The Differences Between Financial Analysis And Security Analysis
Thomas E. Berghage
Financial Analysis is historical in nature while
Security Analysis is focused on the future.
In research terms, the differences between Financial Analysis and Security Analysis are quite evident, it’s the differences between the criterion, or dependent measures. In Financial Analysis the dependent measures that analysts use are things like revenues, revenue growth, earnings, and earnings growth, along with other corporate performance figures. These are not direct measure of security performance, and are considered to be intervening measures when it comes to Security Analysis. The use of intervening measures is acceptable in forecasting as long as you can demonstrate a strong connection with the true dependent measure (security performance or total return), and as we have seen in Chapter 3, that connection for the traditional accounting measures used in financial analysis has not been made.
Historically, analysts have tried to explain this lack of relationship between traditional financial variables and security performance by assuming that markets were extremely efficient and were incorporating new information as soon as it became available. Certainly efficient, friction free, incorporation of new information would impact the information’s predictive value, and there is good reason to believe that markets do incorporate new information very quickly. The problem is that the new information by itself is not predictive, it is just part of the market noise. The new information adds another piece to the puzzle, but you still have to be able to put the pieces together to understand what is going on. In the intelligence community this process is called “connecting the dots.” Unfortunately, the puzzle is not a two or three-dimensional puzzle that humans are so good at assembling or solving, but it is rather a “K” dimensional puzzle that exceeds our human abilities. The future course of a company’s stock price will be determined by the corporation’s behavioral pattern as represented in a complex mix of performance measures.
I am not suggesting that the Efficient Market Hypothesis is wrong, it may very well be true. It may be the reason that financial measures are poor predictors of future stock performance. All I am suggesting here is that, for whatever reason, simple discrete financial figures taken by themselves are not good predictors of future stock performance, and we should stop suggesting that they are. If investors need a story to motivate them to buy a stock, tell them some other fairy tail that is not based on accounting figures. The accounting profession has problems of their own and they don’t need some broker or security analyst adding to their tribulations.
As we pointed out in Chapter 7 the behavioral science community long ago abandon the idea that one could use discrete measures such as I.Q. to predict future human performance. Instead they now look for behavioral patterns among a constellation of measures to try to identify and understand future human behavior. The financial community is faced with a very similar problem in that they are dealing with a living breathing, ever changing entity called a “corporation.” Made up of hundreds, sometimes thousands of human decision makers the corporation is impossible to define with a few discrete measures. Things like corporate culture and unofficial lines of communications are not reflected in corporate governesses or public press releases, yet they are the essence of how the corporation functions and behaves. It is this underlying corporate behavioral pattern that we need to identify and understand if we are ever going to be successful at forecasting future stock performance.
Intelligence agencies around the world are in a constant search for patterns of behavior that can be used to forecast future events. The behavioral patterns being evaluated could be for countries, organizations, or even individuals. The number of dollars spent on these activities depends on the value of the information obtained and the potential cost or lost that could potentially occur by not having good foresight.
Unlike the dissemination of corporate information and the need to provide equal access to all, the national intelligence program does not have the requirement to maintain a level playing field. Military intelligence is designed to provide a dominant exploitable position that puts the adversary at a disadvantage. The laws regarding corporate information and security transactions, however, are a little different; they are designed specifically to eliminate unfair advantage. The release of corporate information and subsequent transactions are closely monitored to ensure that a level playing field is maintained.
While national intelligence services have the liberty and flexibility to access information from numerous sources ranging from published and media reports, satellite images, remote controlled cameras, wire tapes, and inside sources; security analysts are restricted to the use of public information. That does not mean that some of these illegal clandestine sources are not utilized, but they are severely restricted and are generally confined to a limited number of companies at any one time. Any large-scale attempt to use these kinds of information sources would certainly be detected and prosecuted in the courts.
