Printer Friendly

The application of expert systems to securities analysis.

The Application of Expert Systems to Securities Analysis

The single most important factor determining the value of a share of common stock is the level of the expected future earnings per share (16). Earnings can be paid out in dividends to shareholders or reinvested by the firm to generate a higher level of earnings and/or dividends in the future. Unfortunately, for stock analysts and investors, the reported earnings of firms can vary widely from firm to firm solely due to different accounting methods, different reporting methods, and/or different financing methods (3).

The problem of what value to place on a particular firm's earnings is complicated by the fact that a significant portion of the reported earnings of that firm can be "cosmetic" compared to that of another firm's earnings, which may be derived from continuing operations and are more indicative of future earnings power (2,18). However, it is important to point out that "quality" is not an absolute measure, and "in the context of earnings evaluation, is one of comparative integrity, reliability and predictability" (1, p. 73).

Hence, the earnings quality of a firm is a subjective concept relating to the reported earnings of the firm and describes the ability of the firm to sustain earnings through continuing operations. Many firms have a wide latitude in the level of earnings they can report to stockholders even within the boundary of Generally Accepted Accounting Principles (GAAP). Firms that use very liberal accounting methods can report substantially higher earnings than firms that utilize very conservative accounting policies (1, pp. 72-73).

In some cases, cash flows generated can be less for the firm with higher reported earnings than the firm with lower reported earnings. In such cases, the firm reporting the higher earnings can actually be more risky or less profitable. For example, such a case would exist for a firm using the first-in-first-out (FIFO method of inventory valuation in place of the last-in-first-out (LIFO) method during inflationary periods. Under FIFO, reported cost of goods sold would be less, reported earnings would be higher, and the tax paid would be higher. Alternately, use of accelerated depreciation can result in higher cash flow via less taxes paid on the lower earnings reported.

In essence, it is when reported earnings accurately reflect the future earning power of the firm that a high earnings quality is reflected. Since many factors other than accounting treatment, such as type of business, financial plan, and general economic conditions, affect the earning power of a firm, it would be very beneficial to a security analyst to be able to systematically sift through the many different influences comprising a firm's reported earnings in an attempt to ascertain more definitively the firm's true underlying earning power.

Expert Systems as a Solution

Expert systems technology is a possible solution, for it utilizes the ability of the computer to simulate the human thought processes. An expert system has been defined as "a computer system that could perform at, or near, the level of human expert" (8, p. 259). Rather than the traditional techniques of data processing and information processing, expert systems, as a subset of the larger area of artificial intelligence (A.I.), have been described as a method of knowledge processing (10). These systems have been distinguished from traditional computer approaches in their ability to: * Focus on problems that do not respond to standard

solutions or have inexact, missing, or poorly defined

information * Capture and manipulate the significant qualitative features

of a situation rather than rely on numeric methods * Use large amounts of knowledge specific to an area to

solve problems (15).

Earnings quality meets these criteria. The first two criteria suggest that the qualitative elements of earnings quality, even though buried in a financial statement or not readily available, can be analyzed using expert systems technology. The last point implies that an appropriate application is one which is limited to a specific problem area or "domain," yet has a depth of knowledge required for the solution of the problem. The numerous decisions and judgments involved in securities analysis lends itself to such a computerized approach using expert systems. For example, in securities analysis, much of the expert analysis of the performance of a firm focuses on whether earnings are categorized as coming from continuing earnings indicative of higher earnings quality. The difficulties encountered in categorizing earnings require knowledge and experience, and classification of earnings is where "expert" securities analysts can utilize an expert system to assist them in this effort.

Classifying Earnings

Since earnings that result from liberal accounting methods, liberal reporting, and non-sustainable debt levels reflect poor earnings quality, they should be worth substantially less in value to investors than earnings from continuing operations. Table 1 presents a brief outline of some of the sources that can be scrutinized to ascertain earnings quality in these areas. As an example, conservative depreciation or inventory methods would include accelerated depreciation and LIFO inventory, because these approaches reduce current earnings rather than future earnings.

