in to January. The Qstick peaked on 1/03/

FICUR.E 4.7 Candlestick analysis o the September, 1993, deutsche

mark futures conlract wilh upper and lower shadows ploned separately.

CXll

CXll
(XI)

June and July) produoed long shadows on both sides. For

exam ple, a long lower shadow showed support as the market
bounced weakJy; a long upper shadow showed resistance. The
DM weakened after thls long upper shadow, making new
lows.
You can meas~ the upper and lower shadows and plot
the measurements m graphs for trading decisions. A graph of
the up~r and lower sbado~ lets you compare today's shadows with recent market action to evaluate its significance.
';'ou could use the upper or lower shadows to find local res1stance and s~pport, selling near resistance and buying near
thc support, wtth a close stop to protect your position.

ollJ (XI)

350
251'.l

150

50
, Sil

Jun

JUI

l4JJ

St

Oct

"""

Dtc

92

F"tb

Mar

Juo

FI GURE 4.8 A weekly candlestick chart o Amgcn in 1991-92 wilh

a 5-week Qst1ck .

86

QSTICK: Thc Quanlitlivc Candlestick

93, abead ofthe actual peak in prices two weeks later. Qstick
dropped below its own moving average tbe wcek of 1/ 17/ 92,
wben Amgen closed at 70. Thus, tbere was ad vanee notice of
a significant top in the sbarp peak in Qstick and its decisive
break below its own moving average. The sell signa! was in
force for three weeks before Amgen began its slide into ApriJ,
1993.
Figure 4.9 shows weekJy Amgen data from July, 1992
through Jul y, 1993. Note again the sharp peak in Qstick in
thc weck of 11/ 13/ 92, before tbe stock peaked at 78 the week
of 12/~~93. This w~ a double top in Amgen. Tbe Qstick
was dec1s1vely below lis own movingaverage in January 1993
before Amgen prices fell bclow the 10-wcek average
collapsed. Qstick bottomed witb prices m Marcb and accelerated above its moving average close to the act~al bonom in

and

go..w eeklr 0111sm C-35 lf,O 500 0-35 500 HooJ6 7SJ L"35 DDO

Analyzlng Thrce

Slocks_::
w:_::ll:h...::
:.: Q~s::.:
lic::.:k:..__ _ _ _ _ __

_ _ _-"-87

mid-March. This cxample shows tbat Qstick provides early

warning oftrend changes, particularly when the market malees
sbarp moves.
Qstick also proves valuable in analyzing Ionger moves, as
you can see in Figure 4.1O, a chart of week.ly General Electnc
(GE) data from February, 1988 through November, 1990.
This period includes the move following thc October, 1987
crash into the next significant decline. We plotted the 30-wcek
Qstick witb its 30-week simple moviog average. Note how
the Qstick made a sharp peak on 6/ 1/ 90 and broke down the
week of 8/ 3/ 90, well before the big slide in October. GE was
at 69'1 when Qstick broke below its multiyear upward channel. The Qstick analysis providcs a useful insight into the
trend change in GE.
The GE analysis continues in Figure 4.11 coveriog July,

v.11sn
cm

otQ+
+o

DDO

oco
'll.DDO
O<i<k 000 -036 000

"l<k o35-12

o Jl'j oIXXJ)

200

050XI

O.DO

O.Dl'.l

2DD

.,..

Od

...

IODO

00

o.e

93

... .... "'

FIGURE 4.9 A weekly candlest1ck chan of Amgen m 1992-93 w 1th

a 5-week Qslick.

M A M J J A S 0"081!F M AM

FIGURE 4.10 A muh1year weekly candles11ck chart for General Electric along w11 h a long-term Qstick and its moving average.

88

QSTICK: The Quantlltl~ Candlestidc

"' k 0360 16 0 00

olll

...........

060

Summuy

89

Finally, Figure 4.12 is example o f intennediate tenn analysis usi ng a 20-day Qs ck aod its 20-day moving average. We
used Philip Morris (MO) daily data from its top in September,
1992 through May, 1993. Tbe Qstick fcll below its average in
mid-September, signalling tbe coming top. The 20-day moving a verage turned negati ve in mid-October, suggesting lower
prices to come. The 20-day a verage moved up to tbe zero line
following tbe rally in mid-December. It stayed in oegative
territory ali tbe way to late May, 1993. The Qsck also did
oot 1urn positive from the secood week of December, 1992
through the first week of May, 1993. The Qstick indicator
called the top in MO and stayed oo the short side of tbe
market throughout tbe decline. Using Qst.ick, you could have
avoided a big decline in Philip Morris, o r mo re tban doubled
your mo oey with a shon postioo.

lll
J

S ON

D91 F

MAMJ

SO N

0 92

MA

fl~URE 4.11 A weekly candlestic k chart for Gene ra l Elec tric a long

w11h

Ph1lMofflOMy 07113193 <>-49125 31!> 0-9 f'm H=S0.500 L"l 750 """29-'92

-~"'1--

a Jong-term Qstick.

1991 through .April, 1992. This figure shows a 30-week Qstick

s~oothed by 1ts 3D_-week moving average. A trend lioc analys1s qu1ckly r~veaJs 1mponant cba~gcs. Tbe key turning points
occurred a bit bcfore key moves m GE stock. From this you
~ ~ '!1at Qstick analysis can be used for long-tenn an~lysis
of mdivtdual stocks.
Fisu.n: 4. 11 also shows tbat Qstick can be used to confirxn
candlestick patterns. Tbe bullish eng~lfing l?ttern in January,
199!, was co~firmed by a bottom in Qsttck. Simitarly, the
bullish mommg star . pattei:n in December, 1991 was confirmed. by an upt1ck m Qsuck. The peak in Ma y, 199 1, did
not qwte produce a dark cloud covcr pattem; it did, bowever,
~ur near tbe top ~f tbe Qsck rall y, and broke tbe uptreod
bne. Thus, tbe Qsuck chart belped to clarify a potential top.

FI GURE 4.12 lntermedia1e-1erm analysis using daily data for Phillp

Moms w11h a 2o-day Qshck and its 20-day simple moving average.

90

QSTICK: The Quantilli~ C.ndlestick

SUMMARY
The purpose of quanti fy ing candlesticks is to improve identification of candlestick patJerns. We quantied the patterns
using tbe candle body, via Qstick (a movi ng average) and the
interna! momentum indcx (an oscillator). We also quantified
tbe sbadows by sepa rating upper and lowcr shadows. Tbcsc
indicators can be uscd by themsclves, or to clarify interpretation of candlestick pattcms. We showed bow thcsc ideas can
be used for shon, intermcdiate, and long-tenn anal ysis of
stocks or futures.

TUTORIAL ON QSTICK

Calculating Qstick

TutoriI on Qstick

Table 4.1 Sample Qstick Calculattons wth Deutsche Mark,

March 1993 Contract

Day
1
2

3
4

5
6
7
6
9
10
11

12

Table 4. 1 shows data for the deutsche mark, Marcb, 1993

contract. We did not use any decimal points in the data. The
fust day was 01 / 18/ 93 and the last day was 02/ 02/93. We
calculated thc daily Qstick value as the difference between
the close and the open; in the last column we have taken a
5..<fay simple moving average of the daily Qstick values, but
you can change the number of days as you wish.
Let's assumc you are a trend follower. You would lhen go
short on day 11 since the Qstick value had crossed O and
becomc ncgative on day 10. But wbat if you werc an aggressive trader? Then, you would take the sum of daily (C-0 )
changes at the start of day 1O. This sum is 35 ticks
(+59- 27-7+ 10); bencc, you recognize thatifDM closed
35 ticks or more below the open, that would push Qstick
below O. You could plan a short sale, say, 40 ticks below thc
open of day 10. Thus, you could have sold shon on clay JO
at the 6,209 leve!. Your profit would have been over $2, 100
on this trade over the next two days.

91

Open
(0 )

H1gh

6063
6t5 1
6152
6185
6155
6241
6295
6287
6256
6249
6166
607t

6162
6176
6191
6207
6053
6322
6338
6290
6314
6281
6t73
6t05

low

Close
(C)

Daily
(C-0)

Qstic k
5-day
Average

6057
6t29
6152
6132
6t51
6235
6274
6251
625t
6156
6066
6038

6153
6163
6185
6149
6241
6300
6288
6260
6266
6166
6071
6055

90
t2
33
-36
86
59
-7
-2 7
10
81
-95
-1 6

37.00
30.80
27.00
15.00
24.20
-9.20
- 40.00
- 4 1.80

Table 4.2 Sample Calculations for IMI Using Deutsche Mark Data

Open Close
(C)
(0)
6063
6t51
6t52
6185
6 155
6241
6295
6287
6256
6249
6166
6071

6153
6 163
6t85
6149
6241
6300
6288
6260
6266
6168
6071
6055

5-Day
Down Sum of
Up Oay Day Up Day
(C - 0) (C - 0)
(E)

90
t2
33

86

59

o
o

tO

o
o
o

5-0ay 5um of
Sum of Up&
Down Down
Oay (r) Day (G)

IMIE/G
100

o
o
o
36

o
o

7
27

81
95
16

221
190
178
145
t55
69
10
10

36
36
43
70
34
11 5
210
2t9

257
226
221
215
t89
184
220
229

85.99
64.07
80.54
67.44
82.01
37 50
4.55
4.37

92
QSTICK: The Quantitative Candlestick

Calculating IMI
Let's look at 1MJ cal ti

Table 4 2) Wt fir cu a ons usrng deu1scbe mark data (see

e st separatc the daily (C- 0) diffie
.
up and do d
On
rence 1010
lhe ope . wn days.
up days, tbe markct closed higber than
We sho!' a~-d own days thc market closed below tbc open.
dav (column F)ai~:::::; of the up-day (column E) and down( h
ay momen1um Thcn we show th
?L t os~ momcntum numbers (Col~n G) The IMJ . ~ suml
u1e rabo of columns E and G.
.
is sunp y

5
New Momentum
Oscillators

A large group of 1ecbnical indicators falls into the general

category of momentum oscillators, which show market momentum over a fi.xcd time period on a fixcd se.ale. We examincd many such momentum indicators in Cbapter 1, such
as the relativc strength index (RSI), the stochastic oscilla1or,
and the commodity channel index. The diffcrences in tbese
indicators arisc out of their definition, thc smoothing scheme
and scaling approach.
Rcgardless of tbeir differcnccs, these oscillators llave the
following limitations in cornmon:
None ofthem is a "pure" momentum oscillator that measures momentum directly.
Tbe time period of the calculations is 6.xed, giving a diffcrent piclure ofmarket action for different time periods.
They ali mirror the price pattern; bence, you may benefit
more by directly trading prices themselves.
Tbey do not consisten ti y show extremes in prices because
they use a constan! time period.
93

94

New Momentum Oscllaton

The smoothing mechanism introduces lags and obscures

short-tenn pricc extremes lhat are actually valuable for
trading.
This chapter introduces a package of oscillators that addresses and mitigates the weaknesses of the RSI and other
widely used momentum oscillators.

CHANDE MOMENTUM OSCILLATOR

One ofthe most powerful measures ofprice cvolution is momentum, d7fined as the difference between today's close and
sorne preVJous close. The Chande momentum oscillator
(CMO), introduced in 1he previous chapter, is a pure momenium oscillator that plots momenturn on a bounded scale.
lt helps ~o spot extremes in market momentum, and you'U
see that 1t has many uses in technical analysis. Thc CMO is
a variant of the RSI, but diffcrs from it in the following ways:
l. 1t directly measures momentum, because it uses data
for both up and down days in its numerator. The RSJ

uses data for up days only in its numerator.

2. 1t does not use built-in smoothing in its calculations,
and hence does 001 obscure short-Uved ex tremes in momentum. Calculations are perforrned on unsrnoothed
data, then, once calculated, CMO values can be
smoothed like any o ther iodicator. The RSI has builtin smoothing (calculations are performed on smoothed
data), which has a great effect on its values.
J. 1t varies from - J00 to + 100, so you can see net market
momentum at a glance on a bouoded scale. RSI vares
from O to 100, so you must use the 50 level to assess
net market momentum.
You can more clearly see the differences between CMO and
RS 1 in the following defi ni ti o ns:

9S

Chande Momentum Oscillator

Cbande momeotum ocillator = 100-(S.

unsmoothed RSI - 100-(SJ (S. + S. ))

S4 )/(S.

+ S.)
(5. 1)

Here S is thc sum of up-day momentum over x days, and

s. is 't h; sum of down-day momentum over x days. (Re~er to

the tutorial on RSI for a detailed look at thcsc quanunes.)
The CMO numerator has the differencc betwccn the u~ay
momentum and down-day momentum. The numerator is actually the x-day momentum because the values of S. ~d s.
are not smoothed before being used for the CMO calculauons.
The denominator gives the indicator its ~unded scale. Consequently, C MO is a net momentum osc1llator.
The RSI calculati ons smooth s. and s. before they are used
to calculate today's RSI value. This built-in smootbmg mechanism has a grcat influence on the actual RSl values, be~u~e
it often obscures short-lived extremes in momcntum. fbs is
an important difference between CMO and RSI. We will ~lar
ify these ideas by showing sample CMO and RSI calcula11ons
using T-Bond or S&P-500 data.

CMO Calculations: T-Bond Data

Let's use daily T-Bo nd 06/93 contract data and begin by calc ulating the up-day and down-day daily momentum. Next,
we calculate the 10-day sum of the up-day momentum (SJ
and down day momentum (SJ. We calculate the 10-day CMO
using Equation 5.1. Using the data from Table 5.1 , we show
the calculation of the first two values as:
CMO, - 100'"(2.0937-0.375)/(2.0937+0.375) = 69.62,
CM02 - 100*(2.25 - .375)/(2.25 + 0.375) - 71.43. (5.2)
We use the subscripts 1 and 2 to indicate t.he first and second
days. Note that we did not smooth the da1ly totals for s. and
s. We prefer to calculate CMO with unsmoothed totals. You

Ntw Mo-ntum Oscilltors

Table 5.1 Calculating 10-day CMO Using T-Bond 06/93 Data

Close
Up-Mtm Dn-Mtm
S(u)
S(d)
IO-Day

101.0313
101.0313
101. 125
101.9687
102.7813
103
102.9687
103.0625
102.9375
102.7188
102.75
102.9063
102.9687

o
0.0937
0.843 7
0.8126
0.2187

o
0.0938

o
o

0.0312
0. 1563
0.0624

o
o
o
o

Up-Mtm Dn-Mtm
10-Day 10-Day
Total
Total

0.125
0.2187

0.375
0.375
0.3 75

69.62
71.43
71.08

can ~hen sm_ooth the CMO values using a simple or exponcnual moving average.
A be~er way to visuaJize the CMO is to rewrite it in terms
of rela~ve and absolute momcntum. Relati ve momcntum is
thfc dail~ s1gned momcntum, and is given by the difl'erence
o today s close and yestcrday's close:
CMO - 100- (Momentum)/ (jMomentump.

also can calculate a CMO usiog daily highs, lows, or opening

prices.

RSI Calculations: S&P-500 lndex

o
o

2.0937
2.25
2.2187

97

CMO

0.0313

o
o
o

Chand Momontum Oscill.itor

{5.3)

f ~e CMO simply shows the nct momentum as a fractioo

o . e abs?lu_1e momentum changcs over the desired time
penod. This. is another reason for calling CMO 3 pure momeotum osc1llator._ lf the ID3;f~et moves up or down for severa! days, CMO. will be posrnve or negative. You can then
tell~ a glance ~ th~ i:narket action is one-sided. This is a
use . feai;ure, smce it is aJso a measure of trendiness of the
closmg pnce. A market that is trending stroogly has a high
CMO value. We dened CMO using closing prices, but you

Lct's illustrate the impact of the smoothing scheme on indicator values by showing 14-day RSI calculations, in Table
5.2, using daily closing data for thc S&P-500 indcx. Figure
5.1 also shows the daily close for the S&P-500 index and uses
a pon ion of these data in Table 5.2. Note the panicular pattem of closing prices for the index, espccially the sequcnce
of closes and their positions relative to one another, beca use
we'U refer to them later.
Table 5.2 shows the up-day and down-day momeotum. The
up-day momen tum is Oiftoday's close is less than yesterday' s;
otherwise, it is the absolutc difference belween thc two closes.
The down-day momentum is zero if today's el ose is greater
than yesterday's close; otherwisc, it is the absolute difference
between the two closes.
Thc oext two col umns ha ve the 14-day sum of the up-day
(SJ and down-day {S.) momentum. Figure 5.2 shows the unsmoothed 14-day sum of up-day and down-day momentum.
Notice that these lines also show price extremes. The two
lincs cross one another when tbe net momentum over 14 days
. When thc 14-day net momentum is postivc, the line representing the up-day sum {S.) is above thc down-day sum
{Sd). Thc reverse is true when the 14-day net momentum is
negatve. The CMO shows precisely this relationship, sincc
the numerator is the differcnce between S. - Sd.
The RSI calculations, described by J. Welles Wilder, J r. in
his book New Concepts in Technical Trading Systems, smooth
the s. and Sd data from day 2 forward. This smoothing adds
1/ 14 of the ncw vaJue to 13/ 14tbs of the previous value of
these variables, wbich amounts to roughly a 27-day exponential moviag average.

98

New Momentum <>Klllort

Ow>d~ Mo~tum

Otdlltor

Table 5.2 RSI Calcula1ions for S&P-500 lndex

Date
01/04/93
01 / 0S/ 93
Ot/06/93
0 1/07/93
01/08/93
01/11/93
01/12/93
01/13/93
01/14/93
01/15/93
01/18/93
01/19/93
0 1/20/93
01/21/93
01/22/93
0 1/25/93
01/26/93
01/27/93
0 1/28/93
01/29/93
02/01/93
02/02/93
02/03/93
02/04/93
02/05/93

UpS&P-500 Day
Close
Mtm

435.380
434 340
434.520
430.730
429.050
430.950
431.040
433 .030
435.940
437.150
4 36.840
435.130
433.370
43S.490
4 36.1 10
440.010
439.950
438.110
438.660
438.780
442.520
442.5SO
447.200
449.560
448.930

0 .00
0.18
0.00
0.00
1.90
0.09
1.99
2.91
1.21
0.00
0.00
0.00
2.12
0 .62
3.90

O.DO

o.oo
O.SS
0 . 12
3.74
0.03
4.65
2.36
0.00

DnDay
Mtm

14-Day
Up Down RSI
Sum
Sum Smth

RSI
UnSm

460-000 - - - - - - - - - - - - -455.000
S0.000
S.000

- - - - -- ,

0.000

435.000 ,.__ _
430.000

1.04
0.00
3.79
1.68
0.00
0.00
0.00

25.000
20.000
415.000

L..::==-------:-:--::-:::-:::-.:--::"'.::-:;-::;--;:;-::;-;:~
~ ~ iE. ~ ~ ~ ~ ~ ~ ~;;? p' ~ ~ ~ 7' ~ ~ ~ ~ !:: ~ 5 ~ ~ ~ ~
-~~,..I ~~~r

;::,.

