A School of Business The University of
A Review of Research Related to Financial Analysts’ Forecasts and Stock Recommendations Sundaresh Ramnath * McDonough School of Business Georgetown University [email protected] edu Steve Rock * Leeds School of Business The University of Colorado at Boulder Steven. [email protected] edu Philip Shane * Leeds School of Business The University of Colorado at Boulder Phil. [email protected] edu June 15, 2005 * We greatly appreciate the research assistance of Kevin Hee and comments and suggestions from Zhaoyang Gu and Rick Johnston.
Sundaresh Ramnath is assistant professor of accounting at the McDonough School of Business, Georgetown University, G-04 Old North, Washington, DC 20057, USA; fax: +1-202-687-4031; Tel. : +1-202-687-3812. Steve Rock is assistant professor and Philip Shane is associate professor of accounting at the Leeds School of Business, The University of Colorado at Boulder, 419 UCB, Boulder, CO 80309, USA; fax: +1-303-4925962; Tel. +1-303-735-5009 (Rock), +1-303-492-0423 (Shane). A Review of Research Related to Financial Analysts’ Forecasts and Stock Recommendations Abstract: This paper reviews research regarding the role of financial analysts in capital markets. The paper builds on the perspectives provided by Schipper (1991) and Brown (1993).
We categorize papers published mainly since 1992 and selectively discuss aspects of these papers that address or suggest key research topics of ongoing interest in seven broad areas: analysts’ decision processes, the determinants of analyst expertise and distributions of individual analysts’ forecasts, the informativeness of analysts’ research outputs, analyst and market efficiency with respect to information, effects of analysts’ economic incentives on their research outputs, effects of the institutional and regulatory environment (including cross-country comparisons), and the limitations of databases and various research paradigms. A Review of Research Related to Financial Analysts’ Forecasts and Stock Recommendations 1. Introduction This paper reviews research related to the role of financial analysts in the allocation of resources in capital markets. Two important papers published in the early 1990s provide perspectives on the literature in this area, one appearing in Accounting Horizons (Schipper, 1991) and the other appearing in the International Journal of Forecasting (Brown, 1993). Our paper begins by summarizing the perspectives and directions for future research suggested in Schipper (1991) and Brown (1993). We then take a look at the highlights of what we have learned and new questions that have emerged since 1992. Our goal is to provide an organized look at the literature, with particular attention to important questions that remain open for further research. Larry Brown did not restrict his review of earnings forecasting research to the role of financial analysts. We focus more narrowly on research related to analysts’ decision processes and the usefulness of their forecasts and stock recommendations.
Kothari (2001) provides an excellent overall review of capital markets research, and we refer the reader to that paper for a broader perspective. Since 1992 no less than 250 papers related to financial analysts have appeared in the nine major research journals that we used to launch our review of the literature. We also note that during the six months ending March 15, 2005 alone 105 new working papers with the word “analysts” in the abstract have posted to the Social Sciences Research Network, so the task for the next authors of a review article in this area will be even more daunting.
In our review of papers published during the past 12 years, we find much progress in 1 Also see Givoly and Lakonishok (1983) for a review of analysts’ forecasting research prior to 1983. 1 some of the areas identified by Schipper (1991) and Brown (1993) and less progress in other areas. In particular, the research has evolved from descriptions of the statistical properties of analysts’ forecasts to investigations of the incentives and decision processes that give rise to those properties.
However, in spite of this broader focus, much of analysts’ decision processes and the market’s mechanism of drawing a useful consensus from the combination of individual analysts’ decisions remain hidden in a black box. Furthermore, we still have much to learn about the relevant valuation metrics and the mechanism by which analysts and investors translate forecasts into present equity values. For example, with renewed popularity of the earnings-based valuation model in the early 1990s, research turned toward an investigation of the model’s role in the market’s conversion of analysts’ earnings forecasts into current equity values.
Given the unexpected result that this model does a relatively poor job of explaining the variation in market prices and analysts’ price forecasts and recommendations, researchers have turned their attention to examining heuristics that might better explain analyst and market decisions about firm value. We still have much to learn about the heuristics relied upon by analysts and the market and appropriateness of their use. The rest of this paper draws attention to these and other issues that have arisen in the last 12 years.
The next section provides a summary of the questions identified in Schipper (1991) and Brown (1993) and the directions for future research suggested by those authors, as well as the authors of the four papers commenting on Brown (1993). Section 3 reviews papers published mainly since Schipper (1991) and Brown (1993) and also identifies new research questions that emerge from our reading of the literature. Section 4 provides concluding comments highlighting the areas we consider most promising for future research. 2 2.
Perspective from Schipper (1991) and Brown (1993) Katherine Schipper’s (1991) commentary makes two major points. First, she suggests that research regarding analysts’ earnings forecasts tends to focus too narrowly on the statistical properties of the forecasts, without considering the full decision context and economic incentives that may affect those properties. She takes the perspective that the analyst’s job is to provide buy-sell-hold recommendations and provide research reports to support those recommendations.