Also implied in the classification of earnings quality is an assessment of earnings which generate cash flows. It is from cash flows that dividends are paid. In addition, cash flows are used for capital expenditures as well as debt service.

Cash flows of corporations are reported in three basic categories: (1) from operations, (2) from investments, and (3) from financing, which are included in Table 1. The earnings of a firm are greatly enhanced if the firm generates sufficient cash flow from operations to meet all current and planned future cash needs. The evaluation of these cash flows could result in a good assessment of the firm's ability to meet future dividends, capital replacement, and debt service requirements. On the other hand, if dividends and/or debt service were being paid out of external financing, the future earnings of the firm would be in jeopardy (18).

Earnings quality can be analyzed by examining its components, which are listed in Table 1. Although existing computer technology can provide the traditional ratio analysis and computations and thus address the performance of a company in some categories, existing computer technology is inherently weak in examining missing or qualitative information, which may be included in these categories and thus impact on earnings quality. Expert systems can provide such a qualitative analysis.

Table : Table 1 Sources of Reported Earnings * Accounting treatment

* * Depreciation

* * Inventory valuation

* * Inventory valuation

* * Pension funding

* * Goodwill amortization * Reporting of non-operating income/investments

* * Interest income (non-financial corporations)

* * Capital gains/losses

* * Extraordinary gains or losses

* * Special items * Financing

* * Earnings from financial leverage

* * Potential dilution via options or conversions

* * Share repurchase

* * Solvency assessment * Continuing operations

* * Sales

* * Margins

Earnings Quality Determination

Two expert systems techniques applicable to securities analysis include: (1) rule-based expert systems and (2) induction systems with automated knowledge acquisition. Rule-based expert systems are often thought of as if-then systems which encode the knowledge of a human expert. On the other hand, induction systems with automated knowledge acquisition are relatively unpublicized in financial applications and draw their power from the ability to "reason backwards," i.e., start with a goal or conclusion and then find the rules which led to the achievement of that goal. Once the rules are determined, they may be used to achieve the goal in the future by being used in a rule-based expert system.

Rule-Based Expert Systems

Rule-based expert systems are perhaps the most common form of expert systems. A number of financial institutions, such as Chemical Bank, Arthur Anderson, Price Waterhouse, Goldman Sachs, and Citicorp, use expert systems (17). However, often companies do not disclose much about their use of expert systems for reasons of competitive advantage (11,20).

In addition to the advantages already cited, another advantage of rule-based expert systems is the ability to capture the knowledge of human experts that otherwise would be lost when the expert dies, retires, or leaves the company. Furthermore, expert systems can be used to accumulate expertise from more than one human expert and package it in one place, where the expertise can be revised and improved over time.

The widespread use of expert systems has resulted in computer packages, called "shells," which can be used to generate expert systems without the aid of a computer programmer. Whether written in a programming language or using a shell, an expert system encodes heuristics (rules of thumb) of experts in the form of IF condition - THEN result. The system can combine the rules and arrive at a result. One way in which the expert system differs from tradional programming is in its ability to use "fuzzy logic" or uncertainty, such as might be found in the statement.

If the depreciation method is accelerated, then we are 50

percent sure that the company's accounting treatment is

conservative.

A strength of fuzzy logic over traditional programming is its ability to function with incomplete information. An example follows in which the analyst wishes to determine if the firm is using a conservative accounting treatment:

If the research-and-development expenditures as a percent

of sales are greater than the average-over-the-past-three-years,

then we are 75 percent sure that the accounting

treatment is conservative.

If the information above is not available or if the analyst or computer program is unsure that the numbers determined are not accurate, an alternative can be generated by the computer expert system:

If depreciation is accelerated and inventory-valuation is

LIFO, then we are 85 percent sure that the accounting

treatment is conservative.

Rules can be added until a reasonably clear picture of the accounting treatment is obtained. This determination can then be combined with the other subcategories, i.e., reporting of non-operating income, financing methods, and continuing operations, to determine the overall earnings quality. The certainty factors provide the ability to compensate for missing information and to describe the strength of conclusions (e.g., strong, weak, normal, uncertain). If a decision tree approach is used on the above example, as shown in Figure 1, traversing another branch may reduce the overall level of uncertainty. Unlike traditional computer programs, expert systems examine all potential branches, which could lead to the determination of a question.