~~-.._.se:

~~~

bo,o:"""

~~~V':; 8!:::<~

,-..~-S?.-..

~~~~~~~~a~~GGS

OC'

AGURE S. I Oaily close of the S&P-500 index.

o.oo

0.00
0.31
1.71
1.76
0.00
0.00 0.79
0.00
1.01
0.06 0.94
1.84
0.87
0.00 0.85
0.00 0 .80
0.00
1.01
0.00 0.94
0.00
1.20
0.00
1.28
0.63
1. 19

s.
s.

3000

..--- -- ----,---- -- - -- ----,

25.00
20.00

0 .74
0 .68
0.64
0.72
0.67
0.62
0.58
0.54

o.so
0.46
0 .48

s.
s.

15.00

51.71
59.66
S9.SO
54.60
55.77
56.04
63.44
63.SO
70.62
73.46
71 .48

51.71
61.73
6 1. 29
66.70
72.91
70.40
75.13
72.80
74.89
76.10
75.09

We used the sm oothed vaJues of and

for RS I caJcufor CMO. Th.is
lations and un smoothed values of and
is tbe critica! dilfere nce between RSI and CMO. T he o ther
imponant dilference is in tbe oumerator: Tbe RSI uses just
the up-day sum, S,,, whereas CMO uses the x day momentum,

10.00

6.00

14oiv ' '"'" ot ctowndv momfl'lt1.1m

o.oo L-------~-:--_..:,~::-=:_::-.;:_"""..=-::_-:
~::-::~:---:-;:;"'.~--:::-::;:;-;:;:-;::i~~-::
r;;t~o' ..... .e... ""Q'>CI'~
f7'~~

. ;:.': .

~-~~ss~~>~~-.a~~-i~~
~~
~o""'
- - .t:::.,...._ ~......._......__,
--~I
.J;::::.;:=;.
:;;
S~ci
e.a
o ...
o ,.,..;;-~,..._.,
.......o ~ ""

....

_..

......

<"11

.....

~~s-~
,
-. i!:.~g"" i.s

FIGURE s.2 This shows plots of the 14-day sum of upday _mo'.11en tum and down-day rnomentum for the S&P-SOO index data in Figure
5.1.

given by (S. - s.). For exam ple, o n 1/25/93, the smoothed

vaJues of s. and s. are caJculated as follows (see Table 5.2):

s. s. -

(0.79t 3 + 3.90)/ 14 - l.0 121,

(0.7413+0)/ 14 - 0.6871,
RSI - 100*(1.0121/(1.0121+0.6871))

(5.4)
~

59.56.

100

New Mornentum OKillton

The 01/25/93 value of RSI in Table 5.2 is 59.66. Thc difference between our sample RSI value of 59.56 and tbe tabulated
val~e is caused by the rounding of tbe numbers.
Figure 5.3 shows the 14-day smootbcd RSI and thc 14-day
unsmoothcd RSI (RSI'). Notice how tbe RS I looks similar to
the price changes (price partero) of the S&P-500 index. The
RSI' dfers markedly from the RSI. It looks more likc mome~turn, as you will see below. Note how the RSI' was above
80 tn Mar~h! even though th e RSI barely reached 70. Your
tradmgdec1s1oos would be quitedilferent in the two instances
w~ch highligbts the importance of smootbing in indicato;
des1gn.
Figure 5.4 shows th e 14-day unsmootbed CMO and the 14day RSI. Note that there are significant dilferences in tbe
appearance ofthe two curves, even though CMO and the RSI'
are re~at~d, _sin~e CMO = 2 * RSI' - 100. You can clearl y
~ee this s1m1lan_ty by comparing thc appcarnnce of thc CMO
m Figure 5.4 with the RSI' in Figure 5.3.
. Figure 5.5 plots the 14-day momentum of tbe S&P-500
mdcx over the same time period as in Figures 5.3 and 5.4.

Chande Momentum Oscilltor

''

6000

40.00

<RI

~)

o 00 '>--"-----~:.. ;.---.;;,J"T---.,p:i....:.+-----1

~~ p'\p ~ \:t~r
~ooo '-----------------;:;.........:._ ___.

20.00

14 doy CMO

t!'tf~

01 /22193 02J05.193 02122193 OJ/08193 03/22193 04'05193 0/20193 05/04/93

ACURE S.4
day CMO.

A comparison or the 14-day smoo1hed RSI and the 14-

25.000 - - - - - - - - - . - - - - - - - - - - - - - - ,
20.000

15 .000
10 000
5000

0.000
-5.000
10.000

1oooo
r-------------------~
9000
8000

70 00
00 00
50.00
4000

h~~
. \:
-.~--

....

..\.

...

f\..;

14day RSI

15.000 ' - - - - - - - -- - - - - -- - - - - - - - '

01/22/93 02/05193 02/22J93 03/08/93 OJJ22/93 04'05/93 04/20/93 05104/93

FIGURE S.S
Figure 5.1.

The 14-day momcntum lor the S&P-500 1ndex data in

3000
2000
14--dav 1moothed RSl
10.00
000 , __ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ __ j

01 '22193 0:!Al5/9J 02122193 03!08/93 03'22193 0.!05193 0.120!93 05~ 93

FIGU RE 5.3 A comparison of the 14-day smoothed and unsmoothed

values of lhe RSI for the S&P-500 lndcx data in Figure 5.1 .

The momentum figure is identical (except for scali ng) to the

unsmoothed C MO. Howcver, the momentum calcula tion (Co
- C.) is, by definition, unbounded. Hence, we cannot readily
transform momentum into a bounded oscillator. But, this is
prec iscly the advant.age of thc CMO dcfinition. Figure 5.4
shows that unsmoothed CMO is a "pure" momcntum oscillator t.hat plots momentum on a bounded scale from + 100
to -IOO(or + l to -1).
Next wc'll show tbe eff'cct of varying the lenglh of tbe

Osdllator>

Chande Momentum O~sc~lll:,:t~or'....__ _ _ _ _ _ _ _ _ _ _ _

103

smoothing period on RSI values. Figure 5.6 compares tbe

effect of a 9-<lay exponential smoothing versus the tractitional
27-<lay smootbing. Using 70 and 30 as reference values for
overbought aod oversold condjtions, the lightly smoothed
RSI reached price extremes more often than Lhe heavily
smoolhed versioo. For example, the 9-<lay smoothed RSI was
overbought in January and severely oversold in April. The
27-day smootbed RSJ barely oudged 70 and 30 al lhat same
time. Thus, your interprctatioo of RSI wouJd change significantly as smoothing cbanges. And, because the CMO does
not ha ve any built-in smoothing, it couJd show price extremes
that RSI may noL
We'll show you sorne actual numbers to show how smoothiog makes a difference to RSI values. Table 5.3 shows the
details of the RSI calcuJations with a 9-<lay exponen tia! moving average (index - 0.20). For this example, look at the
calculations for 0 1/25/93:

Table 5.3 RSI Calculations with 9-day Exponenhal Moving Average

S(d)
RSI
S(u)
S&P-500 Up-Oay Dn-Day
~
SmoothedMlm
Mtm
Close
Date

102

New

Mom~nlum

s. =

0.23.9+0.so.19 - 1.412 - 1.41,

s4- 0.20 + o.so. 14 = o.592 - o.59,
RSI = 100*(1.41 /(1.4 1+0.59)) = 70.50.

(5.5)

Tbe difference from the tabulated vaJue of 70.57 is a resuJt

100.00 . - - - -- - - - -- - - -- -- - -- -90.00
9dtv )lpOMntlal amoothlno
80.00
70.00
80.00
5000
40 00
3000
2000
o.oo ....__ _ __ _ __ __ __ _ _ _ __ _ _ ___,
1000

"1

~~~~~~~~~~~~~~~~~~~~~-~~~~~

~~~$~~s~~i~~~s~ss~ss~~~ss,~s
-~R~~~~~~
~~~~~-~C~~~N~*-
OO$SSSSSs
ga~aa~~~~~~~sa

FIGU RE 5.6 A chart of RSI values with the usual 27-day exponential
smoothing and a short 9day exponential smoothing.

01/04/93
01/05/93
01/06/93
01 /07/93
01/06/93
01/11/93
01/12/93
01/13/93
01/14/93
01/15/93
01/18/93
01/ 19/93
01/20/93
01/21/93
01/22/93
0 1/ 25/ 93
01/26/93
Ot/27/93
01/28/93
01 / 29/93
02/01/93
02/02/93

435.360
434.340
434.520
430. 730
429.050
430.950
431.040
433.030
435.940
437. 1so
436.840
435.130
433 .370
435.490
436.11 0
440.0 10
439.950
438110
438.660
438.780
442.520
442.SSO

0.00
0.18
0.00
0.00
1.90
0.09
1.99

2.91
1.21
0.00
0.00
0.00
2.12
0.62
3.90
0.00
0.00
O.SS
0.12
3.74
0.03

1.04
0.00
3.79
1.68
0.00
0.00
0.00
0.00
0.00
0.31
1.71
1.76
0.00
0.00
0.00
0.06
1.84
0.00
0.00
0.00
0.00

0.79
1.41
1.13
0.90
0.83
0.69
1.30
1.05

0.74
0.59
0.48
0.75
0.60
0.48
0.39
0.31

51.71
70.57
70.04
54.48
57.97
58.83
77.10
77.20

ofthe rouncting to two decimal places. The unsmooth~d RSI

vaJue is 61. 73 as shown in Table 5.2. Thc RST value 1s 59.66
with a 27-<iay' cxponential smoothing. This smootbing often
obscures shon-lived market extremes, which may mean the
diJference between closing a position or revcrsing it, closing
it at a smaller profit, or even a loss.
Becausc there is no built-in smoothing in the CMO calcuJations, you can obtain a smoother indi~ator by i~creasing
the number of days in thc CMO calculauons. For mstance,
by choosing, say, a 14-day time framc, we will have a

104

New Momentum Oscillators

smoother CMO curve versus a 9-day time frame. Figure 5. 7

illustrates this idea by showing a 9-day aod 14-day CMO for
thc S&P-500 indcx. Theo, wheo the 9-day CMO is overlaid
on the daily close of the index in Figure 5.8, you can see that
it was more volatile than the 14-day CMO. The 9-day CMO
clearly showed price extremes in the index by rising above
0.70 or falling below - 0.70. The 14-day CMO in Figure 5.7
did oot show the price extremes as obviously. Thus, you could
use the 9-day CMO for short-tenn trading of the S&P-500
iodex.

CMO as Measure of Trendiness

One of the useful features of the CMO is that it can also be
used as a measure of trendiness, and trendiness measures are
very important because you can use them to open trendfollowing trades when they confirm the presence of a trend.
We discussed trendioess in sorne detail in Chapter 2 on linear
regressioo, and we'll use a trendiness measure-vertical hor-

100

r------- - -- -- - ------- ---.

080

14-day CM O

080

O.O
0 . 20

o 00 'f...~~---+--
020
.()40

.() 80
.() 80

9-dayCMO

..

Chande Momentum

Osc~
il~
la::;
IO
~
r

60 000

1 00
080
0 .80

0 .40

r.

"'

0.20

o.oo
-0.20
-0.40

.() 60
.() 80

~r

100 ~----------------~'-----"'

55000

450 000
5.000

:!

:~= ~~
30.000

25 000
420000

FIGURE 5.8 The values of the 9-day CMO supenmposed upon the
S&P-500 daily close.

izontal filter (VHF)-from that chaptcr for our test of the

CMO.
First, we'll compare thc defi.nitions of VHF and CM9. to
point out sorne similarities between thcm. In the defiruuon
of VHF below, let R be the range o ver x days, where R = H
- L H being the highest high and L being the lowest low.
Lct
be the sum of the absolute value of the daily close-toclose ~hanges (momcntum) over thc x. days. We now define
VHF as:

VHF - R/P.

...

_ __ _ _ _ _ _ __ _ __ 105
_

(5.6)

In calculatiog CMO and RSI, we separatcd the daily closetCH:lose changes into those for up days and down days. We
can take the sum of thosc changes over x. days to get (S. +
s.). You will recogniie that P. is essentially cqual to (S. +
Sd).
.
At key turoiog points, the close is not far from lhe h1ghe~t
h.igh or lowest low of the period. Thus., to a good apprmumation, we can replace the range, R, w1th the absolute dif-

106

New Momcntum Oscllatol'!

Chandc M omcntum Osc;.;;

lllc:.
al:.:
O.:...
_ _ _ _ _ _ __ __ _ _ _1_0_7

ference KC.O - C.)I, the momcntum of the clase taken over x

days. When the market makes a swing move, today's clase
could be at one extreme, and thc clase, x days ago, at the
othcr. The momentum of thc clase could theo be very clase
to thc actual range.
In effect, the absolute value ofCMO is similar in construction to the VHF and should be a good indicator oftrendiness.
Tbe denominator is essentially tbe same, and the numerator
will tcnd to converge at key turniog points. Wc alrcady verified tbis during the discussioo oo linear regression analysis,
where wc found that the smoothed VHF, r' and absolute
CMO (absCMO) were excellent indicators of trendiness.
In Figure 5.9 we compared a 14-<lay VHF to thc 14-<lay
unsmoothed absolutc CMO. The calcuJations are for the S&P500 index data used earlier. \Ve used the absolute values of
CMO (absCMO) since the VHF has only positive values.
Notice bow absCMO and VHF converge at a key rurning
point in early March. Thcre are dilferences in the absCMO
and VHF because thc CMO uses momentum rather than the
range in thc numerator.
The VHF falls as the markct cntcrs in to a period of congestion; that is, the range narrows. Thc denominator ofthe VHF
is a scaling refcrence; the numcrator, or range, determines the

value ofVHF. The price range can widen without rising momentum, because of small progress on a closing basis. The
VHF will show trendioess when absCMO does not. Thus, we
can say absCMO is a measure oftrendiness on a closing basis.
Tbe fact that absolute momentum shows trendiness is reinforced in Figure 5. 1O, which used weekJy data for GE. We
bave plotted Lhe absolute 10-week momentum and its 5-week
simple moving average. Note how the absoJute momentum
increases during a rapid move in priccs and falls when prices
enter a trading raoge. Also ootice how momeotum was flat
ali through the trading range in 1992. The momeotum peaks
at differeot values each time because its scale is not fixed.
In shon, absolute momeotum, and therefore absCMO, can
be used to measure treodiness in data just like VHF. Wheo
Lhcse indicators show that a trend exjsts, you can pul on trendfollowing positions with greater coofideoce of success.

.IXll
IXll
V A BS M T M 963 8 80

3 00
01122193 0 2'5193 02122193 03~3 03122193 04J05193 04120193 05'04193

ACURE S.9 A companson of the 14-day VHF and the 14-day absolute CMO for the S&P-500 index d ata.

FIGURE S.10 The 14-penod a bsolute momentum for General Electric

using weekly data.

108

New Momentum Oscillators

109

Chande Momentum Oscillator

Useful Features of CMO

The beauty of the CMO approach is that it converts momcnrum into an oscillator that you can plot on a fixed scale.
By using absCMO to plot the absolutc momentum on a
bounded scalc from O to 100, you can get a consistent scale
to compare different rnarke1 moves.
You can also use CMO 10 define overbough1 and oversold
levels a1 + 50 and - 50 where the ratio of s. to s. is 3: 1 or
1:3. By simple substitu1ion you can find that a ratio of 3: 1
corresponds to a value of +50. This means tha1 the up-day
momen1um is three times the down-day momentum, a rare
occurrence. The reverse is also true; ifthe up-day mornentum
is only one lhird of the down-day momentum, then this oscillator has a value of - 50.
Ano1her convenienl feature of CMO is thal we get positive
and negative values. Hence, wc can tell if the net market
action is falling or rising. Figure 5.11 shows a 14-day RSI and
14-day CMO for a direcl comparison using tbe July, 1993
Wheat contract. The reference levels for CMO (upper iodicator) are +60 and - 60, and they correspond 10 RSI values
of 80 and 20. Note how 1he CMO stayed in ncgative territory
during the downtrend in prices from mid-January througb
March. The divcrgence between the CMO and prices as tbe
markct made new lows in June was a harbinger of the rally
in July. The RSI did no1 show any significant market extremes
during this pcriod, nor did it show a significant divergence
witb thc price as the contract made new lows.
In addition to RSI, we can compare CMO to other iodicators and find new uses for it. For iostance, CMO resembles
both the true strength index {TSI) and the average directional
indcx (ADX). But, as you'U see, gjvcs us additional information. The TSI, described by William Blau, uses doubly
smoothed exponential averages for the relative and absolute
momentum. Hence, its vaJues diffcr significantly from tbe
CMO and will show grea1er sensitivity to price changes. On
the other hand, because of the exponcntial smoothing, the

3111'6

S 12 lOOO 8000 8l00

..,_7'>'.,,..J"'...____
'V'/"
93

F..

....

0-, .

..

~~

00
:IJ.00

...

FIGURf s. t 1 A comparison of che 14-day unsmoo1hed CMO 10

RSI using lhe September, 1993 CBT wheat futures contract.

the

TSI rarely approachcs extremes of - 100 and + 100. Conscqueotly, thc CMO can show tbe extremes more readily. Thc
choice of CMO or TSI will depcnd on your trading borizon
and your prcfcrcnce for indicator sensitivity.
In Figure 5.12, you can see 1hat thc average absolute value
of the smoothcd CMO closel y tracks the ADX of the directional movement system. The ADX is the thin line, and the
averaged absolute CMO is th c thick line. We used the Septernber, 1993 US Bond contract to show tbe similarity between the 14-period ADX and the 14-period averaged absolute CMO. These indicators rose when the market was
trending, and declined whcn the market was going sideways.
Both indicators measure treodiness, but the absCMO was
more responsive than the ADX io this example.
The ADX is a popular measure of trendiness io data. Many

110

New Momentum Oscillators

Chande Momentum O~::::ci.:::

ll:.:':::'
~-------------'1-11

..

... ...

~-----------------..,.-----,

112

"
108

......

.A

106

....

.. ....

~ _ __;._ _ ___;_...;_--"-- - - ' - - - '- - -----'

102 L,__..:_
.;
FIGURE S.13 The daoly close of the June, 1993 Treasury Bond futures
contract.

FIGURE S.12 The absolute CM O and the 14-period ADX for lhe
September, 1993 T-Bond contract.

traders use it to verify that a trend is underway or about to

end. Jfyou use the CMO and its absolute values, you can use
a single indca tor to do the work of many.

Volatility-Based Composite CMO

In lhe foregoing CMO calculations, we used a fued time period, but you can overcome this limitation by combining lhe
CMO values from several time pcrods to define a single,
composite CMO. One approaeb is simply to average CMO
valucs from different time periods witbout changing tbe relative weights. Altemately, you can use a volatlity-based
weighting to bave a truly market-sensitivc composite CMO.
We will use tbe June, 1993 T-Bond contrae1 to explain tbese
ideas (see Figure 5.13 for tbe daily closc).

We calculated the CMO as usual, using 5, 10, and 20 daytime intervals. We cbose these lengths arbitrarily, using popular round numbcrs. You can see ali three CMO lines in Figure 5.14. The 5-day CMO is the most volatile, aod easily
reaches values of + 1 and - t. Ths happens when there are
five up days or down days in a row. The 20-day CMO is the
smoothest curve, showing the least volatility. The 10-day
CMO has a volatility bctween tbat ofthe 5-and 20-day CMO.
Different time pcriods are better suited to meet lhe needs
of certain types of traders. The 5-day CMO and its volatility
may suit a short-tcrm trader, while tbe JO- and 20-day CMO
volatility may suit an intermediate-term trader. The frequeney of sgn changes decreases as the time period increases,
which just meaos that there is more smootbing in the data
as tbe calculation period increases.
You can achieve a more seositive alternative to the 20-day
CMO by calculatiog the average CMOA. This is simply an
aritbmetic average ofthe tbree values and is plotted in Figure
5. 15. Tbe 5-, 1O- and 20-day CMO values are designated as
CMO,, CM0,0 and CMO As a result

112

New Momenlum Osdllalors

CMOA - ((CMO,

o.e
08

O
02

-o 2
-O

-o 6
-o.

5-dovCMO

..

..

-'

10..W, CMO
"!'

...

20-dovCMO

::'!!

FIGURE 5. 14 Chande momentum oscillator shown using three different tome honzons for the T Bond data in Figure 5.13: 5 days, 10
days and 20 days.

+ CM0,0 + CM020)/3).

(5.7)

and each has an equal weight (of 1/ 3) compared with the

others.
This average CM0 4 givcs you a composite view of CMO
over d.ifTerent time periods. You can combine other time periods a.s you desire. You would buy when the CMO rumed
positive, and sell when the CMO turned negative. This strategy would have worked weU in this trending T-Bond market.
For an even more scnsitive composit.e CMO, look at the
volatility-weighted CMO in Figure 5.16. Firsl, find the 5-day
standard deviation of the 5-, 10- and 20-day CMO, calling
the respective standard deviations S,, Sz and S,. Then, define
the volatility-wcigh ted CMOv by using them as the weights:
CMOv - (S, CMO, + S, CM0 ,0
+ S3* CM0 10)/(S , + 5z + S3).

(5.8)

The volatility-weigbted CMOv is more sensitive to market

acti on than CMOA. Furtherm ore, CMOv leads CMOA at kcy

0.8
0.6

O.
0.2

o.a

0.8

O.

-0,2

02
-O.

-0.6

-0.2

ti

:;;:

:;
~

A GURE 5. 15 The average of the three time periods shown in Figure

-O.

-o 8

'

\
Avtrt09 CM O

-08

5.14.
A GU RE 5.16 A companson of the average composite CMO shown
in Figure 5 1S wil h a vola11h1y-we1gh1ed composite CMO.

114

New Momentum Oscillators

115

Chandc Momenlum Os<:lllator

crossings ifthe market moves rapidJy. Jt also reaches a higher

or lower extreme than lhe CMO,.. When CMOv reacbes a
more extre!De value t~an the C~O,., a markct reversal may
be near. Usmg CMOv 1s a way to mclude multiplc time ftames
and market volatility in lhe analysis of market momentum.

its average, triggenng the sell signa!. The samc approach

wouJd have put you back inlo Lhc market, on lhe long side,
11 days la1er, but you would have misscd a pan ofthe move
from 106 to 108.

Trading CMO with a Moving Average

Filte ring Trading Noise from CMO

To gcnerate trading signals, you can use the CMO with its
simple ~ovi.ng average (SMA). (Figure 5.17 shows the 10 day
~. with l!S 10 day SMA.) This approach allows you to
IDJttate a trade before a zero crossing of the CMO. Be aware,
however, that it sometimes gives early signals. You could buy
when the CMO crossed above its average, and sell when it
feU below its moving average. Sucb an approach would have
a given a good cntry pcint in January, 1993 at the start of
the rally. Howevcr, the shon signal in early February was
premature; thc market went sideways and lost sorne upward
momentum. The loss ofmomentum pushed the CMO beJow

So far we ha ve used the daily momentum change in our CMO

even when the market moved by a small amount. Hence,
sorne of the daily noise of trading crecps into !he indicator.
By filtering the CMO 10 ignore small changes in daily momen1um, you will be beuer able to focus on the big picture.
One way 10 ftl1er the noise is to use the daily difference in
the C MO calculations only if it exceeds a threshold value.
For examplc, consider the actively traded Treasury Bond fu1ures contrae!. Assume that a daily closing difference of2 ticks
(2/32 - 0.0625), or less, signals an unchanged market. We
then can set the daily momentum (whcther up or down) to
O i f the change is "smalJ." We can show !he difference in
approach as follows:
Mtm(up) - C0

Ch if C0

- o
Mt m(up,f) - C0 -Ch

- 0

FIGURE S. 17 A t 0-day CMO wth ts l 0-day smple movng average.

The T-bond data used to c.ikulale CMO are in Fgure 5.13.

C, (unftcred),

ir c.i < e,,

(s.s>

if (Co - C,) > x (filtered),

if(Co - C,) <=X.

You can writc similar equations for down-day momentum,

and choose x arbitrarily. Or, you can use 100 days data 10
find average values for daily momentum changes on up days
and down days. lf you find the standard deviation of those
daily changes over 100 days, you can use thc average change
plus one standard deviation as the value of x in the above
fter. This approach can ignore sorne seemingly large values.
l n Figure 5.18 you can view the CMO calculated with and
without filtering, using the same data as in Figures 5.13

116

New Momenlum Oscillalors

08

CMO

06
04
02

i!\

.() 2
-O.
-O&

-08
I ' - - - - - - - -- - - - -- - - - - - -....--~

ACURE 5.18 A comparison o a fihered CMO anda 10-day CMO.

through 5.17. We calculated the filler using 100 days data and
the average plus one standard devfation of the daily momentum changes. The filter for up-day momentum was 20/
32 and 14/32 for down-day momentum. This means that, if
the daily change on up days was less than or equal to 20/32,
theo t.hat day's up momentum was set to O. Similarly, if the
down move was less thao or equal to 14/32, the down-day
momentum was set to O. The daily momentum equation is
as follows:
up-day mtm - f( (C-o- C,)
down-day mtm - if( (C, - C.,,)

> 20/32, (C.,,-C,), O),

> 14/ 32, (C,- C.,,), O).
(5.9)

Here, C.,, is today's close and C, is yesterday's close.