Schipper (1991) describes analysts’ earnings forecasts as one component of their research reports and a means to an end rather than an end in themselves. She suggests that a more complete description of analysts’ economic incentives and the role of earnings forecasts in the full decision context of analysts should lead to richer hypotheses regarding the statistical properties of the earnings forecasts. The second major point (related to the first) is that research regarding the statistical properties of analysts’ earnings forecasts focuses on the outputs from rather than the inputs to the analyst’s decision process.
The commentary calls for more research into how analysts actually use accounting information and their own earnings forecasts in making decisions. From Larry Brown’s (1993) review paper, we glean four key points. First, he points out that models producing the most accurate forecasts of an earnings variable should also produce the best proxies for the market’s expectations, assuming market efficiency and assuming the research design correctly models the valuation implications of the earnings variable.
Under these assumptions, in Brown’s words, “predictive ability and association are two sides of the same coin. ” Brown notes mixed results on this issue and calls for future research to sort out whether the apparently conflicting results stem from research design problems or market inefficiency. Second, Brown admonishes researchers to carefully consider whether summary files of I/B/E/S 3 consensus forecasts make sense for their studies.
Although the date of the I/B/E/S report and coding of the forecast horizon indicates a timely consensus, the consensus may contain stale forecasts not updated since the information event on which the study intends to condition the forecasts (O’Brien, 1988). Brown suggests that using the I/B/E/S detail files can avoid this problem, and that forecast timeliness is a crucial attribute for effective proxies for analyst earnings expectations when evaluating the accuracy or price-relevance of those expectations. Third, Brown (1993) calls for research to better understand the role of analysts’ forecasts in postearnings announcement drift. In particular, he calls for research into reasons for variation in the degree and speed of forecast convergence following earnings announcements and the effect, if any, of forecast convergence on post-earnings announcement drift. Finally, like Schipper (1991), Brown calls for research to better understand the decision processes of analysts and the roles of analysts’ earnings forecasts, macroeconomic and industry factors, and other information in formulating stock price forecasts and recommendations.
In terms of research methodology, both Brown (1993) and Schipper (1991) indicate that behavioral research can play a more prominent role in understanding the uses of accounting and other information to make stock recommendations, within the full context of the analyst’s decision environment and economic incentives. In Brown’s words, “joint research by capital markets researchers and behavioralists to examine these issues more thoroughly would considerably enhance our understanding of the role of analysts in the price formation process. It will be interesting to see the extent to which the role of behavioral research has expanded over the last 12 years. 2 Most of the studies reviewed by Brown (1993) relied on either I/B/E/S consensus or Value Line data. With less frequency, studies also used Merrill Lynch’s Opinion Alert, Standard and Poors Earnings Forecaster and Zack’s Investment Research. Some used detail files from I/B/E/S and Zacks which, as the paper points out, only became readily available towards the end of the period reviewed. 4
Four authors commented on Brown (1993), and each provides interesting insights and suggestions for future research. For example, O’Hanlon (1993) calls for investigations of the degree to which financial analysts’ earnings forecasts effectively distinguish permanent from temporary earnings changes. Thomas (1993, p. 327) suggests that the importance of research into how analysts make earnings predictions depends on the answers to several questions, including: whether analysts’ forecasts influence the marginal investor; whether they seek to predict reported earnings or a ‘core’ arnings number that will persist in the future; and whether their incentives are consistent with producing the most accurate forecasts possible. Philip Brown (1993) calls for research into whether some analysts are better forecasters than others, whether the market’s earnings expectations reflect these differences, and the degree to which consensus forecasts drawn from analyst tracking services such as I/B/E/S reflect investor expectations. Zmijewski (1993) focuses on the need for investigations of cross-country variation in properties of earnings forecasts and their roles in price formation in capital markets.
Based on our reading of Schipper (1991) and Brown (1993) and its related comment papers, along with an initial look at the research published over the last 12 years, we decided to organize our review around the following seven broad research areas: (1) What is the nature of analysts’ decision processes and how do analysts rationalize the forecasts and recommendations contained in their research reports? (2) What is the nature of analyst expertise and what are the characteristics of distributions of individual analyst earnings forecasts? 3) How informative are the outputs from analyst research (including earnings forecasts, target price forecasts, stock recommendations and conceptual analysis)? (4) Do analysts’ forecasts and recommendations efficiently impound information about future earnings? Do stock prices efficiently impound the information in analysts’ forecasts and recommendations? (5) How do analyst and management 5 incentives affect the statistical properties of analysts’ forecasts? (6) How does variation in the regulatory environment (over time and across countries) affect the behavior of analysts’ forecasts and the role of analysts in capital markets? 7) What are some research design and database issues that threaten the validity of inferences from studies of the behavior of analysts and their forecasts and recommendations? The next section is divided into seven subsections that selectively review research papers addressing these questions, with selective focus on papers published since Brown (1993). 3. Selective review of research related to the role of financial analysts in capital markets The questions at the end of section 2 above naturally arise from the picture of the analyst’s reporting environment shown in figure 1.
