Stock Analysis in the Twenty-First Century and Beyond

By Xlibris US

Stock Analysis in the
Twenty-First Century and Beyond

For years, financial analysts have struggled with the fact that practically all the financial measures used to analyze corporate performance lack predictive power when it comes to forecasting the market performance of the companys stock. Numerous academic studies have documented and reported this lack of predictability. Correlation coefficients close to zero have been reported for the relationship between stock market performance and such critical financial measures as earnings growth, sales growth, price/earnings ratio, return on equity, intrinsic value (models based on discounted cash flow or dividends), and many more.

It is this disconnect between traditional financial measures and the performance of stocks in the marketplace that has led to the now-famous efficient market hypothesis, the cornerstone of modern portfolio theory. To accept the idea that the future performance of stocks is unpredictable is to say that nothing a company does will affect the future performance of its stock in the market, and that is absurd. It would be more accurate to say that everything a company does will affect the future performance of its stock in the market. The problem with this statement is that it makes the forecasting of future stock performance so complex that it removes it from the realm of human solution.

Confident in the belief that something other than chance and irrational investors determine future stock prices, several research groups around the world have started exploring the use of intelligent computer programs (programs that self-organize based on environmental feedback). Early results are very promising and have provided a glimpse of the economic forces described by Adam Smith as the invisible hand that guides economic activity.

Stock Analysis in the Twenty-First Century and Beyond describes the stock analysis problem and explores one of the more successful efforts to harness the new intelligent computer technology.

Many people mistakenly classify Artificially Intelligent (AI) computer systems as a form of quantitative analysis. There are two distinct differences between advanced AI systems and traditional quantitative analysis. They are (1) who makes up the selection rules and weighting and (2) what information is used to discriminate between good- and poor-performing securities.

In most quantitative systems, even in an advanced expert system form, humans make up the investment rules and mathematically derive the weightings associated with the rules. Computer systems that depend on outside human intelligence to program their actions are not inherently intelligent. In advanced AI systems, the computer makes up its own rules and weightings. The computer learns from examples of good- and poor-performing stocks and determines its own ways for discriminating between them. The procedures that are derived by the computer are often so complex that they defy human understanding.

In addition to making up its own rules, advanced AI systems look at corporate financial data differently. Just like in the human brain, where information is not stored in the brain cells but rather in the connections and relationships between cells, so too is corporate performance information stored in the relationships between financial numbers. Assessing the performance of companies is not so much in the numbers as it is in the connections between the numbers. Financial analysts recognized this early on and have used first-order relational information in the form of financial ratios for many years (price/book, debt/equity, current assets / current liabilities, price/earnings, etc.). Now with advanced AI systems, we are finally able to look at and evaluate high-order interrelationships in financial data that have been far too complex to analyze with less sophisticated systems. These then are the fundamental differences between what has been used in the past and what will be used in the future.

• Language: English
• Publisher: Xlibris US
• Release date: Jul 30, 2014
• ISBN: 9781499049077

Book preview

Copyright © 2014 by Thomas E. Berghage.

Library of Congress Control Number:   2014912720

ISBN:      Hardcover      978-1-4990-4905-3

                Softcover         978-1-4990-4906-0

                eBook              978-1-4990-4907-7

All rights reserved. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage and retrieval system, without permission in writing from the copyright owner.

Any people depicted in stock imagery provided by Thinkstock are models, and such images are being used for illustrative purposes only.

Certain stock imagery © Thinkstock.

Cover design by Maria Rosario Legarde, Xlibris

Rev. date: 07/25/2014

Xlibris LLC

1-888-795-4274

www.Xlibris.com

636338

CONTENTS

List of Figures

List of Tables

Preface

Background—The Basis of This Book

Section One—The Stock Analysis Environment

Chapter 1—Introduction

Chapter 2—Inherent Stock Market Risk

Section Two—Current Capability

Chapter 3—Sell-Side Analyst

Chapter 4—Buy-Side Analysts

Section Three—The Problem

Chapter 5—The Metric Problem

Chapter 6—The Human Element

Section Four—The Current Alternatives

Chapter 7—Traditional Approaches to Stock Analysis

Chapter 8—The Appropriate Use of Traditional Financial Analysis

Chapter 9—Portfolio Theory

Chapter 10—The Search for the Rational Investor

Section Five—The Science of Investing

Chapter 11—The Science behind Stock Analysis

Chapter 12—The Complexity Hypothesis

Chapter 13—Financial Hyperspace

Chapter 14—Neural Networks

Section Six—Building an Intelligent Stock Analysis System

Chapter 15—Basic Requirements of a Stock Selection System

Chapter 16—Fundamentals of the Stock Analysis System

Chapter 17—System Design

Chapter 18—System Performance

Chapter 19—Financial Quality Control

Chapter 20—The Future of Stock Analysis

Appendix A—Neural Network References

Appendix B—Neural Network Vendors and Consultants

Glossary

References

LIST OF FIGURES

Figure 2-1 Actual Daily S&P500 Returns Compared To Normal Distribution

Figure 2-2 Change In Standard Deviation Based On Investment Time Horizon

Figure 2-3 Effects of Time Horizon On Investment Return Statistics Distribution Proportion Relative To Standard Deviation

Figure 2-4 Distribution of Stock Returns

Figure 2-5 Probability Of Loss On Random Selections Held For One Year

Figure 2-6 One-Year Total Return (Percentage) For Stocks In The Ford Investor Service Database During The Period 1996-2000

Figure 2-7 Risk Associated With Market Capitalization And Various Hold Periods

Figure 3-1 Distribution Of The Buy and Strong Buy Recommendations Made By Analysts (Years 1992-1999)

Figure 3-2 Recommendation Categories Used By Analyst, 1992-1999

Figure 4-1 The flow of Dollars Into Domestic Index Funds (Information provided by the Vanguard Group)

Figure 4-2 Percentage of Equity Mutual Fund Assets Invested In Equity Index Funds

Figure 4-3 Percentage of Mutual Funds that beat the Performance of the Vanguard Index 500 Fund for the Trailing 10 years (O’Shaughnessy, 1984)

Figure 4-4 Russell Study Of 162 Institutional Money Managers

Figure 4-5 Callan Associates Study of 498 Institutional Money Managers

Figure 5-1 The Holy Grail Of Stock Analysis

Figure 5-2 The Best We Can Hope For

Figure 5-3 Illustration of the meaning of various levels of a Correlation Coefficient

Figure 5-4 The Concurrent Relationship Between Earnings Growth And Total Return (Years 1999-2001)

Figure 5-5 Relationship Between Discounted Future Earnings And Total Return (Years 1999-2001)

Figure 5-6 The Relationship Between Concurrent Growth Of Sales And Total Return (Years 1999-2001)

Figure 5-7 Relationship between Price/Earnings Ratios And Stock Total Return (Years 1999-2001)

Figure 5-8 Relationship Between The Price/Sales Ratio And Stock Total Return (Years 1999-2001)

Figure 5-9 Relationship between the Best Ten Metrics And Stock Total Return (Years 1999-2001

Figure 6-1 Factors Impacting Thinking and Decision Making

Figure 7-1 Linear Separator

Figure 7-2 The Problem With Linear Regression

Figure 7-3 Complex Stock Universe

Figure 8-1 Growth of ETF Managed Assets

Figure 10-1 Probability Density Functions Of Noise and Signal Plus Noise

Figure 10-2 Analyst Operating Characteristic (AOC) Curve*

Figure 11-1 AOC Curve for Data in Chapter 3

Figure 13-1 Picture of Hypercube

Figure 14-1 A Typical Neural Network

Figure 17-1 Investment Returns For NeuWorld Financial’s Neural Network System

Figure 17-2 The Interaction Between P/E And Market Capitalization

Figure 17-3 Relationship Between P/S Ratio And Market Capitalization

Figure 17-4 The Relationship Between High And Low P/S Ratios and Market Capitalization

Figure 17-5 Relationship Between Relative Market Strength And Market Capitalization

Figure 18-1 Monthly Variation In The S&P500 Over a 56 Year Period

Figure 18-2 S&P500 56 Year Returns

Figure 18-3 S&P500 & Eagle12 Fifty-Six Year Returns

Figure 18-4 S&P500 Volatility As Measured By the VIX Index Deviations From the Mean

Figure 18-5 Performance Figures For Eagle12, Berkshire Hathaway and the S&P500

Figure 18-6 Market Performance for Various Neural Network Selection Scores

Figure 19-1 Probability The S&P500 Will Have A Down Month (Based on 60 years of data)

Figure 19-2 Three Investment Quality Control Tools

Figure 19-3 Risk Control Through Diversification

Figure 19-4 Risk Reduction By Diversification In Space And Time

Figure 19-5 Portfolio Risk Reduction Using Three Quality Control Tools: Diversification In Space & Time, And Advanced Analytics

Figure 19-6 Deviant Statistical Patterns That Signal That Action Must Be Taken

LIST OF TABLES

Table 2-1 A Priori Probability of Loss For Various Sectors Of The Market Along With Potential Returns

Table 2-2 A Priori Probabilities of Loss For Various Market Capitalizations

Table 2-3 Probability That A Random Selection From Various Market Sectors Will Exceed A 5% Annual Return

Table 2-4 Risk Associated With Various Industry Sectors*

Table 2-5 Risk Associated With Market Capitalization And Various Hold Periods

Table 3-1 Recommendations By Wall Street Analysts

Table 3-2 Probability Of Loss And Mean Return For Analysts Recommendations

Table 3-3 Mean Returns For Different Time Horizons

Table 8-1 Stock Funds, Index Funds & SPDRs

Table 10-1 Matrix of Response Alternatives

Table 18-1 Estimated S&P500 Returns

Table 18-2 Estimated Eagle12 Returns

Table 19-1 Length Of Market Memory In Months

DEDICATION

To Kathryn and Tyler Carpenter who I Love dearly

PREFACE

I n the great stock analysis crusade, the search for the golden challis has been the search for that bit of information that will allow investors to identify winning stocks, stocks that appreciate substantially over some period of time, and the shorter the period of time, the better. The search for this valuable piece of predictive information has led analysts to scour annual reports, balance sheets, profit and loss statements, and even illegal insider information. Analysts have combed through every conceivable source of information that could give them a competitive edge. Even imperfect pieces of information that can increase the probability of selecting winning stocks have been sought with crusading vigor.

The problem is that in stock analysis, as with many complex problems, there is no single piece of information or even a group of independent pieces of information that is going to substantially improve the forecasting of stock appreciation potential. The golden challis of stock analysis is not in the data items themselves, but in the behavioral patterns created by the interaction among all the individual pieces of information generated by living, breathing corporations.

If you read the fundamental stock analysis texts used in most American colleges and universities, it is like stepping back in time more than eighty years. They are not that much different than what Benjamin Graham and David Dodd, the fathers of financial analysis, wrote in their first book Security Analysis published in 1934. It is little wonder that 80 to 90 percent of domestic money managers can’t beat the performance of a stupid market capitalization–weighted index like the S&P500.

Researchers in the academic community, in the new founded field of behavioral finance, have spent the last ten years documenting the limitations of human investors. I would now like to suggest that the human is the weakest link in the stock analysis process and needs to be replaced or, at the very least, supplemented with intelligent computer technology. Some have referred to these artificially intelligent (AI) investment systems as black box approaches to investing, with the term black referring to the unknown nature of what goes on in the box rather than its color. I would suggest that the real black box is the human mind and that we know a lot more about what is going on in our computer black boxes than we do about what is going on in the human black boxes. The thought process that leads an individual or group of individuals to fly a plane into the World Trade Center is far more confusing and a lot less rational than what goes on in the computer AI systems.

We may not fully understand how the AI computer analyzes and selects stocks, but we can be confident that its approach has been rational and based on the known facts. Its decisions have not been reached based upon some human’s misperception, misunderstanding, preconceived bias, or alternative competing objective or agenda. I would like to suggest that the stock analysis problem is so complex that it is beyond the capability of us poor humans and that AI computers are capable of performing a far more sophisticated level of analysis.

We are now about to embark upon that age-old struggle of man versus machine. We have all heard the stories before in the folktales of John Henry and the laying of railroad track and Paul Bunyan and the tree-cutting competition. More recently the real-life competition between Chess Master Garry Kasparov and IBM’s computer Deep Blue has captured public attention. Now we are faced with the economic challenge of Eagle12, the intelligent computer that analyzes stocks better than humans. Eagle12, although only in its third generation, has already evolved to a point where its investment portfolios are surpassing the performance of 90 percent of those formed by humans. Despite the fact that we have only just scratched the surface of this exciting new technology, it has developed to the point where it must be taken seriously.

In the area of stock analysis and portfolio management, the most dominant theory in the last four decades is the capital asset pricing model (CAPM) and its associated efficient market hypothesis. These concepts are being taught in every school of business across the country despite the absurdity of their underlying assumption of a rational, well-informed investor. The recent research efforts by the members of the behavioral finance community have cast so much doubt on this critical underlying assumption that it has seriously damaged the credibility of the model.

To reestablish the usefulness of this important building block in financial theory, we need to build and employ a totally unemotional, unbiased, rational investor, one that can efficiently assimilate new information, thoroughly evaluate it, and act on it in a totally rational way. Fortunately, we now have the technology to build such an entity, and we are in a position to reestablish the preeminence of the CAPM.

BACKGROUND

The Basis of This Book

I n 2002 I gave a presentation to the CFA Society of San Diego, where I had been president and board member for a number of years. Following this I was invited to speak at the Los Angeles CFA Society. The following year, I was invited to join the CFA Institute’s Speaker Retainer Program, where I participated for three years. It was a wonderful opportunity to go out and talk about the exciting new area of artificial intelligence and its future impact on the securities industry. The title of the presentation was Security Analysis in the Twenty-First Century and Beyond and is the basis of this book. The presentation has now been given to over one thousand CFAs in thirty-five CFA Institute chapters in the United States, Canada, Ireland, Scotland, and England.

The CFA Institute’s Speaker Retainer Program described the presentation in their brochure as follows:

Academic researchers in the new founded field of Behavioral Finance have spent the last ten years documenting the limitations of human investors and security analysts, but little has been done to provide the tools necessary to compensate for the identified weaknesses. Berghage, in his presentation separates financial analysis from security analysis and documents why the analytical tools developed for financial analysis do so poorly in forecasting security performance. He points out that the security analysis problem is so complex that it is beyond the capability of we poor humans; in the future, intelligent computers will be used to identify the subtle corporate behavioral patterns that are necessary for forecasting security performance. Berghage outlines the fundamental characteristics of the security analysis problem and talks about the intelligent computer technologies that are currently available for addressing them. He shows how computers learn and how they can detect emergent properties within corporate performance data that are far too complex for humans to perceive or understand. Finally, Berghage presents performance figures for intelligent computer systems that are currently in operation.

When I was first invited to give this presentation, I was a bit worried that it might be perceived as job threating to the CFA community that had spent three years of their life studying to become chartered financial analysts. The exams for the three levels of the CFA charter are very difficult and require great devotion for those that achieve the designation. The level of understanding of financial markets and investment instruments achieved through this program is the most thorough available anywhere in the world and is essential if we are ever going to make progress in adding value to client portfolios.

Following the warm reception I received from the CFA Society of San Diego, I was encouraged but thought that they might just be polite because I had been a past president and board member, so I thought I would sample individual opinions with a simple yes/no feedback questionnaire after each subsequent presentation. The following are the fourteen questions in the feedback questionnaire and the response I received:

1. Before the presentation, did you know that the probability of loss from a random selection from the equity markets was close to 50%?

Yes—50%         No—50%

2. Before the presentation, did you know that most of the metrics used in financial analysis have close to zero correlation with stock market performance?

Yes—48%         No—52%

3. Before the presentation, were you aware of how poorly the recommendations publish in Barron’s Research Report section actually performed?

Yes—48%         No—52%

4. Before the presentation, were you aware that only about 20% of mutual fund managers consistently outperformed the S&P500?

Yes—88%         No—12%

5. Have you ever attended a presentation on intelligent computer technology?

Yes—65%         No—35%

6. Before the presentation, have you heard of Neural Networks?

Yes—30%         No—70%

7. Before the presentation, were you aware of how Neural Networks operate? *

Yes—30%         No—68%

8. Before the presentation, were you aware that Neural Networks were being used to analyze equities and manage investment portfolios?

Yes—36%         No—64%

9. After the presentation, do you now feel that you can explain to someone how a Neural Network operates? *

Yes—74%         No—24%

10. After the presentation, do you now feel that you can explain to someone what emergent properties are? *

Yes—64%         No—35%

11. After the presentation, do you now feel that machines can solve some problems that humans cannot? *

Yes—89%         No—8%

12. Do you feel that you learned something from this presentation?

Yes—98%         No—2%

13. Would you be interested in learning more about intelligent computer technology?

Yes—95%         No—5%

14. Would you recommend the CFA Institute have workshops on intelligent computer technology at their annual meetings?

Yes—92%         No—8%

*Not everyone answered every question.

It is obvious that I did not reach everyone in the audience with my message, but I was encouraged by the positive responses I received and the willingness of the CFA community to embrace new ideas. From the answers to question 11, it is obvious that I need to do a better job of demonstrating the power of artificially intelligent systems. When 8 percent of the responders don’t think that machines can outperform humans on some tasks, I have not done my job.

I was especially pleased with the answers to question 12, and it is because of these responses that I have been motivated to write this book. Most of the concepts covered in the presentation are dealt with in the following chapters with the exceptions of the machine learning demonstrations and videos that were used. Many of the figures used in this book are from the original presentation and are consequently a little dated, but the principles are the same, because the financial community is very slow to change and the analysis techniques used today are still based on the procedures developed by Graham and Dodd back in 1934.

Before leaving this section, I should tell you a little bit about my background and why I felt qualified to attempt these presentations and the writing of this book. I served as a navy research psychologist working on human performance problems in advanced military systems for twenty-four years. After the navy made the mistake of sending me to San Diego, California, for four years of duty, they wanted me to return to Washington, DC (Disneyland East), to the R&D command. By that time, my family had been exposed to the wonderful