The large pools of investment capital that reside within institutions require numerous individual investment options in a number of different investment categories. Trying to gather detailed intelligence on all of the investments needed for an institutional size portfolio is just not practical and the legal probations against such activities takes it outside the realm of reasonable possibility.
Although the sources of corporate intelligence are limited, closely monitored, and controlled, the analysis of the information is open for exploitation. What you do with the information contained in public reports, databases and government filings is unrestricted and open to creative analysis. National intelligence gathering organizations have found that most of the information they need already exist and is readily available. It is not so much having to go out and collect the information, as it is being able to decipher what you already have, and being able to “connect the dots” and develop the picture. The same holds true in security analysis. We have more information on companies than we can possible use. The problem is deciphering what we have and being able to identify the corporate behavioral patterns that are meaningful for forecasting future stock performance.
Up until recently humans were the most effective pattern recognition system we had, both for military intelligence and for corporate security analysis. We put pilots in our fighter aircraft because they could not only make on the spot decisions, but also could recognize and identify targets on the ground. As opposition forces became more sophisticated in the use of camouflage we have had to resort to using satellite imaging and remote infrared sensors with built-in intelligence to identify targets.
Many of the same engineering problems that faced military planners now face the analyst community. The human pattern recognition system that we have relied on for so many years is just not working and we need to reengineer our investment analysis systems to take advantage of the new advanced technology. Having the ability to rapidly integrate numerous complex inputs and detect patterns has had to wait for the development of advanced computer chips and new software, but now we have them and the technology is going to open up an entirely new way to look at and analyze some of our most complex problems.
The search for solutions in higher dimensional space is not confined to security analysis. Theoretical physics has for many years explored multi-dimensional space for answers to some of the most complex problems faced by mankind. Once Einstein opened the box by incorporating a fourth dimension in his theory of relativity and Kaluza added a fifth dimension to combine the field theories of Maxwell (light) and Einstein (relativity) into a single field equation, the number of dimensions has been expanding. Physicists have found that the laws of nature are simpler and easier to explain if they use higher dimensional space in their theories. Currently, theoretical physics is exploring the secrets of the universe and developing mathematical theories that incorporate up to ten dimensions. How many will they eventually end up with is an open debate.
Just like our brethren in the field of physics are seeking solutions to the mysteries of nature in hyperspace; so too should we in the financial field seek answers in multidimensional space. As we venture into financial hyperspace where we are dealing with the human behavior of millions of independent decision makers the number of dimensions that will be needed to explain events could be quite large, even more than that required to solve the problem confronting theoretical physics.
In classical set theory the universe is described in terms of sets, subsets and the intersection and the union among sets. Sets, of course, are just groups of items or entities and this is a reasonable way of describing categories of items and their relationship to each other. In the security analysis world, the universe we are talking about is the universe of all publicly traded stock. Within this universe there is a set of stocks that will produce a positive return over the next twelve months. Our job is to identify the members of this set. Classic set theory uses what are known as crisp sets, which means an item or a stock is either a member of the set or it is not. This is the black/white, yes/no, true/false, ones/zeros world that the computer is based upon. There is another world however, and it encompasses the vast region between one and zero.
The world of crisp sets and binary logic has dominated our thinking from the time of Aristotle and probably before. In security analysis we create crisp sets all of the time, we talk about large cap stocks, low P/E ratios, and high or low return on equity. Even though the establishment of these categories or sets is rather arbitrary we use them for screening stocks like their boundaries are fixed and ridged. In our world of ones and zeros the screening process looks something like this:
(Market Capitalization >= $5 billion) equals Large Cap (Yes=1, No=0)
(Return on Equity >= 20%) equals High (Yes=1, No=0)
(Price/Earning Ratio <= 30) equals Low (Yes=1, No=0)
We could add more variables or dimensions to this screen, but lets limit it to three so we can visualize the problem. For a stock that did not meet any of our criteria the computer would see a string of zeros (0,0,0) where the first position represents the stock’s market capitalization, the second position its ROE, and the third position its P/E ratio. To visualize these three dimensions let put them in the form of a cube (Figure 11-1), where the eight corners represent the eight possible crisp sets for our stock screen. Only at one corner have we met the criteria (1,1,1) for all three dimensions. At the other seven corners, one or more of the criteria have not been met and the binary code has one or more zeros.