Figure 1 puts the analyst’s overall reporting environment in perspective. The analyst obtains/develops information from various sources including: earnings and other information from SEC filings, such as proxy statements and quarterly and annual reports; industry reports and reports describing macro-economic conditions; and conference calls and other management communications. From this information the analyst produces earnings forecasts, target price forecasts and stock recommendations, along with a conceptual report describing the firm’s prospects.
Investors use these outputs from analyst research to make trading decisions that affect market prices. If the analyst forecasting process and capital markets are efficient, then market prices and analysts’ forecasts immediately reflect all of the information described in the figure. Inefficiencies create predictable analyst forecast errors and stock price changes. The decision processes and analyst research output pictured in figure 1 depend on overall governing forces including: regulatory and institutional factors that vary over time and across countries, and analysts’ economic incentives.
Finally, limitations associated with archival databases and econometric/analytical research technology create 6 research design issues that constrain our views of the forces that ultimately drive market prices. We launched our review by listing and categorizing all papers related to analysts and published since 1992 in the following nine major research journals: The Accounting Review, Contemporary Accounting Research, International Journal of Forecasting, Journal of Accounting and Economics, Journal of Accounting Research, Journal of Finance, Journal of Financial Economics, Review of Accounting Studies, and Review of Financial Studies.
However, we expanded the set of papers as needed, given references to papers outside of the initial timeframe or published in other journals. We discuss papers as they apply to each of the seven research areas described at the end of section 2 above. These seven subsections are also shown in figure 1 and form our outline below. Our goal is not to provide exhaustive reviews of (or even references to) all of the papers published since 1992, but rather to selectively discuss aspects of papers that we think capture the pulse of the research and that suggest new questions that might be addressed during the foreseeable future. . 1. Inputs to analysts’ earnings forecasts and stock recommendations This section reviews research regarding the role of earnings and other information in the broader context of the decision processes analysts use to produce their research reports and stock recommendations. 3. 1. 1 Inputs to research reports Learning about the information analysts use and understanding analysts’ decision processes is no easy matter. Researchers have used surveys to simply ask analysts how they process information (e. g. , Block 1999), protocol analysis to record analysts’ thought processes as they process information (e. . , Bouwman, et al. 1995), content analysis of analysts’ research reports to infer the information analysts’ rely upon to make forecasts and recommendations (e. g. , 7 Rogers and Grant 1997), and laboratory experiments to study how analysts use information (e. g. , Maines, et al. 1997). As opposed to the research methods mentioned above, archival studies potentially offer more generalizable results, but are limited in their ability to penetrate the black box containing analysts’ actual decision processes.
The challenge is that analysts have a contextspecific task that is very difficult to model. The rest of this section reviews research from each of the paradigms mentioned above. Previts et al. (1994) examine a broad cross-section of 479 sell-side analyst reports (from Investext) to ascertain the information that analysts apparently use to make decisions. The content of the reports suggested heavy analyst use of earnings-related information and a strategy of disaggregating firm-level information into segments beyond the disaggregation level provided in GAAP-based segment reporting footnotes.
The study also finds evidence of substantial analyst effort to extract non-recurring items and focus on “core” or “adjusted” earnings as a basis for forecasting future earnings. The study also reports heavy analyst reliance on management for information, emphasizing the intermediation role of financial analysts. Interestingly, the authors also observe that analysts prefer following firms with effective strategies for presenting smooth earnings streams. The paper reports (p. 3) that “analysts most frequently refer to accounting earnings quality in terms of a company’s ability to manage earnings through the establishment and adjustment of conservative, discretionary reserves, allowances, and off-balance-sheet assets, which provide analysts a low-risk earnings platform for making stock price forecasts and buy/sell/hold recommendations…” It would be interesting to know if analysts continue to endorse such views in the wake of Reg. FD and Sarbanes-Oxley. Rogers and Grant (1997) extend Previts et al. y using “context-specific” content analysis to examine 187 sell-side analyst reports issued between July 1993 and June 1994 (obtained from 8 the One Source database). They report that only about one-half of the information in the analysts’ research reports could be found in the corresponding corporate annual reports, consistent with analysts also using other, external, information . Further, even within the annual report about one-half of the information seems to have been obtained from the narrative sections (e. g. MD&A and president’s letter) rather than the basic financial statements. Thus, examining analyst reports based solely on quantitative information may not capture the complex nature of the analyst’s task. Using protocol analysis, Bouwman et al. (1995) confirm the aforementioned findings from content analysis of research reports that analysts rely on a variety of sources in their analysis. 3 They further find that, in making investment decisions, the nature of information used by buy-side analysts varies among the different stages of the decision process.
In the familiarization stage, which involved getting to know the company and its general financial condition, subjects relied heavily on the GAAP-based materials, especially performance ratios and the historical summary provided in the 10-K. They relied most heavily on segment information (with more emphasis on non-GAAP materials) while performing an in-depth analysis of the company. Finally, while integrating material from the other phases and reaching a final screening decision, the analysts relied more heavily on non-GAAP general company information.
