Table Of Contents

Title
Copyrights and Disclaimers
Books by Jackie Ann Patterson
Introduction
Chapter 1: Moving Parts of MACD
The Very Basics of Technical Analysis
MACD Basics
Advanced MACD Signals
Chapter 2: Baseline for Reference
Why A Baseline
The Strategy Evaluation Process
The Baseline Strategy
Baseline Strategy Backtest Results
Chapter 3: Using MACD Divergences to Find Big Bottoms
MACD Divergence Strategy Under Test
MACD Divergence Long Entry Backtest Results
MACD Divergence Exit Backtest Results
Conclusions on Original MACD Divergence Backtest Results
Out-of-Sample MACD Divergence Backtest Results
How To Apply MACD Divergences
Chapter 4: Divergence Count Indicator
Overview of Divergence Count Indicator
Strategy Development Methodology
DCI Win Rates Backtest Results
DCI Exit Backtest Results
DCI Exits Without Stop Losses Backtest Results
Conclusions From DCI Backtesting
How to Apply DCI Zones to Your Trading
Chapter 5: Intra-Day Divergence Count
My Intra-Day MACD Divergence Scans
Hypothetical Performance of Intra-Day Divergence Count
How to Do Intra-Day Divergence Scanning and Counting
Chapter 6: Resources
Bonus Materials for Book Owners
Divergence-Alerts.com
MACD Divergence Detection Software
 Related Reading
Chapter 7: Selling Short MACD Negative Divergences
What is a Short Trade
Strategies Under Test
MACD Divergence Short Entry Backtest Results
MACD Divergence Short Exit Backtest Results
Chapter 8: Anticipating the Cross with MACD Buy Signals
Strategies Under Test
Basic MACD Entry Backtest Results
Conclusions About Non-Divergence MACD Entry Signals
Chapter 9: Catching the Wiggles with MACD Sell Signals
Strategy Under Test
Non-Divergence MACD Exit Backtests Results
Conclusions for Non-divergence MACD Exit
Chapter 10: The Missing Link With EMAs and MACD
MACD MA Crossover Entry Backtest Results
MACD MA Crossover Exit Backtest Results
In Summary
Acknowledgements
Beat The Crash Volume II
 Truth About MACD:
What Worked, What Didn’t
Work, and How to Avoid Mistakes
Even Experts Make
Second Edition

By Jackie Ann Patterson

Editor, BackTesting Report
Copyright and Disclaimers

All rights reserved. No portion of this work may be reproduced in any

manner whatsoever without written permission except for brief quotations
in critical reviews.

Published by:
Own Mountain Trading Company dba BackTesting Report
P.O. Box 620427
Woodside, CA 94062-0427
support@backtestingreport.com
www.ownmountain.com
www.backtestingreport.com
www.truthaboutmacd.com

Despite best efforts to produce a useful and correct report, this book may
contain errors. Much of the data relies on backtesting, which is a historical
simulation with significant differences from real live investing and
trading. No claim is being made that you will make money from this
book, and in fact you may lose money from investing. The author is not a
licensed financial planner. This book is intended for educational purposes
and is not a recommendation to buy or sell securities. You must consider
carefully what is right for you and consult professionals as needed.

TradeStation® and RadarScreen® are trademarks of TradeStation Group,

Inc. Neither TradeStation Technologies nor any of its affiliates has
reviewed, certified, endorsed, approved, disapproved or
recommended, and neither does or will review, certify, endorse,
approve, disapprove or recommend, any trading software tool that is
designed to be compatible with the TradeStation Open Platform.
HYPOTHETICAL OR SIMULATED PERFORMANCE RESULTS
HAVE CERTAIN LIMITATIONS. UNLIKE AN ACTUAL
PERFORMANCE RECORD, SIMULATED RESULTS DO NOT
REPRESENT ACTUAL TRADING. ALSO, SINCE THE TRADES
HAVE NOT BEEN EXECUTED, THE RESULTS MAY HAVE
UNDER-OR-OVER COMPENSATED FOR THE IMPACT, IF ANY, OF
CERTAIN MARKET FACTORS, SUCH AS LACK OF LIQUIDITY.
SIMULATED TRADING PROGRAMS IN GENERAL ARE ALSO
SUBJECT TO THE FACT THAT THEY ARE DESIGNED WITH THE
BENEFIT OF HINDSIGHT. NO REPRESENTATION IS BEING
MADE THAT ANY ACCOUNT WILL OR IS LIKELY TO ACHIEVE
PROFIT OR LOSSES SIMILAR TO THOSE SHOWN.

Self-discipline not included.

Books by Jackie Ann Patterson

Truth About ETF Rotation: Fund Your Retirement By Investing In Top

Exchange Traded Funds in One Hour Per Week. Beat The Crash Volume I

Truth About MACD: What Worked, What Didn’t Work, and How to
Avoid Mistakes Even Experts Make. Beat The Crash Volume II

Administering Wills and Trusts: A Layperson’s Guide for Executors and

Trustees of Mid-Size Estates. Truth About Estates Volume II
Introduction
Some people say I’m obsessed with the MACD technical indicator. True,
my involvement with it is long and deep. Yet it is just one tool of many
for objectively analyzing price charts of various markets.

I didn’t set out to be an expert in technical analysis. I’d say I’m probably a
lot like you, dear reader, in that I wanted to trade to make money with as
little effort as possible.

I wanted to quit my “day-job” – actually a demanding career in high tech –

and retire comfortably while profitably trading a couple hours a week. My
vision of “the good life” certainly didn’t include hundreds of hours of
backtesting!

What I was willing to do was take trading classes from Dr. Alexander
Elder (www.elder.com). His trading camps not only cover the intellectual
material on trading, they are an introduction to a lifestyle of freedom and
travel.

The camp I took was on the island of Cyprus. I learned not only why Dr.
Elder called MACD divergences the most powerful signal in technical
analysis, and not only how Gerald Appel, the inventor of MACD, used it
to find the bottom in 2003, but also enjoyed intriguing travel around
Cyprus including festive group dinners at Greek tavernas in the local
villages.

Adventures like that bonded me with my classmates and apparently to the

MACD indicator as well.

The thing was, when I came home to trade using the MACD divergences,
it was not as easy for me as it seemed in the classroom. You see, I made
the almost-fatal mistake of treating all MACD divergences alike.

More research showed me that some signals had a chance of helping with
trading and others not so much. Needing to recover my composure and
my income for trading, I decided to research in depth and sell that research
as written BackTesting Reports, videos, and a web service called
divergence-alerts.com that flags MACD divergences on US stocks and
ETFs. In fact, the first edition of this book was packaged as a series of
reports that sold online for more than $100.

An interesting thing happened as I shared my research: a deeper

understanding came back to me. The audiences at the MoneyShow and
Traders Expo challenged me with hard questions.

A divergence-alerts.com member, Joe Gruender Jr. of InvestorsHELP.net,

made a brilliant suggestion. He said that as long as I have an automated
MACD Divergence scanner and as long as I am scanning the US stock
market every day for divergences, keeping a count of those divergences
day-to-day may shed light on the overall stock market.

I refined and implemented Joe’s suggestion and then backtested to see

what we might expect. Once that was running, other customers requested
intra-day scans. After over four years of scanning, I think I have enough
data to make meaningful backtests and report the results in this book.

Read on for a complete explanation of the MACD. The first chapter

covers the basics and all the moving parts. The second chapter is
background information on my back-testing methodology and a baseline
for comparison. From there we dive into the more useful aspects of the
MACD: positive divergences and the Divergence Count Indicator (DCI).
The latter section of the book covers the sketchy side of MACD: short
selling, MACD crossovers, and MACD Histogram ticks.

While researching the original BackTesting Reports on MACD, I found

widespread confusion about how to calculate the MACD Histogram. Here
I lay the facts out for you early on.

By the end of this book, you’ll have a complete briefing so that, when you
see a MACD on a chart, you can tell at a glance how it was made, what it
indicates about the market, and most importantly, how much or how little
to count on it based on its track record.

In general, my research on the MACD applies to stocks, exchange traded

funds (ETFs), and futures. In Chapter 3, I also discuss my take on trading
options based on MACD divergences.

I make no effort to address the forex markets with MACD. In fact, I

believe other more trend-oriented indicators are a better fit for the
persisting trends in forex. In my opinion, MACD is more useful as a
reversal indicator.

The first book in the Beat the Crash series, Truth About ETF Rotation,
presented a completely mechanical system for investing in Exchange
Traded Funds (ETFs). This second book in the series begins the long and
arduous path of learning tools to help with discretionary trading and
investing.

If you come away with only one thing, let it be ideas for finding big
market bottoms. Beating the crash is as much about picking good times to
buy as knowing when to sell. It is also about resilience to re-enter the
market after a healthy downturn.

Of course to be in a position to enter near the bottom, you need money

management and sound exit strategies to have the funds safe on the
sidelines looking for a bottom. Those are topics for later books in the Beat
the Crash series.

Overall, this work is NOT meant to convince you to use the MACD
indicator or even to trade the stock market. Only you can make those
decisions for yourself. This book is meant to give you information to
consider while you weigh your choices and point out areas that appeared to
perform better than others in historical simulation.
Chapter 1: Moving Parts of MACD

The Very Basics of Technical Analysis

You can make objective decisions about the market with the help of a
technical indicator. Of course, you could just look at price alone, however,
price is notoriously fickle, and identifying price patterns as subjective as
picking out shapes in the clouds in the sky.

The calculation of a technical indicator necessarily abstracts information

from the price and focuses attention on key aspects of its behaviour.

For example, technical indicators may follow trends, highlight reversion to

the mean by oscillating between overbought and oversold, warn of trend
reversals, or report on the broad market.

The MACD is a complex indicator that can do all of the above. This
chapter explains the different parts of the MACD and shows you how it is
typically interpreted. Later chapters report on the performance of each
aspect of the MACD.
Before getting complex, let’s cover the basics of technical analysis. The
first thing to understand is that the price of a stock, ETF, or futures
contract is commonly expressed as charts of bars. Each bar represents a
time interval such as days, weeks, months, hours, 5 minute intervals. Most
of my research focuses on charts of daily bars, plus key elements of
weekly and hourly charts.

Each bar on a chart has four prices associated with it:

Open – the first trade of the interval

High – the highest price traded during the interval
Low – the lowest price traded during the interval
Close – the last trade of the interval

I often plot price charts in candlestick format, with the rectangular candle
body drawn between open and close prices. Wick lines extend to the high
and low prices outside the candle body. In color-coded charts, green
candles represent up days where the market closed at a higher price than it
opened, and red candles represent down days. In black and white or
grayscale, down days are darker. See Figure 1 below for an example
plotted with the TradeStation® charting software.

Larger versions of this figure as well as all tables and figures in this book
are available online at truthaboutmacd.com.

Many patterns of candlesticks are studied by traders which are beyond the
scope of this book. Click here to learn more about candlesticks.

In order to analyse price despite is often abrupt movements, it can be

useful to average the price over many bars. A simple technical indicator
called Moving Average (MA) does just that.

To construct a moving average, take each price bar, look back a fixed
number of bars, say 12, calculate the average of all of the past 12 bars, and
plot that average by the price bar.

Two methods of calculating the average are popular. The Simple Moving
Average (SMA) equally weights each bar and takes the arithmetic
average. Most sports statistics are calculated this way: earned run average,
rushing yards per game, etc.
The other popular method is an Exponential Moving Average (EMA)
which weights the more recent bars more heavily than the past bars. This
has the obvious advantage of being more sensitive to recent price, and the
obsolete advantage of being easier to maintain an EMA on a chart by hand.

Now computerized charting packages calculate EMAs, SMAs, and about

every technical indicator you can imagine. See Figure 1 for a candlestick
price chart with 12-day and 26-day EMAs plotted.

Figure 1–Price as candlesticks with 12-day and 26-day Exponential

Moving Averages (EMAs). Created with TradeStation. ©
TradeStation Technologies, Inc. All rights reserved. (Click here for
larger figure.)

One more piece of background information: one of the earliest and most
popular technical signals was the crossover of two moving averages. For
example, when the 200-day simple moving average crosses up through
(i.e. surpasses from below) the 50-day simple moving average it is called
a Golden Cross, which is rumoured to be a buy signal for institutional
investors. Similarly, when the 200-day plunges below the 50-day SMA it
is called the Death Cross.

MACD Basics
This section covers the MACD in two ways. First it describes the basic
signals of the MACD roughly following the story of their evolution. Then
it walks through those same basic signals in order of their appearance in
example stock price movements.

The first thing to know about MACD is what the acronym stands for:
Moving Average Convergence Divergence. That’s too much of a
mouthful for polite company so let’s just use MACD. All the letters in its
name are often spoken – “M-A-C-D”, although “Mac-D” is easier.

MACD is the brainchild of Mr. Gerald Appel. He started writing his

investment newsletter, Systems and Forecasts in 1973, and began
managing money for individual clients shortly thereafter. His company,
Signalert Asset Management (www.signalert.com), is now led by his son
Marvin, and remains dedicated to using the tools of technical analysis to
help clients and subscribers grow their wealth while managing risk.

The Intention of the MACD

The basic idea to keep in mind about the MACD is that it started as a way
to anticipate a Moving Average crossover and jump ahead of other traders
waiting for the MA crossover signal. MACD anticipates MA crossovers
by tracking how the two exponential moving averages come together
(converge) and drift apart (diverge).

Moving Averages Form Backbone of MACD

The MACD Line tracks the relative position of two moving averages by
measuring the difference between them. See Figure 2 below.

As the EMAs move together the MACD line converges or moves towards
zero. MACD approaches zero from the above when anticipating a sell
signal and comes up from below for a buy signal.
Figure 2- MACD with dotted lines added to mark EMA crossovers
and MACD zero crossings. Created with TradeStation. ©
TradeStation Technologies, Inc. All rights reserved. (Click here for
larger figure.)

MACD Line as Trend-Follower

The MACD Line crosses zero at the exact same point as the MA
crossover, as illustrated in Figure 2 above. When the MACD Line is
above zero, the MAs have crossed over into the bull market configuration
of the fast MA above the slower MA. When the MACD Line is below
zero, the MAs are in the opposite bear market configuration. In this way,
MACD functions as a trend-following indicator.

MACD as Oscillator

The “D” in MACD stands for divergence, but it is not the divergence you
seek. As the MACD Line moves away from zero, it shows that the
underlying moving averages are moving further apart. That is all. The
more important MACD Divergence is described in the Advanced MACD
Signals section below.

Some traders use the MACD as an overbought/oversold oscillator by

measuring the distance of the MACD Line above or below zero. I have no
experience with that technique.
Earlier MACD Signals

If the MA crossover was all we wanted, we could just stick with the two
MAs and call it good. In fact, some traders do exactly that. But,
remembering the intent is to anticipate the crossover, MACD gives
definite signals in advance of the crossover. In addition to the zero
crossing, three more ways evolved to interpret MACDs, each giving
progressively earlier signals.

The MACD Signal Line is an exponential moving average of the MACD

Line itself. It smoothes (and slows) the MACD Line in the same way that
an exponential moving average smoothes price. See Figure 3 below.

Figure 3- MACD Line and its Signal Line. Created with TradeStation.
© TradeStation Technologies, Inc. All rights reserved. (Click here for
larger figure)

When the main MACD Line crosses up through the MACD Signal line, it
was originally said to indicate a buying opportunity. To be meaningful at
all, the MACD must be below zero (indicating an oversold market) at the
time of the crossover. When the MACD crosses down through its signal
line, it was originally considered a sell signal, especially when it happened
above the zero line which indicates an overbought market.

The crossing of the signal line gives advance notice of a potential moving
average crossover as illustrated in Figure 4.
Figure 4 - Relationship between MACD Lines and MAs. Black line is
12-day EMA, brown is 26-day EMA, blue is MACD, and purple Signal
line. Created with TradeStation. © TradeStation Technologies, Inc.
All rights reserved. (Click here for larger figure.)

To catch an even earlier signal, be there when the MACD line turns. With
the naked eye, it’s easier to see the turning point drawn as a histogram.

Figure 5 shows a histogram plot highlighting the uptick before the MACD
Lines cross.
Figure 5 – Appel’s MACD Histogram. Arrow highlights a buy signal
as the histogram ticks up from below zero. Created with
TradeStation. © TradeStation Technologies, Inc. All rights reserved.
(Click here for larger figure)

With a histogram you simply compare the height of the current bar to the
previous bar. If the current bar is below zero but dropping less low than
the previous one, it can be considered a buy signal. If the current bar is
above zero and shorter than the previous, it can be considered a sell
signal.

Histogram, Smistogram,What’s the Difference?

Here’s where things get a little crazy: Two different MACD histograms
float around in the world.

Gerald Appel thinks it started with the original monograph on MACD he

wrote decades ago. He says, “Everybody who has written about MACD, I
think, has taken what they wrote from that monograph, which I [now]
think was inaccurate.”

Mr. Appel draws a histogram under the MACD line representing the
difference between the two moving averages. From here forward, let’s
call this Appel’s histogram. Most texts and some prominent web sites
describe the MACD Histogram in this manner.

However, most chart packages and texts actually calculate the MACD
Histogram another way. They plot the histogram as the difference
between the MACD Line and its Signal Line. Let’s continue to refer to
this as the MACD Histogram, a.k.a. MACDH.

See Figure 6 below for a graphic comparison of the two histograms.

Figure 6 - Two types of MACD Histogram. Appel's Histogram is on

the bottom. The commonly drawn MACDH is in the middle of the
screen. Created with TradeStation. © TradeStation Technologies,
Inc. All rights reserved. (Click here for larger figure.)

You can tell which kind of histogram you have on any chart by looking at
how the zero crossings of the histogram relate to the zero crossings of the
MACD lines. If the bars in the histogram cross zero at the same time as
the lines cross zero, it is Appel’s histogram. If, however, the bars cross
zero when the MACD lines cross each other, then it is the MACDH.

The MACDH uptick or downtick is the earliest signal. It is the first to

detect when price is moving in the direction that could cause a moving
average crossover. The MACDH is the most sensitive signal and that
means it may also be prone to having the most false signals.

When you see a histogram on a chart, take a quick moment to check what
type, so you know if you’re getting the early signal or the later one. You
can use the results from this report to see the performance track record of
both types of histogram.

With either histogram, remember the main point is to see when the bars
change direction. In fact, a change in direction of the MACD Histogram is
the first MACD signal to emerge in a trend change.

Just because the MACD Histogram changes direction, it doesn’t mean that
the price trend will reverse. There is the rub. In general, the more
sensitive or fast a signal is, the more false or fruitless signals it gives.