Note in Figure 5.18 how the filtered CMO identified two
key overbought positions and two key oversold conditions in
the market. You can see that tbe two CMO calculations converge wben the market makes big or significant rnoves. Tbe
filtered CMO nauens when tbe market makes small rnoves

Chande Momenlum Osclllator _ _ __ _ _ _ _ _ __ _1_17

in a shon-lived trading range. lt also gives a clearer picture

of market action when used with the unfiltered CMO.
Zero crossing of botb types of CMO were significant. The
early warning of market CJtlremes provided by the filtered
C MO would have been useful to sorne traders. The amount
of tbe filler was set for intennediate-tenn trading, but you
can eJtperiment with djfferent arnounts of filtering.
We will cxplain the calculation of tbe filtered CMO using
the data in Table 5.4. We set the filters at 5/ 32 for the T-Bond
data for both the up-day and down-day momenrum. ln coturnos 2 and 3, we show the usual calculation of up-day and
down-day momcntum. lf today's close is bigher than yesterday's, then the up-day momentum is the diJferenc.e between
tbe two closes. lf today's close is less than yesterday's, then
h ble 5.4 Calcula ting 10-Day Filtered CMO

Close
101.03
101.03
101. 13
101.97
102.711
103.00
102.97
103.06
102.94
102.72
102.75
102.91
102.97
10313
103.72
104.22

Up
Mtm

On
Mtm

o
o
o
o

0.0937
0 .8437
0 .8 126
0.2 187

0.0938

o
o
o

0 .1563
0 .0624
o.1563
0 .5938
0 .5

0.0313

0. 125
0 .2187

o
o
o
o
o
o

Filtered
Up Mtm
(5/32)

Filte red
Dn Mlm
(5/32)