Analysts also often verbalized a desire for information not included with the materials, such as more segment information, a 10-year rather than 5-year historical summary, 3 Bouwman et al. (1995) asked 12 buy-side analysts to “think out loud” as they examined a variety of information to determine whether a particular company merited further consideration as an investment opportunity. The analysts took an average of 52 minutes at the task and generated an average of 4,000 recorded words, which the researchers transcribed into nearly 400 pages of text.
GAAP-based materials included the company’s most recent 10-K, 10-Q and proxy statement; and non-GAAP materials included a COMPUSTAT report, an S&P stock report, an industry level report and current stock price information. 9 more forward looking information about management’s plans and one or two research reports by sell-side (“Street”) analysts. Maines, et al. (1997) use experimental methods to examine the importance of segment disclosures to analysts’ decisions. Subjects were presented with a hypothetical firm’s financial statements, which included results for two segments, one of which included two divisions. The study manipulated two factors: congruence/incongruence between the segment definitions for internal and external reporting purposes; and similarity/dissimilarity between the product lines of the divisions combined in the segment with two divisions. As long as the analysts were told that the company’s external reporting conformed to its internal reporting, there was no evidence to suggest that dissimilarity between divisions combined into a segment diminished analyst confidence in their forecasts and recommendations.
The results suggest that analysts suspect companies of obscuring the definitions of segments in order to hide performance issues, and that the FASB’s new conformity rules for segment reporting should alleviate these suspicions. The same experiment was conducted with 60 MBA students. Unlike the experienced analysts, the MBA students were unable to discern from information in the MD&A section of the annual report whether or not the two divisions combined into one segment were similar or dissimilar.
This calls into question the use of student subjects as surrogates for financial analysts and also suggests the need for research to better understand the development of analyst expertise. Several experimental studies have evaluated whether classification issues affect expert analysts’ judgments. For example, in an experiment involving 56 experienced buy-side analysts who specialize in banking industry stocks, Hirst et al. (2004) evaluate whether classification of gains and losses on financial assets and liabilities affected analysts’ judgments. The researchers 4 Their subjects averaged 5. years of experience and included 42 buy-side analysts, 8 sell-side analysts and 6 analysts involved in other investment-related jobs. 10 manipulated two variables: the reporting of gains and losses due to interest rate risk (full-fairvalue versus piecemeal-fair-value); and the banks exposure to interest rate risk (fully hedged versus exposed). In the full-fair-value condition, fair values of all financial assets and liabilities were recorded on the balance sheet and all gains and losses were reported on a performance statement immediately following the income statement.
Under piecemeal-fair-value accounting the same information was available, but deposits and liabilities were not marked to market on the balance sheet and gains and losses on those instruments were reported in the footnotes. The results indicate that the analysts were only able to effectively adjust their risk assessments and valuation judgments for the higher risk of exposed banks when the financial statements of those banks applied full-fair-value accounting.
The study also examined the effect of the analyst’s workload and found that analysts following less than 40 (the sample median) stocks in their normal work environment performed significantly better in distinguishing the risk characteristics of exposed versus fully hedged banks. This suggests the need for further research into the effects of spreading analysts’ workloads across larger numbers of stocks. It also suggests an important reason why buy-side analysts have an interest in the reports of sell-side analysts who typically follow far fewer stocks.
Other experimental studies showing similar effects of accounting method or classification on the judgments of experienced analysts include Hopkins (1996), Hirst and Hopkins (1998) and Hopkins et al. (2000). A number of archival studies suggest that complexity affects analyst forecast accuracy (e. g. , Brown et al. ; Haw et al. ; Lang and Lundholm ; Duru and Reeb ; Clement ). Perhaps the most direct test of this proposition is in Plumlee (2003), which examines the effects of six aspects of the Tax Reform Act of 1986 (TRA86) on errors in Value Line analysts’ effective tax rate forecasts for 355 firms.
The six tax law changes vary in 11 complexity from a simple decline from 48% to 36% in the corporate statutory tax rate to the highly complex implementation of an alternative minimum tax. The results indicate a significant positive relation between the absolute values of errors in forecasting effective tax rates and the research proxies for the expected effects of the three most complex tax law changes; and an insignificant relation with the three least complex tax law changes.
Plumlee interprets her results as indicating that higher information complexity reduces analysts’ use of the information, due either to analysts’ processing limitations or time constraints. Since the research design did not predict the direction of the forecast errors, an alternative explanation is that analysts obtained and efficiently processed all possible information regarding the effects of the more complex tax law changes, but because those effects were highly uncertain, forecast errors were large in absolute value for firms most affected.
Further research is needed to distinguish between these explanations. Research also suggests that analysts are more likely to cover firms that provide them with more (better quality) information required for analysis . Lang and Lundholm (1996) find that the quality of corporate disclosures (as rated by expert analyst committees) affects analysts’ coverage decisions and the accuracy of their forecasts. In a more detailed study of the disclosure items that analysts value most, Healy and Hutton (1999) identify 97 firms whose disclosure quality ratings jumped significantly during the period between 1978 and 1991.