In the example above, we are attempting to determine whether accounting treatment is conservative or liberal as part of our determination of earnings quality. The expert system will see where in the decision tree the accounting treatment is evaluated and then search backwards to see if all intervening steps can be shown to be true. Frequently, expert systems will ask the user questions as necessary, resulting in a highly interactive system. In our simplified example, accounting treatment is determined in two places. The system could possibly find that we are both 75 percent and 85 percent sure that the accounting treatment is conservative, depending on how the intervening questions are posed and answered. If so, the expert system could be designed to provide the result that we are 85 percent sure, in other words, the higher percentage. Table 2 shows the expert system rules which were used in a simulated system and a sample interaction between the expert system shell and the user.

Table : Table 2 Rules and Results from an Expert System (ES)

Rule 1:

If (1) depreciation-rate is known, and

(2) depreciation-rate is greater than the industry

average, then, there is strongly suggestive evidence (75%) that the accounting treatment is conservative.

Rule 2:

If (1) R&D is fully expensed, and

(2) depreciation is accelerated, then, there is strongly suggestive evidence (85%) that the accounting treatment is conservative.

Results

Run 1:
ES: Is the depreciation-rate known?
User: Yes
ES: Is the depreciation-rate greater than the industry-average?
User: Yes
ES: There is strongly suggestive evidence (75%) that
 the accounting treatment is conservative.


Run 2:
ES: Is the depreciation-rate known?
User: No
ES: Is the R&D amortized or fully-expensed?
User: Fully-expensed
ES: Is the depreciation accelerated or straight line?
User: Accelerated
ES: There is strongly suggestive evidence (85%) that
 the accounting treatment is conservative.


Application to Stock Valuation

Another strength of expert systems over traditional programming is their ability to answer a question such as: "Is company A a timely investment?" The expert system will search through its rules to arrive at the conclusion of whether a company is a good investment at this time. In the process, the expert system may pause to ask questions concerning the current state of the economy, interest rates, or current trends in the company's industry. If the investor has no particular company in mind, the investor may tell the system to: "Find a company with low price-to-value ratio, good performance, and from an industry with above average long-term potential," in which case potential investments would be sought and ranked in order of meeting the criteria.

Expert systems can be combined with database screens as well, where databases (e.g., Value Line or Compustat) of company information are available. Models (e.g., Benjamin Graham's stock valuation model, and dividend, earnings, and/or cash flow models) can be computed using traditional computer technology on all stocks with the result that some models will be best at choosing certain kinds of companies. For example, some models favor growth stocks whereas others favor cyclical or income stocks. Expert system rules which combine models to choose "mispriced" stocks can add to the strength of the screens. Furthermore, expert systems can be used to analyze the qualitative aspects of selected stocks.

Automated Knowledge Acquisition

Another form of expert system has been termed an induction system(20). Induction systems are a form of automated knowledge acquisition, where rules can be generated from data. Induction systems are occasionally included in expert system shells, but often their potential is overlooked. Most people think of an expert system as going through a deductive process to arrive at the best investment, but automated knowledge acquisition provides inductive reasoning. For example, say that you have screened several companies that had met certain of your standards and that sizable returns in excess of other companies of similar risk, what might you learn from those companies that would help you pick future investments? An induction system could compare the uniqueness of the companies with other companies on the basis of ratios, models, growth patterns, size, etc. and suggest the rules to use in choosing future investments.

A securities analyst could use such a system to forewarn of higher risk or potentially poor earnings performance. An analysts could evaluate earnings or a category of earnings quality, such as accounting treatment, for several companies and put the data into a data table, such as in Table 3. Tools, such as Knowledge Maker or VP-Expert, will examine the data table and generate decision trees and rules for use in rule-based expert systems. n1 The induction system automatically generates both the decision tree and the rules (see Table 4). In situations where the data conflict, the induction system will assign probabilities or certainty factors to the rules. The rules can then be used as a rule-based expert system, as previously discussed.