Filtered
10-day
CMO

Regular
10-day
CMO

o
o

o
o
o
o
o
o
o
o
0.2187
o
o
o
o
o
o

0.79
0.81
0.81
0.72
0.67
0 .73

0.69
0.71
0 .71
0 .60
0 .55
0.61

0.8437
0.8 126
0.2187

o
o
o
o
o
0 .1563
o
o 1563
0 .5938
0 .5

118

New Momentum Osdllators

fotorial: Defining RSI

119

~~~~~~~~~~~~~

the up-day momentum is O. A similar logic applies to the

down-day momeoturn.
In columns 4 and 5, we added a filler to the daily momentum calculations. 1f today's close is greater tban yesterday's close by 5/32 or more, then the up-day momentum is
however, today's
thc difference between the two doses.
closc is higher than yesterday's close by lcss than 5/32 of a
point, then wc set today's up-day momcntum cqual to O. We
assume that thc change is not significant if it is less than 5/32.
As bcfore, if today's close is less than ycs terday's close, tben
the up-day momentum is also O.
For example, on day 3, tbe up-day momentum was 0.09375,
or 3/32. Since this is less than 5/32, we set the filtered upday momentum equal to O. Here we treat an upward close
by 3 ticks in tbe bond market as no change" oo a trading
basis. The same logic applies to down-day momeotum. The
up-day and down-day filters can be different, aod could be
based on market volatility. Tbe filtering process is a way to
reduce noise io the daily momeotum calculations.

rr.

CMO Summary
As you havc scen, the CMO is a flexible and versatile momentum oscillator tbat you can casily adapt to a variety of
trading stylcs. ln sum, you can:
Use C MO to identify overbought or ovcrsold conditions.
Use CMO as a trendiness indicator.
Combine CMO with a moving average to generate trading signals.
Use CMO with a different amount ofsmoothing for trading.
Combine CMO values calculated for differcnt time periods into a composite CMO.
Fil ter the CMO to ignore small changcs in market values.

In short, CMO is one indicator that can do the job of many.

TUTORIAL: DEFINING RSI

In technical analysis we define momeotum as the difference

between two prices. Tbe time period between the two prices
varies; they can be as little as one period apart, where the
period can be any iime length from minutes to months. Most
commonly, we use one trading day as the unit of iimc. With
a one-day time period, the closing prices define momentum
as shown below:
momentum = C(today) - C(yesterday).

(5.10)

Since today's close can be abo ve or below yesterday's close,

momentum can be positive or negative. We will call this "relat.ive momentum." We can define "absolute momentum," in
wbicb we ignore tbe sign of the dail y momentum, as
JmomentumJ = IC(today) - C(yesterday).

(5.11)

The val ues of absolute mome ntum are always positive,

while the values of relative momentu m can be positive or
negative. Lel's do a small exercise. Consider five trading days
using values for the S&P-500 index from 11/29/9 1 to 12/ 15/
91. Now, let's calculatc the daily relative momentum and
absolute momentum. T be ca.lculations are shown in Table
5.5.
Note that the sum of the relative momcnrum over four
days is equal to the momentum over four days. This occurs
because the intermediatc tenns cancel cach other wheo we
add them. ln the equation below, subscripts refer to the numbcr of days back from today (0):

New Momentum Oscillators

120

We can conrinue this exercise in a modified format by separating 1.be daily momentum into values for days wben tbe
market closes up or closes down. In eacb instance, tbe momen1um values wiJI be posive. On an up day, tbe momentum is 1oday's close mfous yes1erday' s close. On down days,
momentum is yesterday's close minus today's close.
Table 5.6 gives us two key results that will ultimately be
used to define RSI. The relalive momentum over four days
Table 5.5

S&P-500

11/29/ 91
12/ 02/ 91
12/03/ 91
12/04/ 91
12/05/91

375.22
381.40
380.96
380.07
Jn.39

6.18
-0.44
-0.89
- 2.68

6. 18
0.44
0.89
2.68

2.17

2.17

10.19

Total

s. -

Date

11 /29/91
12/02/92
12/03/91
12/ 04/91
12/05/91
Total

Relati ve
Absolute
Up-Day Down-Day
S&P-500 Momentum Momentum Momentum Momentum

377.39375.22 2.17

0.5*(momen1um

+ lmomentuml).

(5.13)

Calculating RSI

Momentum Separated for Up and Down Days

375.22
381.40
380.96
380.07
377.39

(5. 12)

You can vcrify this result from the data in Table 5.6 (S. O. 5 ( 10. 19 + 2.17} = 6.18). We will use 1hese resulls to define
Wilder's RSI (see Bibliograpby).

Rela11ve momenium - dose o today - close of yesterday.

Absolutc momentum = lclose of today - close o yesterdayj.
4day mornentum - 377.39 - 375.22 - 2.17.

Table 5.6

sd,
s.

Wc can solve tbese two equalions for s. using algebra Simply add thc two equations and gather terms 10 find:

Relalive
Absolute
Momenlum M omentum

Date

is the dilference between the up day momcntum and down

day momentum over four days (2.17 - 6.18 - 4.01). Also,
the absolute momentum is tbe sum ofthc up-day momentum
and down-Oay momentum (10. 19 - 6.18 + 4.01). Let s. and
S4 be the sum of the up-day and down-day momcntum over
a given number of days. We can summarize tbese results in
the following equations:
momentum - s. tmomentuml = S. +

Relative and Absolute Momentum

121

Tutorial: Oefining RSI

6.18
- 0.44
-0.89
- 2.68

618
0.44
0.89
2.68

6.18
0.00
0.00
0.00

0.00
0.44
0.89
2.68

2.17

10.19

6.18

4.01

(S.)

(S.J

Thc rcla1ive strengtb index measures the proportion of momcn1um change over a given period caused by momentum
on up days. Tbis will become clear from the following definition:
RSI - 1OO*(RS/ (1 + RS)).

(5.1 4)

Here, RS is the ratio of average momen1um over tbe last n

days on up days versus down days. The revised equation looks
like this:
RS = AJA.,.

(5.15)

New Momentum Oscllators

122

Since the number of days in both averages is the same, we

can multiply both averages by the number of days in the
average. This converts the averages into the sum ofthe momentum on up days and down days. Hence, we can rewrite
RS in terms of s. and s., as shown below:
(5.16)
We can now rewrite tbe RSI definition using the new RS
definition. This gives us an RS! defioitioo in terms of the
sum of tbe momentum oo up days and down days.
(5.17)
We can continue our process of adjustment to reOect tbe
relationship between relative momentum, absolute momeotum, s. and s. For instance, we can rewrite RSI using absolute momentum and relative momeotum:
RSI

= 100*(0.5*(momentum + momeotumJ))/(lmomentum),
= 50*(momentum

+ lmomentuml)/(lmomcntum).

(5.18)
We have found an important result bere 1hat directly ties
RST inlo rclative and absolute momentum. Over x days, the
RSI is tbe proportion of absolute momentum beca use of up
days. Hence, the x-day RSI is equivalent to the x-day momeotum.
In Table 5.7, we did tbe RSI calculations using the sum of
up-day and down-day momentum. Next, we will redo the 4day RSI calculations using tbe relative and absolute momentum to show that we gel the same value using eitber method
(see Table 5.8). Now it is easier to understand what the RSI
is trying to capture by using the ideas ofrelative and absolute
momentum.
Rather thao use Wilder's original method, we lik.e to use

Stochastic RSI Oscillator

123

Table 5.7 Calculating 4-Day RSI Using Sum of Up-Day and

Down-Day Momentum

Date

S&P-500

Up-Day
Momentum
(SJ

11/29/91
12/02/92
12/03/91
12/04/91
12/05/91

375.22
381 .40
380.96
380.07
377.39

6.18
0.00
0.00
0.00

0.00
0.44
0.89
2.68

6.1 8

4.01

Total

Down-Day
Momentum
(S.)

RSI

60.65

RSI - 100(6.18)/(6.18+4.01) - 60.64769382.

Table 5.8 Calculating 4-Day RSI Using Relative and

Absolute Momen1um
Date

S&P-500

11/29/91
12/02/92
12/03/91
12/04/91
12/05/91

375.22
381 .40
360.96
380.07
377.39

To tal

Relative
Momentum

Absolute
Momentum

- 2.68

6.1 8
0.44
0.69
2.68

2.17

10.19

6.18
- 0.44
-0.89

RSI

60.65

RSI = 50' {2.17+ 10.19)/(10.1 9) - 60.64769382.

al! available data to calculatc RSI and lhen smooth t with a

simple moving average. Wilder's method effectively uses a
27-day exponential moviog average to smoolh daily s. and
Sd values before computing RSI, but tbis smoolhing ofteo
masks lhe underlying price extremes. It also makes the RSI
take on lhe appearaoce of the price curve.

124

New Mo~ntum Oscillators

STOCHASTIC RSI OSC ILLATOR

A.s an RSI uscr, you may ha ve been frustrated whcn the RSI
did not reach .an extreme value (above 80 or below 20) for
months .at a time. Perhaps you wanted ao eotry point into
an oogomg trend, and wcre lookiog for a price extreme but
couldn't fin~ one using RSI. The solution to your probl;m is
the Stochast1c RS I (StochRSI). wbicb we will now discuss as
thc second clc~ent in our package ofmomentum oscillators.
. The st~chasttc RSI osc1 llator com bines the two popular
ideas. behm? thc RSI and thc stochastic osciUator. The stoc~as~1c osc1llator !Deasures thc location of closing prices
w1th1n thc rece~t bigh to low range. SimiJarly, StochRSI measures the locatton of RSI withio its rccent range, showing
shon-tenn momentum extremes. lt can be used as an antitrend or a trend-foUowing tool.
T he sen~itivity of the StochRSI overcomes thc disadvantages of us1ng a fixed num ber of days in its calculation and
1!1e ten~ency of the built-in RSI smootbing to mask shonIJv~ pncc extremes while showing swing failures in RSL Tbe
~b~hty to S_POt shon-tenn extremes in RSI (and momentum)
1~ 1ts P1:111C!PaJ advantage. The stochastic RSI is a more cons1stent rnd1cator of overbought and oversold conditions simply becausc wc are measuring its position with in the most
rcoent range. T he stochRS I is defined as fo!Jows:
stochRSI - (RSI - RSIJ/( RSIH - RSIJ.

(5.19)

Here RSIH and RSIL are the highest and lowest valucs ofRSI
over a gvcn look-back peri od. Whenever RSJ makcs a new
lo~, ~e stochRSI wi!J be at O. Tbis is an example of a failure
swmg .m a down move. During an upmove, thc RS I will make
new highs (ovcr thc caJculation period) and the stochRSI wiU
be near 1.0. Thus, you could use stochRSI both as an overboughtfoversold oscilJ~tor and to follow trends in RSI. Divergences are also eVIdent on the stochRSI plots, which

Stochas1ic RSI Oscillator

125

means you can combine ali the different elements of RSI

analysis into the single stochRSI ind.icator.
For symmetry, we use the same numbcr of days in the lookback pcriod as those in the RSJ calculatioos, but you can
experiment with dilfercnt calculation periods if you wish. For
example, if we caJculate a 14-period RSI , then we will find
RSI 11 and RSILover 14 periods. The vaJues of stocbRSI vary
betwcen + l and O. When RSI is at its highest value, stochRSI
has a value of + 1. Convcrsely, wben RSI is at its lowest value,
stochRSI is at O. You can multiply stochRSI by 100 if you
wish. Again, remember that you have sorne smoothing built
in to the RSI numbers that varies with the software package.

Ca se in Po int: StochRSI, RSI, and the T-Bo nd Market

Figure 5.19 shows tb e Septcmbcr, 1993 T-Bond contract, a
14-day RSI, anda 14-day stochRS I. Thc SuperCharts graph ics
package plots lines at RSl values of80 and 20 for overbought
and oversold conditions. During the ni ne and a half months
of trading shown, the RSI oever reacbed an oversold or overbought condition. lt did come clase to becoming overbought a fcw times. Note that the RSI pattem is similar to
thc pattem of closing prices, with one notable divergence at
the highs in M arch, 1993: Price made a new high, while RSI
was vinua!Jy flat. Other than that, RSI moved in sync with
prices.
lo the same figure, the 14-day stochRSI sbowed sbon-tcrm
and intermediate-term markct extremes far better than the
RSI. A stochRSI value of O provided excellent entry points
in to lhc uptrend. Thc di vergence betwecn stochRSI and prices
is notewortby at the peaks in M arch. During stroog, brief
uptrcnds, stochRSJ remained pinned at 1.0, showing the
strength of the move. Note the seU signa! in April when
stocbRSI first feU away from its bighs. Note also thc buy
signals in January, February, aod April, as thc T-Bond market

126

New Momentum

Oscillators

Stochastlc RSI Oscilla'""t"or_ _ _ _ _ _ _ __ _ _ _ _ _1_27

00

ro

t---------------------1 ~00
o

..

..

FIGURE 5.19 A comparison of the 14-day RSI to the 14-day stochRSI

for the T-Bond 09/ 93 con1rac1.

corrected or consolidatcd. These signals are unrnistakably evident in the stochRSI plot, but not as obviously cvident in
the RSI plot. Thus, thc stochRSI analysis is more useful in
this example.
In Figure 5.20 we show a close-up oflhe T-Bond September
1993 contract from Fcbruary throush June, J 993. The
stochRSI showed each significan! overbousht and ovcrsold
condition over this period, Oagsing a sbon-terrn tradeable
rnove. The divergence in March that we spotted in Figure
5.19 can be seen on an expandcd scale bere. Buying around
lhe 0.2 leve! and selling around lhe 0.8 leve! would have been
profitablc. lt's importan! to reenter the market if stochRSl
again rises above 0.80 or falls below 0.20. Say you sold the
bonds when the stochRSI feU below 0.80 in early June; the
uptrend resumed and the stochRSI again moved above 0.80

FIGURE 5.20

A closer look at the stochRSI for the T-Bond contract.

towards 1.0. It would have been best to reverse the short

position to a long trade in kecping with tbe stochRSI rnoving
abovc 0.80.

Case in Point: StochRSI and S&P-500 lndex

Figure 5.21 shows tbe action of the S&P-500 indcx for approximately scven months from October, 1992 througb April,
1993. Thc market was trending up to mid-December, and
tben entered a broad trading range. The 14-period RSI did
not exceed 80 nor faJJ below 20 tbroushout this period. Tbus,
it could not be used asan overbousht/oversold indicator.
The stocbRSI showed eacb major price extreme, thoush it
showed choppy action during sorne periods of the trading
range. A sell signa! occurred when the stochRSI was above

128

New Momentum Oscillators

129

Stochs1ic RSI Oscillator

NEW Sf.ANOAROti1y OlQ1N1 C-310 470 -+ 1 1 1 ()io'367 070 11=370 110 t.El.733

tqi,111111 37110ll

ti

111111 1111'1 1111 11

1 111 1

1,

"1 1

11

,,.,
111

'I

11

1
11
1

EllDJ

3lll OlJ

3400:0
:m.m:J
l:llllDl

11

111 1

310JXIJ

mOJJ

-~~

..
.. ........

-o~llXXI oxeo

........

93

,..

. /\
~

....

FIGURE S.21 A comparison o the performance of the RSI and the

stochRSI over a time interval in which the S&P-500 index moved up
and entered a broad trading range.

0.80 and then fcll below that level. A buy sigoal occurred
when this indicator was bclow 0.20 and then exceeded that
level. Note the reversa! co nditi.on: lf the stochRSI reverses
after crossing abo ve 0.20 or rises after falling below 0.80, theo
you should reverse the bu y or seU signal. The reversa! condjtion occurs because stochRSI can stay above 0.80 during
uptrcnds and below 0.20 during downtrends for the duration
of the trend.
Nexl, we' U look at the S&P-500 index in Figure 5.22 coveriog the pcriod November, 1991 lhrough February, 1992.
Note how the 14-day stochRSI in this figure weakened aU
lhrough December, 1991 as the index moved up to the 330
area and then entered a light trading range. The stochRSI
settled at O showing a downtrend just before the decline to

""

v.~
91

F..

FIGURE S.22 An illustration of the stochRSI indicator during a trend

ing period in the S&P-500 index.

the 310 area, and it moved offits bottom two days before the
index itself bottomed. The index and stochRSI rose strongly
in January until tbe stochRSl was at 1.00, indicating a strong
uptrend. Tbe stochRSI reroained above 0.80 for the entire
rise into February, again fafog off withio a day of the high
for the move. As the index entered a trading range, the
stochRSI fell off again dropping to O.
The important idea to grasp here is tbat tbe stochRSI feU
when the iodex entcred a trading range. The stochRSI is qwck
to react at both bottoms and tops, so you can trade it only
when it is al ao extreme, showing a trend. Acc.ordingly, you
can buy when it rises above 0.80 and sell when it faUs below
0.20. You can then e.lose your posilion (or reverse) when the
values fall below 0.80 or rise above 0.20. Or, you can trade
ooly when the stochRSI falls below 0.50 or rises above 0.50.

130

New Momenlum Oscillators

You also can use sorne smoothing to reduce whipsaws or

require stochRSI to closc abovc or below 0.50 for three days
before putting on pcsitions. This migbt delay entry, bul it
may also reduce whipsaws. ln this way, you can adapt this
indicator to your trading style.

Trading Strat egies with Stoc hRSI

The versatiliry ofthe stochRSI can be funher iUustrated with
weekly data for Amgen (AMGN). Figure 5.23 shows a 7-week
stochRSI anda 7-week RSI below recent prices. The stochRSI
again picked off significan! ltigbs and lows that the RSI did
not. Note how the fall in stochRSJ values was timely ndication of the high in Amgen; lhe stochRSI fell below 0.20

131

Stochastic RSI Oscilltor

well before thc huge scll off. lt also picked most of the small
rally that followed. This examplc also prove~ the tim_elne~s
and scnsitivity of stochRSl. You couJd expenment with d1fferent time periods of calculations and action Jevels, and
hence use stochRSI for long- as well as shon-term analysis.
But what about using stochRSI to analyze a very short time
frame? Figure 5.24 illustrates the use of the stochRSI with
hourly data using the September, 1993 T-Bond contract. Here
we smoothed the stochRSI heavily to remove the noise nherent in using such a shon time frame for analysis. We computed a 20-period RSI and a 20-period st?Ch~stic oscilla~or.
Then, we smoothed the 20-period stochasuc with a 10-penod
simple moving average and plotted the slow stochRSI over
the hourly price bars shown in this figure.
The momentum rose as prices rose and fell when prices
entered a trading range. You can see thal the stochRSI can
fall even when the market is moving sideways. A sideways

........

...'
...
'

o ,

'

......'"
,...
o.

o.o

U.P

........

111.
111.0

..' .'

'"

IUO
UI

111
11
110.

11

A GUR.E 5.23 The stochRSI os shown overa longer lime period using
weekly data for Amgen

.....,...
.."..'
o

o.

11.

...

...' ..
l lf .O

11e,. ~

U'-~

111.c

111

..

u: 11

111.'t

110

A CURE 5.24 Hourly data for the T Bond September, 1993 futures
contract 1s used 10 illuslrate the achon of the stochRSI.

132

New Momentum Osdllators

move causes a loss ofupward momentum that produces tower

~tochRSI values. Dips in the stochRSJ correlate welJ with Jows
in the hourl y chart. Note Lhc stecp decline in stochRSI and
how it flattened out at Jow valucs (below 0.2) during the market decli ne.
You could have covered your long when the stochRSJ fell
below 0.50 at the 116-0 level, and you hada secood chance
to cover at the 11 6-0 leve) when a rebound to that leve! kept
the stochRSI below 0.2. Or, you could bave used Lbe bounce
to. go sbort near the 116-0 level. When using the stochRSI
wub hourly data, you should use other indicators to coofinn
your analysis. For examplc, you could use a daily bar chan
with a short-term moving average to set the trend.
To show RSI's performance in analyzing hourly data, we
plotted ~ 20-:period RSI for thc hourly US 09/ 93 T-Bond
contract 10 Figure 5.25. We superimposed a 10-:period simple

133

Vriable Length Dynamic Momentum

moving average on thc RSI numbers. The 20-:period RSI

peaked bcfore the acrual high, producing a classic divergence
bctwecn momentum and prices as they rose to a new high.
The RSI declined slowly as prices feU ovcr the next thrcc
days. Tbe stochRSI also peaked bcfore prices, and showed
the classic divergencc as prices made new highs. Howevcr, it
declined faster than the RSI itself.
You can directly compare the 20-:period RSI (solid line) to
the smoothed 20-:period stochRSI (doued li ne) in Figure 5.26.
Note how the stochRSI moves faster thao the RSI. This
makcs i1 va luable for signalling trend chan,ges. For example,
the stochRSI fcll below its halfway point of 0.50 while the
RSI was still near 65. The stocbRSI also pea.ked bcfore the
RST did. lt is a more sensitive measure of momentum changes
than the RSI.

.'

-......
,..
~

..
.."
e

IH.a
11111.0

..

'
111.0
1

.o

11>.
11-..
lll.9
u1.o

llf.

11

u .o

.....
1111.11

12.t

tu.o

111.1
111

'5.

11

::~ : ~~IUlbi.M.&lilll.lllbtLlllJ.lliltllMdNt 1wln.IQltSll!JibJI

n

..

..

111.0

110.1
110.0

FIGURE 5.25 The 20-peod RSI for Lhe hourly T-Bond September
1993 contraC1.
'

01 ...u ..)

....
......
....
....
....
....
....
........
....
... ,.,.
..,.

.1''

LOO

::

~ _#."l.
~

..'

.
'

.,..'.].

....
....

:. !'

'.

...

.o

...

"

"

,.

'

'..;'

..

"

"

......
..,.
......,
....
....
...."
....

"

FIGURE 5.26 A comparison of lhe 20-period RSI and the stochRSI

(do tted hne) for the hourly Sep tember, 1993 TBond contraCL

134

New Momentum OKilltors

Vui.ible

~nglh

Dynamic Momenlum

135

VARIABLE LENGTH DYNAMIC MOMENTUM INDEX

!he ~d indicator in our package of momcntum oscillators
IS a vanable leng1h RSI, which we'll show you how 10 builcl.
Y?u'IJ get the mos1 out of this section if you ha ve software
Wllh a ."ca!Jablc" RSI function, that is, a built-in function for
RSI wtlh .the number of days as itS iopul. The dynamic mo~entum rndex (DMI) ~ spccifical~y designed to use a changt~g number of day~ rn_ LtS <;alculanons; in fact, its length can
g1vc Y?U valuable ms1gh1 mto market action. Thc variable
l~ngth idea o~crcomes the effects of smoothing to somc degree
srnce smoothing o~en obscures short-lived markct extremes.
In DMI <;al~ulaoons, you use a longer period as volatiliry
dcc~ses, ~YJng a longer range view when the marke1s are
trad1~~ qmelly. Thcn, you use fewer pcriods of data when
volat1hty mcreases, shortening the horizon for finding overbought or oversold regions in active markets. The variable
length distinguis.hes DMI from CMO and slOchRSI. Freed
~!TI th~ constramt o f ~ fixed number of days, DMI gives you
ms1ght rnto the dynam1cs of ma rket bchavior, as prices speed
up ar slow down . In that sense, extremes in DMJ are more
l~kely to provide successfuJ entries for aotitrend trades than
e1ther CMO or stochRSI.

lndexing DMI to Volatility

One approach for indexing DM1 10 marke1 volatility is to first
calcula te the 5-day standard deviation of closing prices. Theo,
wc ta~e a IO-day rnoving average of the standard dcviations.
W~ p1ck the numbcr of days in thc cquivalent static RSI ; in
this case, 14 days. Next, we define the following equations 10
calcula1e the numbcr of days in the variable tength DMI:

Std,. ... Average,0(Sld(C,5)),

V, - Std(C,5)/ Std..,
T 0 - INT{l4/V,).

(5.20)

The Std,. is a 10-day simple moving average of the 5-day

standard deviation of the close. The volatility index V, is the
ratio oftoday's value oflhe 5-day standard deviaon divided
by its average vaiuc over lhe past 10 days. V, increases with
rising volati lity. The T D divides 14 by tbc volatility indcx.
The notatioo INT cnsures that we use integer values for the
number of days in DMl calculations.
lf the index is grcater than 1, then T 0 is less than 14. Thus,
rising volati lity reduces lhe length of tbe DMJ. 1f the iodex
is less than 1, then T 0 increases. We a.Isa define lower and
upper bounds for the numbcr of days in thc calculations.
(T o)MAx - 30
(To)MIN ... 5

(5.21)

We arbitrarily restricted the maximum and mnimum numbcr of days used for DMl calculations to 30 and 5. These
limits fit our trading hori.zon, but you can use other Iimits as
you wish.
Be aware that the conversion is nonlinear, arising from
lhe deftnition of the volatility index itself. The values of tbe
percentage change in V, will give you a sense of the nonlincarity. The conversion is more sensitive wheo the index is
less than 1 and less sensitive to cb.anges when lhe index is
greater than 2.
As shown, we buih the variable length DMI around the
static RSI length of 14 days. Heoce, when the V, is approximately 1, the DMI and RSI will have s.i milar values. As 1be
index drops bclow 1, DMl and RSI values will diverge
quickly. As the index increases above 1, DM I and RSI values
will diverge slowly. Hence, the exact narure o f the DMJ-RSI

136

New M omentum OS<lillators

curves will depend on tbe volatility in tbe data over the given
test period. Tbe divergence will increase as volatility increases.

137

Vriable length Oynamic M omcntum

3050 .-~~~~~~~~~~~::;:-~~~~-,

3000

2950
2900

Case in Po int: DMI and Dow Jones Industrial Average

Figure 5.27 shows a dircct comparison of a 14-period RSI
and DMI for th~ Dow Jones Industrial Average (DJIA). Tbe
DMI leads RSI mto overbougbt and oversold regions.
We'll use DJIA data from tbe trading range in early 1991
to show how tbe length of DMI vares. Figure 5.28 shows the
scale~ l~ngth ofDMI plotted below the DJIA daily close. We
~ultJ plted the days in thc DMI by 10 and added to 2550 to
g1ve the scaled valucs. You can now directly compare the
length of DMI wth market action.
Evcry period of markct volatility reduced the lcngtb of
~MI. Wben the markct traded quietly, the leogth of DMI
mcreased. Note thc shon length used for DMI calculations
100
90
OVER60UOtlT REGION

80

?O t---''fli~:--~~--17;--~~~-tt~.-::;-~~-::"iti-j
80

so
40

RSI

2850
2800
2750

2700
2650
2600 '-~~-=~...:.~~~__:;.;.;.;~~..-:::....~~~~~-'

910201

910304

910402

910430

910&29

910626

9107U

FIGURE 5.28 The Oow Jones Industrial Average and the eective
length of the OM l.

in early March, la te March, mid-April, lat~ May. and midJune. lo eacb instance, the market was makiog qwck moves.
This example makes it clear that you can use the DMI to
identify overbougbt or oversold conditions. DMI usually
leads RS! into the extreme regions by one or more days (see
Figure 5.27), and lbjs lead time can prove valuable 10 many
you. Because DMI lengtb is more closely tied to. markct dynamics, 1be extremes tba1 DMI shows are more likely to lead
to profitable antitrcnd trades tban the extremes shown by
stochRSI. In o tber words, DMI is more sclective about sbowing extremes. You can also use DMI to develop flexible parameter trading models.

JO r-~~~~;f-~~~__:~~~~~l'--~---l

20

OMI

10

o
910201

910304

OVCASOlO REGIOH

910402

910430

810529

910828

Ca se in Point: DMI and the S&P-500 lndex

910725

FIGURE 5.27 A comparison o f unsmoothed RSI and the OMI for the
Dow Jones Industrial Average data.

Figure S.29 compares a 14-day RSI to a 14-day DMI for the

S&P-500 indcx. You can see that the RSJ was more or less
llat over thc test period, moving in a narrow range. Tbe D MI,

138

New M<>m<'nlum O..Oilltors

139

Vuible Lenglh 0yn..nic Momenlum

8000
80 000 .....-~-:-:--~~~~---::--~~~""1
8000

55 000
7000
60.000
8000
~

i\~

445 000

:i ..oooo
~

'tvl

35.000

30000
S&PS~ 1na...

425 000
Oyn1mlc momencvrn lnde:c (OMI>

5000

CIOll

20.000

tS.000

/_

da~ l./'

00
L~...::..:.:.:....~~-:--:~~::::-::;-:;-:;:;;-;-:;;~
"""""...,""" .... -....,;t""""""2~...,~~
""""""~~~~~~~r~~~~~~~~~sss~~

~a~~sSS~~is~S~aaa~a~aaoa

on the other hand, sbowed overbougbl and oversold conditions on four occasions over this period.
In Figure 5.30, we overlaid the DMJ on the S&P-500 index
daily close. You can see that tbe extremes in DM 1 correspond
to extremes in the index as well. One trading stra1egy is to
buy when the DMJ goes below 20 and then rises up. You
would sell wben the DMI is above 80 and then crosses below
tbis level. This approach would have given tradeable signals
three times in February-March, 1993. Each signal preceded
substantial moves by about two days. T his provides you with
is a significan! edge in trading.
Fina lly, Figure 5.31 shows the e ffective length, that is, nurnber of days, used for these D MI calculations. Low values
correspond to periods of higb volatility. Convcrsely, high valucs fo r the effective lcngth imply periods of low volatility.
You can see from Figure 5.30 that quick moves in thc market
reduced the effective length of DMI calculations.

FIGURE 5.30 The OMI fOf the S&PSOO index.

35.00
3000

.,,

2500
20.00
1500
10.00
5.00

2000
10.00

~~s-es-~J~ - ~~~8----~~~
FIGURE 5.29 A companson of 1he 14-day RSI and the 14-day OMI
for the S&P-500 mdex data 1n Figure 5.30.

0.00
30.00

.,

..

!' \

---

1 '

'

140

Ncw Momentum Osdllators

Tra ding Strategies with DMI

There. are a myriad of possible tradfog strategies that you can

u.se ~th DMI. In this section, we give briefsummaries ofthe
snuauons that lend themselves to trading with DMI.
Let's say trendiness indicators such as ri and absCMO do
not show any trends, mplying that an aotitrend strategy has
the g~~ater chance of success. You would use ao overbought
condit1on to go shon and an oversold condition to go long.
For cxamplc, when the DMI signa Is an ovcrbough t condition
today, you could go short tomorrow a bit below thc Jow of
today. The reverse is true for a long position. You could use
a tighl stop above the lghest high of thc last five days for
short tradcs. For long trades you could place a stop below the
lowest low of the last 5 days.
In coou:ast, if the market is trending, as measured by the
A!?X or linear ~egressi~n, Y?U could use DMI as the entry
pomt for trades tn the ~rec~on of the trend. For example, jf
the ma~ket were trendng lgher, you would wait for an oversold s1gnal from the DMI. You would then assume that this
":'s a m!nor correction within an uptrend and, thus, a Jow
nsk-buyng opporturuty. The opposite is 1ruc in a downtrend.
You would scll after an overbought co ndition wilhin a downtrcnd. For example, you could sell below the lowest Jow of
the last threc days with a stop at the ltighest h\gh of the Jast
three days..A cautionary note: You should adequately test
both strateg1es and be comfortable with the odds of success.
In a markct trendng lghcr or lower, many traders use
DMI as an cxit too! by exiting on an overbought condition.
Note that the market does not have to reverse directioo automatica!Jy simply because it is overbought or oversold. An
extreme price condtion often shows that thc market has lgh
momeotun;i in that directioo. Thus, an overbought market
coul.d go lgher and an oversold markct could go Jower.
. Fmally, rcmember that the maio advantage of using DMI
is that you do oot have to specify lbe number of days in the
calculations. DMI will adjust the number of days based on

Summary

141

market volatility. The construction of your volatiliry index

detennines the sensitivity of DMI.
SUMMARY

Our new momentum oscillators make upa group ofpowe~

tools to analyze momentum. Tite Chandc momentum oscillator (CMO) is a pure momcntum oscillator .t~at s~ows net
momcotum change on a bounded scale. lts cnt1cal <l:iffe~en~e
from the RSl is that it does not use built-in smoothing 10 1ts
calculations. The CMO can be uscd to find market extremes,
oflen showing extremes that the RSI docs not. 1t can also be
used to trade momentum changes using the zero-crossing of
the indicator.
You can combine CMO values from different time intervals
to find a composite CMO: The combin":t~on ~ be crc:a~ed
by simple averagiog or we1ghted by ~?lat~ty. This vo~~ty
based composite CMO offers a scns1uve v1ew of muln-m1erval momcntum changes. You can also filt.er !he CMO. to reduce ooise in 1he data and locate truly s1gnificant pnce extremes. Thus, the CMO is a powerful and Oexible momentum
oscillator.
The stochastic RSJ indicates whcrc the current RSI values
les within its recent range. lt can. be used both asan overbought or oversold indicator, as wcll asan in~icatoroffailure
swings in the RSJ. lt can, therefore, combme the two approaches to analysis of momc~tum using RSI. ~ particular,
it is an excellent short-term osc1llator, often shoWJ.Dg extremes
that RSI does not.
Lastly, the Dynamic Momentum lndcx Iets us calcul~te RSI
without spccifying the oumber of days ~~ the ~culat1ons. 1t
adjusts its Jcngth based on market votaulity. Tls helps o~er
come a key limitation of RSI. Thc DMI often leads RSI rnto
overbought or oversold regions by several days. Such carly
wamings could prove extremcly useful to you.

6
Market Thrust and
Thrust Oscillator

This chapter introduces powerful new indicators that combine ilems of data unique to the stock market and unavailable
for the futures markets. Tbese four data itcms are the number
of advancing (AJ) and declining issues (DI) and advancing
(AV) and declining vol ume (DV). Tbe Al and DI count issues
without regard to the extent of price change or the market
capitalization.
Market analysts usually look at thesc data in a variety of
ways. On days when the market is up strongl y, Al > Dl by
more !han 1000, and AV> DV by a factor of 3:1 or more.
Thus, the advancing volume could easily be four times (or
more) than the declining volume. Similarly, there is general
agrccment that, on days when the market is very weak, Al
< DI by severa! hundred issues, and DV > AV by a ratio of
3:1 or more.
On days when thc markcts move decisively in either direction, wc see a pattern in which Al > DI, AV > DV or Al
< DI, AV < DV. There also can be other combinations of
these variables, which produce ambiguous cases bccause it's
common to analyze tbe advanciog and declining issues as one
143

144

M.trket Thrusl ~d Thrusl Oscillator

block, and thc up and down volume as anotber block. These

blocks, thcn., do n~t aJways move togelher because of the
randomness m tradmg.
. Tbe popular Arms index devised by Richard Arms comb1~es the t';'o.blocks of data in to a single indicator. He devised
th1s trader s 1_n dex, (com monJy caJJcd TRIN) to show when
~bnormally h1_gh vol u me was goi.ng in to advancing or declining s tocks. lt 1s defined as
Arms lndex =

(Al/ DI) = (DV/Dn

(A V/ DV)
(AV/ Al)

(6. 1)

whcre Al - # Advancing issues, AV = Advancing Volume

DI = # Dcclining issucs and DV - Dcclining volume.
'
.T_h e ~s lndex 1s thc ratio ofthe ave rage volume in dech ning 1ssucs ~o the average volumc in advancing issues; it
mcasures rclattve volume flows. A reading of LOO is neutral
~ vaJue gr~atcr than 1.0 indicates more volumc in declining
1ssues, whil~ a ".alue bclow 1.0 shows more advancing vol~me. The d1r~hon and speed of the changing indcx values
~s o fien more 1mportant than the absolute valuc of the index
nsclf.
On a day wh~n the m~ket is weak, !he Arms index is
grcater than 1.0 sm?e therc 1s more volurne in declining issues.
When the market is up strongly, thc index is less than J.O.
Note that_thc scales for up days and down days are not the
samc. This occu.rs because the Arms index is bounded betwccn Oand 1 for up days, but unbounded bcyond 1 on weak
days. For exa~ple, rea~ings greater than 4.0 have been record~d many times. This makes it difficult to use TRIN with
moving av~rages. We'll rearrange tbe terms in the Arm's l ndex
o nc more ume to clarify lhis idea:
Arms Index -

(AJ/ DI) = (Al DV)

(AV/DV)
(DI A V)

Marl~I

Thrusl and Thrusl Oscillator

145

Because the iudex muJtiplies Al by DV, and D I by AV, it can

produce unusual effects in " rnixed" markeis, that is, when AJ
> D I but AV < DV, or Al < DI but AV > DV. lt is intuitively
contradictory to have the index driven by the product of
Al*DV (and Dl*AV) rather than Al*AV (and Dl*DV). When
we havc " one-sidcd" markets, we expect greater volume, or
more stocks, or both moving in that direction (up or d own).
The primary application of TRIN is to detect overbought
or oversold conditions in the markct. Since the daily data
tend to be noisy and apparcntly trendless, ihe daily TRIN is
usually smoolhed with a 10-day simple moving average
(SMA) to isolate the underlying trend. When the 10-day SMA
of TRIN rises above 1.20, the markct is thought to be oversold. The market is thought to be overbought if the 10-day
SMA of the TRIN faJJs below 0.80. The market is expected