They find that the following factors apparently play an important role in analysts’ evaluations of firms’ disclosures policies: segmental reporting quality, quality and candidness in the management discussion and analysis (MD&A) section of annual and quarterly reports; publication of supplemental disclosures outside of required periodic reports; and availability of management to analysts. Consistent with Healy and Hutton (1999), Botosan and Harris (2000) find that analyst 12 following increased with firms’ decisions to include information on segment activity as part of their quarterly (as opposed to only annual) reports.
In other examples of studies that relate disclosure quality to analysts’ forecasting decisions, Bowen et al. (2002) and Barron et al. (1999) find that analysts’ forecast accuracy depends on conference call information and the quality of managers’ MD&A disclosures, respectively; and Williams (1996) finds that analysts’ reliance on management earnings forecasts depends on the reliability of the forecast as measured by past management forecast accuracy. Chandra et al. (1999) find that analysts effectively use industry trade association disclosures to estimate the persistence of firm-specific sales changes.
Ely and Mande (1996) find that analysts’ earnings forecast revisions reflect the corroborative information in dividend and earnings announcements, particularly when the earnings information is noisy. Thus, analysts appear to rely on detailed information in corporate reports, and their incentives to cover firms and report accurately appear to be related to the quality of corporate disclosures. We expect archival research to continue identifying associations between information variables and analysts’ forecasting decisions.
However, archival association studies cannot determine whether analysts actually use the specific information variables or whether analysts’ private information production activities produce signals that are correlated with the public signals. Experimental research and protocol analysis can play important roles in evaluating the relative importance of each public information variable, as well as the importance of analysts’ private information development activities, to analysts’ actual forecasting decision processes. 3. 1. The role of earnings forecasts and other information for analysts’ recommendations Bandyopadhyay et al. (1995 ) examine “the decision-usefulness of analysts’ earnings forecasts for their price forecasts” (p. 430). They rely on a sample of 128 firms on the database. 13 They obtain analysts’ price and earnings forecasts for 128 firms from Research Evaluation Service (RES), and Value Line price and earnings forecast data for the 46 firms in the group that are also covered by Value Line. RES supplies forecasts of prices over a 12 month forecast horizon and Value Line supplies price forecasts over a 3-5 year forecast horizon.
Both databases provide current year and next year earnings forecasts, and Value Line also provides a 3-5 year earnings forecast. The primary results show that revisions (from June 30 to September 30) in RES forecasts of the next year’s earnings explains about 30% of the variation in revisions in RES’s 12-month ahead price forecast. In contrast, revisions in Value Line’s 3-5 year earnings forecast explain about 60% of the variation in the revisions in Value Line’s 3-5 year price forecasts.
Thus, the authors conclude that analysts’ earnings forecasts provide important inputs to their price forecasting decisions, and the relation between the two variables increases with the forecast horizon. While these results provide insight into analysts’ decision processes, they do not provide information as to the mechanism used by analysts to convert earnings forecasts into price forecasts. Block (1999), in a survey of 880 members of the Association for Investment Management and Research (AIMR), finds that nearly half of the respondents never used present valuation techniques, while only 15% said they always used present value techniques.
Demirakos et al. (2004) use content analysis to examine sell-side analyst research reports (from Investext) and also find that analysts overwhelmingly refer to simple P/E multiples to support their stock recommendations. Bradshaw (2002) analyzes research reports of analysts and he too finds that analysts do not refer to present value techniques as a method they rely on to support their recommendations. 5 However, Bradshaw (2004, p. 7) points out that analysts may choose 5 Consistent with the view that analysts rely more on multiples to value stocks, as opposed to relying on variants of DCF valuation models, only 23% of the reports contain earnings forecasts with horizons beyond the next fiscal year. 14 to communicate with investors in terms of simple heuristics that correlate with more sophisticated multiperiod present value models underlying analysts’ valuation and recommendation judgments.
He finds that a simple heuristic based on analysts’ consensus longterm growth rate forecasts explains 23% of the variation in analysts’ consensus stock recommendations (both variables from First Call over the 1994-98 time period). However, this simple heuristic (i. e. , buy stocks with high long-term growth rate forecasts) is negatively correlated with value-to-price ratios derived from more sophisticated residual income valuation models. Furthermore, the long-term growth rate forecasts are also negatively correlated with abnormal returns over the year following the publication of the consensus forecasts and recommendations.
This evidence is consistent with analysts pushing stocks whose high longterm growth rate forecasts have already been overpriced by the market. Bradshaw also finds evidence to suggest that the value-to-price ratios are positively associated with future abnormal returns but negatively associated with the analysts’ recommendations. All in all, Bradshaw’s evidence suggests that analysts do not use their own earnings forecasts efficiently in making recommendations. However, Bradshaw’s sample period corresponded to a time period when the market was overheating, perhaps due to analysts’ pushing their long-term growth forecasts.