TABLE : Table 3
 Accounting Treatment Evaluation
Treatment Depreciation Inventory R & D Goodwill
Conservative Accelerated LIFO ? Amortized
Conservative Accelerated LIFO Expensed ?
Liberal Straight FIFO Amortized Unamortized
Liberal Accelerated FIFO Amortized Unamortized
Conservative Accelerated LIFO Expensed Amortized
Conservative Straight LIFO Amortized Amortized
Conservative Straight LIFO Expensed Unknown
Liberal Straight FIFO Expensed Unamortized
Liberal Straight LIFO Amortized Unamortized
Liberal Accelerated FIFO Unknown Unknown
Conservative Accelerated Unknown Expensed Amortized
Conservative Accelerated FIFO ? Amortized
Conservative Accelerated LIFO Expensed ?
Liberal Straight FIFO Amortized Amoritized
Liberal Straight Unknown Amortized Amortized
Conservative Straight LIFO Expensed Amortized


TABLE : Table 4

Rules Generated from Table 3

Rule 1:
If INVENTORY is Unknown
and DEPRECIATION is Accelerated
then ACCOUNTING-TREATMENT is Conservative


Rule 2:
If INVENTORY is Unknown
and DEPRECIATION is Straight-Line
then ACCOUNTING-TREATMENT is Liberal


Rule 3:
If INVENTORY is FIFO
and GOODWILL is Unamortized
then ACCOUNTING-TREATMENT is Liberal


Rule 4:
If INVENTORY is FIFO
and GOODWILL is Unknown
then ACCOUNTING-TREATMENT is Liberal


Rule 5:
If INVENTORY is FIFO
and GOODWILL is Amortized
and DEPRECIATION is Straight-Line
then ACCOUNTING-TREATMENT is Liberal


Rule 6:
If INVENTORY is FIFO
and GOODWILL is Amortized
and DEPRECIATION is Accelerated
then ACCOUNTING-TREATMENT is Conservative


Rule 7:
If INVENTORY is LIFO
and GOODWILL is Unknown
then ACCOUNTING-TREATMENT is Conservative


Rule 8:
If INVENTORY is LIFO
and GOODWILL is Unamortized
and DEPRECIATION is Straight-Line
then ACCOUNTING-TREATMENT is Liberal.


Rule 9:
If INVENTORY is LIFO
and GOODWILL is Unamortized
and DEPRECIATION is Accelerated
then ACCOUNTING-TREATMENT is Conservative


Rule 10:
If INVENTORY is LIFO
and GOODWILL is Amortized
then ACCOUNTING-TREATMENT is Conservative.


This tool cuts short the difficult task of formulating and writing rules, often called knowledge acquisition, which is frequently done by interviewing an expert(11). Such a system acts not only to quicken the development of a rule-based expert system but also to show patterns that the analyst may have difficulty discerning. Often, analysts may not even know what rules they are using as they look at information and form opinions(20). Alternately, an expert may not be available to formulate the rules. Finally, induction systems provide the ability to respond to rapidly changing knowledge in domains such as finance and securities analysis, where long system development times may cause the expert system to become obsolete before it is used(11).

Conclusion

The implications of A.I. for securities analysis are substantial. Rule-based expert systems and induction systems with automated knowledge acquisition are techniques within A.I., each of which has practical application within securities analysis. Rule-based expert systems are widely used at this time and are immediately applicable to such an analysis. Induction systems with automated knowledge acquisition is a technology which is in its early years of development and offers potentially great rewards in its application to securities analysis.

The use of these two techniques, coupled with traditional computer technology, provides a useful tool for the securities analyst. Each technique has its strength and proper application. Neither of them is a replacement for existing computer technology; each is a supplement in its own right. By using these techniques in combination, the judgment and expertise of a securities analyst can be encoded and captured by the computer, and the securities analyst can use the computer to further his or her expertise and enhance productivity.

Endnotes

1. Knowledge Maker is the registered trademark of the

induction system made by Knowledge Garden, Inc. and

VP-Expert is the registered trademark of Paperback

Software.