to trend higher af\er an oversold TRIN condition, and trcnd
lower af\er an ovcrbought TRIN reading. The 10-day SMA
is considered an intermediate-terrn indicator.
There are three difficuJtics with using TRIN:

l . Thc ratios used in TRIN often obscure market action.

This is panicularly true when we have "mixed" market
a.c tion.
2. The very process of smoothing TRIN with moving averages distorts the picturc of relative volumc flows.
3. The TRlN has a bounded scale for up-sidc activity, but
an unbounded scale for down-side aetivity.
We developed our market thrust indicators to overcome these
wea.knesses. You'U see thesc features more clearly in the following discussion on markct thrust and tlie thrust oscillator.

MARKET THRUST ANO THRUST OSCILLATOR

(6.2)
A more consistent approach to analyzing Al, DI, AV, and DV
data is to look at the product of thc number of shares ad-

146

Market Thrut and Thrust Osdllator

vancing or declining and the volurne going into lhose shares

lhat is, using the pr~ucts AlAv and 0 1ov. The lhrust, o;
power ofthe move, 1s measured by lhe number ofstocks and
the volume going into lhosc stocks.
For example, if 5 stocks advanced on 100 shares lhe lhrust
is 500; next, if 7 stocks advanced on 90 sharcs, the thrust is
630. Thus, we would say there was greatcr market thrust the
secood day. This d efinition is intuiti vely more satisfying beca use "one-sided" action meaos greater volume, or more
stocks, or both, moving in tbat direction. lf we had used
average volume, on the first day it was 100/ 5 = 20 versus
90/7 - 13. Using ratios, as in TRIN, we would say there was
more activity on the lirst day. This smaU example illustrates
the djlfereoce between TRIN and market thrust (MT). We
define the market thrust MT as:
MT - (AlAV - DI DV)/1,000,000.

MT-linc,.,..,..,,,

(6.4)

We can d efine a thrust oscillator (TO) to compare relativc

volume flows:
TO -

(Al *AV - D t DV)

100
(A l. AV + 01ov)

tween + 1 and - 1 or + 100 and - 100. There is importan!

differencc between MT and TO: TO values are always betwecn + 1 and - 1 (or + 100 and - 100) but MT values are
unbounded.
The thrust oscillator definition has many advantages. The
biggest advantage is that it has the same scale for up and
down days. The range of values is bounded between the ~e
numbers on either side. Tb is is in contrast to TRJN, which
is bounded for up days but unbounded for d own days.
The second advantage of TO, is that it depicts upthrust
(AIAV) and d owndraft (Dl DV), so that a strong up day or
strong down day is identified consistently. TRIN, as shown
in Table 6.1 sometimes obscures a strong one-sided up or
d own action'. A third advantage of TO over TRIN is that it
shows the net balance betwcen bullish and bearish activity
on a gi ven day. Lastly, it prescnts consisten! information; TO

(6.3)

We divide the thrust by 1,000,000 to givc reasooably small

numbers for con venieoce and s implicity. You do 001 have to
divide by 1,000,000 at ali; you co uld cboose any other scaling
constant. The dail y number may be used to spot Jarge rnoves
in one direction. The thrust is c umu lated to identify underlying trends, and thcsc can be smoothed or summcd for trading. purposes.
Thus, you can llave a cumulative MT-line'
.
wh1ch 1s a vo lumetric advance-declinc lin e:
MT-line,..., - MT

147

Market Thrust and Thrust Oscillator

(6.5)

Tbe multiplication by 100 is optional, so tbat TO varies be-

TABLE 6.1

Da y
1
2
3
4

6
7
8
9
10
11
12
13
14
15
16
17

Slmulatcd market data comparing TRIN, MT and TO

Al

AV

DI

DV

1000
100
1000
100
800
600
800

1,000,000
100,000
1,000,000
200,000
800,000
600,000
700,000
900.000
700,000
600,000
500,000
500,000
600,000
700,000
400.000
500,000
400,000

100
1000
100
1000
400
600
400
400

100,000
1,000,000
200,000
1,000.000
400,000
600,000
500,000
300,000
500,000
600,000
700,000
700,000
600,000
500,000
800,000
700,000
800,000

800
600
700

600
700
500
500

soo

400
400

600
500

600
500
700
700
700
800
800

TRl'J

MT

TO

990 0.98
1.0
- 990 - 0.98
1.0
980
0.96
2.0
0.50 - 980 -0.%
480 0.60
1.0
1.0
0.47
1.40
3&0
0.71
0.67
600
0.71
120 0.17
120 0.17
1.40
1.40 - 120 -0.1 7
o
1.96
o
0.71 -120 -0.17
o
0.52
o
1.43 - 360 -0.47
0.70 -360 - 0.47
t.00 - 480 -0.60

o.o

148

Market Thrusl and Thrusl Oscillator

provides normalized volume tlows and can be used as an

overbought/ oversold indicator.

Varialions of MT and TO
Consider the special case wben thcre are uni t volume tlows
in!o advancing and declining issues, where AV - DV = J. In
this case, the AV (or DV} 1enn mereJy acts as a constant
muJtiplier, and lhe MT-line simply collapses to 1he usual cumulative Advance-Decline (A-0 ) line, since:
MT - (A l - D l)(AV- DV - 1),
MT-line_, - (AJ - DI) + MT-line,.......r

(6.6)

Therefore, we can expec1 a broad similari1y betwceo the cumulative MT-line and the A-D line. Tbc two ofteo diverge
at tuming points when volumc rotales predominantJy into
advancing or declining issues. T herefore, the MT-line is more
useful an the A-D line at 1uming points.
Ncxt, considcr the special case wben AV = DV which converts the TO into a variant of tbe advance-declin~ ratio since:
TO - (A l - 01)/ (AI + 01) (if A V- DV}.

(6.7)

lf AV = DV, lhen TO beco mes thc net advaneing or declining

stocks as a fract.ion o f total stocks traded. TO may aJso be
considered a variant of a volume oscillator beca use if Al=
DI, then:
TO - (A V- DV)/ (A V+ OV} (if Al - DI).

(6.8)

This makes TO an effective net volume oscillator: it works

as a combination of an issues oscillator and a vol~e oscillator at the same time. A plot ofTO looks similar to a volume

149

Mark<t Thrust and Thrust Osclllator

oscillator plot oran issues oscillator plot, wbich is an unusuaJ

but useful featurc.
Our market thrust idea captures most of the variations
misscd by TRIN. The product of AI*AV and DI*DV shows
the amount of up or down thrust. We can sort out the dilferent
combinations of AJ, D I and AV, OV more accurately by using
markel thrust than with TRJN. We'll compare and cont.rast
TO and TRJN in the nexl section.

Case in Poinl: Comparin g MT, TO, and TRIN

We developed fictitious data to illustrate each of lhe different
combinations of AJ, DI, AV, and DV 1hat couJd occur in actual
trading, Thc fictitious data mercly facilitate our discussion of
diffcrent market conditions using similar numbers for each
occurrencc. Later, we'll use real data to make our case.
In Table 6. 1, day 1 is stroog up and day 2 is strong down ;
however, TRIN shows they are neutral. In contrast, MT
shows the exceptional strength and weakness of eacb day, and
TO is at positive and ncgative extremes eacb day. Remember
tbat the unbounded MT could exceed the +990 or - 990
va lues recordcd here, but TO will rcmain between + 1 and
- 1.
Day 3 is also a strong up day, but has somewhat greater
down vol u me than day J. TRIN flags itas a bearish day since
thc aver.ige volumc in declining stocks is greater t~an t?e
average volume in advanc ing stocks. The volume dispanty
occurred evcn though advances led declines by 900 stocks.
Note, howe ver, that tbe lhrust value of 980 shows day was
almost as strong as day 1. Thus, MT was more cons1stent
wi th its values on days 1 through 3 lhan TRIN.
Oay 4 is a reversa! of day 3, with 900 more declines and
volume predominantly in declining shares. H ence, it was a
bearish day. Nevenhcless, TRIN calls this an overbought day
with bullisb demaod, because average volume in advancing

150

Market Thrust nd Thrust Osdllato r

shares was greater. Note how Thrust reads -980, and TO 0.96, capturing the beansh mood ofthe day. Day 5 is neutral
says TRlN; MT says it is moderately strong on the upside.
This shows that TRIN does not consistentl y show bearish or
bullish market activity.
TRlN correctly calls day 8 a bullsh day, as do MT and
TO, but calls day 9 more bullish lhan day 8. Thrust correctly
calls day 9 mildly bullish, bccause up volume cxceeded down
volume. Here an equal number of stocks advanced and declined. TRlN is bearisb on day 10, while Thrust is unchanged
from day 9, correctly calling the mildly bullsh day.
Day 11 provides a striking contrast betweeo TRIN and MT
and TO. TRJN calls it strongly bearish; MT and TO call it a
standoff. It is thc same story for day 14: TRIN says great
bullish demand, Thrust is dead neutral. Here thc divergent
interna! dynam ics of TRIN give conflicting values.
Relacive volumc nows, defined as Dl/DV and Al/ AV, simply do noc consisteotly show if lhe market action was evenly
split or onesidcd among the four variables. In contrast, because oftbeir definitions, bolh MT and TO clearly show when
the action was evenly split or one-sided among these variables, and provide a more consisten! picturc of market accion
than TRIN.
In essence, using thc product of Al*AV and Dl*DV elminates tbe confusion that rcsu lts from taking the DV/ DI and
AV/ Al racios found in TRIN. What's more, wc can easily use
moving averages to smooth MT and TO in ordcr to reduce
noise and sec the underlyi ng cbanges more accurately.

Figure 6. l shows the Dow Jones Industrial Average {DJlA)

from June 1987 through Nov ember, 1988 anda 10-day SMA
ofTRLN. This period covers a new record high and the market
collapse of Octobcr, 1987, including the sideways action that
followed. The values of smoothed TRIN rose to nearly 2.50
during tbe collapsc. lt is difficult to scc any other notable
overbought or oversold conditions using TRIN.
Figure 6.2 shows tbe samc market period and a plot of tbe
2 1 day SMA ofTO. You can see immediately tba1
drop~
tojust uoder - 0.30 io OctobeT, 1987. The prcVlous ~km
TO occurred above 0.30 in Juoe, followed by success1vely
lower peaks in August (at the top) and early_ Oc~ober. This
was clear evidence that tbe markct was los1ng tts upward
thrust as it surged to ncw highs in Augus1 and tried a wcak
rally in late Septembcr. The market again rcached overbougbt
readings in January, 1988 and March, L988, followed by small
corrections.

:ro

,_
... .

,._

.""

Now we'll look at real markct data using thc thrust oscillator.
TO is bounded betwccn + l and -1 (or + 100 and - 100),
having the same scale for up and down days. Conversely
TRLN is unbounded on down days.
'

,_ ' -

"

-.
.....
.,... l.t"
-
...._. ~,_
HM O

".

fo
,..

- '-
~

-
U H

-"".
'-' o

.... .... ...

--

U N. O

"' "'

1 ..

... ...

.. ,_,. -

K V. .~

,...,..~

......

1-

1 . ...

UM O

Case in Point; Thrust Oscillato r

and the Crash of 1987

151

Mukel Thru51 and Thrusl Oscillator

~~

'-

"..

H C

.
.

I".., .

.,_ .,- 1,1

l:

..

';'I
~u

~ !/ ,.

...:

11011

- ,_

-.. "-

... "' '"

'f- t:'t:

..

'
"""'

'

1111. . ...

.....-
...

_:

... '" ...

FIGURE 6.1 A 10-day smple movng aver~ge o the Arms index and
the Dow Jones Industrial Average in 1987-88.

152

Market Thrufl and Thrusl Oscillator

Market Thrusl and Thrust Oscillator _ _ _ _ _ _ _ _ _ __ l _SJ

....._
.......

-...
"''"
~

...

""

.....,....
....
..........
'"'.
.... .
...." .
....
,,....
....
.... .
l lft

...

..

..

--

......
...
.......
.. '
...
.....
KO

FIGURE 6.2 The 21-day simple moving average of the Lhrust oscillator for the same period as in Figure 6. 1.

FIGU RE 6.3 The OJIA and the 21-day smoothed thrust oscillalor at
the markel bouom in t 990.

These two figures show that TO bandled the buge volume

lows of the crash just as easily as the quiet period tbat preceded and followed the crasb. lt providcd coosistent indications of overbought (0.2 to 0.3) and oversold areas (-0.2
to 0.3).
Figure 6.2 also shows that long-term rnarket bottoms occur
whcn the 21-<lay SMA ofthe TO falls to, or below, the - 0.3
area. lntermediate bottoms seem to occur wben the 21-day
SMA falls to the -0.20 area, with intermediate term tops
occurring in the 0.2-0.30 area. Minor tops and bottoms often
occur with 21-day TO at 0.1 or -O. l.

markct bottom in October-November, 1990 using the 21-day

si mple moving average of TO and TRIN. Here again, the
smoothed TO made two trips bclow the - 0.30 in September
and October 1990. The smoothcd TO slipped under its own
13-day cxpo~ential average in Dccember, tipping offthe brief
correction into mid-January. Thc smoothed TO values werc
greatcr tban 0.20 during intermediate bottoms.
Figure 6.4 shows the same period witb a 21-day smoothcd
TRlN. It went higher until peaki.og above 1.2 in Octobcr,
then fell steadily into late-December. Tbe 21-day smoothcd
TRIN did not warn ofthe correction until thc cnd ofDecember. lt was more than 10 days slower thao the smootbed TO
in giving this signal.
.
.
In Figure 6.5 you can sce tbe pnce progress 10to 1991 off
tbe October, 1990 Jows. Tbe markct stayed in a trading rangc
for most ofthe year. Note how the 2 1-<lay smoothed TO ga".'c
consstent oversold signals near thc - 0.20 level. The one 1n

Case in Point: TO and DJIA in 1990- 91

The market had another October massacre in 1990 and
moved up ioto a trading range in 1991. Figure 6.3 shows thc

MarkI Thru<I 1nd ThruI Osdllator

.....
-
.....
... ..
.....
.,,, .
.....
,.,...
.... . I~ l~li
.,... fl
...........
...
._

UJI 1

.~. .

._.

..

10

dr1
1111

'

~11'

...

FIGU RE 6.4 The 2 1-day smoolhed Arms index for the same period
peaked al 1.20 al 1he marke1 bottom m 1990.

155

December was panicularly mely. Note also the strong market tbrust in January- February, 1990 tbat took the 21-day
smootbed TO to thc 0.50 area. Sucb market thrusts often
signal highcr prices 12 montbs into the future.
In Figure 6.6, we compare tbe 21-day smoothed TO (lower
curve) to the 21-day smootbed TRIN (upper curve). Notice
that TO and TRIN mirror one another. Tbe smootbed TRIN
sbowed oversold conditions near values 1. 1; TO and TRJN,
bowever, do not peak at the same values, wbicb is what we
expect in a trading range. For example, smoothed TRIN
dropped to the 0.70 area in both February and AugusL However, TO was at 0.50 in February, and only 0.20 in August.
Tbus, thc 1hrus1 was much stronger in February tban in August.
Similarly, 1he smoothed TRlN rose above 1.1 on seven
occasions, whereas 1he TO approached its leve! of - 0.20 level

....

l HH

1 . .. .

......

..........
..........
.""
...
.. l ..

MMt

FIGURE 6.5 The DJIA and the 21 -day smoothed thrus1 oscillator wilh
1he marke1 m a narrow range.

,-.-.

llU

....

il'U '

,..

l.. . ..

f1'

MT

O!fll

f l(

f 'lll

A CURE 6.6 A companson of the 21-day smoothed thrusl oscillator

and Arms mdex (upper graph) for the 1991 period shown in Figure

6.5.

157

_15_6_ _ _ _ _ _ _ _ _ __.:_M
:.:::arket Thrust nd Thrust Oscillator

Market Thrust and Thrusl Osdllator

just three times during the year. For example, the May peak
in smoothed TRIN dropped the TO to just the -0. 1 leveL
Finally, the smoothed TRIN stayed flat in early December,
but thc TO moved up sooner off its bottom.
In shon, this comparison in trending and lat markets
shows that the TO is a better osciUator than TRIN because
its signals are more consistent and timcly.

(EMA) on 1he 5-day SMA ofTO. A buy signa! occurs when

the 5-day SMA firs1 turns up after a downswing, or crosscs
above its own 5-day EMA. A sell signa! occurs when the SMA
crosses under its trailing EMA. These are shon-term buy and
sell signals becausc they las1 only a few days at a time. The
plo1 suggests that sucb a strategy would have been profitable.
TO can also be uscful in providing loog-1erm buy signals.
Assuming we could replicate the Dow Jones Industrial Average (through a suitable mutual fnd), lhe gains and losses
would be similar 10 those in Table 6.2 of DJIA signals using
TO. These calculations pro vide a potentially useful long-term
indicator. This table shows that there has not been a significant correction in the stock market through mid-1993 after
the bottom in August-Octobc r, 1990.
To crea1e the long term indicator 1ested in Table 6.2, we
first calculated TO and then plotted a 21 -day simple moving
average ofthe TO, adding a second 13-day exponential moving average of this SMA. A buy signal occurs when the 2 1day SMA ofTO goes below - 0.3 then crosses above its own
13-day EMA. We enter the market on 1he close the next day.
Though TO does a good job ofshowing long-term bottom,
long-term tops are harder to find using an SMA. Market tops
seem to occur in response to externa! factors, and !hose oc-

Trading Strategies with TO

To see an example of shon 1erm trading using TO, look at
Figure 6.7. Herc you can scc a 5-day simple moving average
of 1he TO (uppcr graph) and the S&P-500 indcx closes for
the firs1 half of 1993. To generate trading signals, we bave
also supcrimposed a 5-day exponential moving average

.,...
~~----'----

_.

__ __
__,

__.

___..__--"

Table 6.2 DJ IA Buy Signals from TO (Rounded)

Date
03-Apr-80
28-Scp-81
23 -Jun-82
24-Feb-84
05-Nov-87
08-Feb-90
27-Aug-90
18-0ct-90

FIGURE 6.7 A 5-day smoothed thrust oscollator for short-term trading

combined woth ots a 5-day exponentoal movmg average.

Average

Buy
784.00
842.00
813.00
1165.00
1985.00
2644 .00
2611.00
2452.00

3 MOS% 6 MOS% 12 M05% 24 MOS%

13.27
3.33
13.78
-5.32
- 3.78
3.37
-2.60
7.91

21.17
-2.26
28.54
6.09
1.76
3.40
10.65
22.31

28.44
9.14
46.71
9.61
8.06
7.03
15.89
25.49

6.89
48.81
39.1 1
45.75
30.83
22.73
24.63
30.02

3.74

11.46

19.05

31 .10

158

Market Thrust and Thrust Oscillator

159

SUMMARY

. . . . . i-

..... ""
.""""'"""
.....
.. ..

Summary

. . . . . i-

f\.

/\

...

V /

\''\ -1-

!"\

t;. ,......

r'\.!;\

\ [f.~

""

rv
V

..... gl$ffii ;~~~m ~i ~~

-~

10

Kt

MC

~l

Mt

, ...

la

H -

MI

. . . H<

,, ... ,_,,

dttttttrtttJ1Y
:::~l
...
Uf'

MI

. . . . .(

l .lt

fU

-1

rlf."

t'IMI.

114

tO

9(1

. . (:

IJUI

f U

ACURE 6.8 A 50-day smooihed TO and ils 50-day simple moving

average (upper graph) at the market top of 1987. The lower graph
shows the 30day RSI and 1he DJIA.

curring since 1980 did so with an overbought condition followed by a drop in thc 50-day smoothed TO below its own
50-day simple moving average. This means that downside
thrust increased after thc top and led to an intermediate-tolong tenn bottom with the 30-day TO average dropping to
tbe - 0.2 to - 0.3 area. In Figure 6.8, you can see the market
top in 1987. In particular, note how the market was overbougbt (30-day RSI o ver O. 70) and the 50-day smootbed TO
(uppergraph) fell below its 50-day movingaverage. Tbis shon
position began in late August, weU before the major selling
episodes. The graph also shows that this approach is not perfect and should be used wi tb caution.

We hope you can see from the previous discussion that market
thrust is a powerful new way to analyze stock market action.
Tbe thrust oscillator TO provides more consistent readings
tban TRrN and can be smoothed witb averages without distortion. The TO combines an ad vanee/decline oscillator with
a volume oscillator to provide a uniquely seositive market
indicator.

7
Controlling Risk:
The Key to Profitability

Risk is inhcrcnt while trading in a dynamic, ever changing

environmcnt. And though there are many factors to risk, the
result is always thc same: unexpected losses to your trading
account. This chapter will focus on some exciting new ways
10 con1rol risk.
The main purposc for risk control is the use of lcverage in
trading futurcs, as levcrage magnilies the adverse impact of
market changcs. Volatility is another reason to try to control
risk: Markcts now tcnd to make large moves very quickly,
requiring evcn grcatcr vigilance oo your part. Of course, an
obvious rcason to control risks is to meet your profit objectives and margin rcquirements.

THE MENTAL STRESS OF TRADING

Still another, albcit implicit, reason for risk control is to light
the mental stress of trading. A series of big losses is enough
to shake the oonfidcnce of most traders. Tbe mathematics of
losng is against you, too. For example, to recover from a 50
161

162

Conlrolling Risk: The Kcy to Profita,bility

percent loss ofinitial capital, you oeed a 100 perceot gainjust

to restore ali your initial capital. Human psychology is such
that winning is exhilarating, and losing is depressing. Eveo
though, often, there are a few externa! reasons for both success
and failure, they are hard to see regardless of whether we are
winning or losing.
The trader also must bear the burden of working in a competitive, rapidly changing, and hostile eovironmeot in which
many variables beyond your control can affect performance.
The trader must also deal with the problem of informatioo
overload: A torrent of incomplete information vying for attentioo. Contrast this scenario to the slow moving, collegial
world of an engineer designing a oew product that wilJ not
go to markct for anothcr nine months. The engineer's world
is usually well structu rcd, with weU defined goals and limited
infonnation sourccs. Couple that with the fact that the enginecr is usually working within a team, which diffuses direct
responsibility and, consequeotly, diffuses both blame and
credit. The engi necr's team also provides emotional support
and serves as a nctwork of resources. The trader, on the other
hand, is flyi ng solo: .lt's your trade, win or lose. The trader's
peer group, on the o ther hand, is the competitor, which makes
making trading a high -speed dogfight.
Within thc trader's un eertain world, there is also the curse
of hindsight. lt is possible to look back upon yesterday or the
day before and determine in ao instant what should have
been done. The same decisions, bowever, are more difficult
to make in real time since you cao't ever know for certain
what the markets are going to do. Thus, trading demands
intense coocentration, which only heightens the emotional
responses to winning and losing.
Anothcr common cause of trading stress is what we caJJ
trend persistencc, akin to persistence of vision. You can use
the statistical test of runs to show that successive trades are
independent events. However, as you trade in real time, your
mind imposes trend persisteoce. It is easy to convince yourself
that winning or losing streaks will persist, which causes mood

Estimating Risk on New Posltions

163

swings between euphoria and depression or overconfidence

and no confidence. Overconfidenoe can gel you into bad
trades, just as lack of coofidence may kcep you out of good
ones.
There are no simple solutions for combatting any ofthese
mental stresses, but a zealous approach to risk control is a
largc part of staying sane while you are in the game.
Ooe way 10 sleep bellcr at night is to minimize your theoretical risk of ruin, as discussed by Nauzer J. Balsara. Your
trading suocess dcpcnds on the perceotage of capital risked,
the probability of succcss, the payoff ratio, and ultimately, on
controlling risk.

ESTIMATING RI SK ON NEW POSITIONS

The best place to bcgin risk control is to define the initial risk
on every ncw position. You should havc a clear idea of how
much money you are willing 10 lose in each situation. Selecti ng this initial risk oumber should be your first step.
A good first estimate is the amount of thc initial margin
requircd to open the position. Thesc margins change often,
and your broker may require an amount greater than the
exchangc minimum margins. Thcsc margins are a measure
ofexpcctcd volatility, with thc more volatile contracts usually
requiring a larger margin. We recommend that yotJr ioitial
risk not exceed the required initial margin. A good choice is
betwccn 50 percent 10 70 perccnt of initial margin.
Another way to select thc initial risk number is as x percent
ofyou r current or initial account equity. We recommend using
a number between 1-2.5 pcrccnt of the account. For example,
if your initial equity is SI 50.000, you could use $1, 500 as
your initial risk.
A third way to select thc initial risk is from lhe historical
testing of your model. John Swceney, the technical editor of
Stocks & Commodi11es magazine suggests an approach called
Maximum Ad verse Excursion or MAE. He plots the worst

164

Controlling Risk: Thc Key to Profitability

loss or ma.ximum adverse excursion for winning trades. In

most useful systems, winning trades do not ha ve MAE greater
than an amount such as $1 ,000-$1 ,250 per contracL You
should place your stop jusi beyond the worst loss shown by
a majority of winning trades. lf this dollar amount exceeds
2.5 percent of you equy, then set the stop at 2.5 percenL
Once you pick a number, you should place a stop loss order
with your broker to close out the new position. It is essential
that you maintain the loss cent.rol discipline by placing an
order every day or one tbat is good for longer tban a day.
Your stop will gel you out if the market moves against you.

165

Estimating Risk o n New Posltions

20

..

15

....~

12

'5

1
:z

- ~---1

o
<500

----

501750

751-tOOO

tOC>11250

t2511500

Mwmum IOss ol p<ofrtable ~

Maximum Favorable Excursion

Sometimes you will see trades lhat start off profitably, but
then quickJy develop into losing positions. We can the measure of the best profit of an cventually losing trade the Maximum Favorable Excursion (MFE), the opposite ofthe MAE.
lf you anal yze your winning and losing trades, you will find
that the losing trades often showed a small profit duri.ng the
course of the trade. The market thcn reversed and thc trade
ended ata loss. Similarly, winning trades usually went beyond
the "small" profit and stayed there.
To get a bctter idea of how MAE and M.FE work, look at
data in Figure 7.1. This figure shows the MAE for profitable
trades using a model for the British Pound. You' ll see the
resull for 38 tradcs, of which 28 (74 percent) had a loss of
less than $750; almost 90 percent of the trades had a worst
loss of less than $1 ,000. Therefore, an initial rnoney management stop of $1 ,050 would allow most profitable trades
to continue. In addion, you must decide ifthis $1 ,050 limit
also meets your risk exposure criterion: For example, you
would like $1,000 to be lcss lhan, say, 2.5 percent of your
equy.
To illustratc MFE, wc show an analysis oflosing trades for
the same model in Figure 7.2. Out of 104 losing trades shown,

FIGURE 7.1 Max1mum Adverse Excursion for a system that trades

British Pound utures contract.

67 (64 percent) showed a maximum profit ofless than $1,000.

As man y as 87 percent of the tradcs showed a maximum profit
of less then $2,000. We surmisc thcn that there is sufficient
volatility in this markct to stop us out cven after a good profit.
Thus, during trading, you must manage your stops aggressivcly whcn thc profit is less Iban $2,000. You must try to
avo id quick rcversals that will stop you out at a loss. After
thc profit exceeds $2,000, the trade is likely to mature into a
profitable one. Theo you can move the stops to the breakcven point, and perhaps even looscn thcm a bit to prevent
being stopped out by random fluctuations.
In short, M FE will hcl p you manage trailing stops during
tbe inilial portien of thc tradc, when you are most likely to
be stopped OUI.

Trailing Stops for Open Trades

Trailing stops can be used to protect open positions with a
signicant profit. The stop will gct you out should the market

166

Contro lling Risk: The Key to Profit..bility

60

'!! 40

20

ol__~~~~~====::::::::====~
c.1000

10~2000

2000..3000

3000-4000

Ellimating Risk on New Postions

167

You can set unique, difficuJ1-to-gun slops, using VIDYA, as

shown in Figure 7.3. Remember, VIDYA is a movi ngaveragc
that adapts to market volatility, which we discussed in Chapter 3. Sincc tbe moving average is based on closing prices, it
does not account for tbe PoSition of the most recent highest
high or lowest low relative 10 today's close. Therefore, VIDYA
stops are relatively "'loose," and best uscd for long-term trades
on a closing basis. Funber, such stops may not protect open
trade cquity in volatile markets as cffcctively as tbe volatilitybascd stop discussed below.

>-4000

Maximum profit of loslng tr1d<l1

Maximum favorable excursion for a system that trades

the Bn11sh Pound futures contract.

Volatility-Based Trailing Stops

FIGURE 7.2

make a sudden move against your PoSition. Tbe exact location of a stop is often critical to tbe outcome of tbe trade.
Because oftbis, we'll discuss many different ways to set stops
1ha1 you'll find useful at one Point or another.
One of severa] methods for setting trailing stops is to change
the stop jusi once a week. Say you use the Wednesday close
10 calculate the profit over thc past weck. You can theo advance the stop x percent ofthe weekly prolit, say 40 percent.
This mechanical stop will advance when you have a profit,
but will not step back when you bave a weekly loss.
Another simple approacb is to use a fixed do llar trailing
stop. For instance, you couJd use a $1 ,500 trailing stop, measuring tbe amount from tbe highest high or lowest low during
tbe trade. The exact dollar amount could be an arbitrary
amount from $500 to $5,000 based on your trading style.
One classic method is 10 set tbe stop j ust below a swing
low or high. A look at the price chart will show tbe most
reccnt significant high or low. Tbougb sucb a stop can be just
bcyond the recent high or low, be aware that such stops can
be gu n ned by floor brokers.

Volatility stops are another approach to setting smart stops

that are not easily gunncd. This approach is particularly useful
for intermediate-term tradcs to preven! being shakeo out by
market turbulence. These stops are derived from thc most
recent higbest high or lowcst low, and they move farther away
(loosen) from these extreme prices when volatility increases.
Tbe loosening of the stop, whjch prevents prematuro shak11 , - - -- - - - -- - - - - - - - - -- - - - .
US Bond 06/ 93 COntract Oaily Cklae

11 2 ! -- - - - - - > , 110
108

FIGU RE 7.3

A lrailng stop using VIOYA.

168

Controlling Risk: The Key to Prota.bility

eout, also means tbat you must give up a sizeable chunk of

your pro.lit if the stop is hit. Our solution is to modify this
approach in such a way that tbe stop can only advance witb
prices, not retreat. This will lock in a greater portien of potential profits, but still allow the market room to wiggle.
We begin by calculating a 10-day simple moving average
ofthe average true range (ATR). This function is available in
most technical analysis software. If il is not available, you
can use the absolute daily closing changes over the Last 10
days and average them instead. We'll call this quantity ATR 1oFor long trades, subtract the amount (3*ATR,0) from the highest high over the last 10 days. We'll call this quantity our
preliminary long stop. Finally, use the highest value of the
preliminary stop over the last 20 days as the actual stop for
long trades.
For short trades, add tbe amount (3*ATR,0) to the lowest
low of the last 1O days, calling this quantity our preliminary
short stop. Finally, use the lowest value of the preliminary
short stop over the last 20 days as the actual stop for short
trades.
You can experiment with the ATR,0 multiplier to get a
looser (4*ATR,0) or tighter (2*ATR, 0 ) stop. Or, you can use
a diflerent time period to calculate tbe average of the ATR,
or to calculate tbe highest (lowest) value of the preliminary
stops.
Figure 7.4 shows a light (l.5*ATR io) and a loose
(3.5*ATR ,0) volatility stop for a long trade in tbe September,
1993 Japanese Yen contract. Note how the stops llattened out
during volatile, sideways periods whiJe advancing witb rising
prices. There are two ways to use tbese stops. As an intermediate-term trader, you could use the looser stop as lhe
trailing stop. Jfthis stop were hit intraday, you would use the
tighter stop for reentering long trades. For instaoce, if you
were stopped out in early May when prices toucbed the loose
stop, you would have gooe long when prices closed above the
tighter stop. This wouJd have provided an excelleot reentry
point.

169

Oeveloping a Trade Template

Juo

FIGURE 7.4 A volatility-based stop used with the Japanese Yen fulures con1 raC1.

As a short-term trader, you would use tbe tigbter stop on

eilher an intraday or closing basis. Notice bow tbe prices
stayed above the tighter stop during strong up swings. As
expected, the tighter stop would have exited tbe position
sooner tbao tbe looser stop near the June higbs. However,
the tighter stops would have produced whipsaw trades during
tbe price consolidation in April and May. Hence, your trading
style would determine your choice of stops.

DEVELOPING A TRADE TEMPLATE

One of the toughest problems in managing lhe open trade is
deciding when LO take profits. Usually, there is no simple
answer since you can encounter so many different market

170

Controllng Risk: The Key lo Profitability

conditions. We suggest that you use the historical trading performance of your system to plot a template for assessing
trades.
You can plot the day-by-<lay evolution of each trade from
your system by ploning the best and worst equity of each day
on separate charts. If your system has provided more than
30 trades, you can develop a fairly represen ta ti ve trade profile.
First, you would average the best equity of eacb trade on tbe
first day of tbe trade to get a single numbcr for day J. Tben
you would do so for day 2 and so on. (Thcrc may be fewer
tradcs open as the number of days incrcascs.) Then you would
average th e equity across only open tradcs.
Your plot sbould be similar io the one shown in Figure 7.5,
whicb is from a system for trading 10-year Treasury notes.
The middle line shows the best equi ty for tbe average trade,
which peales nine trading days into its cxistence at about
S 1,500. We also plotted the lines one standard deviation ( + 1
sigma and - 1 sigma) away on cilher side of the average.
Tbese plots should account for about 67 percent ofthe trades
from tbis model. The upper line shows that the more prof-

Average winning Hd

2500

1&00

1000

1- - ---i,:-,,,- ,.-::--_- ____

--

Oeveloping a Trade Template

171

itable trades peak 9-10 days from inccption atan equity of

$2,500. lf the trade is moving slowly but is profitable, the
lowcr line suggests it would peak at about 11 days with a
profit barely above $500.
Th is plot now serves as a template suggcsting bow the trade
may evolve in time and price. Using Figure 7.5, let's assume
the market is strong and you had a $4,000 profit after just
three days in the trade. We would urge you to take your profits.,
or at lcast, tighten your trailing stop since the template suggcsts tbe trade is ahead of itsclf. lo contrast, if you had a
$1 ,000 loss on day 4, we would suggest closing the trade since
it secms unlik:ely to succeed.
We show the evolution oftwo trades iJl Figure 7.6 in wbicb
tbere is one big winncr and an average trade. Note how the
big winner's equity peaked at ninc days as expected. The other
trade peaked at 11 days. By showing the potential cvolutioo
of trades in price and time, a templa te gives you an estima te
of the probable course of a trade. Actual trades may develop
faster or slower, aod could last longer. You can conclude from
this that it's also important to check your recntry rules after
exiting a trade using the tcmplate.
Justas tbe trade templatc is an anticipatory tcchnique, you
can use price projcction methods to devclop dilferent scenarios for trade planning.

......=:~-- ...- - - -

..... - ,,,.....:;.._-:;;:::::::::::7 - -"""--

10

t1

12

13

O.yo In Trst

FIGURE 7.5 A typical trade template for a system trading the 10-year
Treasury Note contract.

10

11

12

13

01y1 WI 1t1de

FI GURE 7.6 Two actual trades overla1d on the typical trade template
for the 10-year Treasury Note contract.

t72

Controlling Risk: The Key to Profitability

~~~~~~~~~~~~~~

ANTICIPATING PRJ CES FOR A RISK CONTROL PLAN

Thus far in this chapter we used passive risk control strategies,
such as setting an initial risk or using ditrerent trailing stops.
lmplicitly, we assumed little about the next day's possible
range of prices. The trade template anticipates prices in a
general way, but does not explicitly account for the acrual
oat11rc of recent market action.
In this section, we propose a more proactivc approacb
toward risk control that is based oo anticipating tbe possible
range of priccs for tbe next trading day. Since the anticipated
range is only a guess, tbe trading plan develops specific actions
Cor exit, cntry, or reentry, wbetber tbe markct trades inside
or outside the expected range. ln tbis sense, the plan looks at
risk as weU as reward.
SpecificaUy, we will develop two target prices (H 1 and H2)
that are bigher tban today's close, and two target prices (Ll
and L2J tbat are lower tban today's close. Ll and L2 represen!
two levels of risk for a long position, and H 1 aod H2 are tbe
two levels of reward. The reverse is true if we ha ve a short
position. Our goal is to develop specific orders to be entered
under the following conditions duriog the next trading day:

Price exceeds H2.

Price trades between H 2 and H l.
Price trades between H 1 and Ll.
Price trades between LI and L2.
Prices falls below L2.

In actual trading, literally tbousands of different paneros

could occur bctween tbe limits H2 and L2, and the possibilities just listcd covcr only a few sit11ations. They are, bowcver,
a good starting point. First, tbough, let's discuss how to obtain
the price targcts.
First, you calculate tbe absolutc daily differeoce betwecn
today's price and yesterday's pricc. Then, take a 10-day simple

t73

Antlcipating Prices lora Risk Control Plan

moviog average of these daily differe nces. Note that tbis is

similar to the ATR,0 number used in volatility stop calculations. (In fact, you could use ATR,0 if you wish.) We' U denote
tbis average by tbc tener A, and project tomorrow's price
raoge by adding sorne multiplc of A to today's closing price,
as shown in tbe foUowing equations:
momentuml = c;,- c ,,
A - 10-day simple moving average of momentuml.

Hl - C0 + A,
H2 - C0
L l - Co
L2 - Co

+ 2*A,
- A,
- 2*A.

(7.1)

We denote tbe absolute momentum by lmomentum~ today's close by C,,, and yesterday' s close by C ,. The estimat~s
for tomorrow's highs and lows are H 1, H2, Ll, and L2. This
calc11lation says our best estmate for tomorrow's close is today's close. The total range is four times A ( 4*A}, being a
span of 2*A on eitber side of tbe close. You can vary tbe
n umber of days used in the calculations as well as tbe multiples used to project the span. This approach has tbe advantage that, as market volatiUty changes, the projected span will
change as well.
Now that we know how to co mpute price targets, we'U
illustrate how you can develop scenarios and risk control
plans. Assume tbat the markct has been trending lower and
you have a sbort position. A strong rally tomorrow wo11ld
close near or above H2. You also feel that such a strong rally
probably signals a reversa! of tbe dowotrend. In sucb a situation, you could cboose to pul in a stop order to close out
your shortand go longa fewticksabove H2. But, iftbe ~~et
traded between H2 and H I, you would hold tbe short poSJtJOn
and do nothing. Similarly, you would continue to hold your
short position if tbe market tradcd between LI and L2. In
addition, however, you feel only a selling clmax could pusb
the market below L2, so you could decide to cover your sbort

174

Conlrolling Risk: The Key to Profitability

PoSition (but not go long) if the market traded below L2 at

any Point during the day.
You could use the pricc targets as entry Points. For instance,
L1 ~~uld !><: an entry Point into an eltisting uptrend for long
Pos111ons. Similarly, you could look for shon en tries near H J
with a stop at H2 or between H 1 and H2.
'
Anotber application ofthe projected price targets is to visualize how tbe daily chan would look should the market close
~ithin the inner band or at any othcr point inside tbe proJCCted range. Such a close mighl com plete or start a pce
pallcrn lhat may havc predictive valuc. In this way, you can
gct ajump on your chan analysis by using the projected range
as a template for tornorrow's price bar.
These examples show tbat you can work out many dferent
scenarios w_ithio objective projecons of tbe next trading
range. The llDPonant Point to note is that you can idcntify
speci6c price levels for concrete trading decisions.
How weU do the projected price targets work? Figure 7.7
shows the projected price targets for the Japanese yen September, 1993 contract. Note how this market often found
supPon and resistance near the inncr bands (H 1 and Ll). The
yen casily reached or exceeded the outcr bands (H2 or L2)
on strong up and down days, and found SUPPort and resistancc at the outer bands at major turning points. You can see
how thc bands widened in response to thc increasing volatility
from carly June to carly July. Thus, thc projected price targets
would have been quite useful for devcloping a risk control
plan.
. Thc August, 1993 Cornex Gold (Figure 7.8) contrae! prov1des another example from the futures markets. Gold often
traded outside the (HI-LI) in.ner band. During the intermediatc top in June, 1993, an open outside H2 was a good profittalcing opponunity as you may expect Tbc H J band provided
good entries for shon trades in late May. Similarly the Ll
band provided good en tries for long trades in Ju ne. Thls chan
also shows how the bands narrow and widen in response to

175

Practica! lssues in Risk Control

m
m
m

m
...10.0..Y AllEAAGE Of AESOtllll: OM.Y CHANCE
Cjl I CLOSE Of YESTERDAY

NER 6'HlS; ctll+M

OOJTE~ 8.t.\OOS

q 14

,,

2""
21

12

FIGURE 7.7 The Japanese Yen September. 1993 contract is shown

with the projected pnce targets.

changing volatility. Tbis fcature is useful in tradiog the volatile futures markets.
In surn, it is clear that our proactive risk control stratcgy
based on projected price ranges can be successfully implementcd in the futures markets.

PRACTICAL ISSUES IN RISK CONTROL

We ha ve discussed a nurnber of dilfercnt approacbes for risk
control without touching u pon sorne ofthe operacional details
of making them work. In this section, we'U mention a few
iteras that may help you execute your risk control strategies
more smoothly.

176

Controlling Risk: The Key lo Profitability

177

Praelical lssues in Rlsk Control

HIGH GAAOE COPPER CSl.l) D~Y 0711 Wl C"85 15 1 05 OoeS.15 -

00 L=82 D

EO

e
Jul

FIGU RE 7.8
contrad .

93

Projecled price targets or August. 1993 Comex Gold

FIGURE 7.9 High Grade Copper Seplember, 1993 conlrad showing

Fibonacci retracemenls.

Wben you devclop your price targets H2, HJ , LJ, and L2,
observe whether thcy faJI ncar any significant retracemcnt
points such as 33 pcrccnt, 50 pcrcent, or 67 percent for the
current move. Thcse are callcd Fibonacci retracements, and
often provide points of support and resistance. Figure 7.9
shows these Fibonacci retraccmcnts in the Higb-Grade Copper September, 1993 con tract. The major move from A to B
produced a 50 percent retraccmcnt to C. Smaller moves from
A to g produced a 64 pcrcent rctracement to h. Similarly, !he
move from d to f produced a 47 pcrcent retracement to e.
The conclusion? Pay particular attcntion if your projected
targets happen 10 be near Fibonacci retracement levels.
Aftcr you calculate a 1railing stop, avoid placing it near a
round number such as 62.00 or 62.50, but place i1 7-33 ticks
past the round number. For cxample, set your sell stop at

60.87 and hope tha1 the market will find support near 61.00.
You may notice that reversals in the US T-Bond market usually oocur less than 7 ticks beyond the previous higb or Iow.
Thercfore try 10 place your stop say 11 ticks beyood the
previous high or low. Often, tbosc fcw extra ticks may be the
dHfercncc belwecn having a winning trade or avoiding a losiog onc.
Evcn though wc have continuaJJy mentioncd stops (stop
ordcrs) for risk control, using market orders may be tbe best
way to cxit a pcsitioo during the trading day. Once you decide
to closc a pcsitioo, get out as fast as pcssible.
Study the type of orders your broker wilJ accept, since they
can influence how you actuaJJy enter orders for your risk
control plan. A well-placed order will help you trade bener.
Anotber ordcr placement tactic is to use tbe facsimile machine

178

Controlling Rlsk: The Key to Profitbility

to send. orders, since it avoids confusion and simplifies error

correct1on.
One fi~~ o~der entry tip: Next mc you have to roll over
your pos111on rn an expiring contract, use market orders but
on. a " not-held" basis, gjving your broker the discretio'n in
filllng your trades at a time when thc spreads between the
two contracts are favorable.

8
How to Use This Book

CONTROLUNG "INVISIBLEH RISKS

\)ur discussion on risk contr0I was orientcd towards pricensk, but be aware that there are many other forros of risk,
a.nd you sbould recognize tbeir existence as part of your larger
nsk control .strategy.. We call tbese risks "invisible," beca use
Y?u ca~ easily los~ s1ght of them in the heat of trading. Our
discuss1on. here will be brief, tbough, because otber authors
bave proVIded detailed discussions elsewhere (see Bibliography).
. Bccause many market.s are correlatcd, you incur portfolio
nsk fro~ the combination of markets you trade. Remember
~t trading co~elat~ markets is equivalcnt to trading mulUple contracts m a srngle market. Market characteristics such
as gaps, volatility, liquidity, aod order exccution are other
fact~rs ~at inllueoce risks arising from your portfolio. Divers1ficatJoo over six or more markets is one approach to
contr0Uing these risks.
".'our asset aUocation and use of leverage are other forros
o.f nsk. There are man y ways to balance the amount of capital
nsked to tradrng with the desired risk-return tradeoffs. We
refer Y?U to ~he books by. Nauzer J. Balsa.ra and Ralph Vince
for a d1scuss1on of these 1ssues. ~opular rules include risking
2.5 percent or less of total equ11y per trade, and using 30
percent or less of total equ1ty toward margin requirements.
In short, you assume many "invisible" risks when trading
tod~y's markets. As a result, you should address these forros
of nsk as part of your overall risk control strategy.

The best way to use this book is to integrate our ideas into
your own trading proccss. We'U give you examples ofhow it
can be done by combining indicators to find new, elfective
trading systcms. For instance, we'll show you how to devclop
an adaptivc trading systcm combining the momentum oscillator, CMO, with VIDYA. We'll also show how you can use
the ideas of composite momentum and linear regression
analysis for market rotation-moving out of quiet markets
i11to those experieocing major moves. We hope these examples may stimulate otber ideas in your own mind on how to
combine tbe new indicators in this book.

A CMO-DRIVEN VIDYA TRADING SYSTEM

Tbe varying volatility of markcts causes particular difficulty
for long-term systems: They tcnd to generate more false signals. Obviously, then, a system that adapts to markct volatility would be highly desirable for long-term treod following.
Another significant advaotage of an adaprive system versus
IUed parametcr systems is that you do not have to worry
about optimizing system parameters. Optimized system pa179

180

How to Use This Book

rameters onen work wondeully on hislorical data, but fail

miserably in real-time trading.
We wiJI build our system using VIDYA, tbe adaptive moving average. We will index VIDYA to market momentum
using the absolute Chande momcntum oscillator (absCMO)
as discussed in Chapter 3. Our CMO-driven VTDYA based
system wiU anempt to follow long-tenn trcnds, perhaps witb
a wioning percentage around 40 percent aod a payolf ratio
greater thao 2.50.
We want to combine VlDYA and CMO in such a way that
the effective length ofthc moving average increases wheo tbe
market has low momcntum (absCMO vaJues nearO). Furtber.
we waot the effective lcngth to decrease when tbe market gaios
momeotum (absCMO > 0.30). ldeaJJy, this average would
move rapidly whcn priccs make a big move. Then, as momeotum slows, the average will slow down also.
To combine CMO and VIDYA, we'U fust define CMO in
terms of tbe unsmoothed 9-day RSJ because we'll use the
System Writer software from Omega Research for our caJculations, which provides thc unsmootbed RSI as a built-in
function. We'll use that function to calculate CMO, dividing
CMO by 100 to get oumbers bctwcen + 1 and - 1, rather
than + 100 and - 100. We arbitrarily used a 9-day RSI to
calculate CMO of the daily closc. Ninc days is a popular
choice for short-to-intermediate analysis, but, you could use
a 14-day CMO or any other lengtb you wish. Second, we will
take the absolute valuc of CMO to eliminate the negative
sign, aod multiply it by a scaling constaot t. This scaling
constant t is lcss than or equal to 0.50 to provide a smooth
conversion between CMO and VIDYA. We will designate the
scaled, absolute CMO by the symbol A. Heoce, the equations
for calculating CMO and VIDYA are:

A CMO-Orlven VIOYA Trading System

181

Wben the market is trendiog stroogly, VIDYA wiU have

stroog momentum in the trend's direction which will give
high values of the absolute CMO. As A increases, the term
A Close in the definitioo of VlDYA increases. The average
is lheo taking larger chunks out of new data, decreasing tbe
effective length of the average. The opp0site happens when
momentum decreases during sideways periods.
To choose a value for tbe scaling constan! l, we want to
avoid testing over a variety of values to find ooe particular
vaJue that works best. Our design goaJ is for a long VIDYA,
so a " small" value of t, such as 0.05, 0.10, 0.15, or 0.20, is
what we reaJly need. Equation 8.1 for VIDYA shows thal as
the values of t decrease, the term A also decreases, and we
use smaJler and smaller fraetions of ncw data to update VIDYA, increasiog the effective Lengtb ofVIDYA. Now we can
pick a value between 0.05 and 0.20 arbitrarily. Or, we can
use additional calculations to help narrow the choice.
The approach we used to choose a value for t was to test
thc four values, 0.05, 0. 1O, 0.15. and 0.20 over 24 markets
using the data from 01/01/92 to 04/26/93. We again used the
CSI #39 Perpetual contract. The results are in the Table 8.1,
with the columns showing net profit (or loss) for different
values of t. On studying the average profit in Table 8.1 for
different vaJucs of L, we found tbat thc resulls were very similar for t-0. 1O aod 1-0. 15. Consequently, we decided to use
the value 1-0. I as our constan! across ali markets and time
pcriods.
When ICM0 - 1, t -0. 1 translates iot? a !~-da.Y exp0n~n
tial moving average, which meets our des1gn cntenon of usmg
a long VTDYA. We can now define the long and shon entry
rules as follows:
t. The rule to open long positons. lfthe close oftoday and

c Mo - K2Rs 1. - 100)1q
A - t4CMOj (t <- 0.5),
(8.1)
VIDYA, - A. Close + (1 - A) . VTDY\..

yesterday are both above tbe VTDYA oftoday, theo buy

tomorrow oo a buy stop order. Place the order 1 tick
above today's hlgh.
2. The rule to short posilions. If the close of today aod tbe

182

How to Use Thit 8ook

Tabl~

Market
Bean o il
Bn1ish pound
Canadian $
Cocoa
Coffee
Cotto n
Crude o il
Deutsch mark
Eurodollars
Gold
Heatmg 011
HG copper
Japanese yen
Uve cattle
Uve hogs
Po rk bellies
Sil ver
Soybeans
Sugar
Swiss franc
T no te 10-yr
US bond
USDX
Whcal
Average
Stand. Dev
Avg/stdev
Max
Min
Ma.x/min

8.1 Effect of Changong Scahng Constan! t

o.os
-1078
16682
1480
- 2170
11177
-515
- 790
8000
1025
- 2350
109
4372
6887
-1 548
4175
- 5492
- 4955
3150
2824
15050
6481
2981
1920
2206
2785.04
5529.34

o.so

16682
-5492
3.04

Scale \ for VIDYA -

0. 10
0. 15
302
25706
2230
- 1930
9005
5
- 150
5837
3 125
2410
1293
5082
10287
- 20 13
4716
3704
- 2890
22 13
109 1
22463
11137
661 2
5240
3032
4340.52
71 47.56
0.61
25706
3704
6.94

- 2778
31381
2860
-350
5841
-4510
- 1490
5087
3325
3250
177
4687
134 12
1312
5106
- 163
- 4115
2663
2983
21262
1191 8
- 1087
1470
5431
4306.68
7987.04
0.54
31381
-4530
6.93

0.20
-3598
30537
2860
-2070
718
-7310
- 610
32 12
2525
5550
-2013
5787
14325
820
4900
- 1306
-5075
800
5 10
5775
8812
- 6206
1290
4043
2571.84
7521.35
0.34
30537
-7310
4.1 8

A CMO Driven VIDYA Trding System

183

close of yesterday are both below the VIDYA of today,

then sell tomorrow on a sell stop order. Place the order
1 tick below today's low.
You also could trade on a closing basis for stocks and mu tual funds. We won' t use any specific exit strategy in testing
the model, wbich means a long entry signa! wil l exit a shon
trade and vice versa. We'll use a S 1,500 initial stop, wh ich
can be refined later using the MAE or maximum adverse
excursion idea. We'll increase losses by $1 25 and decrease
profs by $125 to allow for slippage and commissions.
Our data comes from Commodity Systems !ne. (CSI). in
tbe fonn of their #39 Perpetua! Contract, though you can use
any otber perpetua! contract or continuous data you like. We
wiU use ten years of data for the Swiss franc contract staning
05/ 26/ 83 and use the System Writer Plus (SWP) software
package for system testing. The data will be divided into two
approxiroately equal five-ycar blocks. First we'll test the
model over the May, 1983 to May, 1988 period. Tbeo, wc'll
run it over the May, 1988 to April, 1993 period.
Remember, we are hoping our long-tcrm model for the
Swiss franc will have a winning percentage between 3~5
percent and a payoff ratio grcater than 2.50. Let's see then
how our expcctation matcbcs up with the results from the
fust five-year block, shown in Table 8.2. For starters, we know
our test period was not too sbort since there were 33 trades
over this test block. The SWP output shows that 36 percent
of the trades were winncrs and the ratio of average winner/
toser (payoff ratio) was 3.37; the payoff ratio was 3. 18 without
the largcst winner and toser. On average, tbe model cut off
losing trades in 13 trading days, and allowed the average winning tradc to run 79 trading days. Thc average trade made a
profit of more than $700, well above our slippage amount of
$1 75. Tbus. this unoptirnized model did meet our main design goals from which wc can cooclude it is a reasonable long
term trcnd following model.
An anal ysis of the MAE showed none of the winning tradcs

184

How to Use This Book

Table 6.2 Swlss Franc Test Block # 1

Model i'ame
i'otes
Dala
Cale Dates
Commissions
Slippage
Margin

: Z New Vidya CMO Model

. long Term CMO-based VIDYA model
. SWISS FRANC CSI #39 Perpetual Contrilct

:
:
:
:

05/26/83
$50
S75
SJ.000

Total net profit

C ross pro fit
Cross loss
Total ,. of trades
Pcrcent pro fitable
Num ber winning tr.1dcs
Number losing trades
largest winning trade
largest losing trade
Average winning lrade
Average losing trade
Ratio avg win/avg loss
Avg trade (wm & loss)
Max consecutive winners
MJX consecutive losers
Avg # bars in winners
Avg " bars in losers
Milx closed-out drawdown
Max inlraday drawdown
Profit factor
Max # of contracts held
Account size required
Return on account

05/30/88

$23,400.00
$48,662.50
$ - 25,262.50

33
36%
12
21

58,262.50
S- 2, 175.00
$4,055.2 1
$- 1,202.98
3.37
$709.09
2

5
79
13

$-5,950.00
$-6,850.00
1.92
l

$9,850.00
237%

A CMO-Ori~n VIOYA Tr~ding System

185

had an MAE of more lhan $1 ,000. Thus, in actual trading,

we could use an initial money management stop of $1 ,050
instead of $1 ,500 used in the testing. Nevenheless. wc will
continue 10 use lhe $1 ,500 number when we test the model
over the second data block to maintain consistency.
An anal ysis of the maximurn favorable cxcursion (MFE)
shows that:
There were 12 losing trades with a maximum profit of
less thao $1 ,500.
Another seven losing tradcs had a maximum profit of
between $1 ,50 1 and $2,500.
There wcre just 1wo losing 1rades with a maximum pro6t
of more than $2,500.
T hus, you could move your stop to the break-even p0int after
a maximum profil of $2,501. You could also manage your
slops more aggressively ifthe profit were more Iban $1 ,501.
Now that we have a model that worked on our 6rst test
block, we must do sorne "forward testing", that is, test it on
aoolhcr contiguous data block without changing any variables. This is called an "out-of-sample" test. The goal of the
second set of tests is to check if thc model performance de~
grades in any way. This might occur if we "curve-fitted" the
model on the tirst set, by adding so many conditions that we
picked up nuances in thc test set that are unlikely to rcpeat
in actual trading. The test over thc second data block might
tell us how weU the model would do in real trading. Jf the
performance were about the samc on both blocks, wc would
ha ve greater confidencc in the modcl than if the performance
were poor on the second set.
Wc next tested the Swiss franc data from 06/01/ 88 through
04/27/ 93; the results are in Table 8.3. We had 27 tradcs in
this out-of-samplc test, which we will generously round up to
30. Our results in this period were as good as the results in
the tirst test period: Our winning pcrcentage was the same

186

How lo ~ This Book

Table 8.3 Swlss Franc Test Block #2

Model Name : Z i'ew Vidya CMO mode l
No tes
: O ut-o-sample test, long-term CMO-based VIDYA
Data
: SWISS FRANC CSI # 39 Perpetua! Contract
Cale Dates
; 06/01 / 88 - 04/ 27/ 93
Commiss1on
: $50
Shppage
: $75
Margin
: $3,000
Total net profit
Cross profit
C ross loss
Total # o trades
Percent pro itable
Number winning trades
Number losing trades
La rgest winning trade
l argest losing trade
Average winning trade
Average losing trade
R.1110 avg win/ avg loss
Avg tradc (win & Joss)
Max consecutive winners
Max consecutive losers
Avg # bars in winners
Avg # bars in losers
Max d osed-out drawdown
Max intra-day drawdown
Pro it aao r
Max # o contracts held
Account size required
Re turn on a ccount

S49,775.00
$77,737.50
$ - 27,962.50
27
37%
10
17
$1 7,012.50
$- 3,037.50
$ 7,773.75
S-1 ,644.85
4.72
$1,843.52
3
4

102
10
S- 8, 112.50
2.78
1

$11 ,712.50
424%

CMO-Ori~n

VlOYA Trding SyJtem

187

(37 percent), and the payoff ratio was somewhat higher al

4.72. We recalculated the payoff ratio al 4. 1O withou1 the
largest winning and losing 1rades. The average losing trade
!asted about 10 trading days, with winning trades, on average,
open for 102 trading days.
In all, our trend-following model did reasonably well over
a l 0-year lime framc. Our 1988-93 test period had se ven
winning trades with an MAE less 1han $1,000, and 11 losing
1rades with an MFE less than $1,500. Tbere were four losing
1rades wi1h an MFES 1,501-$2,500 and only two losing trades
with an M FE greater tban $2,500. This shows tha1 you eould
move up your stop to break even after a maximum profit of
$2,501. The Swiss franc has sufficient volatility to knock out
our initial stop before tbe model can give an opposi te entry.
We also tested tbe model with continuous contracts developed using the Continuous Cootractor software from
TecbTools over the same time period using CSI #39 Perpetua!
Contract. T here was no significaot difference in results with
the two sets of data. We oow had even more coofidence io
the model. A comparisoo for a four-year period is io Table

8.4.
As you can see, this model does follow long-term trcnds,
judgiog by the oumber oftradiog days io the average winning
trade. There are oo guarantees tbat it will work as well io the
future io every market, bowever. It is simply a good 1reodfollowing tool thal adapts to market volatility, and il's probable that this model could be profitable io trending markets
that have occasional volatile periods.
We could now test tbe model over successive 12-month
periods to develop data on the variabilily in tbe payoff ratio
and winning percentage. Or, for planning purposes, we could
assume a 35 percent winning percentage and a payolf ratio
of 2.50.
Two weaknesses in this model deserve mention. One is its
perfonnance in sideways markets or trendless markcts tbat
produce whipsaws. The other is its performance in very volatile markets, which can stop the trade out at a loss before

188

Table 8.4

How lo Use This Book

Comparison: Perpelual Vs. Continuous Contract Data

Swiss franc ContraC1

Model Name
Notes
Cale Dates
Comm1ssion
Slippage
Margin

:
:
:
:
:
;

l. New Vidya CMO model

Checks on CSI and Con1inuous ConlraC1
01/26/88 - 03/25/92
S50
S75
$3,000

Data
Total net profi1
Cross profit
Cross loss
Total # of trades
Percent profitable
Number winning trades
Numbcr losing lrades
Largest winning lrade
Largest losing trade
Average winning trade
Average losing 1rade
Ratio avg win/avg loss
Avg trade (win & loss)
Max consecu live wlnners
Max consecutive losers
Avg . bars in winners
Avg # bars In losers
Max closed-out d rawdown
Max 1ntraday drawdown
Profit factor
Max # o con1raC1s held
Account size required
Return on account

CSI 39

Continuous ContraC1

SJ 1,237.50
$56,900.00
$ - 25,662.00

$29,737.50
$57,912.50
$- 26, 175.00

29
27%
6
21

30
23%
7
23

$17,300.00
S-2,900.00
$7, 1 12.50
s - 1,222.02
5.82
$1,077.16

$17,312.50
$ - 2,900.00
$8,723.21
$ - 1,225.00
6.75
$991.25

3
7
98
6
S- 8,687.50
S-9,275.00
2.21
1
$12,275.00
254%

3
7

109
9
$- 8,637.50
$-9,225.00
2.05
s 12,225.00
243%

Market Rotallon

189

the model can generatc an opposite signal. The volatility in

the fall, 1992 was a good example of a period in which this
model did not do well , which is not surprising since the model
must be made insensitive for long-term trend following. Markets that have long trading ranges also inevitably produce
losses, though VlDYA does help improve performance in volatile and sideways markets over other moving averages.
You can test other variations of this model, such as increasing its sensitivity by using a value of t greatcr than 0.10.
Or, you could select a time period shortcr than nine days to
calculate the CMO uscd 10 drive this model.
You can now procced to integrate risk control with this
model. For example, you could use VTDYA itself as a trailing
stop, test a volatility-based stop asan exit strategy, rigorously
testing each model. Be sure to review the results grapbically
to beuer understand model performance.
Before trading with the model, you could develop a typical
tradc template as discussed in Chapter 7 to understand trade
evolution and considcr discretionary exits. ll should be apparent by now that the model calculations are only the starting
p0int for building the complete system, which includes risk
control and money management.
We hope this discussion has giveo you a feel for the process
of developinga trend -followi ng model that combines different
indica1ors from the book. You'U probably agree that there are
many tradeoffs and no one model can do it ali.

MARKET ROTATION

Which markcts should you be trading this week? The ones

most likely to have big moves, ofcourse, ifyou could identify
them early eoough. You'U nd an answer to this eternal question if you use our new indicators to rank markets by their
price action. You can theo move out of ioactive markets in to
active o oes, a process we caJI market rotation, which is often
the key to big pro6ts io the ever-sbifting futures markets.

190

How to Use This 800.k

Table 8.5 ranks 25 major futures markets on composite

momentum aod trendiness measures. Marke1s teod 10 stay
near tbe top or bouom of the list for sevcral days ata time,
and hence market rotation provides significant trading opportunities. To rank markets, we found composite momentum, wbicb was the sum of 5-<iay, 10-day, and 2Ck!ay momentum convened into dollars. We also found tbe 18-<iay
slope and converted it into dollars. This shows thc cxpected
change in the daily close from a best-fit Line.
There are two measures of trendiness: the 18-day ri of the
best-fil line and tbe 18-<iay A DX. The trend was significant
only if value of r2 was grcater than 0.22.
Table 8.5 shows the markets witb tbe strongest uptrends
near tbe 1op of the rankings; the markets witb stroog downtrends are ncar the bottom. The markets at the top or bonom
may be surging, and good candidates for short-term trades.
The markets in the middle are changing markets, and good
candidates for new trading ideas. Because these markets may
be consolidating witbin long-term trends, they may provide
good entry points in tbe direction of tbe long-term trencl
Further, since sorne oftbese markets may be seeing a change
in the long term trend, they may be good candidates for taking
antitrend positions or early positions in the direction of the
new trend
Table 8.5 presents a systematic method of surveying market
trends and obscrving changes in trends. lf you are a discretionary trader, you can rotate into markets that are strong. A
systems trader could reject a signa! from a trend-foUowing
model if the market is not trending. Or, you can wait for the
trendiness indicators 10 strengtheo before taking a new signa!
from a trend-following model. We ha ve observed that markets
stay near the top or bottom for many days at a time. Thus,
these rankings are relatively stable from time to time, providing excellent opponunities for market rotation.
Sorne of you have probably guessed that the idea ofmarket
rotation could be generalized to stock rotation. Table 8.6
shows j ust such a sample stock trading List where we have

19 t

M<arket Rottion
Table 8.S

Ranking Markets by Momentum Using 'learby Contracl

(Data at close of; 04/29/92)

Market
Swoss franc
British Pound
Solver
Coffee
Gold
Japanese yen
Deutsc he mark
Sugar
Caule
Heating 011
Crude 011
Eurodollar
Cocoa
Soybeans
Frozen O J
Corn
Hogs

Conon
W heat
us bond
Canadian $
Copper
US S index
Bellies
lumber

Composile
Momen1um

SIo pe
18-Day

($)

($)

8288
7550
6875
611 1
5340
4736
4525
3069
1092
916
530
94
- 190
-338
- 645
-938
- 1497
-1560
- 1875
- 1906
-2180
- 4763
-6350
- 8496
- 14850

232
234
97
98
97
205
99
91
12
-17
- 23
60

- 3
- 14

- 68
- 28

- 40
- 51
- 48
74
- 24
- 152
-181
-138
-173

R-Sqd
18-Day
0.74
0.73
0.60
0.29
0.67
0.63
0 .49
0.41
0.09
0.26
0.44
0.28
0.0 1
0 .20
0.61
0.67
0.62
0.39
0.76
0. 11
O. IS
0.73
0.77
0.56
0.26

ADX
18-Day
48.72
54.67
34.94
30.61
45.89
40.61
41.72
18.22
31.33
12.00
18.44
25 .22
14.33
22.06
28.17
45.00
2 l.11
15.28
34.78
26.33
17.94
64.83
51.44
11.78
56.33

Compo~otf' Momentum: Sum of 5-, 10- and 20-day momentum n dollars.

Slope: Slo>e of 18-day regression In dollars.
Rsqd: R-squared; lrend exists 11 R-squared > 0.22.
.
ADX. Average Oirectional lndex; trend exists if this is moving up and os greatcr
than O 20.

Table 8.6

Rank
1

2
J

4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

21
22
23

24
25
26
27
28
29
30

Stocks: Ranked by Composue CMO

Stoc~

T uscon Elec Pwr

Mexico Fund
l.aidlaw B
Texas U1 ilities
Hansen Ple
South Wstrn Bell
BioTechGeneral
GTE
US Surgical
Philip Morns
Bristol Myrs Sqb
lllinoos Pwr Co
!'\evada Power
Mellon Bank
Amer Home Prod
Abbor Labs
Heinz
Amgen
Boeing
lllinois Pwr Co
Varity
Johnson & Johnson
Pfizer
Waste Mgmt
Gap Stores
Pepsi
Merck
Hong Kong Telcm
Laidlaw B
Chemical Waste

0 7/ 13/ 93

Com
poste CMOIO CM020 CM030 CM050

39.30
37.12
32.48
24.36
15.95
14.89
13.96
13.65
3.58
- t.56

33.33
55.56
46.15
47.83
63.64
37.93
18.18
4.76
24. 14
0.00

- 1.95
-2.17
- 2.90
- 3.53
-8.26
- 9.49
- 12.16
- 13.25
- 15.56
- 19. 15

12.50
- 36.84
- 14.29
- 30.30
-t.96
- 7.14
- 8. 11
- 41.67
- 10.00
-42.86

- 19.41
- 20.47
- 24.09
- 24. 16
-27.96
- 28.23
-29.40
33.96
- 4$.74
- 55.56

-1 4.29
- 12.82
- 31.25
- 37.78
- 49.02
- 65.22
- 33.33
- 73.91
66.67
- 53.85

37.50
14.81
42.22
10.38
46.84
6.90
41.46
1.62
12.50
1.34
5.66
3.60
8.11
4.67
21.95
7.40
20.63 - 6.96
-4 62 - l .21
-1 0.59
15. 15
7.69
8.94
- 12.62
-8. 20
- 18.64
3.23
- 24.44
-17.81

- 23.33
- 34.69
- 34.94
-36.08
- 28.74
-20.00
- 47.37
-29.63
- 52.38
-77.27

- 1.97 - 3.81
2.32
6.02
- 1.67
0.00
1.68
2.19
- 3.66 -7.50
- 3.79 - 1 1.27
-6.67 - 1.89
-.43 - 13.29
-6.22 - 3. 17
- 5. 13
- .56
- 7.74 -16.83
-8.21 - 9.76
-7.79 -6.81
- 6.37 - 3.78
-10.97 -1.22
- 4.53 - 14.10
-9.10 - 9.62
- 11. 12
1.02
-13.14 - 24.53
- 18.53
35.56

Rank Ranked by average o all CMO values

CM010 = 10-day Chande momenlum osclla1or
CM020 - 20-day CMO
CM030 - 30-day CMO

CMOSO = 50-day CMO

Composile - (CMOI O + CM020 + CM030
Overboughl al CMO > - 50
Oversold at CMO < - -SO

192

41.94
19.57
16.22
3.28
- 8.33
5.17
15.53
5.66
- 9.59
7.98

+ CMOS0)/4

Mari.el Rolation

193

rankcd the stocks b y a composite CMO. We first calculated

the 10-, 20-, 30-, aod 50-day Chande momentum oscllator,
and used a simple aritbmetic average of tbese values !O rank
the stocks. You would short the weakest stocks and buy the
strongesL lf you owncd these stocks, you would tighten stops
as the stocks weakencd.
We calculated these rankings using daily data. lfyou wish,
you can use weekly data for the rankings, wbich would reduce
the frequcncy oftradcs since you would buy ooly those stocks
with a positive CMO.
You also can use this ranking for antitrend tracliog by looking for overvalued stocks ncar the top of the ranlcings and
undervalucd stocks near the bottom. You also can look for a
bounce in oversold stocks and a corrcction in overbought
stocks. Table 8.6 includes four time periods, so you can study
the numbers to sec where cacb stock is in diffcrent time
frames. Sucb a tally sheet is particularly useful when you study
it for several weeks in a row. What's more, it's faster than
reviewing the equivalen! chans.
Lct' s review the " rotation" tables: We built them by combining the new indicators from momentum oscillators aod
linear regression. Wc also uscd the idea of combinng data
from differeot time interva ls into a composite inclicator. It
should be apparent now how combining our ideas can improve your trading.
To furtber stimulate your creative processes, Table 8.7
shows bow our indicators may be combincd to gain an cdgc
under different tracling styles. You have already seeo the de
tailcd discussion of C MO-driven VIDYA and market rota
tion. One combination for stock market timing could combine Qstick, the StochRSI, the thrust oscillator, and pricc
targets. For instance, you could tak:e profits at the LI price
target if overso ld conditions on Qstick and StochRSI are confirmed by extremes in the thrust oscillator. Here we combined
intraday momentum (Qstick), with interday momcntum
(stocbRSI), and advance/decline data (thrust oscillator).
Tools for open trade management include tbe MFE, volality

194

How to

Us~

This Book

Tibie 8.7 Examples of Combining lndicators for Analysis

llem
Slo""
r'
VIOYA
ns tic k
CMO
St"rhRSI
MT
Thrust Ose.
ITOl
MFE
Volatility
SIOD

Price Tantets
Trade
template

CMOdriven Market
VIDYA Rotation

Stock
Marke1
Timing

Open
Trade
ManagePrice
Trend
ment Extremes iness

.
.

Bibliography

stop, pricc projection, and the trade template. Tbey can belp
you place unique stops critica! to profitability.
Dcpending on your time frame, you could identify price
extremes using the regression slope or momcntum oscillators
(CMO,. stocbRSI, or DMI). On occasion, you may want to
determine presence or absence of trendiness in a market by
using the linear regression r2, the absCMO, or even a volatility
stop. You could derive many other combinations by adapting
thcse flexible and powerful indicators to your trading style.
We encourage you to do so, for your unique combination
could boost your profits by giving you a mucb oeeded analytical edge in today's tougb markets.

Altman. Roger. ''Relative M omentum lndcx: Modifying RSl." Techmcal Analysis o/Stocks & Commod11ies, Vol 11, February, 1993 p .

30-35.
Arms. Richard. The Arms lndex: An lntroducuon to the Volume Analys1s of the Stock and Bond Markt'ls. llomewood. lL: Dow Joncslrwin, 1989.
Babcock Jr., Bruce. Business One lrwin Guide to Trading Sysrems.
Ho mewood, ll.:, Business O nc lrwin , 1989.
Balsara, Nauzer J. Money Managem ellf Strategiesfor Futures Traders.
New York: John Wiley & Sons, lnc., 1992.
Bla u, William. "Tbe True Strength Tndex," Tec/111ical Analysis ofS1ocks
& Commodities, Vol 9 No. 11, November, 199 1 pp. 18-31.
Burke, Gibboos, "Gaio without Pain: Money Management in Acon,"
Futures Dec 1992, (Vol XXI 1114) pp. 36-38.
Cbande, Tushar S. " Adapng Moving Avcrages to Market Volality."
Technical Analysis of Stocks & Commodities, Vol 10, No. 3 ( Mar
1992), pp. 108- 114. Copyright 1992 Tecbnical Analysis, lnc. Used
with permission.
Cbandc. Tusbar S. "Forecasttng Tomorrow's Trading Day." Technical
11nalysis of Scocks & Commod111es. Vol. 1O. No. 5 (May 1992), pp.
220-223. Copyright 1992 Technical Analysis, lnc. Uscd v.itb permission.
C hande, Tushar S. " Market Tbrust." Technica/ llnalysis of Suxks &
195

196

Bbliography

Commod111es. Vol. 10, No. 8 (Aug 1992), pp. 347-350. Copyright

1992 Techn1cal Analysis, lnc. Used with pemmsioo.
Chandc, Tusbar S. "Sman Stops." Techmcal Analysis o/Stocks& Com
mod111es. Vol. IO, No. 12 (Dcc 1992), pp. 507-509. Cop)ngbt 1992
Technical Aoalysis, loe. Used wttb perrn1ss1on.
Cbandc, Tushar S., aod Stao.ley Kroll. "Stochastic RSI and Dynaroic
Momentum lndex: Technical Analys1s o/ Stocks <l Commodities.
Vol. 11 , No. 5 (May 1993). Copyright l 993 Technical Analysis, lnc.
Used with permission.
Kaufman, Perry J. The New Commodiry Tradmg Systems and Meth
ods. New York: Jobo Wiley & So ns, !ne., 1987.
Knig.ht, Sheldon, "Tips, Tricks, and Tactlcs for Developing Trading
Systcms." Futures, Jan 1993 (Vol. X.JI, #1), pp. 38-40.
KniglH, Shcldoo. "Trading System Redux." Futures, Feb 1993 (Vol.
Xll, #2), pp. 34-36.
Kroll, Stanley. Kroll on Fwures Trading Strar~. Homewood, lL; Dow
Joncs-lrwin, l 988.
Kroll, Stanley and Micbael J. Pauleooff. The Business One Cuide to
The Futures Markets. Homewood, lL; Business Onc lrwin, 1993.
Le Bcau, Charles aod David W. Lucas. Techmcal Traders Guide LO
Computer Analysis o/the Furures Markns. Homcwood, LL: Business
One lrwin, 1992.
Meibuhr, Stuan. "OEX and the Thrust Oscillator." Technica/ Anafysis
o/ Stocks & Commodities. Marcb, 1993, p. 58.
Morris, Gregory. Candlepower. Cbicago: Probus Publishlng, 1993.
Murphy, John J. Technical Ana/ysis oflhe Fwures Markets. New York:
New York lnstitute of Finance, 1986.
Nison, Steve. Japonese Candlestick Forecastmg Tec/111iques. New York:
Ncw York lnstitute of Finance, 1991 .
Pardo, Roben. Design, Testing, and Opli1111zatio11 ofTrading Systems.
New York: Jobo Wiley & Sons, Inc., 1992.
Schwngcr, Jack D. A Complete Guide to thl' Futures Markets. New
York: John Wiley & Sons, lnc., 1984.
Snedccor, Gcorgc W.. aod William G. Cochran. S1at1s11ca/ Me1hods
81h Ed. Ames, LA: lowa State University Prcss, 1989.
S"'ecney. John. "Where to Put Your Stops," 1'echmcal Ana/ysis of
Stocks & Commodmes. Deccmbcr, 1992 (Vol 10. 1113), p. 30.
Rotclla. Roben P. The Elements o/Succesifu/ Tradmg. New York: New
York l nstitutc of Finaoce, 1992.
Vince, Ralpb.. The Mathemarics of Money Management. New York;
John Wiley & Soos, lnc., 1992.

197

Bblogrphy

Vinoe, Ralph. Portfofio Management Formulas. New York: John Wiley

& Soos, lnc., 1990.
d.
.. fi
White, Adam. "Tuning ioto TrendinC$$ with tbe VHF In 1ca1or, U
rures Aug. 1991, p. 20-23.
.
,
J
Wells
New
Conceprs
111 Technical Tradmg ysrems.
er
r.,
.
.
Wild
Grec:nsboro, NC: Treod Research, 1978.

lndex

Advancing lssues (Al), 143-45

Advaocing Volume (AV), 143-45
A mgeo (AMGN), 85-87, 130-31
Analysis of Variance (ANOVA)
Table, 45
Antitreod Trading
and dynamic momentum
indcx (DMI), 134, 137, 140
in futurC$ markets, 190
and linear rcgrC$sion analysis,

25, 28
and StochRSI measurcs, 124
and stock rounioo, 193
and VTDYA bands, 54, 68
Arms, Richard, 144
Arnu Jodex, ix, 144
See a/so TRIN
Average Dircctional lndcx
(ADX)
and Chande momcntum
oscillator (CMO), 108- 10
and dynamic momentum
index (DMI), 140

and futures markets, 190

and linear regression analysis,
29-34
Average True Range (ATR), 168

B
Balsara, Nauzer J., 163, 178
Blau, William, 108
British Pound Futures, 164-65,
166

e
Cand.lestick Analysis, 73-75
and quanfying candlcstick
shadows, 82- 84, 90
See a/so Quantitative
candlestick
Cbandc Momentum Oscillator
(CMO), 11
fea1ures of, 108- 14, 118
199

200

lnde

CMO (Contmued)
and lhrust oscillator, 151-58
filtcnng tradc noise from,
See a/so Stock market
115-18, 141
Oynamtc Momcotum lndcx
and rclat1ve strcnglh index
(DMI). 12
(RSI), 94-95, 141, 180
and Dow Joncs 1ndustrial
and S&P-SOO index, 97-106
Average, 136-37
and stock rotation, 192-94
and S&P-500 index, 137-39
and treasury bonds, 95-97,
trading strategies with, 14()....41
109- 18
and volaLi li ty, 134-36, 141
nnd trendiness, 104-7, 140
and VIDYA, 58, 60-62, 68-70,
179-89
E
a nd wheDL fu lures, IOS-9
Clinton , Bill, 65
Engineers, 162
Cotree Futures, 25-28, 79-81
Commodity Channel lndex
(CCI)
F
limitations of, 93-94
and similarities among
Failurc Rates of lndicators, 1- 3
indicators, 3, 5- 7, 16-17
Fibonacci RctraccmentS, 176
Commodity Systems, lnc. (CSJ),
Flexible Paramcter Trading
7, 9, 25-27, 181-83, 187
Models, 137
Compctition, 162
Forecast Oscillator, 4D-42
Contingcncy Planning, 13-14,
39-40
Copper Futures, 176-77
G
Can on Fu lures, 74
General Elcctric (GE), 87- 89,
107
o
Gold f'uturcs, 20-23, 174-76
Dcclining lssucs (DI), 143-45
Dcchning Volume (DV), 143-45
Deutscbc Mark ContraCtS, 81 ,
83-84, 90-91
Dircctional Mocmcnt System,
3
Dow Janes Industrial Average
and dynamic momentum
indcx (DMl), 136-37

lntcl Corporation (lNTC), 29-34

lntraday Momcntum lndex
(! MI ). 80-81 , 91-92
Intcrday ll'3ding
See Stoch:mic RS! O:s<:illator
and hourly data trading

201

lndex

J
Japancse Yen Furures, 168-69,
174-75

L
Lcverage, 16 1, 178
Linear Regression Analysis, ix
and colfee futures, 25-28
and oorrelation amoog
indicators, 7-9
and dynamic mornentum
index (DMI), 140
and lntel Corporatioo stock,
29-34
and rnarlcet rotatioo, 179, 190,
193-94
and price fo=m, 1O, 20,
34-42
and trcndiness, 106
tutorial, 43-48
use of, 19- 25, 43-48
and VlDYA, 60-63, 70-72
wcaknesses in, 34

Market Rotation, 179, 189-94

Market Thrust, ix
compare<! to lhrust oscillator
and TRJN, 149-50
dcfincd, 145-49
aod stock marl<ct advance
decline data. 12
Maximum Adversc Excursion
(MAE)
and risk control, 163-64

and settiog stopS, 183-85, 187

Muimum Favorable Ex=ion
(MFE)
and losmg tradcs, x, 13
and nsk control, 16')5
and sctting stops, 185, 187,
193
Momentum Oscillators, ix
defined, 93
limitations of, 93-94
and rotatioo tables, 193
and VJDYA, 49
See a/so Chande roomentum
oscillator (CMO)
Moving Average ConvergenceDivcrgence (MACO)
and similarities among
indicators, S-6, 9, 17
and VIDYA, 54

N
New Concepts in Technical
Trading Sysiems (Wilder),
97
Noise Filters, 115-18, 13 1, 141

p
Philip Morris Stock, 4-9, 89
Plus Direeonal Movement
(DX+),5, 16
Price Oscillator, 5. 7-8, 17
Price Paneros, 17-18
a,nd limitations of mornentum
oscillators, 93
and nsk control plans, 174
Price Targets, x

lndex

202

Prioc Targcis (Conunued)

and linear regression analysis,
20
and risk conlrOI plans, 172-76
and stock market ti ming, 193

Q
Quantitative Candlestick
(Qstick)
calculation o, 90-91
dcfined. ix, 11
and momentum, 75-77, 798 1, 90
a nd 1987 cras h, 77-79, 87
and stock prices, 85-89, 193
a nd trading strategics, 76-77,
84
tutora!, 90-92

R
Ra ndom Prices, Generating of
tutoria l, 17- 18
Rc lati ve SlrCngth Lndcx (RSl)
nnd Amgen stock priccs,
130-31
and Chande momcntum
oscillitor {CMO), 94-95.
14 1, 180
and correla on among
indicators, 9
dcfincd, 119-23
a nd Dow Jones Industrial
Average, 136-37
a nd d)namic mo mc ntum
index {DMl), 134-36, 141
a nd intraday momcntum
i.ndex (!MI), 80-81

limitations of. 11-12, 93-94

and S&P-500 indcx, 97-104,
127-30, 137-38
and similarities among
indicators, 3-6, 15-16
static nature of. 49
and stochastic osci llator, 12433, 141, 193
and thrust osci llator, 158
and treasury bonds, 125-27,
131-33
tutorial, 119-23
and VlOYA, 58
and wheat fu turcs, 108-9
Resistance
and can.dlcstick nnalysis,
83-84
and price cbangcs, 2-3
and VTDYA bands. 54, 63
Risk Control
and anticipating pnccs, 3,
172- 76
and correlated markcts, 178
and esti mates o n new
pos itions, 163-69
and \everagc, 161, 178
and li near regrcssion analysis,
35
and mental stress. 161-63
and trade tcrnplates, 169- 72,
189
and VIDYA, 58, 189

s
S&P-500 lndex
and Chande momentum
oscillator (CMO), 104-6

203

lndelC

and dynam1c momcntum

index (DMI), 137-39
and relative strength mdex
(RSO, 97- 103, 119-21
and StocbRSI, 127-30
and thrus t oscillator, 156
Stocbastic Oscillator
and correlation amon.g
indicators, 9
limitations of, 93-94
and relativc strength index
(RSO, 124-33, 141, 193
and sinularitics among
indicators, 3-6, 15-16
Stochastic RSI Oscillator
(Stoch RSI), 124-33
and bourly data (i ntraday)
trading, 131-33
and RSI and thc T -bond
markct, 125
and S&P-500 index, 127-30
and uading strategjes, 1~33
Stock Marlcct
analyzing advanccs a nd
declines in, 143-45
and market tbrust, 12
and l 987 crash, 77-79, 87
and Qstick analysis, 85-89
and stock rotation, 190,
192- 93
See a/so Dow Jones Industrial
Average
Stops, Sctting of
and candlcstick analysis, 84
and linear regrcssion analysis,
25, 35, 40
and maxi mum adverse
excursion (M AE), 164, 185
and maximu m favorable
excursioo (MFE). 165, 185,
187

and pnoc iargcts, 173

and s1ock rotauon, 193
and traihng s1ops for open
tradcs, 165-71 , 176-77, 189,
193-94
and VIOYA, 54, 55, 58, 62,
63, 68, 167, l8l, 183
Stress, 161-63
Suppon
and candlestick analysis,
83-84
and pncc changes, 2-3
and VIOYA bands, 54. 63
Swcency,John, 163
Swiss Franc Futures, 183-89

T
Thrust Oscillaior, 12
compa.red to market lhrust
and T RJN, 149-50, 159
dcfined, 145-49
and Dow Jones Industrial
Average, 150-58
and S&P-500 index, 156
trading stratcgies with, 15658, 193
Trade Template, 169-72, 189,
194
Trading Bands
and candlcstick analysis, 74
and VlDYA, 54-55, 58-59, 63,
67-68
Trcasury Bonds
and candlestick analysis,
82-83
and Chandc momentum
oscillator (CMO), 95-97,
109-18

204

Treasury Bonds (Conti11ued)

forecasts for, 34-42
and relative strength index
(RSI), 125-27, 131-33
se1ting s1opS for, 177
and VlDYA, 55-62, 63-72
Treasury Notes, 170-71
TRIN, 12
and Dow Jones lndustnal
Average, 151, 153-56
and thrus1 oscillator, 144-47,
149-50, 159
True Strength Index (TSI).
108-9
Tu1orials
on linear regression analysis,
43-48
on Qsticlc, 90-92
on Random prices, gcnerating
of. 17-1 8
on RSI , defining, 119-123
on spreadshee1s for VlDYA,
63-72

V
Variable lndcx Oynamic Average
(VTDYA)
and advan1ages over s tati c
indica1ors, 49
calculation of, 63-72
and Chandc momcntum
oscillator (CMO), 11- 12, 58,
60-63, 68-70, 179-89
defined, ix, l I
dynamic range of, 52-54, 63

lndex

and linear regression analysis,

60-63, 70-72
and 1rading strategies, 54-55,
58, 63. 66, 68
and trailing stops, 167
and treasury bond markel
analysis, 55-62, 63-72
tutorial, 63-72
and volality, 51-52, 54, 55,
58, 63-65, 67-68
Variable Lengtb Moving
Average, 10-11
See a/sQ VlDYA
Vertical Horizonial Filler
(V}IF), 29-34, 104-7
VlDYA
See Variable lndex Oynamic
Average
Vincc, Ralph, 178
Volatility lndex
aod asscssing price action, 5152, 63-64
aod dynantic mnmentum
indcx (DMT), 135, 141

w
Wheat Futures, 108- 9
Whipsaw Trades, 130, 169, 187
White, Adam, 33
Wilder. J. WeUes, Jr., 29
and calculation of relative
strengtb index (RS1), 97,
121, 122-23
William's %R
See Stochastic oscillator

11
fcontinued lrom ftOl'I flPI
irnpr()'.1! idl>nl1lic.ition

d QllClestick p;lltems.

W~h Qslick. vou can q.iaOOly bcth the intemal

momcntOOI and sl\adc1.V5, and proOJce OOJec

tive ~ IO look al rather than a p;lltem IO

ponde<

ExplOit powcrlul new incicalors loe risk cootrol

'tbu'll fine! the hottest new indicators for: crea!
re prK;e-time templates to dose out losers and
cash oul ol ast sirig winners, leaming how to
aggr~ssivcly manage op<;?n trades using the

malom~m f;ivorable excu(s1on, performng

what-l simulatlons with the proji:cted price

ranga to prepare lor any marl<et eventuality, and

usir,g volJ!ility.based slO>S to get you in or take

yooout.
In IOday's fterccly compelitive lutu.-es and linancial maii<el,. 'IOl6 success depends on masterir6
fast-~ marl<et informatioo and ~ it IO
maloe s"ifl, smart, strategic ~ and irMst
ment decisoons. Desigrled IO ~ ~'OU harde the
~d ~ and the bad. The New Technical
T1ader IS the proacli\-e, single sourre guide for
pl~ the wirri1g percentages and reapSig ~
ser trad1ng prof'its with minin'a.Jm risk.
Aboot the authors
Tl.JSHAR $. CHANDE is a scie1'llis4 ifl\~lo~ and,

since 1967, an lnnovative tectmica( anal)'St and

trading system dcwlo;>et He has. a PrD in eng
neering from the Uovfslty of ilinois, and an MBA
frorn t~ U1ivmty of PittsblJ'gh. He has pub
lished many sdentific papers in engineering P.-
nals, and hokls nine U.S. patents for creative
soluOOn> IO flexible rnanufact~ problems ~
hi8h-power lasers and optical fbess. He is a prolific contribotf18 edror <X the lr.lder's magaziie,
Technical Analysis of Stocks and Commodi
ties. He ~ il the proce.s c:J sen~ ~ !-is own
money mana~'l'Tletll ~

STANLEY ICROLL is one d the most respecte<!

ridividu;\ls in the furures industry and was promi

neotly lrotured 1n the bestselling The Money
Ma~tef'S, as one ol the nine l~igg)Qbal r:noney
mah,1ge>rs. /\ iveye<1r vetetan of te Finaoi;ial
World's qo1modty column, he i~ t~ i!llthor Q(
The Prfcssionnl Commodily Ttader aM the

classic Kroll on Ftures Trading Stracegy. His

most recent book Is The 8usiness One Guide

10 the Fuwres Markets, written w~h Michae! J.

PaJlenof( He is (u.-rently with Lippo Futures Lid.
in~Kor'@.

In The New Technical Trader, trading system wizard Tushar

Chande and 1'11qney management expert Stanley Kroll present a
l:otd new array of dynami~, price-b<ised, and risk control indicators
thal pVide ti!1'1ly, reliable answers to the dilcult analytic challenges of pattern recognition, variabl indicator length, and price
projectiOn. Developed by th authors to overcome the specific
limitatioos of existing technical indicators, they can be used in
vrtually any market-futures, commodities, stocks, indices, and
rrutual funds. Each new indcator is supported by step-by-step
tutonab and real-world trading scenarios that tllustrate the best
strategies for tts use, and show how to adapt it to y"OlJ" trading style.
Now you can develop a successful, comprehen5ive personalized

trading pldn ~ng:

New momentl.Jm oscillators-the Chande momefltum oscllato~
~todasti~ RSI OS\:ltato~ variable-length dynamic rnomentum
ifl(lei<-lh~t overcome the limitations of the relatiV ~trength inde><
Qstick, a quantitatlve candlestick lhat generales objectve num
bers instead of subjective patterns
Linear regression analysis thal quantifies trendiness, projects
price~, and prompts the development of a better game plan

lf you need to reliably project prices under different market conditiorb. estmate risk on new positions, create s~ to cut your losses,
0< plot momentum swing failures, The New Technical Trader is
your fist place to tum. lt's the frontline guide for developing objective consistency in your decisioo making and gaining a valuable
edge in today's commodities, currency, and stock markets.