It will be interesting to examine if heuristics used by analysts to generate recommendations change over time, as well as the effects of these heuristics and recommendations on stock prices in different time periods. If analysts’ valuation judgments do not conform to finance theory, what models do analysts use to convert their forecasts into value judgments? Bradshaw (2002) examines approximately 100 analysts research reports from Investext, dated (primarily) in 1998 or 1999, Also, consistent with this view, the study finds that only 13% of the analyst reports refer to any variation of a DCF valuation model in computing price targets. 5 and finds that analysts most frequently justify their recommendations with references to P/E ratios and long-term growth rate forecasts. Thus, it appears that analysts combine their longterm growth forecasts with the firm’s P/E ratio to reach a valuation and recommendation decision. The PEG ratio, a popular Street heuristic, suggests that a firm’s forward P/E ratio should equal 100 times its long-term growth rate forecast (Lynch 1989, p. 198). Bradshaw (2002) uses this heuristic to create pseudo price targets and finds that these pseudo price targets are highly correlated (r=0. 9) with the level of analysts’ buy/hold recommendations and with analysts’ reported target prices (r=0. 50). On the other hand, he finds smaller correlations of reported target prices with pseudo target prices based on industry P/E multiples (that do not consider differences across firms in long-term growth rates) and no correlation between these P/E multiples-based pseudo target prices and recommendations. He concludes that the PEG ratio is an important heuristic used by analysts to convert their earnings forecasts into target prices forecasts and recommendations.
While Bradshaw (2004) finds that consensus analyst recommendations based on analysts’ consensus long-term earnings growth rate forecasts do not predict abnormal returns, he does not examine the association between the relative accuracy of an individual analyst’s earnings forecasts and the profitability of the analyst’s stock recommendations. Loh and Mian (2005) address this issue. Specifically, they compare the profitability of stock recommendations of relatively accurate earnings forecasters to those of poor earnings forecasters in any given firmyear.
Relying on I/B/E/S earnings forecasts and recommendations related to over 32,000 firmyears between 1994 and 1999, they find that monthly abnormal returns on hedge portfolios based on recommendations of analysts in the top (bottom) quintile of earnings forecast accuracy are, on average, approximately 0. 74% (-0. 53%). The differences are highly significant, both statistically 16 and economically. The authors infer that efforts by analysts to produce accurate earnings forecasts pay off in terms of the profitability of their stock recommendations.
Thus, it appears that analysts use their earnings forecasts to produce stock recommendations, with more accurate forecasters providing more profitable recommendations. However, the model translating the earnings forecasts into valuation and recommendation judgments remains an elusive issue for further research. 3. 2. The nature of analyst expertise and distributional characteristics of individual analyst earnings forecasts In this section, we discuss research related to two commonly studied properties of analysts’ earnings forecasts: forecast accuracy of individual analysts and the dispersion in forecasts provided by all analysts for a firm. . 2. 1 Forecast Accuracy Studying earnings forecast accuracy of individual analysts is important for at least two reasons. First, investors benefit from identifying more accurate forecasts (and forecasters). Earnings forecasts are an input to analysts’ stock recommendations; more accurate forecasters may provide more profitable stock recommendations, i. e. , better input leads to better output (Loh and Mian ). Second, from a researcher’s perspective identifying more accurate forecasts is important because in an efficient market, the market’s expectation should reflect the best (most accurate) information available at any point in time.
Studies that use analysts’ forecasts as proxies for the market’s earnings expectation should take into account investors’ ability to identify and differentially weight earnings forecasts of individual analysts (Maines ). Forecast accuracy is also important to analysts. More accurate forecasters are likely to be rewarded and less accurate forecasters may be forced to change brokerage houses or leave the profession. The reward to accurate forecasting may be in the form of recognition (for example, 17 being selected to the Institutional Investor All American team) and/or career advancement (Hong and Kubik ).
Mikhail et al. (1999) examine the relation between forecast accuracy and analyst turnover and find that analysts who contribute forecasts to the Zacks database are more likely to change brokerage firms or leave the database altogether when their forecast accuracy is lower relative to their peers. They find that the profitability of analysts’ stock recommendations is unrelated to analyst turnover, suggesting that analysts may have more of an incentive to issue accurate forecasts than to provide profitable stock recommendations.
Forecast accuracy therefore seems important to both analysts and investors. Research on analysts’ earnings forecast accuracy has focused on two main attributes: 1) the nature of the forecast itself, for example, whether the forecast reflects new information or whether the analyst is merely herding with the consensus; and 2) characteristics of the analyst including his/her affiliation, for example, the analyst’s prior experience and the brokerage firm that the analyst represents. 3. 2. 1. 1.
Forecast characteristics Analyst forecasts for a firm differ on a number of dimensions like age (time between the forecast date and the related earnings announcement) and implied information in the forecast. It is fairly well established in the literature that recent forecasts are more accurate (O’Brien 1988). Gleason and Lee (2003) find that the price impact of forecasts depend on whether the analyst brings new information to the market. Analysts’ earnings forecast revisions may bring their previous forecast closer to the current consensus (generally referred to as herding), or they may diverge from the existing consensus.
Gleason and Lee (2003) show that forecast revisions are more informative, i. e. , elicit bigger price responses, when they diverge from the consensus. Clement and Tse (2005) find that one reason why bold (diverging) forecasts have bigger price 18 impact is because they are more accurate. Further, they find that bold forecast revisions tend to improve the previous forecast of the same analyst to a greater extent than herding forecasts. This is consistent with bold forecasts conveying more of the analyst’s private information about the firm.
Consistent with the predictions in Trueman (1994), they also find that smaller forecast revisions are more highly correlated with forecast errors (after the revision); herding analysts revise their forecasts in the direction suggested by their information, but are less likely than bold analysts to fully incorporate this information in their forecasts. Analysts may have to make a trade-off between timeliness of their forecasts and forecast accuracy, i. e. , they could quickly issue forecasts in response to new information, or wait for additional information/analysis to provide more accurate forecasts.
Cooper, Day and Lewis (2001) study the market response to forecast revisions by lead and follower analysts in one hightech industry (semiconductors and printed circuit boards) and one low-tech industry (restaurants). They find that the price response to forecast revisions of lead analysts, defined as analysts who provide timely forecasts, in both industries is higher than the price response to follower analysts. Their results also suggest that timeliness is valued more by investors than ex post accuracy in the forecasts.
Mozes (2003) also finds some evidence that timely forecasts are less accurate ex post, but do improve on the accuracy of the existing forecasts. From an investor usefulness standpoint, however, the timeliness of the forecast seems to be at least as important a criterion as forecast accuracy. Together these results suggest that while accuracy may be an important criterion in evaluating forecasts, analysts could easily improve their accuracy by herding with (or improving on) existing forecasts of superior analysts.
Thus studies that evaluate analyst forecast accuracy using ex post data may identify analysts as “superior” in terms of forecast accuracy, but these 19 analysts may actually not be bringing any new information to the market place. Reliance on ex post accuracy measures may also explain why forecast accuracy increases as the age of the forecast decreases. Analysts providing forecasts later in the period have the advantage of observing the predictions of other analysts in addition to other information about the firm. Sinha et al. 1997) recognize the effect of forecast age on accuracy and find that forecast accuracy differs across analysts, but only after controlling for the relative age of the forecasts. They further find that analysts identified as superior ex ante, on either firm-specific or industry levels, continue to provide more accurate forecasts in the subsequent year. However, Sinha et al. do not control for forecast age in their predictive tests, which would affect their conclusions if for example, superior forecasters systematically issue forecasts later in the holdout period.
Interestingly, they find that inferior analysts do not provide poorer earnings estimates over the next year. However, their evidence is consistent with analysts differing in earnings forecast accuracy even after controlling for ex post bias in identifying accurate forecasters. If analysts have superior information and bold forecasts are valued more by investors, why do some analysts choose to herd (and not fully convey their private information)? 6 Trueman (1994) suggests that forecast boldness is related to analysts’ self-confidence.
Analysts who have confidence in their forecasting abilities are more likely to issue bold forecasts while analysts who have lower confidence in their abilities are likely to herd. Hong et al (2000) find that analysts with less experience are more likely to herd, suggesting that career concerns may inhibit analyst boldness. Clarke and Subramanian (2005) argue that prior forecasting performance should be related to the degree of boldness in future forecasts. Specifically, they It is also possible that analysts issue similar forecasts (i. e. , appear to herd) because they possess the same information.
Welch (2000) in a study of analyst recommendations finds some evidence that herding towards the consensus does not appear to be information driven. Specifically, he finds that analyst recommendations do not herd to the consensus any stronger when the consensus recommendation turns out to be a good predictor (ex post) of future stock returns. 6 20 analytically demonstrate a U-shaped relation between forecast boldness and prior forecasting performance, i. e. , analysts with either relatively good or poor performance in the past are more likely to issue bold forecasts than analysts whose past accuracy is closer to the average.
This relation is driven by compensation and career concerns and is unaffected by the ability of the analyst or the quality of the analyst’s private information. Consistent with their model’s predictions, they find that I/B/E/S analysts (1988-2000) with relatively superior or poor prior performance are more likely to issue bold forecasts than analysts in the intermediate group. Another reason why analysts may herd is because they may be concerned about their reputation, or their private information may be inconsistent with contemporaneously available public signals (Graham ).
It is also possible that more uncertainty regarding a firm’s future performance may lead to herding among analysts. An interesting extension of this stream of research will be to examine whether forecasting difficulty is associated with herding behavior. For example, is herding behavior more prevalent for firms that have higher variance in actual earnings? Higher dispersion in analysts’ forecasts (which is inverse of herding behavior) is generally seen as an indication of analyst uncertainty with respect to firm earnings.
It could well be the case that analysts may choose to herd when earnings are more uncertain, leading to lower dispersion in forecasts for firms with less predictable earnings. 3. 2. 1. 2. Characteristics of the forecaster Given that more accurate forecasts are more value relevant, identifying expert forecasters is a profitable strategy for investors. Characteristics that are associated with analyst superiority should also be of interest to brokerage houses (employers), in trying to enhance the quality of their output. If the quality of analyst forecasts and recommendations differ systematically based 1 on analyst characteristics, then researchers could also use these characteristics to compute superior measures of earnings expectations. One way to identify expert forecasters is by relying on external agencies, such as Institutional Investor (II), which annually recognizes, by industry, expert forecasters as All Star analysts. In fact, Stickel (1992) shows that II All Stars tend to outperform other analysts in terms of forecast accuracy. Alternatively, one could identify expert analysts by studying analyst characteristics including the brokerage firm the analyst represents. Mikhail et al. 1997) use the Zacks Investment Research database and document that analysts’ forecast errors decrease as their firm–specific experience increases, consistent with analysts learning over time. They also find that experienced analysts’ forecast errors are more closely related to the market reaction around earnings announcements than the forecast errors of less experienced analysts. This is consistent with the market understanding the effect of experience on accuracy. However, they do not find consistent evidence that analysts with greater firm-specific forecasting experience also issue more profitable stock recommendations.