References

1. Bernstein, L.A. and Siegel, J.G. "The Concept of Earnings

Quality." Financial Analysts Journal, July/August 1979, pp.

72-75. 2. Cohen, J.B., Zinbarg, E.D., and Zeikel, A. Investment Analysis

and Portfolio Management. Homewood, Illinois: Irwin,

1987. 3. Collins, D.W., Kothari, S.P., and Rayburn, J.D. "Firm Size and

the Information Content of Prices with Respect to Earnings."

Journal of Accounting and Economics, Vol. 9, 1987, pp. 111-138. 4. Dowen, R.J. "What are Analysts' Forecasts Worth?" Financial

Analysts Journal, July/August 1989, pp. 71-74. 5. Foster, George. Financial Statement Analysis. Englewood

Cliffs, New Jersey: Prentice-Hall, Inc., 1978. 6. Freeman, R.N. "The Association Between Accounting Earnings

and Security Returns for Large and Small Firms." Journal

of Accounting and Economics, Vol. 9, 1987, pp. 195-228. 7. Friemark, H.A. "Adverse Opinions - A Reporting Wrinkle."

CPA Journal, Vol. 56, September 1986, pp. 100-101. 8. Harmon, P. and King, D. Expert Systems Artificial Intelligence

in Business. New York: John Wiley & Sons, Inc., 1985. 9. Harrington, S.J. "Implementing Expert Systems: Approaches

and Caveats." Journal of Systems Management, November

1988, pp. 26-29. 10. Holsapple, C.W. and Whinston, A.B. Business Expert Systems.

Homewood, Illinois: Richard D. Irwin, Inc., 1987. 11. _______, Tam, K.Y., and Whinston, A.B. "Adapting

Expert System Technology to Financial Management."

Financial Management, Autumn 1988, pp. 12-22. 12. Keyes, J. "Wall Street Speakes English." Al Expert, July 1988.

pp. 50-55. 13. Kormendi, R. and Lipe, R. "Earnings Innovations, Earnings

Persistence, and Stock Returns." Journal of Business, Vol. 60,

1987, pp. 323-345. 14. Lipe, R.C. "The Information Contained in the Components of

Earnings." Journal of Accounting Research Annual, Vol. 24,

1986, pp. 40-55. 15. Luger, G.F. and Stubblefield, W.A. Artificial Intelligence and

the Design of Expert Systems. Redwood City, California: The

Benjamin/Cummings Publishing Company, Inc., 1989. 16. Miller, M. and Modigliani, F. "Dividend Policy, Growth and

the Valuation of Shares," Journal of Business, Vol. 34, No. 4,

October 1961, pp. 411-433. 17. Newquist, H.P., III. "AI and Financial Services: The Real

Wheel of Fortune." Al Expert, April 1988, pp. 23-25. 18. O'Glove, T.L. Quality of Earnings. New York: The Free

Press, 1987, pp. 126-146. 19. Pari, R., Carvell S., and Sullivan, T. "Analyst Forecasts and

Price Earnings Ratios," Financial Analysts Journal, March/April

1989, pp. 60-62. 20. Shaw, M.J. and Gentry, J.A. "Using an Expert System with

Inductive Learning to Evaluate Business Loans." Financial

Management, Autumn 1988, pp. 45-46.
COPYRIGHT 1991 St. John's University, College of Business Administration
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1991 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Harrington, Susan J.; Twark, Allan J.
Publication:Review of Business
Date:Mar 22, 1991
Words:3469
Previous Article:The effect of the Brazil suspension announcement on returns to large U.S. bank stocks.
Next Article:The learning curve and production standards : learning implications.
Topics:


Related Articles
Can your business use artificial intelligence?
Keeping an eye out for risk.
Catching network chicanery.
ISS EXTENDS DATABASE VULNERABILITY MANAGEMENT SOLUTION.
NetIQ releases new version of application monitoring software.
CATC OFFERS EAS PROTOCOL COMPLIANCE REPORTING SOFTWARE.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters