Surrogate Expertise Indicators of Professional Financial Analysts.

The study of professional financial analysts' forecasts of earnings is important since: (1) the markets use these forecasts in making assessments of share price, (2) researchers use these estimates as surrogates for market earnings expectations, and (3) professional financial analysts generally overestimate earnings. The study of analyst expertise can help market participants, researchers and even managers to discount the effects of such overestimates. Further, since expert behavior is deemed to be desirable behavior (Bouwman and Bradley, 1997), identification of experts and their decision-making processes could improve the efficacy of training and education, as well as facilitate the development of expert systems.

The term "expertise" has been the source of much confusion. While it is difficult to define expertise, most researchers would agree that experts are presumed to make high quality decisions. Prior researchers have examined expertise on the basis of individual attributes, experience, knowledge, decision processes, and decision quality, yet a comprehensive model of expertise has proven elusive. It seems as though a parsimonious linear model of expertise might map psychological characteristics and cognitive processes to decision quality as a way to classify expertise and predict outcomes. However, even with such a model, it would be difficult to gather reliable data from experts along all of these dimensions. Hence, we often resort to the use of surrogate indicators of expertise.

The objective of this research is to identify and validate a surrogate measure of expertise for professional financial analysts that can be readily obtained from public sources. Hunton and McEwen (1997) examined experimentally various nonpublic experiential factors (e.g., age, years as financial analysts, years with firm) and cognitive factors (e.g., cognitive search strategy, time initially searching for information, time subsequently reviewing information) that might be used to classify financial analysts as relatively high or low expertise forecasters. Using an external measure of expertise provided by brokerage firm management, the researchers found that the analysts' cognitive search strategy distinguished high from low expertise analysts. They used a sophisticated retinal imaging system to classify the analysts' cognitive strategy. Since this technology is expensive, extremely difficult to obtain and complicated to use, we re-examine Hunton and McEwen's research findings in an attempt to identify and v alidate a parsimonious, publicly available surrogate indicator of expertise for financial analysts.

Using sixty "sell-side" financial analysts and a retinal imaging data collection method (called the Integrated Retinal Imaging System or IRIS), our findings suggest that experiential factors, commonly used as proxies for expertise (Affleck-Graves et al., 1990; Anderson, 1988; Biggs, 1984; Bouwman, 1982; Bouwman et al., 1987), are poor indicators of expertise, while membership on the All-America Research Team appears to be a reliable surrogate that is publicly available. These findings are consistent with Stickel (1992) who provides empirical evidence that membership on the All-America Team and accuracy are positively related. The current study extends Stickel's work by examining the relations between various surrogates for expertise, All-America team membership and accuracy. Unique to the current study is a factor-analytic, logistic approach that classifies accuracy by experiential factors, cognitive factors and membership on the All-America team.

The next section of this paper describes surrogates for analyst expertise that are found in the current literature. Subsequent sections provide the research design and report the results. The final section summarizes and concludes the findings of the study and indicates areas for future research.

SURROGATE EXPERTISE INDICATORS

It is generally assumed that higher levels of expertise involve more elaborate and complex cognitive processing of informational inputs which results in higher quality outputs (Yates, 1990). Since, in most cases, direct observation of decision processes is impossible or impractical, researchers often rely on certain decision-maker attributes and behaviors as surrogate indicators of cognitive processing sophistication. These surrogate measures are frequently used to predict or validate decision quality. In essence, rather than examining the inputs, processes, and outputs of expert decision making, researchers often associate observable attributes and behaviors of decision makers with decision quality, while treating cognitive processing as a "black box."

This epistemological approach assumes that expertise is accurately reflected in decisions and that the quality of decisions can be reliably evaluated. In situations where decision quality can not be readily assessed, researchers must rely solely on decision-maker attributes and behaviors to determine expertise. For example, in the tax and audit areas of accounting, a performance-based metric is not always feasible since there is often no way to validate decision quality. Thus, surrogates such as decision-maker experience, knowledge, and consensus have been used as proxies for expert decision making (Bouwman and Bradley, 1997). However, it is difficult to estimate the reliability of these surrogates. In the domain of financial analysts forecasts of earnings, expertise can be reliably assessed using forecast accuracy as an indicator of decision quality since forecasts can be judged using actual firm earnings as a means of external validation. Thus, one can compare decision quality to observable decision-maker characteristics and identify dependable surrogates of expertise.

Hunton and McEwen (1997) provide evidence that certain observable characteristics of financial analysts' are related to an historical, external indicator of forecast accuracy. In their studies, brokerage firm management provided an ordinal ranking of 60 financial analysts, where a "1" indicated the most accurate analyst and "60" represented the least accurate analyst. Management's historical accuracy ranking was based on the absolute value of the difference between forecast and actual earnings, divided by actual earnings over multiple time periods (although management wanted to keep the precise time frame confidential). The researchers note that experience (i.e., years as a financial analyst), a frequently used surrogate of expertise, was not significantly related to historical forecast accuracy. However, they document a significant relation between historical forecast accuracy and cognitive information search strategy. Specifically, relatively high accuracy forecasters employed a directive information searc h strategy while relatively low accuracy forecasters used a sequential cognitive search strategy.

Hunton and McEwen (1997) had access to brokerage firm management who were willing to provide historical accuracy rankings of participating analysts. Information of this nature is confidential and difficult to obtain. The researchers utilized a retinal imaging system (IRIS) system to study and classify cognitive search strategies as either directive or sequential. Technology of this nature is not readily available to behavioral researchers. Therefore, our objective is to find a parsimonious, publicly available surrogate for expertise that appropriately reflects accuracy assessments. Given that Hunton and McEwen (1997) validated an indicator of expertise (i.e., a directive cognitive search strategy) against a reliable, external measure of decision quality (e.g., historical forecast accuracy), we re-examine their data in our search.

Based partly on the findings of Stickel (1992), we propose that use of All-America team membership will provide a parsimonious, publicly available indicator of professional financial analyst expertise. Stickel suggests that some analysts view the poll as a "beauty contest" and these analysts tend to discount team membership as a meaningless honor afforded to those who influence voting by visiting money managers at the time of voting. However, Stickle provides empirical evidence that team membership has meaning in that a positive relation exists between forecast accuracy and team membership. We complement and extend Stickel's work by examining the behavioral relations between various surrogates for expertise and team membership.

RESEARCH DESIGN

A major brokerage firm in New York agreed to sponsor the current study, although the firm's primary objective was to test an Integrated Retinal Imaging System (IRIS, patent # 5,572,596) for physically disabled financial analysts. We informed subjects that their participation would contribute in a meaningful way toward improving working conditions and widening job opportunities for physically disabled individuals, and that we would use their responses in a research study on how financial analysts make forecast decisions.

The respondents communicated with the computer via eye movements. The IRIS system used pen-cameras, integrated into the computer screen, that mapped eye focal-points. Use of this system allowed a totally computerized, icon-driven experiment. When subjects focused on an icon for three seconds, the IRIS system activated the icon as if the subject had clicked a mouse button. The researchers trained representatives of the brokerage firm who then conducted the experiment. Sixty sell-side professional analysts agreed to participate in the experiment.

Management of the brokerage firm supplied relative accuracy rankings for the subjects based on a multi-period earnings error metric. Analysts were ranked from "1" (most accurate) to "60" (least accurate). The subjects' average age was 35.8 years, ranging from 24 to 54. Seventy-two percent of the subjects were male, 83% were Chartered Financial Analysts, and 68% held a master's degree. The average number of years as a financial analyst was 9.0 (ranging from one to 29 years), and the average number of years with the firm was 7.4 (ranging from one to 23 years). Fortythree percent of the subjects self-reported membership on the All-America Research Team.