Figure 11-1
Picture of Hypercube
In the binary logic world only the eight corners of our cube are evaluated. The vast space within our cube, despite its proximity to the desired category or crisp set, goes unattended. Stocks represented by the letters “A” and “B” in Figure 11-1 are lost in hyperspace. The artificiality of this categorization process becomes even more absurd when you realize that a stock’s membership in a given set can change minute by minute as it trades in the market place during the day. For the example shown above we were only talking about three dimensions or variables. In the real world, stock evaluation is based on numerous variables so we are talking about “K” dimensional space with vast regions of fuzziness. To navigate in this hazy hyperspace requires the use of tools that are not limited by three-dimensional linear thinking or the crisp sets of classical set theory.
The New Paradigm
Every financial analyst in the world has been taught that a company’s stock appreciates in value because of the underlying earnings growth of the company. In the last five or ten years analysts have had to change this view somewhat because a number of the new technology companies have elected to forgo earnings for growth or increasing market share which, in the long run, may be better for investors. It is now apparent that even companies with earnings need to be viewed differently depending on when and how earnings are derived and reported. Analysts have now begun hedging their bets a bit by talking about the quality of earnings. They point out that not all earnings are the same and that their meaningfulness is dependent on a number of other factors. Analysts are now saying that the earnings, or lack there of, reported by accountants are sometimes misleading and not truly representative of the future performance of the company. The problem gets even worse when it comes to international stocks where the accounting rules for deriving earnings are different.
How then do we determine the appreciation potential of a company’s stock? The answer is that you must look at the totality of the company’s performance. Not just isolated independent figures, but all of the figures and all of the relationships between them. It is only by looking at the entire picture that one can hope to find the patterns of behavior, or emergent properties, that are the precursors of good stock performance. The problem, of course, is that the ability to detect and discern the meaning of these very subtle underlying patterns, or emergent properties, is beyond the capability of we poor humans. That’s the bad news. The good news is that the human mind has been able to develop a computer technology capable of finding and evaluating these patterns.
The security analysis system of the future must be capable of operating freely in the dynamic hyperspace of economic markets. To do this it will have to include two key elements, a Cybernetic feedback capability and a Gestalt analysis capability. By Cybernetic feedback capability we mean that the analysis system must be capable of adjusting itself to stay in touch with the ever-changing market conditions. It also needs to be able to make up its own rules for discriminating between good and poor investment opportunities so it is not constrained by the human thought process. The system also needs to be capable of performing what we call a Gestalt analysis. An analysis that is not constrained by available data, but is capable of detecting what is known as emergent information, information that is based on relationships or patterns among data items that are far too complex for humans to detect or understand. The patterns that are present in the whole that are more than just the sum of the individual parts.
A Cybernetic Process
In 1948 Norbert Wiener published his now famous book, Cybernetics: or control and communication in the animal and the machine. Wiener’s book developed the notion that feedback was essential for the design of dynamic systems. Weiner’s ideas regarding feedback are important for future security selection systems for two reasons: first, as Weiner points out, feedback is essential if you are dealing with a dynamic problem such as market change; and second, feedback is also important in detecting and responding to the hidden patterns to be found in the dynamic data item relationships.
If you are dealing with a static problem where the rules and variable weighting do not change, technologies such as Expert Systems are just what you need. They allow for the consistent application of the same logic and analysis to similar reoccurring problems. These systems produce appropriate results for static problems that keep reoccurring, but do not change. If, however, the problem is constantly changing and evolving you need some type of feedback in the system design that will adjust the rules and their weightings to keep the system in touch with the ever-changing environment.
Perhaps one of the most familiar examples of a feedback system is the thermostat: it achieves a constant temperature in any enclosed environment by assessing the actual room temperature, comparing it to a desired temperature, and then responding - by turning the heater (or air conditioner) either on or off according to whether the existing temperature lies below or above the desired value. The word feedback describes how the process returns (feeds back) the results of the control action (the temperature) to the compensating mechanism. Cybernetics (the science of control) is a mathematical theory of information feedback. In essence, feedback mechanisms are information processing devices that receive information and then make decisions based on it. Wiener speculated that all intelligent behavior is the consequence of feedback mechanisms; perhaps by definition, intelligence is the outcome of receiving, processing, and acting on information.
The other important aspect of feedback is it provides a mechanism whereby a system can change its structure. Genetic Algorithms (GA), Evolutionary Programming (EP), and Neural Networks (ANN) all use feedback to change their structure. The first two, GA’s and EP’s use performance feedback to eliminate under-performing systems while allowing surviving systems to generate new offspring. The disadvantage of this approach is that some information tends to get lost with the killing off of under-performing systems. Neural Networks on the other hand retain all of their information and restructure themselves by adjusting their weight structure. If history does tend to repeat itself, Neural Networks should be superior because they still have available all of the information that may have worked in the past. It just needs to reactivate the information by increasing its connection weights. The structure of the Neural Network with all of its interconnections among neurodes provides an almost infinite number of combinations of data items thus providing a very rich environment for the feedback mechanism to work with. The availability of all of these combinations of data items along with the feedback mechanism allows the network to perform a Gestalt analysis and discover emergent phenomena.
A Gestalt Analysis
There has been a shift-taking place in science over the last few years. A shift from reductionism, the idea that understanding comes from reducing everything down to its smallest component parts, to the realization that the whole is often greater than the sum of its parts. This belief is now being incorporated into security analysis. To describe this phenomenon we would like to borrow a term from the early German psychologist, Max Wertheimer. He used the term, Gestalt, to describe the experience his subjects had during a perceptual experiment in which they reported movement of a light when no movement actually took place. This and several subsequent experiments demonstrated that the whole perceptual experience reported was more than the sum of the individual stimuli presented to the subject. The term Gestalt in German has two meanings; first, a Gestalt is an object, which has shape, an entity in itself that has form, as a chair or table; second, a Gestalt is a property of things as in squareness or triangularity. Taken together then, Gestalt is both the object and the form characteristics of that object, its essence. Sometimes the words “configuration,” “structure,” and “whole” are used as English translations for Gestalt, but the un-translated term is preferable, since none of these words capture its complete meaning.
We experience Gestalts in everyday life all of the time. Motion pictures contain no real motion, but you would have a hard time understanding that in some of the new Cinamax theaters. When we perceive a piece of music we hear the melodic form, not a string of individual notes. The melody seems to emerge from the serial pattern of notes. The term emergent seems to be popping up more and more as people describe situations where the properties of the whole seem to be more than the sum of the individual parts.
In biochemistry many complex molecules have the same atoms but combine in more than one way (pattern of connection). The properties of the compound are found to depend on the relationships formed in combination; that is to say, the properties of the compound depend in part on the relationships, which are shown in the structural chemical formula, a formula that can vary while the structural atoms remain the same. The relationships between elements in the compound produce an emergent property that is more than the sum of the parts.
It does not take much imagination to take this concept a few steps further to suggest that the modern day corporation is more than its individual parts and that its analysis requires the ability to detect emergent patterns. No single measure or group of measures in isolation can tell you about a company’s current performance or future potential. You must be able to evaluate the whole and detect emergent patterns or the corporate Gestalt.
Many people mistakenly classify intelligent computer systems as a form of quantitative analysis. There are two distinct differences between advanced intelligent 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 advance 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 intelligent 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 intelligent 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 intelligent 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.