Systems Approach To Security Analysis
Thomas E. Berghage
The word “productivity” has a rather recent origin. Frederick W. Taylor (Kanigel, 1997) introduced the term in the late 1800’s, but it wasn’t picked up by most dictionaries until the turn of the Century. Taylor focused on “how” best to perform a task and as Peter Drucker (1992) points out, he never ask “what” task should be performed, or “why” a task should be performed. All three aspects of the problem need to be addressed if we are going to truly advance the process of analyzing securities. Any attempt to reengineer or modernize security analysis must look at more than just the actual stock analysis process; it must look at the entire financial organization, its management, organization, training, procedures and finally the analysis process itself.
A lot of what has been passed off, as security analysis in the past was nothing more than marketing support for the investment banking and market making activities of brokerage firms. To reform and modernize the security analysis process we must start at the top. Does the organization really have a “Chinese Wall” between its research department and the rest of its Operations? Are analysts really free to write what they think or does the possibility of losing one’s job constantly hang over their head? These are basic questions that have to be answered before you can even start talking about improving the analysis process. Even if you have the best analysis possible but can’t communicate the results you have accomplished very little.
Perhaps the best training currently available for financial analysts is the CFA program offered by the Association For Investment Management And Research (AIMR). Some have accused the author of being too critical of AIMR and its educational program in the past. On the contrary, I feel their program is the best one currently available and should be the starting point for anyone wishing to work in the field. As a former president and board member of the San Diego Chapter of AIMR I enthusiastically endorse the AIMR program. My problem is that it is just a starting point and that it does not include information and training on the recent advances that have been made in the information sciences. We are arming our financial analysts with M1 rifles while the war is being fought with laser guided missiles and smart bombs.
Task Analysis is a process or method of identifying, describing and analyzing each element in a operational system. It has been heavily used in the design and development of the weapon systems now employed by our military and is now a fundamental process used in setting up just about any large scale manufacturing operation. In fact now a days, just about any operational system, whether it produces goods or provides services, involves the use of both humans and machines and can benefit from a detailed task analysis.
Anytime you establish or modify an operational system, whether it’s in a production line or the analysis of securities, you need to step back and take a look at the total operation from a systems perspective. You need to look at the inputs to the system, the processing that take place within the system and the outputs that are produced, along with the outputs that are desired. You need to look at what role humans play in the system and what roles machines play, where errors are produced and where quality is being compromised. Automation can do a lot for improving system performance, but anytime you add machines to a system you introduce man-machine interfaces that can have unexpected consequences. Because of the importance of these man-machine interfaces they become the main focus for much of what goes on in a task analysis.
To start with we need to define and understand what we mean by the term, “task.” A task is a unit of work, a set of related actions that change or verify a system state. The purpose of task analysis is to obtain explicit, consistent data about the performance of each task, operation, or decision made within a system. The product of task analysis is a database that can be used in the evaluation of each system task, whether accomplished by machines or humans. Each task involves a set of inputs, actions or decisions, and a set of outputs. All three aspects of the task need to be evaluated and it is best done in reverse order. Look at the outputs first and determine if they are what are desired.
We have already seen in Chapter 2 and 3 that the outputs from the security analysis systems currently being used are not producing the levels of performance that the investment community desires or expects. The probability of loss associated with “buy” recommendations is unreasonably high and totally unacceptable. The inability of the security analysis system to discriminate between good and poor investment opportunities and its failure to add value to the investment process means that we need to work our way through the system to discover the source of the errors.
We can identify at least four aspects of the security analysis process; (1) identification of meaningful asset characteristics, features or properties that are measurable and have predictive value, (2) having a fast efficient communications channel for delivering the data, (3) having appropriate analysis techniques, that discriminate and present information in a form that is useable, and finally (4) having objective procedures for interpreting results and structuring the actions to be taken.
Having meaningful asset characteristics, features or properties mean that we must have an accurate reliable means of describing the organization and its performance. Characteristics that are meaningful for the task to be accomplished, whether it is forming an index or finding securities that will produce positive investment returns. Current accounting figures fall far short of providing the necessary measures that are predictive of future market performance. They may provide an approximation of the activities taking place in the company, but they are poorly related to market performance of the company’s securities.
Analysts have felt in the past that corporate performance as measured by standard accounting measures could be used to establish the fundamental value of a company (intrinsic value). Once established this intrinsic value could then be used to determine if the company’s stock was over valued or undervalued. The logic and theory behind this idea is very appealing. The only problem is, it doesn’t work; at least, not for investment time horizons out to two or three years.
As originally proposed the process of arbitrage was supposed to keep a company’s stock price close to its fundamental or intrinsic value. Any deviations from this “true” value were to be quickly corrected by the action of rational investors. The problem with this line of thought is that neither the rational human investor nor the true value of the company exists in reality. Behavioral scientists have destroyed the concept of the rational investor with the numerous research studies we outlined in Chapter 9. Irrational exuberance and pessimism exists for extended periods of time and renders the idea of efficient markets and the value of arbitrage investing almost useless.
One of the main reasons that arbitrage investing fails to produce the desired results is that its base of comparison, intrinsic value, is such a nebulous concept. Defining the true value of a company at any point in time is an impossible task and exists only in the minds of economic theorists and model builders.
We may think that we know what the intrinsic value of a company is, but like everything in a free economic system, a company is only worth what someone is willing to pay for it. The value of a company is found in the eyes of the investor and that changes minute-to-minute and hour-to-hour. Trying to employ arbitrage investing when the standard for evaluation is constantly changing is almost impossible.
If we have no fixed standard for determining if a company’s stock is over valued or under valued, how are we to evaluate and compare investment options? The answer is found in what are called emergent properties or corporate behavioral patterns. An analysis system that can detect and evaluate the patterns of corporate behavior can be used to separate good and poor performing companies and can also be used to discriminate between good and poor performing securities.
Identifying corporate behavioral patterns has been impossible until just recently. The development of sophisticated pattern recognition technology had to wait for the development of semiconductors, fast computer hardware, databases, and advanced software programs. It also required that investigators break out of their three dimensional linear world and start to think about problem solutions in nonlinear hyperspace.
Before we can search for emergent corporate behavioral patterns in hyperspace, however, we need to have some assurances that the corporation’s characteristics and performance are accurately reflected in its reported financial numbers. Although the evaluation of corporate behavioral patterns is much more robust than traditional financial analysis it is still dependent on the honesty of corporate management and their accurate reporting of corporate performance.
Unlike traditional financial analysis that focuses on the individual reported numbers and first order relationships (ratios), analysis of behavioral patterns looks at high order interrelationships and searches for what are called emergent properties. These high order interrelationships and emergent patterns are less susceptible to distortion by corporate management and it is far more difficult to miss lead the investing public if the analysis system is looking at more than a few fudged numbers.
Schilit’s (2002) outlines many of the accounting shenanigans used by corporate managements to alter the appearance of their firm’s performance. There is no question that corporate managements employing these shenanigans are trying to present their firms in the best possible light and in so doing are distorting the true picture of operational performance. They are deceiving the investing public and destroying the effectiveness of traditional financial analysis.
Recording revenue when future services remain to be provided
Recording revenue before shipment or customer’s unconditional acceptance
Recording revenue although customer is not obligated to pay
Selling to an affiliated party
Giving customer something of value as a quid pro quo
Grossing-up revenue
Recording sales lacking economic substance – side agreements
Recording cash received from lender as revenue
Recording investment income as revenue
Recording as revenue supplier rebates tied to future required purchases
Release revenue improperly “held back” before a merger
Recording gains selling assets recorded at deflated book value
Including investment income or gains as revenue
Including investment income as reduction in operating expenses
Creating income by reclassification of investment gains
Capitalizing normal operating costs, particularly if recently changed from expensing
Changing accounting policies and shift current expenses to an earlier period
Amortizing costs too slowly
Failing to write-down or write-off impaired assets
Releasing asset reserves into income
Failing to record expenses (and related liabilities) when future obligations remain
Reducing liabilities by changing accounting assumptions
Releasing questionable liability reserves into income
Creating sham rebates
Recording revenue when cash is received, yet future obligations remain
Creating reserves and releasing them into income in a later period
Improperly holding back revenue just before an acquisition closes
Improperly inflate amount included in special charge
Improperly write off in-process R&D costs from acquisition
Accelerating discretionary expenses into the current period
To overcome these deliberate attempts to mislead, security analysts are going to have to employ some rather sophisticated pattern recognition technology that looks at the entire corporate picture rather than a few isolated accounting facts and financial ratios; a technology that does not use human evaluation techniques or analysis rules; a technology that is dynamic and difficult or impossible for corporate management to decipher or circumvent.
The distortions of the corporate behavioral pattern caused by the shenanigans outlined by Schilit (2002) would, for the most part, be detected by advanced non-linear analysis systems. It is interesting to note that during the recent market down turn that was partially caused by the crises of confidence regarding corporate reporting, none of the companies involved (Enron, Global Crossing, ImClone, WorldCom, and others) were recommended as “buys” by the intelligent computer system described in Chapter 13. Despite the distortions caused by off balance sheet financing and the questionable income figures provided by corporate management, the corporate behavioral patterns produced by these companies keep them from being recommended for purchase.
Regardless of what kind of system is used to evaluate corporate numbers its performance is certainly enhanced by truthful reporting by corporate management. The accurate quantification of corporate characteristics and performance is essential if we are to ever advance the science of security analysis. No system can completely overcome the determined efforts of a dishonest management team, but we can certainly make it a lot harder to fool the investing public.
Before leaving the topic of corporate characteristics and how to describe and report them we need to deal with one more issue, that of future change. The way we describe and analyze corporations cannot be ridge and static because markets and economies change. Market sectors change and the industries of the past are not necessarily the industries of the future. Financial variables that are important for analyzing immerging technologies change as the companies and industries mature. Things that were important in the early life of the steel industry have changed as new technology has been introduced and the industry evolved. As the automobile has changed from an engineering wonder into a marketing commodity the information needed to forecast future security success in the auto industry has changed.
The government also has a major impact on the value of financial information. Corporations have available to them three different vehicles – dividends, capital gains and debt – through which they deliver returns to investors. Historically each of the return delivery vehicles has been taxed at different rates. A change in the tax code alters the after tax return delivered by each of the three different return delivery mechanism.
As we go to press the Congress is considering changing the double taxation of dividends. Such a change could alter the way returns are delivered to investors and alter the predictive value of dividend payout, capital gains, and debt. Security analysis needs to have the ability to adapt to all of these potential changes.
The one area in which we have made tremendous progress is in getting information to the investing public in a timely fashion. Computer networks, the Internet, and new software have made corporate financial information available at speeds only dreamt about a few years ago. Information that was available to only a handful of large investment corporations is now accessible to just about anyone with a computer and access to the Internet. In fact, improved communications has been so effective that it is providing more information that a human can mentally digest. It has put investors on information overload and could actually lead to a denigration of investment decision-making. As we indicated in Chapter 3, a lot of the financial information out there is nothing more than noise in the system and by dumping more of this worthless information into the investment decision process we run the risk of making the situation worse rather than better. Unless you have an analysis system that can effectively separate the signal from the noise, the additional information just increases the noise and the complexity of the analysis task.
There have been a few positive things that have been occurring in the communication process as of late. Some of the government rules regarding corporate reporting have been changed for the better and the time delay in reporting things like insider trading have been reduced. By shorting the time between occurrence and reporting we might even improve the predictability and usefulness of some information items.
By and large, however, the communication channel appears to be working rather well and it is the part of security analysis system that needs the least amount of work. The same cannot be said for the number crunching part of the system that actually does the analysis once the information is received.
Once we have received honest measures of corporate performance we need to evaluate the measures to determine if they have what is known as predictive validity, in other words are the measures actually useful in forecasting the future or are they only useful for describing historical events.
Traditional financial analysis is the study of the past while security analysis is focused on the future. To become a futurist rather than a historian you need new tools and technology. You need to move your thinking out of three-dimensional linear space into "K" dimensional non-linear hyperspace
Forget about all the noise and chatter in the news media, and the bantering and misinformation in the Internet chat-rooms. The real story regarding future stock performance is more likely found in the complex patterns among reported performance and financial figures. These patterns are too complex for humans to detect. To discover and understand these significant relationships requires the use of sophisticated intelligent computers that learn and develop their own rules and forms of analysis.
The research into these complex multi-dimensional domains has been restricted by our computational capability. When humans were the calculators, it was almost impossible to take on the analysis of complex relationships, and graduate students were taught to avoid research problems with more than one or two simultaneous variables for fear of not being able to make sense of the interactions that might exist. It was not until the advent of the computer that we could even start to think about adaptive, non-linear systems and parallel processing, the kind of systems necessary for dealing with complex problems.
In evaluating complex problems there are some common characteristics or attributes that seem to be important. Most of the problems in the complex domain involve the analysis and understanding of multiple simultaneous variables. Humans do not deal well with multiple inputs, especially when they all arrive at the same time. To make the problem worse, most of these variables do not act in isolation, but rather interact with each other and totally confound the picture for the human analyst. On top of all of that, a good many of these variables are non-linear in their effect and defy normal statistical linear analysis. This is what has led many in the academic community to advocate the use of Chaos Theory. To complicate the problem even further, many of the variables used in security analysis are fuzzy in nature, and do not have crisp black and white impacts on the problem. The influence of these variables is vague and is therefore often referred to as being “gray or fuzzy.” To totally take complex problems out of the realm of human comprehension, most of the problems in this domain are dynamic and constantly changing. Assuming one could actually solve one of these complex problems, as soon as you did the problem would have changed and you are back to square one.
Because of the intractable nature of these complex problems scientists have avoided addressing them directly and have opted to include their effects in their statistical error term. It was not until the advent of the computer and some of our new non-linear software solution technologies that we were able to address some of these problems. Now that we have the tools to explore the complex domain we should see rapid progress on many scientific fronts; fronts that use the computer, not as a number crunchier, but as a way to deal with interrelationships and patterns.
Security analysis requires that we have the ability to deal with multiple, interacting, nonlinear, fuzzy, and dynamic variables. To understand the analysis problem we need to address each of these problem attributes.
Few, if any, financial problems can be solved by only looking at a single variable; the complexity of the problem dictates that multiple variables be addressed to even approximate a solution. You cannot expect to outperform the market by buying low P/E stocks or by screening on any other single variable. You must be able to assess the impact of many variables all simultaneously impacting the future for any given stock. Unless the multiple variables can be expressed as a pattern in three-dimensional linear space, humans have a hard time dealing with them. As pointed out by Miller (1969) the contribution of multiple variables in human analysis is not additive, and humans reach a point of diminishing return very rapidly. As additional inputs are added, their contribution lessens. If you go much beyond five or ten simultaneous input variables you have exceeded the practical limits of most human analysts.
Variables do not act in isolation; as they change they have an impact on everything around them. This interrelationship among variables creates a degree of complexity that requires the use of sophisticated analysis techniques capable of dealing with the totality of the problem. The importance of a P/E ratio changes depending on if interest rates are high or low; it changes depending on whether you are looking at a high tech stock or a utility; it changes depending on whether the company is growing rapidly or slowly. The impact of just about any variable changes depending on the value of neighboring variables and how they are changing. Historically, it has been argued that if you hold everything else constant and change an input variable you will be able to see and understand its effects on the output. Unfortunately, that is not the way the real world works. Everything is interacting and changing all of the time, thus making most problems unsolvable using normal statistical analysis.
When a solution cannot be expressed as a simple multiple of the input parameters, it is said to be nonlinear or non-algorithmic. This describes the situation in security analysis. You cannot take several market variables, plug them into a linear equation, and solve for market direction over the next six months with any consistent degree of accuracy. It is quite apparent that we cannot take the Graham/Dodd financial ratios, plug them into a linear equation and hope to solve the investment asset allocation problem. It is equally apparent that the action of the markets is not completely random and does not require an explanation based on Chaos Theory. What is required is a system that can deal with non-linearity
Much of the information processed by the financial community is fuzzy or ambiguous in nature. People tend to think, reason, and speak in a fuzzy way. They talk about fast sports cars, loud music, tall men, happy children, accurate watches, and hard work, all with excellent understanding. Most things, however, are not black and white, but are shades of gray and our language reflects this. Fuzzy set theory is often useful in handling language inputs for analytical systems - including systems for security analysis.
The financial markets are constantly changing and the rules that worked last year will not necessarily work this year. Financial analysts have found and reported that computer models built prior to 1987 are just not producing the same results in the post crash era. In fixed environments, or when dealing with static problems, deterministic equations performs quite well. This is the manner in which standard financial forecasting is usually done. The problem with this is that the world is not a fixed environment, and the world of security analysis is definitely not static. The real world is in a constant state of flux and transition. To maximize performance during periods of transition, one needs the special tools that are not restricted by the boundaries of mathematical science, but act more like the human brain, which up to this point has been the most effective tool in navigating in this complex financial environment. Dynamic environments require systems that respond to feedback and adjust themselves and adapt to the changing conditions
By combining the concepts of multiple, interacting, nonlinear, fuzzy, dynamic variables, you start to gain an appreciation for the complexity of the security analysis problem, and to understand why we say that the problem is truly beyond human comprehension and requires the aid of financial cybernetics.
Even though early researchers understood the basic nature of many of the problems falling into the complex domain, they were helpless to attack them in any meaningful way until the advent of the computer and the parallel development of fast chips and advance software packages. As a matter of fact, the assault on these problems requires a rethinking of our approach to mathematics. Bailey (1996) has outlined the evolution of mathematical thought that has brought us to the point where we can now start to address some of the complex problems that have been beyond our reach. He talks about the three “P’s”. Bailey suggests that the first mathematicians were concerned with place, and developed a geometric based mathematical system to handle their problem. Galileo’s geometry helped establish borders and changed the shape and extent of the known geographical environment. His mathematics was critical in the development of navigational techniques that aided in the movement of people and goods.
During the Renaissance the new mathematics of algebra and calculus emerged as the mathematics of “rate of change” or pace. This new math allowed researchers to expand their investigations in the physical sciences and their efforts have brought us through the industrial revolution and into the dawn of the computer age. Now Bailey suggests that we are starting to explore the new math of patterns, a form of mathematics that was impossible until the development of the computer. The new adaptive math’s presuppose massive amounts of computing power, far more than was available even fifty-five years ago, when people did all of the computing. Because the new math’s are parallel, learning them will change the way we ourselves experience the world and our appropriate role within it. Currently the world is presented to us in books that present information sequentially, so that we see only one set of realities at a time. We see, for example, more isolated facts and fewer relationships and count on the human brain to build the necessary bridges. Words have a linear, discrete, successive order; they are strung one after another like beads. In the world of patterns and relationships there is no single path to a solution.
A basic element of the new evolutionary inter-math’s is the idea of feed-back and self-organization. In fact, these new inter-math’s are sometimes call “Adaptive Computation.” Self-organization is an extension of the Chaos Theory in which complex systems tend to organize themselves out of entropic chaos. "Equilibrium is not only dead, it is death.” To enrich a system you need variance in time and space and constant disequilibria. Feedback is an additional attribute of the new information society. Contrary to the traditional doctrine of diminishing returns, “the more we know, the easier it is to know more; the more we make the easier it is to make more; the richer we are, the easier it is to get richer.” These ideas also run contrary to the old idea of “equilibrium.” Cyberspace is a resource that increases the more it is used. It has been said to be, “a peculiar kind of real estate which expands with development.”
There has been a trillion-fold increase in raw computational power in the last 90 years. We have gone so far that it will take only another thousand-fold increase to match the gross thinking power of the human brain. Through the first fifty years of the computer revolution, scientists have been trying to program electronic circuits to process information the same way humans do. Doing so has reassured us all that underlying every new computer capability, no matter how miraculously fast or complex, are human thought processes and logic. But cutting-edge computer scientists are coming to see that electronic circuits really are alien, that the difference between the human mind and computer capability is not merely one of degree (how fast?), but of kind (how?). Bailey (1996) suggests that computers “think” best when their “thoughts” are allowed to emerge from the interplay of millions of tiny interacting operations. With this idea in mind, the security analysis system of the future needs to be set free to develop and apply its own methods of analysis.
These methods of analysis may be beyond human understanding and thus frightening and intimidating for many. As long as they perform well and can communicate their findings in a meaningful way we should enthusiastically embrace the technology. Developing an intelligent machine that can solve problems that we can not is truly amazing and we stand in awe.
Advanced security analysis can go along way toward identifying good and reasonable investments, but it will never handle the risk associated with general market fluctuations. The only way to protect yourself and your clients against market risk is through diversification. A big trend in institutional money management is “attribution” analysis. Trying to determine the importance of various structural factors in the performance of investment portfolios. Although employing some rather sophisticated mathematics and impressive analytical concepts, I am not sure it does much for improving investment results. One of the problems with these techniques is that they employ linear technology and as we pointed out earlier the security analysis problem is, in all likelihood, very none-linear.
Security analysis involves, not only being able to detect salient features, and understanding their implications, but also being able to respond to them and take action. Designing a security analysis system requires that we have knowledge of the needs of the end user. What kind of information does the end user need to make decisions? Financial analysis is not a singular task and neither is security analysis. There are a number of different types of end users and each has different needs and requires different types of information. Analysts that work at banks and credit organizations are more interested in the credit worthiness of their potential clients. Brokerage firms are interested in companies that have new products, and interesting stories. Their analysts search for companies that will capture investor interest and media headlines. Public pension funds are interested in steady growth with minimal risk to their pension assets. Analysts employed by these organizations spend most of their time trying to get their asset allocation correct so as to outperform a simple market index. Other analysts, are employed by the mutual fund industry and are trying, to produce the highest possible returns they can so as to attract investor attention and dollars.
Each of these end-users has different analysis needs and each needs an analysis system designed to meet those needs. The system must be capable of presenting its findings in a manner that is meaningful to human decision makers. If you assume that people are rational then the only reason for differences of opinion are differences in the amount of information available and the way it is presented. The behavioral finance community has demonstrated the irrational decision making of humans in the presence of uncertainty so the task of the security analysis system of the future is the removal or minimization of uncertainty so that humans can again be an integral part of the investment process. Although humans do not have the multidimensional pattern recognition or computational capability of the intelligent computer, they do bring wisdom and an environmental and historical perspective to the investment process that is critical.