MACD Settings

The MACD can be calculated on any two moving averages. The standard
values are 12-day EMA, 26-EMA, and 9-Day EMA of MACD to create
the Signal Line. The backtesting in this book was done with these
standard settings, plus a few selected alternates.

A trader could use smaller EMA values in the MACD to get quicker
signals and more of them. Use longer EMAs to slow things down.

A notable example of different MACD settings is given in Gerald Appel’s

book Technical Analysis: Power Tools for Active Investors, (FT Press),
2005. Mr. Appel describes a method of using the slope of a moving
average to choose the parameter settings for the MACD. That is also
tested in this book and compared to the basic settings. The “Power Tools”
book reveals more nuances for using the MACD as well as other market
timing tools that are not covered in this report.

MACD Order of Events

To recap, in the idealized case where the market moves from low to high
and back, the MACD signals appear in the order shown in Figure 7 below.

To see Figure 7 larger and animated as I talk through each signal of the
MACD, check out the bonus video for book owners by registering at
truthaboutmacd.com.

Keep in mind that the MACD is not predictive. It is a lagging indicator

due to the nature of its construction. Just because a MACD signal happens
doesn’t mean that price will follow. In later sections of this book, you can
get an idea of how often the MACD signals have worked out historically.
And as always, past performance is no guarantee of future results.

Figure 7 - MACD with various signals annotated. Created with

TradeStation. © TradeStation Technologies, Inc. All rights reserved.
(click here for larger figure)

The potential signals are:

1. Sometimes before the price even moves up, the MACD Histogram
ticks up. By ticks up I mean today’s bar is less negative than
yesterday’s bar. This says that the MACD Line and its Signal Line
moved closer together.
2. Appel’s histogram ticks up. This says that MACDL moved moving
up, which means that the EMAs on the chart moved closer together.
Not that EMAs themselves are moving up, but just that they are
closer together now than last bar.
3. The MACD Line crosses its Signal Line. This isn’t really tied to
anything on the chart – it’s just a way of slowing down signals from
Appel’s histogram. The other thing that happens at this same time is
MACD Histogram crosses zero. That’s only tied to an abstract
mathematical construct as well – a second cousin to price if you will.
 Where price is the thing, the EMAs are first cousins, the MACD
Lines second cousins, and the MACD Histogram third cousins.
4. This signal is the missing link – it’s the MACD Lines crossing zero
which correspond to the original EMAs crossing each other. At this
time, Appel’s histogram crosses zero too.

In this example, which is why I chose it, the price keeps rising as the
MACD signals progress and then it starts down. We see the same order
of signals on the sell side:

5. MACD Histogram downtick.

6. Appel’s Histogram downtick.
7. MACD Lines cross down and MACD Histogram drops below zero.
8. MACD Lines drop below zero and the EMA’s cross down.

That is the classic example. Remember it can be interrupted at any time.

This case was hand-picked to show the whole thing in one screen.

Advanced MACD Signals

No discussion of MACD would be complete without mentioning
divergences. The MACD divergence is a reversal signal whose job is to
detect potential changes in the trend.

These important signals occur when price hits a new extreme while the
indicator is actually showing a less powerful move. See Figure 8 for a
double positive divergence on both the MACD Lines and MACD
Histogram. In a divergence, price makes a lower low, but at each low, the
MACD reading is successively higher.

Backtesting any indicator divergence is challenging and comes with its

own set of parameters to tweak. We explore the performance of MACD
divergences in Chapter 3.
Figure 8 - MACD Lines positive divergence and MACD Histogram
positive divergence. Created with TradeStation. © TradeStation
Technologies, Inc. All rights reserved. (Click here for larger figure.)

Negative divergences look like a mirror image of positive divergences.

When the price makes a new high but the MACD does not confirm with a
new high of its own, that is a negative divergence.

Fuzzy Divergences

Some traders, particularly option traders, define MACD divergences a

little differently. They look for situations where the price approached a
double bottom (or double top), but didn’t quite make that new low (or new
high), with a MACD line that is at a higher (lower) level than the previous
bottom (top). I call these Fuzzy Divergences.
Figure 9 - FUZZY MACD Lines Divergence. Created with
TradeStation. © TradeStation Technologies, Inc. All rights reserved.
(Click here for larger figure.)

Figure 9 is an example of a fuzzy divergence. The price of JCP made a

low in November of 2013 and so did the MACD lines. By December the
price drifted upwards, tilted down to make the bridge of a W, but resumed
its upward movement before ever making a lower low. The MACD lines
confirmed the move by making their own up move. My automated
divergence detection scanner marked the fuzzy divergence with the blue
up arrow and the words MACDL_Div_Buy.

The fuzzy divergences are not within a strict definition of a divergence

because price and indicator are both moving in the same direction.
However, one could argue that the fuzzy divergence is an even stronger
indication of a major price move underway because the price doesn’t reach
an extreme point as far in the opposite direction on the second low or
high. The Out-of-Sample Backtesting section of Chapter 3 gives
performance data on fuzzy divergences.

Divergence Detection Rules

Spotting divergences by eye is actually quite easy. Developing a computer

program to find divergences automatically takes skill and a precise
definition of what to look for. Even when the scanner points out a
divergence, it is necessary to manually confirm its validity and understand
the context. It is useful to detail the divergence detection rules to aid
individual traders.

Three components to algorithmically find a divergence are:

Market context
Divergent indicator set-up, including parameter settings
Trigger event

Market Context for Divergences

The appropriate context for a divergence is a recent price extreme.

Divergences are reversal signals, which can spell the end of the previous
trend and the start of a new trend in the opposite direction.

To make that objective and measurable, look for MACD lines less than
zero to set the stage for a positive divergence or MACD lines greater than
zero to kick off a negative divergence.

Divergent Indicator Set-Up

The basic indicator set-up for a positive divergence is a double-bottom in

price while the MACD indicator makes a higher low.

The basic indicator set-up for a negative divergence is a double-top in

price while the MACD indicator makes a lower high.

A key question is how far apart must the extremes of price be to count as a
divergence? Based on historical backtesting, more compact divergences
with extremes no more than 50 bars apart performed better than looser
divergences with extremes up to 100 bars apart.

To make a double-bottom requires a bounce between the extremes. I

require at least 6 bars separation where the price pulls away from the
extremes. This avoids picking up situations where the price simply falls
day after day. We are looking for a rise in price between the two lows of a
positive divergence.

Dr. Alexander Elder, author of Trading for a Living, also requires that
MACD Histogram divergences “break the back of the bear” (or bull).
What this means in the case of a positive divergence is that the MACD
Histogram starts out below zero at the first price low, and in the bounce
makes it up above zero before price hits the second price low. See Figure
10 for an example of a situation where the MACD Histogram does NOT
break the back of the bull.

Figure 10 - MACD Histogram NOT breaking the back of the bear.

Created with TradeStation. © TradeStation Technologies, Inc. All
rights reserved. (Click here for larger figure.)

In the case of a MACD Histogram negative divergence, to break the back

of the bull, the MACD Histogram starts out above zero at the first price
high, and in the pullback makes it below zero before price hits the second
price high.
It is not necessary for MACD Lines to bounce across zero according to the
rules outlined by Gerald Appel, the inventor of the MACD. Having the
MACD Lines below zero for the first price low, and at some higher point
for the second price low is enough to call it a positive divergence.
Likewise as long as the MACD Lines start above zero and head down
while price hits new highs, it qualifies as a negative divergence.

The back-breaking behavior screens out lessor price moves and reduces
the number of divergences. For MACD Histogram this is necessary to rule
out divergences on lessor price moves. MACD Lines very rarely break the
back across zero and considering the slope of the lines serves to identify a
reasonable number of divergences.

The parameters in the MACD calculation can affect how soon a

divergence appears. For most of the backtesting, I used the default settings
of 12, 26, 9.

Trigger Event

The final necessary ingredient for a divergence is a trigger event –

basically the price moving in the direction of the new trend.

A situation may have all the other parts to lay the groundwork for a
divergence:

MACD lines below zero at first price low

Higher MACD lines which are moving up on the second price low
Appropriate spacing between the lows – not too far, not too close

But only if the price actually closes up for at least one bar do you have a
divergence.

Anticipating Divergences

As you become familiar with divergences you can spot many charts that
could produce a divergence if price makes a certain move. Be very
careful! The divergence might never materialize.

When the situation is ripe for a divergence, it is possible to calculate the

price that would bring a divergence on the next bar. One possibility is to
set a stop entry that would only execute if the price passes that threshold to
make a divergence.
Chapter 2: Baseline for Reference
Why a Baseline
To understand the performance of a technical indicator such as MACD is it
necessary to first understand how the market behaved during the time
period of the analysis.

Performance is relative. For example, it may sound good that a trading

strategy won 60% of the time and produced 10% returns – until you realize
that during that same time, almost all stocks soared more than 20%.
Likewise a strategy that only returned 2% doesn’t sound very good –
except when the broad market takes a dive.

In order to compare technical indicators, I created a baseline trading

strategy and assessed its performance over three historical time periods.
Then I used the baseline to evaluate the MACD trading strategies.

This book presents those performance comparisons, as well as more

recent, out-of-sample testing. This chapter gives you a rundown of the
baseline for comparison.

If you’d like more detail on this baseline of the stock market presented in
this chapter, you can download a free report from
www.backtestingreport.com/BackTestingReportBaseline.pdf

The Strategy Evaluation Process

BackTesting Overview

Backtesting measures the relative performance of a set of trading strategies

on historical price data. Since backtesting relies on past data, it makes no
guarantees about future performance and can’t say whether a strategy will
do as well in the future as it did in the past. However, if a strategy didn’t
perform in the past, there’s no reason to believe it will suddenly turn into a
winner. It can pay to avoid strategies with losing track records.

Although most traders agree that backtesting is useful, many people don’t
do it because of the time, expense, and expertise required. This book can
give you a leg up on the markets without doing all the work yourself.

How to Compare Entry Strategies

Picking entry strategies, which have win rates above the baseline, is a good
start. Also use win rate as a hint on the ease of following a system. For
example, consider if you can really stick to a trading plan that only wins
10% of the time. Most importantly, remember that win rate alone doesn’t
determine if a strategy is profitable – the size of the wins and losses
matters, too.

How to Compare Exit Strategies

Most complete strategies have two types of exit tactics: taking profits and
stopping losses. To keep the exploration to a manageable size, first we
compare profit-taking without stops, with the same criteria to sell the stock
as to buy the stock. This also gives a fantastic what-if scenario to see what
can happen if you run without limiting risk. Next we add stops losses and
compare them both to the stop-less run and to each other. Keep the
following criteria in mind while comparing different exit strategies.
Expectancy versus Percentage Gain

Most investors and many traders are used to measuring their performance
in terms of percentage gain. Compound annual growth rate is a very
meaningful metric when applied to a whole account or portfolio of
investments.

When evaluating a trading strategy, however, it makes sense to consider

individual trades, and as many trades as possible. The number of trades
evaluated can easily outrun the funds in all but the largest accounts.
Making a fictional account that takes all trades and measuring its CAGR is
not very realistic when it might rival the GDP of a small country.

A more useful calculation is the mathematical expectancy of all trades. It

can give you an idea of how a large number of trades might average out
(even though any one trade could do anything).

The mathematical expectancy is calculated as the percent wins times the

average win size minus the percent losses times the average loss size.

Expectancy = (win_rate * avg_win) –

(loss_rate * avg_loss)

In this way, expectancy takes into consideration the probability of

winning, the average size of the wins, as well as the probability of loss and
the average size of loss. This lets us use one number to compare a strategy
that wins often but small with a strategy that wins big but only on rare
occasion.

In general, potentially profitable strategies have a positive expectancy.

Losing strategies have negative expectancies which are color-coded red in
the data tables throughout this book,. Yellow indicates that the expectancy
was slightly positive, but rounded down to zero – indicating caution.

The Baseline Strategy

The baseline strategy is about as simple-minded as it gets. It takes a trade
any time it can, one trade at a time, and holds that position for a pre-
determined number of days. There is no decision-making, no strategy,
only continuous buying.

The number of days to hold reflects different types of traders as shown in

Table 1 below:

(Click here for larger table.)

An example of the TradeStation backtesting engine executing the baseline

strategy for Active Traders is shown in Figure 11 below.

Figure 11 - Long baseline strategy for active investors. Created with

TradeStation. © TradeStation Technologies, Inc. All rights reserved.
(click here for larger figure)

The solid aqua lines highlight simulated profitable trades and the hot pink
lines simulate losses. As Figure 11 demonstrates, according to the
baseline strategy, if the market goes up during the trade, a profit is
counted. If not, it is a loss. Either way, once a trade is over, a new one is
entered without any decision-making.

The purchase price of the trade is assumed to be the opening price the day
the trade begins. The sales price is assumed as the closing price on the day
the trade ends. Slippage – the possibility that one’s own order affects the
market price – is neglected.

That is the simple baseline strategy in a nutshell. Of course no one would

actually trade this way. It simply serves as a point of reference.

Historical Time Periods Tested

Over a decade of data went into the creation of the baseline. 14 years to be
exact, spanning May 1994 - April 2008. In an effort to prevent over-
optimization, the 14-year period was divided into three samples:

May 1994 – April 2004 denoted by darker blue

This ten-year period was chosen to include major up, down, and
sideways movements of the stock market.

May 2004 – April 2007 denoted by medium blue

Out-of-sample data for the original period. At 3 years, it’s 1/3 as

long as original. This period turned out to showcase a strong stock
market performance.

May 2007 – April 2008 denoted by light blue

This full year of data preceded the worst of the credit crisis market
meltdown. However, it was still a down year.

More recent backtest results for the MACD Divergence strategies are also
included in this book, without reference to a baseline.

Stock Lists

The list of 7147 stocks tested for the baseline and subsequent strategy
backtesting consisted of all US stocks and American Depository Receipts
(ADRs) of foreign corporations trading during the time periods above.

This includes delisted tickers, i.e. companies that quit trading on the major
exchanges. A company may be delisted for any of several reasons such as:

Bankruptcy
Stock price below the exchange limits
Merger
Going private

Don’t assume that all delisted companies depress the performance data.
Mergers and leveraged buy-outs have been known to run up the price
before the ticker is delisted.

The historical data used for the baseline and backtesting was purchased
from CSI (www.csidata.com) and read by the TradeStation backtesting
engine. Dividends were ignored.

Commissions were calculated at $0.01 per share each way. For the
baseline, the trade size was always 1000 shares. So the commission was
$20 round trip.

Baseline Strategy Backtest Results

The results in this chapter set the baseline. To be considered useful, an
entry strategy needs to show fewer trades and higher win rate with a
comparable average holding time. Potentially profitable strategies have a
positive expectancy. Losing strategies have their negative expectancies
colored red. Smaller standard deviations are generally better.

Active Investor Baseline Performance

An active investor manages his/her positions and, in my opinion, holds

stocks for about a year. Typically, this investor trades the long side of the
market and puts effort into entry strategies to decide when to buy. Few
active investors practice selling short. A successful active investor catches
a trend, rides it even through minor rough patches, and then exits rather
than give all profits back on an opposing trend.

I modeled the Active Investor with a holding time of 200 trading days.
(Click here for larger table 2.)

Position Trader Baseline Performance

A position trader seeks intermediate-term opportunities. To be successful,

a position trader gets in on a burst of action – perhaps one leg of a longer-
running trend – and gets out before the action fades.

I modeled the Position Trader with a holding time of 20 trading days.

The results to beat for position traders are shown in Table 3 below.
(Click here for larger table 3.)

Option Trader Baseline Performance

Generally, options traders are looking to make a leveraged profit. Because

options are wasting assets whose price decreases with time, it behooves the
option trader to get in and out for a quick move.

I modeled the Option Trader with a holding time of 5 trading days.

(Click here for larger table 4.)

Swing Trader Baseline Performance

A Swing Trader capitalizes on short-term price movements. A swing

trader often holds overnight, possibly for several days, which
distinguishes swing trading from day-trading. However, a swing trader
rarely rides the full long-term trend.

Instead a swing trader is ready with a quick exit the moment the trade goes
the other way. Given the shorter time span, swing traders need entry and
exit strategies that pounce on the opportunities to get in, capture profits,
and get out all the while minimizing the inevitable losses.

I modeled the Swing Trader with a holding time of 2 trading days.

(Click here for larger table 5.)

Short-Seller Baseline Performance

Advanced traders who are pessimistic about the market may choose to sell
short. In this type of transaction, the stock broker arranges the loan of
shares of the ticker of interest. The short-seller then sells the shares on
loan with the intent of buying back the same quantity of the same ticker
once the price has decreased.
The object is still the same: buy low and sell high. However, the order is
reversed to first sell high, then attempt to buy low.

If the stock price doesn’t fall, or worse the price rises, the short-seller is
forced to take a loss. Note that the potential loss of a short sale is
theoretically infinite because the stock could continue to rise to the sky. In
practice, the broker monitors the account margin and forces the client to
end the short sale while they still have funds to cover the losses.

Table 6 shows the performance for a short-seller’s baseline. To create this

table, the same historical data for the same tickers over the same
timeframe as above was used. The only difference is that instead of
buying the stock, the baseline strategy sold short.
(Click here for larger table 6.)

As you can see by comparing the various baseline tables, the long or
regular baseline produced winning trades when the market went up while
the short baseline generally lost. When the market dropped, the short
baseline demonstrated profits while the long baseline did not. Read on to
see how the MACD strategies compared to the baseline.
Chapter 3: Using MACD Divergences to
Find Big Bottoms

This chapter focuses on the aspect of the MACD that produced the best
numbers in my backtesting – the divergences between price and indicator.
In truly obsessive fashion, the pages below take apart the MACD
divergence, trying it with lines and histogram, various settings, and a
smattering of exits.

Here you can find several robust strategies along with data-backed criteria
to help you choose the way to apply MACD that appeals most to you. Oh,
and you can take note of a few pitfalls to avoid as well.

In his Master’s Class, Gerald Appel, the inventor of the MACD, gave
several examples of waiting for a negative divergence to sell. What he
doesn’t tell you, but what is laid out in this chapter, is “Here are the
rewards you might get if you do for wait for a divergence signal.”
I’ve got to admit, early on in my own trading, I understood the concept
rationally but didn’t always wait for the bearish divergences to sell. It
always seemed scary as the price fell away from a high and as the
MACDH started ticking down – I was out of there. Even while coding up
the first negative divergence strategies to test I was shaking my head
believing they wouldn’t turn out and thinking up alternate exits to test.
Boy was I surprised!

Now with years of experience, I hold out even past the first series of
negative divergences. I wait to sell because I believe, based on data, the
negative divergences show up way too soon. A trend often shows a fair
amount of life after negative divergences. Later books in the Beat the
Crash series explore other methods to exit a trade besides MACD
Divergence.

For back-testing, however, a relatively simple and symmetric exit such as

the first negative divergence gives us a way to compare entry strategies
and market timing.

MACD Divergence Strategy Under Test

Before diving in, you need to know the moving parts of the MACD,
including the different histogram types. For more information, refer to
Chapter 1 which has comprehensive explanations.

The basic definition of a positive or bullish divergence is when price

makes a lower low but the indicator makes a higher low. Likewise a
negative or bearish divergence is said to occur when price makes a higher
high but the indicator makes a lower high.

We look at two different types of divergences in this chapter: MACD

Lines and MACD Histogram.

Indicator divergences are sometimes squirrely and subjective however, we

need an objective definition to do meaningful automated backtests. The
algorithms tested here are necessarily approximations of what a human
would pick as MACD divergences. Every effort was made to make them
close approximations and to err on the side of missing opportunities rather
than including bogus signals.

When modeling a subjective heuristic applied by human experts, a key

question is which expert to model. This report based the definition of
MACD Lines Divergences on the Technical Analysis: Power Tools for
Active Investors book by Gerald Appel. For MACD Histogram
Divergences, the teaching of Dr. Alexander Elder served as a guide.

The upshot is that MACD Histogram divergences are required to “break

the back of the bear” by crossing up through zero between the MACD
Histogram humps of the bullish divergence, whereas no such requirement
is placed on divergences of MACD Lines.

To get a clear picture of the various strategies at work and observe an

example of MACD’s ability to find important market bottoms, see Figure
12 - Figure 14 below depicting the 2003 bottom in SPY and how it could
be traded by each strategy.

The testing started with timed exits, comparable to the baseline in Chapter
2. Then I tested with symmetric exit signals of negative divergences. The
screenshots below are of the final steps in the backtesting process – bullish
divergence entry strategies with symmetric exit signals of bearish
divergences plus average true range (ATR) stops – because that gives
more realistic insight into how a person might use them.

MACD Lines Divergences

Figure 12 - MACD Lines Bullish Divergence buy signal with sell signal
of a bearish divergence of MACD Lines. Created with TradeStation.
© TradeStation Technologies, Inc. All rights reserved. (Click here for
larger figure.)

The MACD Lines (MACDL) rise in between the price bottoms of July
2002 and October 2002, forming a positive or bullish divergence as
illustrated by green arrows added to the TradeStation screenshot of Figure
13. Likewise, the MACD Line declines between price tops of June and
Sept 2003, forming a negative or bearish divergence – note the red
arrows. As it happens in this case, the price continues upward after the
strategy exits.

It is educational to plot the Appel’s Histogram as in Figure 13. The

change in height of the histogram bars gives the precise bar where the
bottoms and tops of the divergences occurred. Then the software can
monitor nearby bars for the trend reversal.

Figure 13 - Closeup of MACD Lines Bullish Divergence. Created with

TradeStation. © TradeStation Technologies, Inc. All rights
reserved. (Click here for larger figure.)

Power Tools MACD Lines Divergences

Implementing a “Power Tools” strategy for trading divergences ala Gerald
Appel, required modifications to the software algorithms. In this case, the
buy signal depends on the market trend as defined by the the 50-day
Exponential Moving Average (EMA). For buy signals, the slower 19,39,9
MACD settings – shown in the black histogram at the bottom of Figure 14
came into play when the price dropped below the 50-day EMA. The
standard 12,29,9 MACD settings – green indicator on Figure 14 — were
employed otherwise.

This strategy aims to take profits on bearish divergences of the slower

MACD Lines. Readers of Mr. Appel’s book know that the original
“Power Tools” strategy also used the 50-day EMA as a stop. However,
setting a stop at a fixed multiple of the average true range (ATR) was used
instead for consistency with other strategies in this book.

Figure 14 - BackTesting Report interpretation of Power Tools MACD

Divergence strategy. Created with TradeStation. © TradeStation
Technologies, Inc. All rights reserved. (Click here for larger figure.)

MACD Histogram Divergences

The MACD Histogram (MACDH) – defined as the difference between the

MACD Lines – moves much more quickly than the MACD Lines, and
hence provides more signals. Thus the MACD Histogram may better suit
the temperament of short term swing traders.

In our chosen example in Figure 15, the MACD Histogram divergence

catches the same price double bottom as the MACD Lines strategies, plus
several more. Unfortunately, for this example, the additional trades didn’t
profit but instead whipsawed. From the fact that the last trade did make
good by catching the up move from the eventual 2002 market bottom, we
can see the value of persistence in re-entering after being repeatedly
stopped out.

Figure 15 - MACD Histogram Bullish Divergence buy signal with sell

signal of a bearish divergence of MACD Histogram. Created with
TradeStation. © TradeStation Technologies, Inc. All rights
reserved. (Click here for larger figure.)

Zooming in on the unsuccessful and successful trades in Figure 16 and

Figure 17 respectively gives us some hints on how a human discretionary
trader might improve on the mechanical strategy applied in the backtest.
Subjective variations such as the slope of the MACD Histogram are
beyond the scope of this backtest.

Contrast the steady and steep price decline in Figure 16, where MACD
Histogram got whipsawed with the double bottom in Figure 17, where the
MACD Histogram positive divergence did signal a significant buying
opportunity.
Another interesting feature of Figure 17 is the gap up on the buy day.
Recall that our end-of-day strategy detects a signal after hours and
schedules a buy on the open the next day. This is the proper way to
backtest with end-of-day data because it prevents us from accidently acting
on a signal in hindsight and calling it foresight. A live trader with intraday
data might get the buy signal in time to act before the close on the same
day. While we can’t test this directly, we can and did post-process the
results data to estimate the opening gaps. See the win rate results for
Swing Traders coming up.

Figure 16 - MACDH Bullish Divergence Whipsaws. Created with

TradeStation. © TradeStation Technologies, Inc. All rights
reserved. (Click here for larger figure.)
Figure 17 - Close up of successful entry on MACDH Bullish
Divergence. Created with TradeStation. © TradeStation
Technologies, Inc. All rights reserved. (Click here for larger figure.)

Additional Exit Testing for MACD Divergences

In practice, some swing traders buy on a bullish MACD Histogram

divergence but do not hold until a bearish MACD Histogram divergence
comes about. Instead, they wait for the MACD Histogram to rise above
zero and then sell the moment the MACD Histogram ticks down.

Figure 18 and Figure 19 show this strategy applied to the same 2002-2003
price data on SPY. Notice how the first trade shown is now profitable
whereas before (in Figure 16 and Figure 17) it got stopped out. Also
notice the quick exit on the Oct’02 entry and how that allows the
opportunity to grab another trade at the Mar’03 bullish divergence. This
strategy gives no re-entry signal after that and consequently misses most of
the 2003 bull run.

Remember we now have 20-20 hindsight — in the gloom of early 2003, a

major rise was not so clearly imminent, and a short term trader might have
felt contented to have snagged any profit at all.
Figure 18 - MACDH Bullish Divergence buy signal which exits when
MACDH on a MACDH downtick above zero. Created with
TradeStation. © TradeStation Technologies, Inc. All rights
reserved. (Click here for larger figure.)

Figure 19 - Close up of MACDH Bullish Divergence buy signal which

exits when MACDH downticks above zero. Created with
TradeStation. © TradeStation Technologies, Inc. All rights
reserved. (Click here for larger figure.)

In addition to the strategies presented above, two other runs applied more
tactics to limit losses in conjunction with ATR stops. It applied to both the
basic MACDL and MACDH divergences. Here’s how it works: The value
of the MACDL or MACDH is recorded on the day of the buy signal. If
the MACD (lines or histogram respectively depending on the strategy)
ever violated its value on the day of the buy signal, the position was sold at
the next day’s open. In the results tables, these are called the MACD Div
Low+3ASym runs.

Backtesting Setup Details

Markets: US Stocks and international stocks represented by ADRs on

NYSE, AMEX, NASDAQ including delisted tickers. (Click here for stock
lists.)

Time Periods: May 1994 - April 2008, divided into three samples to
prevent over-optimization.

May 1994 – April 2004

May 2004 – April 2007
May 2007 – April 2008

Direction: Long Only

Entry Strategy: Enter any stock with buy signals as described above, and
when volume criterion is met (more than 500,000 shares daily). All entries
are next day via Market on Open orders.

Exit Strategy: Exit all stocks according to the signals described above.
Profit-taking is done next day via Market on Open orders. Stop losses,
where applied, are simulated as stop orders, which may be executed
intraday. This combination doesn’t exactly model a nimble trader who
may grab profits from an exit signal early in a day in which those waiting
for end-of-day miss out due to a sell-off before the close. It does model
end-of-day trading and investment though.

Sizing: Where no stops are involved, size is fixed at 1000 shares. With
stops, the risk amount was fixed at $1000 and the size computed as the
number of shares that would risk $1000 between the anticipated entry and
the stop.

Backtesting Engine: TradeStation version 8.6, Build 2525

Data Vendor: CSI. This data set was specially selected for accuracy after
extensive testing.

MACD Divergence Long Entry Backtest Results

This section emphasizes Win Rate, which is the metric to beat for entry
strategies. Three time periods are tested separately to help guard against
curve fitting. In the graphs, each set of bars represents the time period
indicated on the horizontal axis. The blue bars are from Chapter 2, the
baseline for comparison. The other colorful bars are the results from this
particular set of tests. See the legend on the graph to map the color to a
strategy.

Active Investors with 200-Day Timed Exit

As shown in Figure 20, the win rate varies with respect to the baseline for
active investors. In more recent years, the MACD Histogram Bullish
Divergence entries beat the baseline. The MACD Lines Bullish
Divergences did not produce a better win rate than the baseline when held
for 200 trading days in any time period.
Figure 20 – Hypothetical win rates of MACD Divergences for Active
Investors with 200-day timed exit (Click here for larger figure.)

Position Traders with 20-Day Timed Exit

The MACD Bullish Divergence strategies edged past the baseline for
Position Traders as shown in Figure 21 below. The differences are slight.
Figure 21 – Hypothetical win rates of MACD Divergences for Position
Traders with 20-day timed exit (Click here for larger figure.)

Option Traders with 5-Day Timed Exit

This backtest bought shares of the underlying stock when it showed a

positive MACD Divergence and sold after 5 trading days. It is named for
option traders because it is an approximation of how long a trader in
options might hold after a signal.

The graph in Figure 22 illustrates the win rate performance of the MACD
Divergence trades compared to the baseline. The numbers are rounded in
the graph and Table 7 below shows the exact amounts.

The upshot is that the MACD Divergence trades came in slightly below
the baseline in all but the one case of MACD Histogram divergences in
2007-2008.
Figure 22 – Hypothetical win rates of MACD Divergences for Options
Traders with 5-day timed exit (Click here for larger figure.)

Swing Traders with 2-Day Timed Exit

MACD Divergences do not get off to a winning start for swing traders
with a 2 day horizon. We can see in the graph below that the test
strategies perform below the baseline for all three time periods.

However, the swing trading results improve by estimating how many

stocks gap up on the buy day. Our strategy is simply buying at the open
on the next day after a signal. However a savvy swing trader might check
for signals near the end of the day and enter just before the close. While
our data gives no insight into the ultimate fate of the gapping trades, it
does show that between 5 to 10% of stocks tested significantly gapped up
on the day after the buy signal.

Catching the opening gap by entering near the close on the signal day
could be enough to take the win rates over 50%.
Figure 23 - Hypothetical win rates of MACD Divergences for Swing
Traders with 2-day timed exit (Click here for larger figure.)

Results for Entry Testing

The table below summarizes the results for all trader types. In general,
potentially profitable strategies have a positive expectancy. Losing
strategies have their negative expectancies colored red. Yellow indicates
that the expectancy was slightly positive, but rounded down to zero –
indicating caution. Expectancies are a very rough guideline at this stage
because we are using timed exits.
(Click here for larger table 7.)

Positive divergences on MACD Lines and MACD Histogram did not

increase the win rate over the baseline with timed exits. To succeed,
MACD Divergences must show larger profits on winning trades and
smaller losses on losing trades.

MACD Divergence Exit Backtest Results

With exit strategies the key criteria are Expectancy and Maximum Adverse
Excursion (MAE), described in detail below. The number of trades
measured how often the strategy had an opportunity to trade. The average
hold time measured how long each trade went on. Between them you can
get an idea of how productively each strategy put its funds to use. Keep in
mind that the average hold times are just that – an average. Winning
trades tended to go on longer while losing trades were often shorter.

How to Use Expectancy

When testing a real strategy with exits, expectancy becomes more

important than win rate. As the name suggests, expectancy guides you an
understanding what to expect by giving you the mathematical average
result from any one trade. As with any average, results of any single trade
can vary but expectancy shows you how they averaged out over time. This
book scales expectancy by the amount risked to make it possible to
compare vastly different strategies across a very broad selection of stocks.

A positive expectancy means the strategy was profitable in the past. A

negative expectancy flags a strategy that lost money during the time period
under test and is something to avoid.

For example, an expectancy of 0.09 means an average of 9% of the

amount risked was returned per trade. If a stop was employed and the
position scaled to limit risk, this means 9% of the total dollar amount
between entry price and stop price.

With no stop, we assume the entire position was at risk, and in that case a
9% expectancy implies a 9% return, on average per trade. Again this
doesn’t mean every trade returns 9%, a big loss and even bigger gain could
amount to an average of 9%.

Obviously, bigger is better and you should seek the highest expectancy
within your risk tolerance. A key question is: how much is enough? Only
you can answer that for sure.

As a rough guide, the graphs of expectancy are set up to plot any strategy
on a color-coded gradient with a red zone for losses, cautionary yellow
zone for strategies which demonstrated a very thin edge, and a green zone
for those that performed better. In the graphs below, you’ll notice the
color scale is shifted such that strategies with expectancies just above zero
are coded red because you need more than break-even in this idealized
case in order to show a real-world profit.

The table below summarizes the backtesting results for exit testing. Note
that we compare to 20-day timed exits as the closest to the actual hold
times of the strategies with real exits.

Table of Results for MACD Divergence Exit Tests

Table 8 – Table 10 collate the data from the exit testing of MACD
Divergence strategies. This data is examined in detail and plotted
graphically in Figure 24 - Figure 27 below. See the following three pages.
(Click here for larger table 8.)
(Click here for larger table 9.)
(Click here for larger table 10.)
Win Rates with Real Exits

Adding real exits of course changes the win rates for each strategy. While
win rate is not as important as expectancy, we still want to know if a
strategy wins often enough to hold our attention and not decimate our
accounts with losses before the winner finally arrives.

It’s not surprising to note that the win rate decreases with the addition of
stops. Nor is it particularly cause for alarm. Adding a stop almost always
reduces the win rate because some trades got stopped out that would have
turned around and later become winners. We’ll miss those but that’s okay.

Figure 24 - Win rates of MACD positive divergences with various exit

strategies 1994 – 2004 (hypothetical) (Click here for larger figure.)

Graphs of win rates for strategies with exits are shown in Figure 24 -
Figure 26. With 13 strategies under test, the graphs get a little crowded.
The data label closest to and even touching each triangle identifies the
strategy by name. Each data point is presented in the same order left to
right as in the data table, top to bottom.

Interestingly enough the fully symmetric strategies, which buy on a bullish

divergence and sell on the next bearish divergence without any stop loss
actually beat the baseline. Glancing at Tables 8-10 above tells us that the
hold times for these fully symmetric strategies were far longer than the
others and longer than our chosen baseline for this test – the 20-day timed
exit.

Let’s reserve final judgment on each strategy until we see the expectancy
and risk factors in the sections below.

Figure 25 - Win rates of MACD positive divergences with various exit

strategies 2004 – 2007 (hypothetical) (Click here for larger figure.)
Figure 26 - Win rates of MACD positive divergences with various exit
strategies 2007 – 2008 (hypothetical) (Click here for larger figure.)

Expectancy of MACD Divergence Testing

Figure 27 - Expectancy of MACD Divergence Strategies
(hypothetical) (Click here for larger figure.)

How to Read the Expectancy Graph

The colorful graph above plots the expectancy for all of the exit strategies
under test plus the baseline. Each strategy occupies one space along the
horizontal axis and is in the same order as in the data table.

All the data for one strategy is aligned in a vertical column. As shown in
the legend, each time period has a particular shape to identify its data
point: 1994-2004 is denoted with a diamond, 2004 – 2007 a square, and
2007 – 2008 a triangle. If a shape is not visible, it is hiding behind a larger
shape, which had roughly the same value.

The vertical axis displays the expectancy. Negative expectancies

(unprofitable) are color-coded red. Notice that the red extends above the
zero line. That is not an accident. A strategy that comes out marginally
positive in hypothetical backtesting is unlikely to be profitable in real
situations. The colors gently fade to yellow to indicate caution. Then they
go on to green to signify strategies that were more profitable in backtesting
and have a chance at real-world profitability.

As the data points spread apart from the zero line, they reveal that
increasing risk goes with increasing returns. It’s relatively easy to find
rules and leverage that increases the expectancy in good times but it
usually comes at the expense of a greater negative expectancy when
conditions deteriorate. Look for strategies that go further into the green
zone and less into the red zone than others.

Effects on Expectancy of Limiting Risk with Stops and Sizing

The baseline strategy as well as the symmetric strategies without stops

produced results clustered around the breakeven point. With stops, the
difference in expectancies between the time periods was accentuated.
Clearly the profits in good times made a nice positive jump. The negative
expectancies (losses) in 2007-2008 also became more extreme as stops
were added.

The extreme jump is largely an effect of the change in sizing and risk
calculations. It illustrates how reducing your risk can improve profits
when the right strategy is employed for the market conditions. However
adding stops is not a panacea for a mismatch between strategy and market
conditions as shown by the losses in 2007-2008.

Analysis of Exit Testing Results

In Figure 27 the expectancies illuminate the differences between the

strategies tested. Analysis brings the following lessons learned:

Trading on divergences brought outstanding profits in good times

(note the 1994 – 2004 and 2004-2007 markers solidly in the green
zone)

The same strategies brought losses in the tough years (2007-2008),

although less.

MACD Lines Divergences (MACDL) got more profits than

MACD Histogram Divergences (MACDH) for the same losses.
 The Power Tools with the longer average hold time hit the highest
expectancy of all. When deciding between the MACDH and
MACDL however, consider also the Standard Deviation. It is lower
for MACDH meaning less violent equity swings, which might be
easier to stomach.

Applying the 200-day MA to the MACD Lines Divergence

unexpectedly hurt results. This is possibly because a stock trading
above its 200-day MA is not at bottom – it either has further to go
down, or if the price is already climbing above the 200-day MA, the
double bottom signaled by a MACDL divergence already happened
while the price was under the 200-day MA and so got filtered out.

Exiting on the first MACD Histogram downtick above zero

succeeded in limiting losses during the tough years, however it
also severely limited profits in good times. This strategy did take
the most trades with the shortest average hold time which signals
productive use of account equity. But its slight edge of 0.05
expectancy leaves no margin for error, slippage or even full-priced
commissions.

Curiously, the 3ATR stop did slightly less poorly in the 2007-2008
timeframe than the tighter 2 ATR stop. This held true for both the
MACDH and MACDL suggesting that at some point tighter stops
possibly bring more losses by getting caught up in the noise and not
allowing the stock enough room to move. Or this may be related to
the sizing algorithm – see the Maximum Adverse Excursion (MAE)
section below for more details on that.

The Low+3 tactic reduced tough year losses while still retaining solid
profits in good years. It consisted of monitoring the MACD Lines or
Histogram value throughout the life of the trade and exiting if that
value was violated.

How to Assess Risk with Maximum Adverse Excursion (MAE)

Equally important to understanding the potential for gain is assessing the

risk of loss. Drawdown is frequently quoted in the industry but, because
most of us are not managing a portfolio of 7147 stocks, it’s not very useful
here. Instead we can gain knowledge of the risks by tracking the
Maximum Adverse Excursion (MAE).

Don’t let the technical term put you off, MAE really just means knowing
how much the position went against you. MAE is not the same as the
biggest losing trade because a stock may wander down for a huge open
loss and come back before the exit.

To stay in the game, the MAE needs to stay under the size of your
account. To be successful, the MAE needs to be limited to a fraction of
your account.

In Table 11 - Table 13 below, the baseline “no-strategy” strategy shows

how far awry trades can go. The 20-day timed exit functions as the
baseline because that most closely matches the average number of days
that the other strategies held their stocks.

In all cases, using symmetric exits such as negative divergences, without

stops gave a worse max MAE than the baseline. See Table 11 - Table 13
below. Keep in mind that the 20-day timed exit baseline was chosen
because it was closer to the average hold times of the majority of strategies
which employ stops and so have shorter average hold times than the
symmetric exits. It’s not entirely fair to compare a 20-day hold time with
the 100- to 200- day hold time of the symmetric MACD divergence
strategies and that is at least partially responsible for the outsized MAEs
on the strategies without stops. Being without stops no doubt contributes.

Rather unexpectedly, the 2 ATR stop on the Power Tools run had a worse
MAE than the 3 ATR stop. One would think that a wider stop would
necessarily lead to a worse MAE. In this case what happened by looking
at one example trade: a buy signal on STEM at 11/21/07. Both trades
wound up being stopped out when price couldn’t follow through after a
strong gap up opened for a buy price higher than expected.

The difference between the two stop values was in the sizing. Because the
2 ATR stop was closer to the price of the stock than the 3 ATR stop, to
maintain comparable risk amounts on the trades, the size of the 2 ATR
stop trade was 5434 shares versus 3623 shares purchased with a 3 ATR
stop. The larger size is what pushed the 2 ATR stop to a higher MAE than
the 3 ATR stop for that particular trade. With the smaller size, that trade
didn’t even have the worst MAE for the 3 ATR stop. Another trade was
the “winner” of that contest resulting in different per share MAE values
between the 2 and 3 ATR stop runs.

MAEs are comparable between Lines and Histogram divergences. The

additional exit criteria of MACDH downticks and Low MACD values
markedly decreased the MAE, in line with the expectancies in the previous
section. What is not in line is the MAE on the MACDHMA. Comparing
the trades for MACDHMA to MACDLMA (Histogram vs Lines) revealed
that the MACDHMA had more gaps up on the open, which pushed more
trades past the expected risk amount of $1000. This undoubtedly
contributed to the higher average MAE but it is unclear if gaps up are the
sole cause.
(Click here for larger table 11.)
(Click here for larger table 12.)
(Click here for larger table 13.)

Select R-Multiple Distributions

Tables of results give a good summary but necessarily leave out detail.
For deeper insight, check out the following results distribution graphs.
Distribution graphs display the number of times a particular range of
results occurs. They indicate whether to expect big losses often or a
multitude of little hits. Likewise, some strategies have many small gains
and others earn their keep in a few large paydays.

To scale the returns by the risk taken, I use R-Multiples which were
popularized by Dr. Van Tharp (vantharp.com). As the name suggests, the
return is expressed as a multiple of the risk, where an R-Multiple of 1
means the return was equal to the risk. As examples, an R-Multiple of
+5R means the profit was five times the amount risked and an R-Multiple
of -1R means a loss of the amount risked.

In these back-tests R, the amount risked per trade, was always $1000.
Then +5R is $5000 profit while -1R is a $1000 loss. All of the trades from
a run of a particular strategy over a particular time period were sorted into
“bins” which correspond to a bar on the graph. The label of the bin is the
mid-point of its contents so the 0.5 bar represents all trades that returned
between 0 and 1 R. Anything more negative than -1 R means those trades
gapped past their stops or jumped the risk limits due to opening gaps.

For more in-depth information about frequency distribution graphs see

http://backtestingblog.com/glossary/results-distribution-definition/

Figure 28 - Results Distributions for MACDL Power Tools with 3

ATR Stop 1994 – 2004 (hypothetical) (Click here for larger figure.)

Figure 29 - Results Distributions for MACDL Power Tools with 3

ATR Stop 1994 - 2004, zoom in the -1R to + 1R range
(hypothetical) (Click here for larger figure.)

Figure 30 - Results Distributions for MACDH with 3 ATR Stop 1994 –

2004 (hypothetical) (Click here for larger figure.)
Figure 31 - Results Distributions for MACDH with 3 ATR Stop 1994 –
2004, zoom in the -1R to + 1R range (hypothetical) (Click here for
larger figure.)

Figure 32 - Results Distributions for MACDL with Low+3ATR Stops

1994 - 2004 (hypothetical) (Click here for larger figure.)
Figure 33 - Results Distributions for MACDL with Low+3ATR Stops
1994 - 2004, zoom in the -1R to + 1R range (hypothetical) (Click here
for larger figure.)

Figure 34 - Results Distributions for MACDL Power Tools with 3

ATR Stop 2004- 2007 (hypothetical) (Click here for larger figure.)
Figure 35 - Results Distributions for MACDL Power Tools with 3
ATR Stop 2004- 2007, zoom in the -1R to + 1R range
(hypothetical) (Click here for larger figure.)

Figure 36 - Results Distributions for MACDH with 3 ATR Stop 2004-

2007 (hypothetical) (Click here for larger figure.)
Figure 37 - Results Distributions for MACDH with 3 ATR Stop 2004-
2007, zoom in the -1R to + 1R range (hypothetical) (Click here for
larger figure.)

Figure 38 - Results Distributions for MACDL with Low+3ATR Stops

2004- 2007, zoom in the -1R to + 1R range (hypothetical) (Click here
for larger figure.)
Figure 39 - Results Distributions for MACDL with Low+3ATR Stops
2004- 2007, zoom in the -1R to + 1R range (hypothetical) (Click here
for larger figure.)

For the 2004-2007 results in Figure 33, Figure 35, and Figure 37, the fat
tail of positive R-Multiples is highlighted in green to make it more visible.
The Power Tools strategy has the thickest fat tail – meaning that it had the
most trades with huge profits. However, the other strategies have more
“everyday” trades bringing in 5R or less.
Figure 40 - Results Distributions for MACDL Power Tools with 3
ATR Stop 2007-2008 (hypothetical) (Click here for larger figure.)

Figure 41 - Results Distributions for MACDL Power Tools with 3

ATR Stop 2007-2008, zoom in the -1R to + 1R range
(hypothetical) (Click here for larger figure.)
Figure 42 - Results Distributions for MACDH with 3 ATR Stop 2007-
2008 (hypothetical) (Click here for larger figure.)

Figure 43 - Results Distributions for MACDH with 3 ATR Stop 2007-

2008, zoom in the -1R to + 1R range (hypothetical) (Click here for
larger figure.)
Figure 44 - Results Distributions for MACDL with Low+3ATR Stops
2007-2008 (hypothetical) (Click here for larger figure.)

Figure 45 - Results Distributions for MACDL with Low+3ATR Stops

2007-2008, zoom in the -1R to + 1R range (hypothetical) (Click here
for larger figure.)

An alarming thing happens with both the MACDL Power Tools strategy
and the MACDH strategy: the -1.5R bin is larger than the -0.5R bin
meaning that many losing trades jumped their stops! It’s also possible
that the risk
value was exceeded if the stock gapped up in the morning.

These results are “risk limited” meaning that trades that would have
increased the risk value more than 20% were thrown out (with negligible
impact to expectancy). That leaves many trades where the initial risk
between buy price and stop price grew beyond the set $1000 risk.

The MACDL Low+3ATR strategy doesn’t exhibit this unruly behavior.

Looking at the Zoom graphs in the odd-numbered figures, we see how
many losing trades are cut short when the MACD (H or L) revisits its
“low” value (at the close on the day of the buy signal). These are the taller
red bars between -0.85 and -0.05. Those trades are cut short and then
don’t go on to either jump stops for a loss or turn around for a profit.

Conclusions on Original MACD Divergence BackTest

Results
Long trades on MACD Divergences bagged attractive returns in favorable
markets and significantly beat the baseline. In the tough years they lost
money but not as much as they gained in the good years.

Having an indicator to determine general market direction would have

been useful to weed out the bad times to take MACD Divergence buy
signals. The 200-day MA is not that indicator – it didn’t provide the
expectancy boost we’ve seen when we paired it with other strategies.

If you’re trying to decide between MACD Lines and MACD Histogram

divergences, consider these factors:

MACD Lines had:

Fewer signals
Higher expectancy
More huge winners

MACD Histogram had:

More frequent signals

Decent expectancy
A larger number of smaller wins
In a baseball analogy, MACD Histogram was oriented towards more base
hits, doubles and triples, while MACD Lines helped traders swing for the
fences to rack up home runs.

Both MACD Histogram and MACD Lines required a trader to let the
winners run for weeks and months to achieve the high positive
expectancies in this backtest. Attempts to exit the MACD Histogram
trades at the first downtick above zero severely limited profits.

A nimble trader that can accurately sense the MACD divergence during
the day it occurs and enter just before the close might have goosed their
expectancy gain by capturing the gaps up on the morning open. No
evidence suggested that this gain affected the expectancy of holding long
term, meaning that the big morning gap up trades had about the same
chance of long term gain as the trades that didn’t gap.

If trading MACD divergences, expect most of your stops to be jumped and

size your positions conservatively. Don’t run per-trade risk at your highest
threshold. Both the MACD Histogram and MACD Lines strategies had
the majority of their losses grow larger than the initial risk value, either
from opening day gaps up or from gaps down past the stop. In either
event, you would be better off setting a nominal risk value that is well
below what you are comfortable losing.

That said MACD divergences did prove out as the most powerful of the
MACD signals.

Out-of-Sample MACD Divergence Backtest Results

Each week I run a backtest of several MACD Divergence strategies on
current data and post the results to divergence-alerts.com. The data is
more recent than the backtests of the previous chapter.

It is also out-of-sample data, meaning that these time periods were not
considered as part of the previous backtests. That is useful for seeing that
the conclusions held up very well even for a completely different set of
data.

BackTesting Setup Details

Markets: The markets tested fall into three types:

DJIS: 141 Dow Jones Indices for sub-sectors of US stock market

ETFS: 354 Exchange Traded Funds
USTKS: 2376 current US stocks and ADRS

No delisted stocks.

Time Periods: The historical time periods under test started from May 1,
2008. The first divergence was spotted on the weekly charts September 3,
2010 and on the daily charts on March 5, 2009. The end date of the
backtests was April 18, 2014. Note that this time period does not overlap
the previous tests at all so it is completely out of sample data that didn’t
even exist when the previous backtests were run.

Weekly chart or daily chart is indicated in the name in the results Tables
14 – 17.

Direction: Both Long and Short trade directions were tested.

Entry Strategy: For this simulation, all the entry signals are MACD
divergences.

All strategies in this test used the standard settings of MACD: 12, 26, 9.
The scanner was set to look for compact divergences with at least 6 and no
more than 50 bars between price extremes. The divergence backtesting in
chapter 3 included larger divergences, with 10 to 100 bars between price
extremes.

The names in the results table spell out the type of strategy tested. Here
are the letters in order:

1. M – MACD
2. L or H – Lines or Histogram
3. D – Divergence
4. L or S – Long or Short
5. Z – fuzzy divergence which did not make a price extreme. Fuzzy
divergence entries are exited after 5 bars.
6. Q – quick exit which ends the trade on a MACD lines crossing or
MACD Histogram uptick/downtick. The exit otherwise is a matching
divergence from the entry signal.
Exit Strategy: The exits for long trades are negative divergences. For
example, enter long on MACD Lines positive divergence, exit on MACD
lines negative divergence. Short trades exit on the uptick of either MACD
lines (for MACD lines divergence entries) or MACD histogram (for
MACD histogram divergence entries). Fuzzy divergence entries are exited
after 5 bars. In addition, all trades have a 2.5 ATR stop loss fixed from the
point of entry.

Sizing: Trades are sized with a constant $100 risk from expected entry
price to stop price.

Backtesting Engine: TradeStation version 9.1

Data Vendor: TradeStation

BackTest Results

Note that the Avg. Gain is per trade, not annualized.

(Click here for larger table 14.)

Table 14 shows the strategies that produced hypothetical profits in

backtesting, sorted from the highest expectancy down. One common
theme among them is that they are all Long-only strategies. No short-
selling strategies made the winner’s list during this time period.

The US Stocks (USTKS) and Dow Jones Indices (DJIS) top the list. ETFs
as a group are challenged to produce profits one way or the other because
they contain both regular and inverse funds.

The largest gains came from waiting for a negative divergence to exit.
Taking the Quick exit (MACD lines cross up or MACD Histogram uptick)
as in the Q strategies or exiting after five trading days as the fuzZy strategy
does, led to reduced profits.

A few profitable strategies managed a win rate over 60%, however, most
came in between 40% and 55%. The highest win rate of 72% came from
entering a Dow Jones Index on a MACD Lines positive divergence and
exiting when the MACD lines crossed down from above zero on a daily
chart. This quick strategy was backtested on index data however tradable
ETFs exist that track those indices.

This highlights a typical trade-off in market operations: larger gains can

take longer to develop and come with more losing trades.

The strategies that weren’t profitable during this backtesting session were
mostly trading in the short direction, as shown in Table 15 below. Note
that Table 15 is sorted by decreasing expectancy so the worst performers
are at the bottom. Those strategies traded short on a negative divergence
on the weekly chart and covered when the MACD lines crossed upwards
for MLxS strats or when the MACD Histogram made an uptick for MHxS
strategies.

Chapter 7 presents more comprehensive backtests of short trades on

MACD negative divergences.

Notice that the win rates of the unprofitable strategies are all under 50%
and generally under 40%.
(Click here for larger table 15.)

Volatility data is included on the next page in Table 16 and Table 17 in

case you are interested in poring through it. Everything is listed in the
same order as Table 14 and Table 15. The expectancy and average gain
data are repeated for convenience.

The new columns are:

Standard Deviation – a typical measure of volatility

%Maximum Adverse Excursion (MAE) – a measure of how far a
trade went against you, averaged across all trades. Note that this is a
loss, even though it’s shown as a positive number
%Maximum Positive Excursion (MPE) – how far a trade went in the
profitable direction, average of all trades.

(Click here for larger table 16.)

The large MPE numbers for the weekly strategies compared to their
average gains indicate that they gave back considerable profits waiting for
a negative divergence to exit.

The MAE doesn’t show any gargantuan losses because it is an average

across all trades. There were a few that gapped past their stops for out-
sized losses.

(Click here for larger table 17.)

Interestingly enough, the MAE of the unprofitable strategies tended to be

lower than the MAE of the money-makers. Even a strategy that often
ended in profits could expose trader’s stomach linings to open losses larger
than they might see in a losing strategy.

In general, the more recent backtests confirmed the earlier conclusions that
trading long on positive MACD divergences could be profitable while
trading short on negative MACD divergences did not produce encouraging
results.

How to Apply MACD Divergences

Two main approaches:

1. Watch your favorite stocks and when you see a MACD positive
divergence, consider jumping on it. This might happen once or twice
a year.

2. Scan the broad market for MACD positive divergences on any stock
or more realistically any stock that meets your pre-set criteria of
volume, fundamentals, etc. To scan the market, you again have two
choices:

1. You could purchase MACD Divergence Detector Software. See

backtestingblog.com/code/macd-divergences/ for my
TradeStation add-in to automatically detect MACD divergences.

2. See my website divergence-alerts.com for more information on a

service that identifies MACD divergences on individual stocks
and ETFs for you.

A mechanical or automated strategy relying solely on MACD divergences

is not recommended.
Chapter 4: Divergence Count Indicator

Overview of Divergence Count Indicator

The movement of individual stocks can be likened to a school of fish: at

times they all turn at once.

The concept of the Divergence Count Indicator (DCI) is simple: keep

count of all MACD Divergences across the stock market and use the count
to guide the timing and direction of trading both individual stocks and
broad market indices.

This chapter describes how the DCI can be used to identify potential
turning points and reports on backtesting to assess the historical
performance of DCI.
Basic Categories of DCI

The DCI tracks the number of divergences that occurred on each day for
daily charts and each week for weekly charts. The INTRAS divergence
count, covered in the next chapter, focuses of intra-day count, specifically
hourly charts. On the general chart shown in Figure 46 below, the positive
divergence count is plotted below the daily prices for the SPY.

The middle pane was created from data from CSI – including delisted
tickers – running the basic MACD Lines Divergence (MACDL Div) scans
that were used in the previous chapter.

The bottom pane is the USTKS data – created from TradeStation historical
data running the MACD Divergence Alerts scans for MACD Lines
Divergences. The “U” data does not contain delisted tickers.

Figure 46 - SPY (upper plot) vs DCI indicator: CSI MACD

Divergence count (middle) and TradeStation MACD Divergence count
(bottom). Created with TradeStation. © TradeStation Technologies,
Inc. All rights reserved. (Click here for larger figure.)

Figure 46 makes a comparison between the two different data sets. In

general, they have similar shapes: both spiking at the same times even if
the intensity varies somewhat.
Another view of the DCI is given by taking the number of positive
divergences minus the number of negative divergences. See Figure 47,
below. When there are more positive divergences, the plot is above zero
and when there are more negative divergences, the plot is below zero. The
plot shows at a glance when positive divergences are overwhelming
negative divergences and vice versa.

Once again, the top plot is SPY, and the bottom plot is created from a
count of MACD Lines Divergences on TradeStation with data from CSI.

Figure 47 - P-N. Created with TradeStation. © TradeStation

Technologies, Inc. All rights reserved. (Click here for larger figure.)

Legend for DCI Naming:

U for USTKS, D for DJIS, C for CSI on US Stock Market

M for MACD

L for Lines, H for Histogram

P for Positive, N for Negative

D for Daily, W for Weekly

Strategy Development Methodology

Even the act of looking at the data plotted allows an opportunity for over-
optimization! Yet with a brand new indicator, it’s necessary to look at the
plot to figure out what to do with it. To deal with this situation, the plots
only show data from 2004 – 2010. The DCI from 1995 - 2004 remains
unseen and thus can be used as out-of-sample data.

The procedure involved manually assessing different strategies using the

2004 – 2010 data. Then those strategies, and exactly the potentially over-
tweaked parameters, were blindly applied by to the 1994 – 2004 data using
automated back testing. Thus any human bias was impartially tested by
computer on out-of-sample data.

Strategies Considered

First, two potential strategies were eliminated. By simple inspection, it’s

clear that the DCI P-N is not a direct indicator of trend because the DCI is
negative while the SPY is in strong up trends. Since that appears obvious
by inspection, it wasn’t backtested.

The second eliminated strategy involved divergences in the DCI itself.

DCI Divergences happen when the SPY makes an extreme price move
while the DCI makes an extreme move in the opposite direction. For
example, when the price makes a new high but the DCI makes a new low
(corresponding to increases in negative divergence count) – that is taken as
a DCI negative divergence.

Again, by inspection it was clear that this strategy was very similar to
looking at MACD Divergences on the market indices. No advantage from
counting divergences expected here.

Strategy Backtested: Peaks and Troughs in the P & N Counts

The strategy that was backtested produced the significant improvements in

win rates and expectancy discussed in the next section.

The approach was to apply a simple moving average to smooth the raw
counts of positive and negative divergences. Then look for favorable
times to buy based on peaks in the positive divergence count and/or
troughs in the negative divergence count. Sells signals and signals to sell
short were not considered in this first pass.
One advantage to this approach is that it’s relatively easy to create
objective signals.

The back testing actually began with an examination of the MACD Lines
Negative Divergence Counts on daily charts (UMLND and CMLND for
TradeStation and CSI respectively).

The basic idea behind the strategy is that times when the negative
divergence count drops very low might be a good time to buy. To quantify
this, the “buy zones” are defined as the 20-day moving average (MA) of
negative divergence counts being below 5. Let’s call these the N-Zones.

The moving average period of 20 was chosen without optimization

because it is often used to smooth volume spikes. The threshold value of 5
was hand-optimized by visually adjusting the threshold line to catch the
troughs in the UMLND for 2005 – 2010.

See Figure 48 for the N-zones marked with green dots when the MA falls
below threshold.

Figure 48 – Buying zones signaled by an absence of negative

divergences. Created with TradeStation. © TradeStation
Technologies, Inc. All rights reserved. (Click here for larger figure.)

Figure 49 below shows the converse case for creating buying zones from
the MACDL positive divergence count. We can call them P-Zones.

Figure 49 – Buying zones indicated by an increase in positive

divergences. Created with TradeStation. © TradeStation
Technologies, Inc. All rights reserved. (Click here for larger figure.)

To create the P-Zone, a 20-day moving average (MA) of the MACD Lines
positive divergence was taken. The zone is defined by the MA crossing up
above the threshold value of 10 – then the buying on MACDL positive
divergences can begin. The zone ends when the MA crosses down below
10.

Obviously neither P-Zones nor N-Zones are perfect – both produced open
buying zones during the spring of 2008, and in hindsight we know the
market had further to fall. Trading during the DCI zones still carries
significant risks and stop losses are recommended.

We also consider All-Zones, which is the union of the P-Zones and the N-
Zones. As illustrated in Figure 50, the buying window is open if either the
count of negative divergences drops below its threshold of 5 or the count
of positive divergences rises above its threshold of 10.
Figure 50 – DCI N-Zones and P-Zones plotted together. Created with
TradeStation. © TradeStation Technologies, Inc. All rights reserved.
(click here for larger figure.)

Backtest Setup

The back test setup and results presentation are similar to Chapter 3.

Those exact trades were then filtered for P-Zones and N-Zones as
described in this report. The total number of zones tested is just above the
minimum of 30 required for a valid sample:

P-Zone = 46

N-Zone = 32

All-Zones = 78

Markets: US Stocks and international stocks represented by ADRs on

NYSE, AMEX, NASDAQ including delisted tickers.

Time Periods: May 1994 - April 2008, divided into three samples to
prevent over-optimization.

May 1994 – April 2004

May 2004 – April 2007
 May 2007 – April 2008

Direction: Long Only

Entry Strategy: Enter any stock with buy signals as described above, and
when volume criterion is met (more than 500,000 shares daily). All entries
are next day via Market on Open orders.

Exit Strategy: Exit all stocks according to the signals described above.
Profit-taking is done next day via Market on Open orders. Stop losses,
where applied, are simulated as stop orders which may be executed
intraday. This combination doesn’t exactly model a nimble trader who
may grab profits from an exit signal early in a day in which those waiting
for end-of-day miss out due to a sell-off before the close. It does model
end-of-day trading and investment though.

Sizing: Where no stops are involved, size is fixed at 1000 shares. With
stops, the risk amount was fixed at $1000 and the size computed as the
number of shares that would risk $1000 between the anticipated entry and
the stop.

Backtesting Engine: TradeStation version 8.6, Build 2525

Data Vendor: CSI. This data set was specially selected for accuracy after
extensive testing.

DCI Win Rates Backtest Results

The first pass of backtesting looks at the win rates of the entry signal using
timed exits. The scale of timed exits represents different trader types as in
earlier chapters:

Active Investor – 200 days

Position Trader – 20 days
Option Trader – 5 days
Swing Trader – 2 days

The following figures compare win rates to two reference strategies:

1. The BackTesting Report Baseline

(the purple bar on the left of each set)

2. MACD Lines Divs from Chapter 3

(the blue bar next to it)

As you can see from Figure 51 – Figure 54 below, the P-Zones and N-
Zones helped the win rates considerably. The out-of-sample period of
1994 -2004 not only confirmed but exceeded the performance of the time
periods used for development.

The longer term holding periods benefited the most from the buying
zones. The advantages to short term traders tended to be smaller and less
consistent.

The N-Zones more consistently increased the win rate than the P-Zone
did.

Figure 51 - Win Rates for buying during DCI open windows and
holding for 200-days show improvement in all time periods
(hypothetical) (Click here for larger figure.)
Figure 52 - Win Rates for buying during DCI open windows and
holding for 20-days generally show improvement in all time periods
(hypothetical) (Click here for larger figure.)

Figure 53 - Win Rates for buying during DCI open windows and
holding for 5-days mixed results (hypothetical) (Click here for larger
figure.)
Figure 54 - Win Rates for buying during DCI open windows and
holding for 2-days generally show decrease in all time periods
(hypothetical) (Click here for larger figure.)
(Click for larger Table 18.)
(Click for larger Table 19.)
(Click for larger Table 20.)
(Click for larger Table 21.)

DCI Exit Backtest Results

More realistic exits were added to the strategies under test and the
expectancy measured. The two types of exits were:
 “Symmetric” – sells when a MACD negative divergence occurs,
regardless of the DCI. It always trades 1000 shares.
“3ATR Sym” – sets a fixed stop 3 ATRs below the entry price and
sells when the stop is hit or a MACD negative divergence occurs,
regardless of the DCI. It calculates the number of shares that risks
$1000 between its expected entry price and 3 ATR stop loss.

The entry signals continue to be filtered by the DCI. The strategy only
buys when there is a positive MACD divergence on the individual ticker
and the broad-market DCI zone is open. The three sets of zones (P-, N-
and All-) were tested independently.

In Figure 55, the DCI-filtered results (column 2-4) are compared to the
Baseline from Chapter 2 (column 0) and the MACD Lines Divergence
with 100-day lookback from Chapter 3 (column 1). The No Zone results
(column 5) are MACD Divergences with a 50-day lookback and no DCI
filter applied.

Figure 55 - Expectancy of MACDL Divs with both P-Zones and N-

Zones (hypothetical) (Click here for larger figure.)

The data from the back tests using 3 ATR stops shows the clear advantage
of the DCI zones because the DCI expectancies in columns 2-4 are all
shifted upwards. This means that in the favorable markets the DCI
strategies showed more positive expectancies.

In the less favorable market of 2007 – 2008, the expectancies are less
negative, with the N-Zone expectancy even venturing into the positive
realm.

The average gain per trade (not annualized) is shown in Table 22 – Table
24. In general, the DCI zones improved on the average gains. The one
exception in 2003-2007 stemmed from a single 200% winning trade,
which fell outside the DCI zones.

The number of trades showed a dramatic reduction for the different zones.

With DCI, the trades are no longer distributed evenly throughout the test
period but instead clustered with start dates within the buying zones. The
end dates could be anywhere as the exit signals used for this set of back
tests was not tied to DCI zones.
(Click here for larger table 22.)
(Click here for larger table 23.)
(Click here for larger table 24.)

The win rates for the DCI zones with stop losses are shown in the
following three graphs, Figures 56 - Figure 58. It is not unusual for stop
losses to push the win rates under the baseline.
Figure 56 – Win Rates for DCI Exit Testing, 1994 – 2004
(hypothetical) (Click here for larger figure.)

Remarkably, the N-Zone win rate was above the baseline for 2004 – 2007
as shown in Figure 57 below.

All other win rates were below the baseline, which was typical for MACD
Divergence strategies.
Figure 57 – Win Rates for DCI Exit Testing, 2004 – 2007
(hypothetical) (Click here for larger figure.)
Figure 58 – Win Rates for DCI Exit Testing, 2007 – 2008
(hypothetical) (Click here for larger figure.)

DCI Exits Without Stop Losses Backtest Results

This collection of data is included for completeness to show the
performance without stop losses. As is usually the case without stops, we
see higher win rates and tighter but lower expectancy. The DCI filtered
strategies still beat out the baseline and complete MACD Lines divergence
runs. The following data used a 100-day look-back period for the MACD
Divergences.

Figure 59 – Expectancy of MACD Divergence runs without stop losses

(hypothetical) (Click here for larger figure.)

As usual, the consistently sized runs show less variation in expectancy

than the risk-sized runs with stop losses. See Figure 59. Still, the DCI
Zones (columns 2- 4) improve the expectancy slightly over the original
MACD Divergence run (column 1).

For the Symmetric runs, the DCI Zones produced win rates over the
baseline in all cases as shown in Figure 60 – Figure 62, below.
Figure 60 - DCI Exit Testing, No Stops, 1994 – 2004
(hypothetical) (Click here for larger figure.)

Figure 61 - DCI Exit Testing, No Stops, 2004 – 2007

(hypothetical) (Click here for larger figure.)

Figure 62 - DCI Exit Testing, No Stops, 2007 – 2008

(hypothetical) (Click here for larger figure.)
(Click here for larger table 25.)

(Click here for larger table 26.)

(Click here for larger table 27.)

Conclusions From DCI Backtesting

The DCI Zones demonstrated great potential for active investors and
position traders. Buying only those MACD Divergences that occurred
during a DCI Zone improved both win rates and expectancy for active
investors and position traders.

The N-Zone, where the moving average of negative divergences fell below
the threshold, showed the most improvement. The P-Zones, where the
moving average of positive divergences increased above the threshold,
also delivered better results than either the baseline or the plain MACD
Divergence strategies.

Both N-Zones and P-Zones severely limit the number of trades. Actively
using both the N-Zones and P-Zones, a process nicknamed the All-Zones,
produced the largest number of trades while still dramatically improving
results.

Short term trading such as options or swing trading did not appear to be
particularly helped by the DCI Zones.

The raw count of MACD Divergences is not directly useful as an indicator

because the number of negative MACD divergences often increased even
while the market kept rising. Since the raw count was not particularly
effective in the past, there is no reason to hope that it improves in the
future.

In summary, when looking to buy an individual stock with a positive

MACD divergence, waiting for a large number of positive divergences
and/or a small number of negative divergences may help improve
performance.

How to Apply DCI Zones to Your Trading

If you decide that the DCI Zones could fit in with your trading style, you
have two choices.

One approach is to attempt to count the divergences on your own. Here

are the steps to take:

1. Count all MACD Lines Divergences across the stock market, each
and every day
 2. Plot 20-day moving averages of the negative divergences and the
positive divergences
3. Consider longer-term purchases on the stock market when the MA of
negative divergences is below 5 and/or the MA of positive
divergences is above 10.
4. Use smaller trade sizes and stop losses to manage risk.

An easier approach is offered on my website where I post the DCI every

trading day. See www.divergence-alerts.com for more information.
Chapter 5: Intra-Day Divergence Count
So far we’ve seen that MACD positive divergences offered a long-term
benefit historically, particularly when many positive MACD Divergences
happened at once.

But what about short-term trading?

Let’s take a look at intra-day scans for a hint.

My Intra-Day MACD Divergence Scans

I scan selected stocks for intra-day divergences in an effort to gain insight
into the daily behavior of the stock market. Let me be clear that I am not
day-trading, but gathering data from a faster timeframe to apply to option
and swing trading.

The time scale I use is a 60-minute chart. The scans run automatically
during the regular trading hours at the bottom of the hour: 10:30, 11:30,
etc.

The stock list contains the component stocks of the Standard & Poor’s
S&P 500® index, plus the underlying ETFs and stocks from the CBOE®
list of options with weekly expiration. This amounts to about 550 tickers
to scan each hour.

The scans look for the following:

Positive MACD Lines Divergence

Negative MACD Lines Divergence
Positive MACD Histogram Divergence
Negative MACD Histogram Divergence
Fuzzy Positive MACD Lines Divergence

When a divergence is found, the stock ticker is recorded and the

divergence count is incremented for that hour.

Hypothetical Performance of Intra-Day Divergence

Count
To measure the hypothetical historical performance of the Intra-day
Divergence Count, I took the data on hand which ran from May 1, 2009
through November 11, 2013.

For each day in that period, the movement in the S&P 500 Index as
tracked by the SPY exchange traded fund was recorded over the next four
days in percentage terms. For a simplistic example, say the SPY went up
1% each day for four days. Then we’d have a Day1 gain of 1%, Day 2 of
2%, Day 3 of 3%, and Day4 gain of 4%.

Seven scans run during the trading day and the maximum number of
positive divergences in any one of those hours became the metric for the
analysis.

The plot below in Figure 63 shows the gain or loss after four trading days
corresponding to the maximum number of positive divergences each day.
Figure 63 - Scatter plot of % gain after 4 days versus number of
positive divergences on 60 minute chart (hypothetical) (Click here for
larger figure.)

For small numbers of positive divergences, the performance clustered

quite evenly around zero. That says that we don’t get useful information
out of just a few divergences.

However, as the number of positive divergences grew large, the

performance tended towards gains. Days which exhibited a substantial
number of positive divergences – greater than 85 – were often followed by
substantial positive gains in SPY price.

With 550 stocks in the scan set, 85 makes 15% of the stocks scanned
showing divergences. As you can see from the graph, only one day in four
and one-half years had over 300 (60%) stocks with divergences.

A more comprehensive summary of the data is shown in Table 28 below.

The performance was analyzed after 2 days and after 4 days from the day
divergences were counted. To create this table, the mathematical
expectancy of the results was calculated as:

(Number of Wins times Average % Gain) minus

(Number of Losses time Average % Loss)

(Click here for larger table.)

The Baseline column shows the expectancy for the complete historical
data set. The market’s upward bias during this time becomes evident from
the fact that holding for two trading days produced a positive expectancy
of 1.1% and holding for six trading days produced a positive expectancy of
slightly more – i.e., 1.6%.

Filtering to find the days which showed high numbers of positive

divergences on the 60-minute charts increased the expectancy to 1.5%
after two days and 2.1% after four days.

While high numbers of negative divergences put a damper on gains, it was

not enough to create negative expectancy. Two days after major bursts of
negative divergences, SPY expectancy dropped to 0.9% after two days and
1.1% after four days. Still it remained positive.

Based on this sample, a strong market may follow on the heels of days
with high numbers of positive divergences on the 60-minute charts, but
shorting the market after high numbers of negative divergences on the 60-
minute charts is not likely to pan out.

How to Do Intra-Day Divergence Scanning and

Counting
Keeping track of the intra-day divergence count is very much like the end-
of-day DCI. You need software to scan for divergences and a means to
track the count.

As an easier alternative, I’ve set up scans in TradeStation to look for

divergences on the component stocks of the S&P 500 index, plus the
underlying ETFs and stocks from the CBOE list of options with weekly
expiration. To scan that many stocks each hour requires 21 scheduled
scans per day. The Intra-Day divergence count is updated each hour on
Divergence-Alerts.com for Platinum subscribers.
Chapter 6: Resources

So far this book has covered strategies that showed promise based on
historical backtesting, namely various forms of positive MACD
Divergences.

Later chapters report on strategies that didn’t do so well.

In case you only want to read about the more useful techniques, this may
be a good place to discuss resources that can help you put MACD
Divergences to use.

Whether or not you seek to trade divergences on individual stocks, MACD

Divergences have the potential to help investors who note:

MACD Divergences on market indices

Positive divergences on large numbers of stocks
 Absence of negative divergences across the market
Intra-day surges of positive divergences on key tickers

Bonus Materials for Book Owners

You can download useful bonus materials at this book’s companion
website at truthaboutmacd.com

The bonuses include:

High-resolution graphics of the figures and tables in this book.

Lesson 1 of The Truth About MACD video series introduces the
MACD by graphically showing how its constructed. Illustrates the
many moving parts of MACD by showing how the MA crossovers,
the MACD lines and the MACD Histogram all relate to each other.
Lesson 2 of The Truth About MACD video series introduces back
testing and gives you a baseline for comparison for all strategies
tested.
Study Guides for Lessons 1 and 2.
Truth About MACD Starter Kit includes 30 min audio presentation
revealing 8 secrets about MACD and Missing Link software for
TradeStation® and StockFinder®. (TradeStation and StockFinder
must be purchased separately.) The sample software offered here is
for Moving Average Crossovers. It is NOT the divergence detectors.

To access all this educational information, go online at

truthaboutmacd.com. You will need to supply a password to prove that
you own the book, and register with your name and email address. I ask
this in order to protect us both from the nefarious characters on the ‘net.
You can unsubscribe from my email list at any time and I neither share the
list nor spam.

Divergence-Alerts.com
My divergence-alerts.com website offers comprehensive reporting of
MACD Divergences of all forms:

Divergence Counts on Daily and 60-Minute Charts

Divergences on Daily, Weekly, and 60-Minute Charts
 Divergences on US Stocks, Sectors, and ETFs including Index ETFs
MACD Lines Divergences
MACD Histogram Divergences
Fuzzy Divergences
Both Positive and Negative Divergences

Recent backtesting results posted each weekend keep you up to date on

what’s been working lately.

Anticipating Divergences

It’s mathematically possible to identify tickers that may have a divergence

coming up and calculate the price which would bring about the divergence.

Divergence-alerts.com reports these situations, giving you both the ticker

and the price where a MACD Lines Divergence could possibly occur.
This allows a trader to watch or even place a stop entry order to get in the
moment a divergence occurs.

Drilling Down to ETF Component Stocks

Stocks on divergence-alerts.com are cross-referenced with the ETFs that

hold them. This can make it easier to take advantage of movement within
a sector.

For example, one can go top-down by clicking on a ticker in the list of

ETFs with divergences to see the divergences on stocks held by the ETF.
Because individual stocks can be more volatile than the ETF or index, this
could offer more profit from a sector-wide move.

Figure 64 below shows a screenshot of divergence-alerts.com displaying

divergences among the components of SPY. Each lists a component stock
of the index. Each column represents a type of divergence labelled using
this shorthand code:

M for MACD

L for Lines, H for Histogram

P for Positive, N for Negative

D for Daily, W for Weekly

Negative divergences are highlighted in red and positive divergences in

green. The date is the last occurrence of the divergence and the color gets
lighter as the date fades into the past.

Figure 64 – divergences among SPY component stocks (divergence-

alerts.com screenshot) (Click here for larger figure.)

Going the other way, bottom-up, is possible too. Divergence-alerts.com

members also get a spreadsheet listing ETFs that hold each stock. That
way you can check if the ETF is diverging at the same time as the stock.

MACD Divergence Detection Software

Traders interested in combing markets on their own for MACD
Divergences can take advantage of the productivity offered by automated
screeners.

My MACD Divergence Detectors plug into popular charting packages to

find several kinds of MACD divergences. The software can scan futures
and international stocks in addition to US stocks and ETFs using the data
that drives your charts. They are fully parameterized to adapt to your own
needs.

Figure 65 shows an example of my MACD Divergence Detectors running

within TradeStation®, the backtesting engine used in this report. See
tradestation.com for more information on the overall software package
which is sold separately.

Figure 65 - MACD Divergence Detector strategy and RadarScreen®.

Created with TradeStation. © TradeStation Technologies, Inc. All
rights reserved. (click here for larger figure.)

Divergences on the historical chart are flagged in green for positive

divergences and red for negative divergences. When a divergence occurs
at the right edge of the chart, the software calculates a trade size and stop
based on parameters you enter.

At the right side of Figure 65 you can see RadarScreen® working with the
MACD Divergence Detectors to monitor many stocks for divergences in
real time.

For more information and larger screenshots visit:

http://backtestingblog.com/code/macd-divergences/

TradeStation clients can try out this software for free within the
TradeStation App Store at
https://tradestation.tradingappstore.com/products/MACDDivergenceDetector

Related Reading
The author’s current reading list is posted at
http://backtestingblog.com/order/books/

A glossary of backtesting terms is posted

at http://backtestingblog.com/glossary/
Chapter 7: Selling Short MACD Negative
Divergences
With this chapter on selling short on negative MACD divergences we
begin our survey of MACD strategies that didn’t prove very useful. Read
on if you need to be convinced that you probably won’t trade your way to
a private island using the techniques that follow.

What is a Short Trade

Selling short is a technique used by traders who believe the market is
about to go down. It involves selling high with a goal of later buying
lower.

A short seller must borrow shares of stocks or ETFs which the brokers
make easy to do. The borrowed shares are then sold at the current market
prices. That is the short sale.

If the market goes down, the trader buys shares at the prevailing lower
prices and pockets the difference. The newly purchased shares are used to
automatically repay the loan that facilitated the short sale. This is called
“buying to cover”.

But what if the market doesn’t go down? A short trade can theoretically
lose an infinite amount of money as the stock price rises. Even without
reaching that theoretical maximum an open short can wipe out a trading
account.

As the market goes up, the short seller experiences an open loss. The
broker may require more collateral for the borrowed shares. If that
requirement exceeds the size of the account, the broker may close out the
trade, or issue a margin call requiring the trader to add more money to the
account.

When a trader buys back at a price higher than they sold short, either of
their own volition or at the hands of the broker, a loss is debited against the
account.

Strategies Under Test

Figure 66 shows an example of a short trade on a negative MACD
divergence. The negative divergence is highlighted by a red arrow drawn
onto the TradeStation screenshot below. As you can see, the MACD lines
made a higher peak in December of 1999 than they did in March and April
of 2000.

The simulated trades by the TradeStation backtesting engine are shown as

red and green lines over the price bars. The first trade beginning in March
of 2000 ended in a loss when the price jumped above the stop.

The second trade went much better as the price dropped significantly.
That trade ended on the positive divergence marked with the green arrow
drawn over the MACD.

You may notice that the MACD sunk lower throughout early 2000. The
strategy waited for the price to actually close at new highs before entering
the trade. This served to weed out “false” sell signals.

Figure 66 - Example of basic short trade on MACD divergences.

Created with TradeStation. © TradeStation Technologies, Inc. All
rights reserved. (Click here for larger figure.)

The second entry method relies on the ADX indicator to select “trendless”
markets before searching for MACD divergences. The thinking here is
that because many uptrends don’t cave in at the first negative MACD
divergence, the ADX can be used to give an indication that the uptrend has
weakened and a reversal more likely. The backtesting runs show how this
compares to using MACD divergence alone.
Figure 67 is a chart of the same time in history as Figure 66. The ADX is
plotted below the MACD indicator. The rule is that the strategy only sells
short on negative MACD divergences if the ADX is below 20 to indicate a
trendless market.

As it happened in this example, the ADX temporarily blipped above 20

just at the wrong moment to prevent the second trade which would have
been profitable. Backtesting can illustrate how often ADX filtered out
good trades versus bad.

Figure 67 - MACD Divergence and ADX. Created with TradeStation.

© TradeStation Technologies, Inc. All rights reserved. (Click here for
larger figure.)

Exit strategies under test are:

The usual timed exits

Symmetric exits on positive divergences

Quick exits on MACD lines cross upwards below zero and MACD
Histogram uptick.

BackTest Setup Details

Markets: US Stocks and international stocks represented by ADRs on

NYSE, AMEX, NASDAQ including delisted tickers.
Time Periods:

May 1994 - April 2008, divided into three samples to prevent over-
optimization:

May 1994 – April 2004

May 2004 – April 2007
May 2007 – April 2008

Direction: Short Only

Entry Strategy: Enter all stocks according to the MACD and ADX signals
described above, when volume criterion is met (more than 500,000 shares
daily). All entries are next day via Market on Open orders. All runs
comparing entry types used the de-facto standard settings of 12-26-9 (fast-
slow-signal).

Exit Strategy: Timed Exits of 2 days, 20 days, 200 days, as well as

symmetric and quick exits as described above. It exits via Market on
Close orders.

Sizing: Where no stops are involved, size is fixed at 1000 shares. With
stops, the risk amount was fixed at $1000 and the size computed as the
number of shares that would risk $1000 between the anticipated entry and
the stop.

Backtesting Engine: TradeStation version 8.6, Build 2525

Data Vendor: CSI. This data set was specially selected for accuracy after
extensive testing.

MACD Divergence Short Entry Backtest Results

The first set of performance data is the win rate with timed exits. Nine
runs were made: three historical periods for each of three different holding
durations corresponding to different trader types. Each run sold short
when a MACD negative divergence was detected and covered after the
specified number of trading days. The short-only runs were compared to
the long-only baseline which is the leftmost blue bar in each group.
Figure 68 – 200-day timed exits from short sales on negative MACD
divergences (hypothetical) (Click here for larger figure.)

For longer term holding periods which model active investors most short
strategy win rates were directly opposite the long baseline. See Figure 68.
When the long baseline won more often than not, i.e. 1994 – 2004 and
2004 – 2007, the short strategies did not produce a winning record. On the
other hand, when the market turned down in 2007 – 2008, the short
strategies come in winners more than 60% of the time.

A glaring exception that struggled in all time periods was the addition of a
5 ATR stop loss to the basic short selling strategy plotted as the rightmost
bar in Figure 68 - Figure 70. Its poor performance in the 2007-2008
timeframe indicates that even in a favorable market environment a short-
seller might face an unendurable open loss.
Figure 69 – 20-day timed exits from short sales on negative MACD
divergences (hypothetical) (Click here for larger figure.)

Position traders with a 20-day holding period could expect a similar

outcome, although not so extreme. Still, the short trades did worse than
the long baseline when the market went up and better than the baseline in
downwards markets. See Figure 69 above.

The final run in this series used timed exits of 2 days. Note that the scale
in Figure 70 is much tighter than that of Figure 68 and Figure 69. That
makes it appear as though the results were spread out more than they really
were. Actually, it is only a slight difference among the win rates with 2-
day exits however it demonstrates the same pattern of running opposite to
the long baseline.
Figure 70 – 2-day timed exits from short sales on negative MACD
divergences (hypothetical) (Click here for larger figure.)

MACD Divergence Short Exit Backtest Results

The next set of Figures (71 – 73) display the win rates for MACD short
trades with various realistic exit strategies. In general, the quick exits of
MACD lines crossing upward and MACD Histogram uptick produced
higher win rates than waiting for a symmetric exit of a positive MACD
divergence.
Figure 71 - Win rates of negative MACD divergence short trades with
exits 1994 – 2004 (hypothetical) (Click here for larger figure.)

Figure 72 - Win rates of negative MACD divergence short trades with

exits 2004-2007 (hypothetical) (Click here for larger figure.)
Figure 73 - Win rates of negative MACD divergence short trades with
exits 2007-2008 (hypothetical) (Click here for larger figure.)

Likewise, the expectancy chart in Figure 74 reveals dismal performance

for shorting on a negative MACD divergence and covering on a positive
MACD divergence – whether MACD Lines Divergences or MACD
Histogram Divergences.

Figure 74 - Expectancy of short MACD divergence trading strategies

(hypothetical) (Click here for larger figure.)

Again, the short strategies with quick exits fared better than waiting for a
positive divergence to exit.

More recent out-of-sample backtesting results as shown in Chapter 3

confirm the conclusions based on these earlier backtest results.

Divergence-alerts.com reports negative divergences primarily as potential

trade exits. It does track MACD lines negative divergences with quick
exits only because that is the only tactic that got any traction in this testing.
Chapter 8: Anticipating the Cross with
MACD Buy Signals
This chapter measures the non-divergence entry signals offered by MACD
– looking for reliable buy signals. It also compares a short list of
recommended parameter settings in search of improvement over the
default 12, 26, 9.
Strategies Under Test

Four different entry signals were tested with timed exits:

MACD lines crossing each other while below zero

Appel’s Histogram Uptick while below zero
MACDH Uptick while below zero
MACDH Uptick while below zero with price above the 200-day MA

Figure 75 - TradeStation screenshot of MACD Histogram Uptick buy

signal with price above the 200-day MA (Click here for larger figure.)

To avoid an overwhelming amount of data we’ll check different parameter

settings on one selected style, chosen by the highest reliability of those
tested. This saves you from wading through a bunch of parameter data on
unused entries.

Three basic parameter settings were tested:

12-26-9
19-39-9
6-19-9
In addition Mr. Appel’s method of using the slope of a moving average to
choose the parameter settings for the MACD was also tested.

Backtesting Setup Details

Markets: US Stocks and international stocks represented by ADRs on

NYSE, AMEX, NASDAQ including delisted tickers.

Time Periods:

May 1994 - April 2008, divided into three samples to prevent over-
optimization:

May 1994 – April 2004 denoted by darker blue.

Ten-year period chosen to include major up, down and sideways
movements.
May 2004 – April 2007 denoted by medium blue
Out-of-sample data for the original period. At 3 years, it’s 1/3 as
long as original.
May 2007 – April 2008 denoted by light blue
Current data. It’s 1/3 of the previous sample and is more out-of-
sample data.

Direction: Long Only

Entry Strategy: Enter all stocks according to the MACD lines and
histogram signals described above, when volume criterion is met (more
than 500,000 shares daily). All entries are next day via Market on Open
orders. All runs comparing entry types used the de-facto standard settings
of 12-26-9 (fast-slow-signal). The MACD parameter settings, which were
tested separately, were (fast-slow-signal): 12-26-9, 19-39-9, 6-19-9

Exit Strategy: Timed Exits of 2 days, 20 days, 200 days, chosen as the
simplest way to make a baseline for popular Trader Types. It exits via
Market on Close orders.

Sizing: 1000 shares for every trade.

Backtesting Engine: TradeStation version 8.3, Build 1631 and Build 1634
(for Appel Histogram only)
Data Vendor: CSI. This data set was specially selected for accuracy after
extensive testing.

Basic MACD Entry Backtest Results

This chapter focuses on Win Rate, which is the metric to beat for entry
strategies in similar exit conditions. Three time periods are tested
separately to help guard against curve fitting.

Results are sectioned by Trader Types for readability.

Active Investors

MACD Entry Types for Active Investors

BackTesting Report tests entries for active investors by using a 200-day

timed exit to assess the win rate after a longer term holding period.

The backtesting results shown in Figure 76 - Figure 78 demonstrate why

so many persist in drawing the MACD Histogram as they do: it performed
better than MACD Lines crossing and beat the baseline in all three time
periods tested. Appel’s histogram fell short of the baseline in two of the
time periods tested, so even though it edged ahead of MACDH in 2004-
2007, it is too inconsistent to be declared the winner.

The most notable results in this chapter’s backtests came from adding the
200-day moving average (MA) as a filter and only buying stocks when
they are priced above the 200-day MA and have an uptick on the MACDH
from below the zero line.
 Figure 76 – Win rates of non-divergence MACD signals with 200-day
timed exits, 1994 – 2004 (hypothetical)(Click here for larger figure.)
Figure 77 – Win rates of non-divergence MACD signals with 200-day
timed exits, 2004 – 2007 (hypothetical) (Click here for larger figure.)
Figure 78 - Win rates of non-divergence MACD signals with 200-day
timed exits, 2007 – 2008 (hypothetical) (Click here for larger figure.)
(Click here for larger table 29.)

Table 29 summarizes the results for active investors. In general,

potentially profitable strategies have a positive expectancy. Losing
strategies have their negative expectancies colored red. Yellow indicates
that the expectancy was slightly positive, but rounded down to zero –
indicating caution. Expectancies are a very rough guideline at this stage
because we are using timed exits.

Even with a longer-term outlook, Active Investors appeared to be better

off jumping in on a fast-moving MACD Histogram uptick, not waiting for
MACD Lines to cross.

MACD Parameter Settings for Active Investors

The MACD Power Tools approach offered a significant advantage in the

difficult 2007-2008 time period. Its win rate exceeded both that of the
baseline and the more basic MACDH entry strategies’ by roughly 9%.
These results indicate that it may help protect assets in difficult times and
that is one of the main points of the Power Tools book. However, the
Power Tools results were less remarkable in the other test periods, even
falling below the baseline during 1994-2004. If this approach can deliver
larger gains than losses, the benefits might outweigh the lower reliability.
That potential is evaluated in the next chapter.

Other than that, varying the parameter settings for the MACD Histogram
didn’t make much difference in the win rates. See Figure 79 - Figure 81
below.
Figure 79 – Varying parameter settings for MACD Histogram entry
signals, 1994 – 2004 (hypothetical) (Click here for larger figure.)
Figure 80 - Varying parameter settings for MACD Histogram entry
signals, 2004 – 2007 (hypothetical) (Click here for larger figure.)
Figure 81 - Varying parameter settings for MACD Histogram entry
signals, 2007 – 2008 (hypothetical) (Click here for larger figure.)
(Click here for larger table 30.)

Position Traders

MACD Entry Types for Position Traders

A position trader seeks intermediate-term opportunities. To be successful,

a position trader gets in on a burst of action — perhaps one leg of a longer-
running trend — and gets out before the action fades. This is modeled
with 20 day timed exits in order to test entry strategies.

As it turns out, the tests for MACD were inconclusive for position traders.
All entry tactics and all setting performed about the same in terms of win
rate. See Figure 82 – Figure 84.
Figure 82 - Win rates of non-divergence MACD signals with 20-day
timed exits, 1994 – 2004 (hypothetical) (Click here for larger figure.)
Figure 83 - Win rates of non-divergence MACD signals with 20-day
timed exits, 2004 – 2007 (hypothetical) (Click here for larger figure.)
Figure 84 - Win rates of non-divergence MACD signals with 20-day
timed exits, 2007 – 2008 (hypothetical) (Click here for larger figure.)

All flavors of the MACD tested out roughly the same as each other and
only slightly better than the baseline for Position Traders.

See Table 31 for the complete results for position traders. Remember,
expectancies are a very rough guideline at this stage because we are using
timed exits.
(Click here for larger table 31.)

MACD Parameter Settings for Position Traders

The various settings for MACD Histogram came out about the same for
position traders. See Figure 85 – Figure 87. During 2007-2008, they all
did a little better than the baseline, perhaps saving a few losses during
these down years.

The real test of the trading strategies is adding exits to see if expectancy
improves by producing larger gains and smaller losses due to the MACD
Histogram signals. This is the topic of the next chapter.
Figure 85 - Varying parameter settings for MACD Histogram entry
signals 20-day timed exits, 1994 – 2004 (hypothetical) (Click here for
larger figure.)
Figure 86 - Varying parameter settings for MACD Histogram entry
signals 20-day timed exits, 2004 – 2007 (hypothetical) (Click here for
larger figure.)
Figure 87 - Varying parameter settings for MACD Histogram entry
signals 20-day timed exits, 2007 – 2008 (hypothetical) (Click here for
larger figure.)
(Click here for larger table 32.)
Swing Traders

MACD Entry Types for Swing Traders

Figure 88 - Figure 90 show the results of backtesting various MACD entry

tactics for swing traders. A Swing Trader capitalizes on short-term price
movements. This is modeled with 2-day timed exits in order to test entry
strategies.

The MACD line crossover and Appel’s histogram uptick consistently

under-performed the baseline. The MACD Histogram, as popularly
constructed, stayed above the baseline in two of three timeframes.

The clear leader of the MACDs for the swing trader is the combination of
price above 200-day moving average (MA) and MACD Histogram uptick,
which beat the baseline in all three test periods. This is consistent with
other backtesting experiments where the addition of the 200-day MA
screen improved the win rate.
Figure 88 - Win rates of non-divergence MACD signals with 2-day
timed exits, 1994 – 2004 (hypothetical) (Click here for larger figure.)
Figure 89 - Win rates of non-divergence MACD signals with 2-day
timed exits, 2004 – 2007 (hypothetical) (Click here for larger figure.)
Figure 90 - Win rates of non-divergence MACD signals with 2-day
timed exits, 2007 - 2008 (hypothetical) (Click here for larger figure.)

Table 33 summarizes the MACD entry type results for swing traders. Win
rates might improve by using MACDH with fast (6,19,9) settings and a
200-day MA filter to get buy signals.
(Click here for larger table 33.)
MACD Parameter Settings for Swing Traders

The MACD Histogram parameter test results (Figure 91 - Figure 93)

suggest that swing traders might be better off with faster settings. In all
our cases, the fastest setting tested — 6,19,9 – did better than the others
tested and beat the baseline for swing traders.

Figure 91 - Varying parameter settings for MACD Histogram entry

signals 2-day timed exits, 1994 – 2004 (hypothetical) (Click here for
larger figure.)
Figure 92 - Varying parameter settings for MACD Histogram entry
signals 2-day timed exits, 2004 – 2007 (hypothetical) (Click here for
larger figure.)
Figure 93 - Varying parameter settings for MACD Histogram entry
signals 2-day timed exits, 2007 – 2008 (hypothetical) (Click here for
larger figure.)
(Click here for larger table 34.)

This chapter tested several different MACD entry styles and parameter
settings other than divergences.
Better results came from taking the earliest buy signal – the MACD
Histogram uptick. Adding a 200 day MA to screen for stocks in an
uptrend improved the win rates for active investors and swing traders.
Overall, position traders didn’t get much benefit from the MACD in any of
the configurations tested. Swing traders did better with faster settings: 6-
19-9 performed well in these tests. Otherwise, the parameter settings did
not significantly affect the results.

Win rates were in the 50-60% range predicted by Mr. Appel, who also
attributes to the MACD larger wins than losses.

The next chapter on MACD exit strategies compares expectancies for the
different MACD strategies. It is wise to withhold final judgment until all
the data is in.
Chapter 9: Catching the Wiggles with
MACD Sell Signals
This chapter is the second in the series about non-divergence MACD
signals. The previous chapter explored buy signals and this one explores
the sell signals. It helps us find out if non-divergence MACD signals can
be profitable, and if so, which exit type out-performed the others.

In setting up the backtests for this chapter, I got mired in the nuances of
MACD exit signals. While the isolated events of MACD Histogram
ticking up/down and lines crossing up/down are objectively obvious, the
actual usage can be quite subjective, especially when using different
parameters for entries and exits.

For example, when you buy on a 12-26-9 MACD Histogram uptick, the
19-39-9 MACD Lines may still be above zero. Do you require them to dip
below zero and rise again before accepting an exit signal from them, or
not?

A human trader may not be consistent in cases like this, but instead wait
for a full cycle in strong markets and be quick on the trigger in weaker
times. The backtest forces consistency so we choose a rule and live by it
in all situations.

It’s tempting to try out many little twists during the backtests to eke out
more gains. However, any one special usage may be the result of chance
and not repeat in the future, especially when the stats come out very close.

However, a bigger payback probably comes from spending energy

exploring big hitters such as MACD indicator/price divergences and
different strategies to complement and compete with MACD.

Keep in mind while reading this chapter that even Gerald Appel, the
inventor of the MACD, recommends using other strategies along with the
MACD. Ultimately our trading strategies must do that too but first we test
each piece independently to find out how it might contribute to the overall
operation.

The goal of this chapter is to help you decide whether to incorporate basic
MACD strategies into your trading plan. In general, the MACD
divergence strategies discussed earlier in the book performed better.

Strategy Under Test

The goal is to characterize the performance of non-divergence MACD exit

strategies. To better understand the merits of each exit strategy, it helps to
isolate exits from the entry strategies by picking one entry to use
consistently throughout the test.

Unless otherwise stated, tests in this chapter were run with the MACD
Histogram Uptick with price above the 200-day moving average. Also
unless stated otherwise, all runs used the de-facto standard settings of 12-
26-9 (fast-slow-signal). This entry proved to have a decent win rate,
although it was not always the most reliable entry signal for all trader
types. For more details on how to correctly identify this entry see Chapter
8, Figure 75.

Basic Exits Tested

The first experiment in this report compared several exit signals given by
the MACD:

Symmetric to Entry: exit when the 12-26-9 MACDH ticks down and
is above zero

Slower Exit: exits when the 19-39-9 MACDH ticks down and is > 0,
(with a 12-26-9 entry)
Gerald Appel Power Tools: basic decision structure from Mr.
Appel’s book, uses Appel’s histogram but without indicator/price
divergences or other chart patterns
Lines Cross: exit when the 19-39-9 MACD lines cross down and are
above zero

These strategies are illustrated in Figure 94 - Figure 97 below.

MACDH_MA_Sym_12-26-9 Figure 94 - Strategy which buys when

MACDH 12-26-9 ticks up (and is below zero) and sells when MACDH
12-26-9 ticks down (and is above zero). MACD lines are ignored.
This shot shows the strategy taking 4 trades, buys denoted by
MACDH_buy and sells denoted by MACDH_MA_Sym. Dark green
lines connect profitable trades and red lines connect unprofitable
trades. Created with TradeStation. © TradeStation Technologies,
Inc. All rights reserved. (Click here for larger figure.)

MACDH_MA_Sym_19-39-9 Figure 95 - Strategy sells when MACDH

19-39-9 ticks down (and is above zero). 12-26-9 is in the middle of the
chart, and 19-39-9 is on the bottom. Created with TradeStation. ©
TradeStation Technologies, Inc. All rights reserved. (Click here for
larger figure.)
Figure 96 - Zoom in on backtesting strategy which buys when
MACDH 12-26-9 ticks up (and is below zero) and sells when MACDH
19-39-9 ticks down (and is above zero). Created with TradeStation. ©
TradeStation Technologies, Inc. All rights reserved. (Click here for
larger figure.)

Comparing Figure 94 and Figure 95 presents a surprise extra trade. Figure

96 zooms in to show that what really happens. Both strategies buy in at
the same time, about June 21 – as well they should, they are using the
same entry strategy.

The difference is that the slower setting of 19-39-9 shown in Figure 95

actually gives a sell signal early! This is because the slower MACD
Histogram has not yet had a chance to get below zero before the faster
MACD Histogram triggers the buy. Then the fast MACD Histogram has
another buy signal (with a better entry point).

In this example, both the faster and the slower histogram tick down at the
same time, showing no real difference in the exits. In fact, that happens
quite a bit. Visual inspection of several charts gives the impression that
the settings impact the zero crossings of the MACD Histogram, which are
the same as the lines crossing. This change in the settings doesn’t appear
to make much difference in the MACD Histogram uptick/downtick
timing. But don’t go by visuals alone, read backtest results for a larger
sample size.

Appel’s Power Tools

Figure 97 - Backtesting strategy based on Gerald Appel's Power Tools
book. Created with TradeStation. © TradeStation Technologies, Inc.
All rights reserved. (Click here for larger figure.)

The “Power Tools” book by Gerald Appel offers complex instructions for
trading with the MACD along with other techniques. In this test, we
isolate the basic MACD-related rules. The case where price diverges from
the indicator is not implemented in this run. This strategy uses the “Power
Tools” entry method, which selects different parameter settings for
Appel’s Histogram upticks based upon price being above or below the 50-
day Exponential Moving Average (EMA). Looking very closely, you can
also see that the parameter settings make a more distinct impact on
Appel’s Histogram than MACD Histogram.

In Figure 97 we can see the strategy is geared towards later exits by

waiting for the second time that the slower lines cross downwards. The
second exit signal is defined as the crossing after a pullback, which
follows the first crossing. A fail-safe exit rule sells if the price closes
below the 50-day EMA.

The trade-off for the Power Tools strategy is evidenced by the Oct
whipsaw as the strategy entered below the 50-day MA, only to be stopped
out when Appel’s histogram ticked down signifying that the MACD lines
reached lower than at the entry price - a rule in the book.

MACDH_MA_Lines_12-26-9
Figure 98 - Strategy which buys when MACDH 12-26-9 ticks up (and
lines are below zero) and sells when MACD 12-26-9 lines cross down
(and are above zero). Created with TradeStation. © TradeStation
Technologies, Inc. All rights reserved. (Click here for larger figure.)

The final strategy tested without stops is illustrated by Figure 98. The aim
of this strategy was to attempt to let winners run longer by waiting for the
MACD lines to cross downward while above zero. To avoid pre-mature
exits, that means waiting for the lines to go below zero before entering.
The side effects are shown above: no re-entry when the trend resumes in
Jul - Sept but a better enter point in the Oct timeframe by waiting for the
stock to be oversold as indicated by the lines under zero. Backtesting
results can illustrate the net effect of that trade-off.

Adding Stop Losses

The first set of runs above experimented with different ways to let profits
grow. The second section experiments with ways to limit losses. For this
series, the MACDH_MA200_Sym1939 served as the basic strategy and
three different kinds of stops were pitted against each other. The stop
types and parameter variations were:

%: 3% and 9% of price
ATR: 2 and 3 times the average true range (ATR)
Histo-stop: a downtick of the MACDH from the entry price
Rather than taking a fixed number of shares and almost unlimited risk as
we do with no stops, setting a stop gives an opportunity to have a fixed
risk on every trade by varying the number of shares. If we assume a
hypothetical $100,000 account and risk 1% of that account on each trade,
that amounts to risking $1000 per trade. The risk is considered the dollar
amount between the entry price and the stop. Then Size = $Risk / (Entry
price – Stop Price)

The benefit is that trades at different share prices now have equal impacts
to risk and reward.

Backtesting Setup Details

Markets: US Stocks and international stocks represented by ADRs on
NYSE, AMEX, NASDAQ including delisted tickers.

Time Periods: May 1994 - April 2008, divided into three samples to
prevent over-optimization.

May 1994 – April 2004 denoted by darker blue.

May 2004 – April 2007 denoted by medium blue
May 2007 – April 2008 denoted by light blue

Direction: Long Only

Entry Strategy: Enter any stock with buy signals as described above, and
when volume criterion is met (more than 500,000 shares daily). All entries
are next day via Market on Open orders.

Exit Strategy: Exit all stocks according to the signals described above.
Profit-taking is done next day via Market on Open orders. Stop losses,
where applied, are simulated as stop orders, which may be executed
intraday. This combination doesn’t exactly model a nimble trader who
may grab profits from an exit signal early in a day in which those waiting
for end-of-day miss out due to a sell-off before the close. It does model
end-of-day trading and investment though.

Sizing: Where no stops are involved, size is fixed at 1000 shares. With
stops, the risk amount was fixed at $1000 and the size computed as the
number of shares that would risk $1000 between the anticipated entry and
the stop.

Backtesting Engine: TradeStation version 8.6, Build 2525

Data Vendor: CSI. This data set was specially selected for accuracy after
extensive testing.

Non-Divergence MACD Exit Backtests Results

With exit strategies the key criteria are Expectancy and Maximum Adverse
Excursion (MAE), described in detail below. The number of trades
measured how often the strategy had an opportunity to trade. The average
hold time measures how long each trade went on. Between them you can
get an idea of how productively each strategy put its funds to use. Keep in
mind that the average hold times are just that – an average. Winning
trades tended to go on longer while losing trades were often shorter.

Basic Exits, No Stops

(Click here for larger table 35.)
The number of trades was far less than the baseline because the entry
strategy selectively picked times to enter. The MACDH_MA_Lines
strategy had far fewer trades than the others, and a longer average hold
time, showing that it stayed in winning trades longer. The
MACD_Power_Tools strategy probably stayed in winning trades the
longest but it had more trades and short average hold times when
repeatedly whipsawed while attempting to find the bottom in pullbacks
under the 50-day MA.

Win Rates

Graphs of win rates are shown in Figure 99 - Figure 101. No surprise that
adding exits pushed the win rates under the baseline for two of the
strategies. However, the real acid test of the exit strategies is expectancy
and MAE.

None of the win rates were so low as to make the strategy overly difficult
to follow. The lowest was 30%, which would average two losses for every
win, although strings of losses could go on much longer. You decide what
is acceptable for you.
Figure 99 – Win Rates 1994 – 2004 (hypothetical) (Click here for
larger figure.)
Figure 100 – Win Rates 2004 – 2007 (hypothetical) (Click here for
larger figure.)
Figure 101 – Win Rates 2007 – 2008 (hypothetical) (Click here for
larger figure.)

Expectancy

A large money-making trade can make up for several smaller losses and
allow a strategy with lower win rate to be more profitable. That’s why
expectancy is a very important metric for a complete strategy.
Figure 102 – Expectancy without stop losses (hypothetical) (Click here
for larger figure.)

At a glance we can see from Figure 102 that none of the basic exit
strategies tested demonstrated great profitability and in fact all results are
clustered around the breakeven point. The next two sub-sections explore
underlying reasons for the lackluster results and then attempt to improve
the results.

Review on How to Read the Expectancy Graph

The colorful graph above plots the expectancy for all of the exit strategies
under test plus the baseline. Each strategy occupies one space along the
horizontal axis and is in the same order as in the data table.

All the data for one strategy is aligned in a vertical column. As shown in
the legend, each time period has a particular shape to identify its data
point. For example, 1994-2004 is denoted with a diamond. If a shape is
not visible, it is hiding behind a larger shape, which had roughly the same
value. The vertical axis displays the expectancy. Negative expectancies
(unprofitable) are color-coded red. Notice that the red extends above the
zero line. That is not an accident. A strategy that comes out marginally
positive in hypothetical backtesting is unlikely to be profitable in real
situations. The colors gently fade to yellow to indicate caution. Then they
go on to green to signify strategies that were more profitable in backtesting
and have a chance at real-world profitability.

As the data points spread apart from the zero line, they reveal that
increasing risk goes with increasing returns. It’s relatively easy to find
rules and leverage that increases the expectancy in good times but it
usually comes at the expense of a greater negative expectancy when
conditions deteriorate. Look for strategies that go further into the
green zone and less into the red zone than others.

Risk as Measured by MAE

(Click here for larger table 36.)

The MAEs in Table 36 above give clues as to why the results weren’t
better. Huge adverse excursions make these strategies untenable. Note,
however, that MACD strategies still did much better than the baseline
MAE. The section on stops shows them doing even better. The Power
Tools strategy consistently controlled risk in the backtests as evidenced by
the smallest MAEs in each time period.

Figure 103 - High MAE for the MACDH Lines Strategy. Created with
TradeStation. © TradeStation Technologies, Inc. All rights
reserved. (Click here for larger figure.)

See Figure 103 for an example of how MAEs can get so large. First off,
the stock is high priced. Secondly, the strategy always buys a fixed
number of shares, in this case 1000. So it makes a huge position in just
one stock. Then the price goes against the position. While this gigantic
open drawdown is taking place, a triple divergence forms on both the
MACD lines and the MACDH. Still the drawdown grows to roughly 30%
before the stock bounces off the 200-day MA and reverses upwards. That
thrust carries price upwards to the breakeven point for this position and an
exit signal occurs slightly above. This example happens to have a happy
ending with a small profit but it could well have been devastating. It
illustrates the need to somehow get out of the position early with minimal
loss and be ready to re-enter at a more favorable time.
Select R-Multiple Distributions

The R-Multiple distributions of all the basic strategies look similar so only
one is presented here for reference. In Figure 104 we see the two tall
spikes containing trades that produced results in the range of -1R to 1R.

Note that only 14 times in 14 years did this strategy more than double the
money invested (see the tiny numbers above the positive bins in Figure
104 and Figure 106). This is characteristic of strategies without stops,
where all the money invested is considered at risk, and so 1R represents a
100% return on the investment.

Remember that R is the amount risked per trade, in our test $1000. R-
Multiple is the number of times that risk is returned, so +5R is $5000
profit while -1R is a $1000 loss. The trades are sorted into “bins” which
correspond to a bar on the graph.

The label of the bin is the mid-point of its contents so the 0.5 bar
represents all trades that returned between 0 and 1 R. Anything more
negative than -1 R means those trades gapped past their stops or jumped
the risk limits due to opening gaps.
Figure 104 – Results Distribution without Stop Losses 1994 – 2004
(hypothetical) (Click here for larger figure.)
Figure 105 - Zoom in on Figure 104 (hypothetical) (Click here for
larger figure.)
Figure 106 – Results Distribution without stop losses 2004 – 2007
(hypothetical) (Click here for larger figure.)
Figure 107 - Results Distribution without stop losses 2007 – 2008
(hypothetical) (Click here for larger figure.)

Summary of Basic MACD Exit Strategy Results

Overall, roughly breakeven, with slight positive expectancy in

favorable markets and slight negative expectancy in down years.
Huge losses due to putting the entire investment at risk at 1000 shares
per stock
Despite trying to slow the exits, still not staying with winners for the
full ride. The average hold time was closest to position traders.

Basic MACD Strategies without stops didn’t deliver a winning edge

because huge losses matched the gains and open losses caused blow outs.

Adding Stop Losses

The next set of runs explores the addition of stops to the strategy. In five
of the seven runs below, the basic strategy entered on MACDH (12-26-9)
uptick while price was above the 200-day MA and exited if the MACDH
(19-39-9) ticked down while above zero or the price dropped below the
200-day MA.

Various stops were added and the results compared. The sixth strategy
used the “Power Tools” entries and exits, with a 2ATR stop added. The
seventh strategy used the basic entry with an exit when MACD lines
crossed down and a 2ATR stop (chosen based on the results of the
previous runs with stops).
(Click here for larger Table 37.)
(Click here for larger Table 38.)
(Click here for larger Table 39.)

Expectancy

Let’s cut to the chase and examine expectancy first. In Figure 108 below,
we can immediately see that adding stops markedly changed the
expectancy. The general pattern seemed to be that during time periods
where the market rose (1994-2004 diamonds and 2004-2007 squares) the
expectancy crept into positive territory. However, in the time period
where the market headed down, the expectancy became clearly negative
(2007 – 2008 triangles).

Observe that as the stops got wider, 3% vs 9% and 2ATR vs 3ATR, the
range of expectancy actually got narrower. The most dramatic case was
the Histo Stop. In most trades, it was the closest stop, coming just at the
price where the MACDH would tick down again from the first uptick entry
signal. In this case, we see that it got better expectancy in the positive
situations while the negative went from bad to worse.

Two interesting points about the MACD Power Tools Strategy: First,
despite a lower win rate than the other strategies, it produced a higher
positive expectancy in rising market conditions. Secondly, the Power
Tools strategy compares favorably to the MACD Lines exits, as
demonstrated by similar expectancies in the relatively favorable conditions
of 1994 - 2007 but not as negative expectancy in the tough conditions of
2007 – 2008.
Figure 108 - Expectancy of MACDH MA with Stops
(hypothetical) (Click here for larger figure.)

Even with stops, the basic MACD strategies lost money in down markets
and didn’t push solidly into the profit zone during favorable times.
Therefore, it’s probably not going to be profitable to blindly follow the
MACD signals. Since they did have some situations with positive
expectancy though, they might fit in as part of a larger strategy.

Notes on the Effects on Expectancy of Limiting Risk with Stops and

Sizing

The baseline strategy as well as the symmetric strategies without stops

produced results clustered around the breakeven point. With stops, the
difference in expectancies between the time periods was accentuated.
Clearly the profits in good times made a nice positive jump. The negative
expectancies (losses) in 2007-2008 also became more extreme as stops
were added. The extreme jump was largely an effect of the change in
sizing and risk calculations. It illustrates how reducing your risk can
improve profits when the right strategy is employed for the market
conditions. However adding stops is not a panacea for a mismatch
between strategy and market conditions as shown by the losses in 2007-
2008.

Win Rates

Graphs of win rates are shown in Figure 109 - Figure 111. Adding stops
pushed the win rate under the baseline in every case. One concern is that
the win rates were consistently below 25% for the histo-stops, which may
make this strategy too demoralizing for most people.
Figure 109 – MACDH MA Stop Loss Win Rates 1994 – 2004
(hypothetical) (Click here for larger figure.)
Figure 110 - MACDH MA Stop Loss Win Rates 2004 – 2007
(hypothetical) (Click here for larger figure.)
Figure 111 - MACDH MA Stop Loss Win Rates 2007 – 2008
(hypothetical) (Click here for larger figure.)

Risk as Measured by MAE

This section shows what we got in return for the lower win rates with
stops. Adding stops trimmed the MAE down considerably. Certainly, the
per-share MAE decreased after adding stops. The MAE in total dollars of
the positions – Viability — was still surprisingly high.
(Click here for larger Table 40.)
(Click here for larger Table 41.)
(Click here for larger Table 42.)

Remember, these runs held to a $1000 maximum risk by sizing the

position according to the distance from the stop. No doubt a few trades
gapped past the stop for a slightly larger loss, but an open loss of over
$32,000 for the histo-stop was outrageous. Probing deeper revealed two
difficulties which actually applied to all stop strategies tested:

1. Entry Price Below the Stop. The decision to enter a trade is based
on today’s end-of-day data but the entry takes place the next day at
the open. The trouble is that sometimes the stock price gaps down
below the stop overnight, before we even had a chance to get in. The
automated strategy goes awry here because it doesn’t detect that
situation, enters anyway, and is immediately stopped out. While this
didn’t account for the biggest MAEs, it’s still not realistic because a
human would thank their lucky stars and either scratch this stock or
recalculate a new stop before entering. An automated strategy
operating in real time would also need to be able to detect and correct
this situation. Having an entry below the stop is an artifact of
historical backtesting.
2. Close Stop Makes Oversized Position. Simply sizing to the stop
means that a close stop on a high priced stock can make for a wildly
large position. Then a gap past the stop can be devastating.

Special post-processing was done to address these difficulties. The

premise is an account size of $100,000 with no margin. Risk is limited to
roughly $1000 per position, and the overall cost of the position can’t be
over the account size of $100,000. Trades from the original backtest that
didn’t meet these criteria were thrown out. The results are not shown here
but the upshot is this: Removing the oversized and scratched trades didn’t
make much difference to the expectancy or the win rate but it did bring the
maximum MAE down under $1200.

Select R-Multiple Distributions

Figure 112 - Figure 115 display the R-Multiple distributions for the basic
MACDH uptick/downtick strategy with 2ATR stops. With stops, 1 R-
Multiple is $1000, the nominal dollar amount risked between entry and
stop. Then 2 R-Multiples is $2000 gained, 3-R is $3000, and so on.

Compared to the distributions with no stops, the distribution graphs with

stops were quite spread out (Figure 112, Figure114, and Figure 115). Also
notice how the red bins down to -2R were larger than the green bins up to
2R. However, at the higher R-multiples, the profitable green bins won
out. That says the money was made by a relatively few large gains.
Figure 112 - Results Distributions with Stops 1994-2004
(hypothetical) (Click here for larger figure.)

In Figure 113 below we zoom in on the action between -1R and +1R for a
close-up of many more small losses than small wins. The taller spike to
4088 losses at label -0.95R corresponds to trades that hit their stops. The
trades that lost less than that either had a better-than-expected entry price
or got an exit signal that took them out below the breakeven point.
Figure 113 – Zoom in on Results Distributions with Stops 1994-2004
(hypothetical) (Click here for larger figure.)
Figure 114 - Results Distributions with Stops 2004-2007
(hypothetical) (Click here for larger figure.)

Figure 115 - Results Distributions with Stops 2007-2008

(hypothetical) (Click here for larger figure.)

Summary of MACD Exit Strategy with Stops

Stop losses reduced risk and accentuated expectancy

Expectancy improved in rising markets versus not using stops
Expectancy decreased in tough markets even with stops
Basic MACD signals need to be combined with another strategy that
identifies market direction

Conclusions for Non-divergence MACD Exit

Simply applying the MACD indicator “as-is, out-of-the-box” didn’t

consistently produce profits in historical backtesting, even when combined
with the 200-day MA. These tests can give no reason to expect it will in
the future either. Without stops, in favorable markets, the MACD
strategies tested delivered positive expectancy but underwhelming profits
and overwhelming MAEs. In difficult markets, the expectancy ranged
from bad to worse.

Stop losses improved profitability as measured by expectancy and helped

to control MAE or open drawdowns. Given the huge MAEs of the basic
exits, some means of limiting losses appears essential to survival in the
markets.

With stops and in favorable markets, MACD did somewhat better than the
baseline. Given a means to identify favorable markets, MACD and
MACDH could ultimately prove useful in timing entries and exits a little
better than chance.

No single MACD exit strategy did the best in all market conditions. In
general, tighter stops and slower exits produced more extremes in
expectancy – more profits and more losses. Histo-stops had the highest
expectancy at the expense of the highest MAE and lowest win rate. Power
Tools moderated the MAEs and still had nearly as high expectancy.

Active Investors

MACD is often used with a longer term outlook, typically associated with
active investors who look to stay in a stock for a year or more. However,
from these tests we see that basic MACD signals didn’t keep an investor in
positions all that long when applied to a daily chart. Even when we
specifically tried to hang in longer by using the MACD lines or skipping a
signal per Power Tools, the average trade length was under 20 days.
Perhaps applying MACD to weekly charts would yield positions that
lasted over a year.

Position Traders

Position trading – grabbing one leg of a longer-running trend - appeared to

be the sweet spot of MACD Histogram on a daily chart. Most winning
trades had a duration measured in weeks.

Swing Traders
Waiting for MACD Histogram to tick down may try the patience of a
swing trader, even as it cuts out before the full extent of the price move.
Skipping a signal or waiting for the lines to cross takes even longer (and
still might not grab the full move). The tighter stops – 3%, 2ATR, and
histo-stop – all had shorter average hold times and might be a better match
for the swing trader’s temperament. Perhaps faster settings would speed a
swing trader in and out a little faster.

How to Apply Non-Divergence MACD Signals

First consider that the backtesting data in this chapter shows that you
probably won’t rack up profits by simply following the basic signals of
MACD Lines or MACD Histogram – even with the 200-day MA as a
guide. However, you may still want to bring MACD up on a chart. If you
have other reasons for believing the stock is of interest, MACDH and the
200-day MA may give you better entry and exit timing than pure chance.
Chapter 10: The Missing Link With
EMAs and MACD
This chapter backtests a multi-faceted strategy: the 12-bar and 26-bar
Exponential Moving Average (EMA) crossover. It is not only an
important range of moving averages but also the moving averages that
form the basis of the MACD.

In fact, the crossover of these two moving averages corresponds to the

MACD line crossing its zero axis, which is often referenced as a trading
signal. I call it the Missing Link because the ties between the MA and
MACD are often overlooked.

After reading this chapter, you should thoroughly understand this strategy,
not only as a buy signal but also as a complete system with symmetric sell
signals. To get a taste of how stops can reduce risk, an additional test run
with a stop loss is included. We take the test even one step further by
applying a 200-day moving average as a filter to select only those cases
where the market has already shown some strength.

Whether or not you elect to add this simple strategy to your trading
arsenal, you can learn and improve your skills by understanding it in
detail.

Strategy Under Test: Moving Average Crossovers

The strategy tested in this report derives its buy signal from the 12-day
EMA crossing up through the 26-day EMA as shown in Figure 116. We
first test with the usual timed exits of 2 days, 20 days and 200 days to
represent swing traders, position traders, and active investors.

To backtest the EMA crossover as a complete strategy, we take the sell

signal when the 12-day EMA crosses down through the 26-day EMA. See
Figure 117.
Figure 116 – Entry testing for 12/26 EMAs crossing up with timed
exits. Note that EMA crossover is the same as blue MACD line
crossing zero. Created with TradeStation. © TradeStation
Technologies, Inc. All rights reserved. (Click here for larger figure.)

Figure 117 - Symmetric entry/exit testing for 12/26 EMAs crossovers.

Note that EMA crossover is the same as blue MACD line crossing
zero. Created with TradeStation. © TradeStation Technologies, Inc.
All rights reserved. (Click here for larger figure.)

We attempt to limit risk by applying stop losses, which are calculated to be

two times the (20-day) average true range (ATR) of the stock on the day of
the buy signal. This stop loss is held constant throughout the life of the
trade.

Seeing how the results of the previous tests came out, a 200-day moving
average filter was applied to give us a backtesting run which attempts to
distinguish better opportunities for this long-only strategy to initiate a
trade.

In this case, it only buys if the price is above the 200-day moving average
(and the buy signal occurs) and sells if price falls below the 200-day
moving average (or the sell signal occurs). Since this was an after-
thought, it was only tested with real exits rather than timed exits.
Figure 118 - TradeStation screenshot illustrates how trades are
filtered out when price is below the 200-day MA. Created with
TradeStation. © TradeStation Technologies, Inc. All rights
reserved. (Click here for larger figure.)

Backtesting Setup Details

Markets: US Stocks and international stocks represented by ADRs on

NYSE, AMEX, NASDAQ including delisted tickers.

Time Periods: May 1994 - April 2008, divided into three samples to
prevent over-optimization.

May 1994 – April 2004

May 2004 – April 2007
May 2007 – April 2008

Direction: Long Only

Entry Strategy: Enter any stock with buy signals as described above, and
when volume criterion is met (more than 500,000 shares daily). All entries
are next day via Market on Open orders.

Exit Strategy: Exit all stocks according to the signals described above.
Profit-taking is done next day via Market on Open orders. Stop losses,
where applied, are simulated as stop orders, which may be executed
intraday. This combination doesn’t exactly model a nimble trader who
may grab profits from an exit signal early in a day in which those waiting
for end-of-day miss out due to a sell-off before the close. It does model
end-of-day trading and investment though.

Sizing: Where no stops are involved, size is fixed at 1000 shares. With
stops, the risk amount was fixed at $1000 and the size computed as the
number of shares that would risk $1000 between the anticipated entry and
the stop.

Backtesting Engine: TradeStation version 8.6, Build 2525

Data Vendor: CSI. This data set was specially selected for accuracy after
extensive testing.

MACD MA Crossover Entry Backtest Results

Throughout the reporting of results, this strategy is referred to as MAXO

12/26 as a short-hand for the full explanation of 12 and 26 exponential
moving average crossovers or MACD zero crossing.

Active Investors

BackTesting Report tests entries for active investors by

using a 200-day timed exit to assess the win rate after a longer term
holding period.

As shown in Figure 119, the win rate is very nearly the same as the
baseline for active investors, especially in the primary test period of
1994-2004.
Figure 119 – MAXO for Active Investors with 200 day exit
(hypothetical). (Click here for larger figure.)

Position Traders

A position trader seeks intermediate-term opportunities. To be successful,

a position trader gets in on a burst of action —perhaps one leg of a longer-
running trend — and gets out before the action fades. This is modeled
with 20 day timed exits in order to test entry strategies.

The Moving Average crossover strategy edges past the baseline as shown
in Figure 120.
Figure 120 – MAXO for Position Traders with 20-day exit
(hypothetical). (Click here for larger figure.)

Swing Traders

Swing traders are in it for a quick profit. We model this with a 2-day
timed exit. Our moving average crossover strategy did not get off to a
winning start for swing traders. It performed below the baseline for two of
three time periods.
Figure 121 – MAXO for Swing Traders with 2-day exit (hypothetical).
(Click here for larger figure.)
(Click here for larger Table 43.)

The average hold times for the 200-day tests wind up being a little less
than 200-day because some trades are still open at the end of the test
period.

MACD MA Crossover Exit Backtest Results

For the next backtesting runs, more realistic exits took the place of timed
exits. Besides the basic exit of moving averages crossing down, this
chapter also includes a run with a stop loss and another with a 200-day
MA filtering opportunities.
The number of trades was far less than the baseline because the entry
strategy was selectively picking times to enter. The use of the 200-day
MA filter strategy further reduced the number of trades.

Keep in mind that the average hold times are just that – an average.
Winning trades tended to go on longer while losing trades were often
shorter. Interestingly enough, the average hold times did come out near 20
days so the position trader’s baseline is used here for comparison.

(Click here for larger Table 44.)

Win Rates

Graphs of win rates are shown in Figure 122 - Figure 124. All exits
pushed the win rates under the baseline for two of the strategies. We
reserve judgment until seeing the expectancy.

Figure 122 – Win Rates of MAXO 12/26, 1994 – 2004

(hypothetical). (Click here for larger figure.)
Figure 123 - Win Rates of MAXO 12/26, 2004-2007
(hypothetical). (Click here for larger figure.)

Figure 124 - Win Rates of MAXO 12/26, 2007 - 2008

(hypothetical). (Click here for larger figure.)

Expectancy
Figure 125 – Expectancy of MAXO Strategies with Different Exit
Types (hypothetical). (Click here for larger figure.)

The benefits of a 200-day MA filter show on the graph in Figure 125 as a

general shift upwards in expectancy for both good times and bad. The
expectancy for 1994-2004 increased into the solid green (profitable) zone.
At the same time, the negative expectancy for 2007-2008 became less
negative. This is a clear demonstration of the positive effects of the 200-
day MA filter.

In contrast, it’s relatively easy to find rules and leverage that increases the
expectancy in good times but it usually comes at the expense of a greater
negative expectancy when conditions deteriorate. Because the 200-day
MA improved results in both good times and bad, we can see that it’s
useful. However, it’s not enough to turn around an unprofitable stretch for
long-only strategies like 2007-2008.

Risk as Measured by MAE

Each strategy in Table 45 does better than the baseline in terms of average
MAE. Only one case, using Symmetric exits in 2007-2008, gave a worse
max MAE than the baseline. The biggest bang-for-the-buck comes from
consistently applying stop losses. The numbers improved even more by
adding the 200-day MA filter to screen out stocks in a downturn.
(Click here for larger Table 45.)

The combination of stop losses and bull market filter topped the categories
of win rate, expectancy, MAE in this chapter.

Select R-Multiple Distributions

Figure 126 - Figure 128 show the R-multiple distributions for the strategy
of EMA crossovers with stops and 200-day MA filter.

The three time periods show similarities, which we’ll discuss by

proceeding from left to right. Stop losses prevented most (but not quite
all) of the really large losses. Nothing is perfect. The stops did generate
lots of small losses and we see that the -0.5R bar dominated – this
corresponds to losses up to the risk amount which means getting stopped
out of the trade.

The +0.5R bar is shorter. In the two profitable time periods, the small
losses outnumbered the small profits by about 3.5 to 1. However, in the
2007-2008 time period, that soared to over 6.6-to-1, which was the first
sign of unprofitability. Also 2007-2008 evidenced few large gains. It was
those large gains, which truly carried the strategy to profitability in the
first two time periods.
Figure 126 - Results Distribution for MAXO 12/26 with stops 1994 -
2004 (hypothetical). (Click here for larger figure.)
Figure 127 - Results Distribution for MAXO 12/26 with stops 2004 -
2007 (hypothetical). (Click here for larger figure.)
Figure 128 - Results Distribution for MAXO 12/26 with stops 2007 -
2008 (hypothetical). (Click here for larger figure.)

Conclusions for MAXO 12/26 Strategy

The 12/26 Moving Average Crossover, which is also the

MACD line zero crossing made a decent performance in good times as

measured by expectancy. However the strategy needs help to stay out of
unsuitable markets such as 2007-2008 where it dove squarely into the red.
Using a 200-day MA as a filter helped but doesn’t do it all.

Stop Losses clearly reduced risk as demonstrated by the MAEs. This

report touched on stop losses with the 2 Average True Range (ATR) stop.
Check out the previous chapter to see how various other stop losses
performed with related signals.

How to Apply MAXO 12/26

First consider that the backtesting data in this report showed that you may
lose money if you trade this strategy in market conditions that are not right
for it. However, you may still want to bring it up on a chart, examine it
with your favorite stock, or backtest it against a completely different
market.

A mechanical or automated strategy relying solely on this EMA crossover

is not recommended.
In Summary

Backtesting revealed MACD Divergence to be a promising strategy for

active investors and traders with an intermediate-term horizon.

Positive divergences performed better than negative divergences leading to

the conclusion that MACD Divergence might signal interesting times to
buy stocks and ETFs.

Looking across the whole stock market to identify times when many
stocks show positive divergences and/or few stocks show negative
divergences appears to be a worthwhile practice.

Non-divergence elements of the MACD such as the MACD Lines crossing

or the MACD Histogram upticks and downticks did not fare so well in
backtesting.

More work is needed to identify rules on when to sell. The next volume in
the Beat The Crash series will investigate exit strategies in detail. Stay
tuned…
Acknowledgements

Thank you to Joe Gruender Jr. of investorsHELP.net, Peter P., Ray

Senesac, and all the members of the MACD Divergence Alerts Service for
the suggestions and inspiration for the Divergence Count Indicator.

Many thanks to Jeff Parker for reviewing the original BackTesting Reports
that became the first edition of this book. I also appreciate the work of
Kathleen Callan and Gail Gauvin in formatting the materials. All errors
are mine.

Thank you to Elizabeth Schramm for posting the figures and tables on the
web site for readability. Thank you to TradeStation for all the screenshots.

I want to show my appreciation to you too, dear reader. Please see the
book companion site at www.truthaboutmacd.com to download bonus
materials such as videos, audio presentation, and more.

If you liked this book, please consider leaving a review at Amazon.

Thank you!