Using analyst forecast data from I/B/E/S, Clement (1999) examines the association between analysts’ experience, affiliation, and specialization on their forecast accuracy. Like Mikhail et al. (1997), he also finds that analysts with more experience are more accurate in their earnings forecasts. Further, he finds that analysts affiliated with large brokerage houses, and who cover fewer firms and industries provide more accurate forecasts. 7 Thus resources available to the employer and specialization also seem to increase forecast accuracy. Mikhail et al. 1997) do not find improvement in forecast accuracy for analysts following fewer industries. However, they use Zacks’ forecasts whereas Clement (1999) uses I/B/E/S forecasts. Mikhail et al. define industry concentration as the number of firms followed by an analyst in the same 2-digit SIC as the subject firm, divided by the total number of firms followed by the same analyst on Zacks. On the other hand, Clement’s measure of industry concentration is the number of 2-digit SICs for which an analyst issues forecasts on I/B/E/S. 7 22 Jacob et al. (1999) argue that certain analyst-company alignments (which hey refer to as “analyst aptitude”) may be more successful (in terms of accurate forecasting) because some analysts may have a natural aptitude in forecasting earnings for particular firms. Increases in firm-specific experience, as measured by the length of time over which analysts have made earnings forecasts for a firm, may merely be a manifestation of analysts’ ability to better forecast earnings for that firm. They reexamine Clement’s findings and find that the positive association between experience and forecast accuracy diminishes after controlling for analyst-company alignment.
They conclude that it is more likely analysts’ aptitude rather than overall experience that explains analyst forecasting superiority. They also find that analysts are more accurate in firm-quarters where they make more frequent forecasts. It is certainly useful to understand why certain analysts are superior forecasters than others. However, if investors are merely interested in identifying superior analysts (in terms of forecasting accuracy), gathering information on these various analyst and brokerage firm characteristics may not be beneficial, especially if other (simpler) avenues to predict future performance are available.
Brown (2001b) shows that a simple model that only takes into account past forecast accuracy of analysts does at least as well as the more complicated analyst characteristics model (Clement ) in predicting future forecast accuracy. Brown and Mohd (2003) also find that forecast age is as good a predictor of forecast accuracy, and is equally representative of investor expectations, as the five analyst characteristics model in Clement (1999). Further if as Cooper et al. 2001) show, timeliness is more important than accuracy when it comes to investment decisions, the importance of studying forecast accuracy for economic reasons may be diminished. 23 Analyst characteristics that are associated with forecast accuracy may also be important for brokerage houses that employ analysts. Notwithstanding the fact that brokerage houses have more information about the analyst (other than just forecast accuracy), Mikhail et al. ’s (1999) finding that analyst turnover is higher when analysts are inferior relative to their peers suggests that brokerage houses do evaluate analysts based on forecast ccuracy. This finding together with Jacob et al. ’s (1999) finding that certain analyst-company alignments are more successful than others suggests that even if employers do not terminate inferior analysts (based on relatively inaccurate forecasts for certain firms), the analyst could be reassigned to cover other firms. An interesting extension of the Jacob et al. finding will be to examine whether analysts who are accurate forecasters for certain companies but are not as accurate with others, continue with the same brokerage house but get reassigned out of the companies in which they are not very accurate. Another interesting issue for future research is to examine why certain firms assign their analysts to cover more firms and industries when analyst accuracy is improved by following fewer firms and industries. Clement (1999) reports that analysts in the 90th percentile of his sample cover 21 (seven) more firms (industries) than analysts in the 10th percentile. While a quick explanation is that these are most likely smaller brokerage firms that employ fewer analysts, the role of such “over worked” analysts in the market is still an interesting issue.
Specifically, what is the role of these “inferior” analysts when other, presumably superior, Hong and Kubik (2003) provide some evidence on this issue. They find that relatively inferior analysts are more likely to be taken out of covering prestigious stocks (defined as stocks followed by at least 20 analysts and/or with market cap over $5 billion) even if they continue to work for the same brokerage firm. 8 24 analysts cover the same company for larger brokerage houses? Do investors respond to the forecasts of the inferior analysts, and if so, why? 3. 2. 2. Dispersion in analyst forecasts as a measure of investor uncertainty Forecast dispersion (measured as the standard deviation in analyst forecasts), which is a signal of the extent of analyst disagreement about a firm’s upcoming earnings, is generally used as a proxy for investor uncertainty prior to information events. The reasoning (similar to the reasoning for using the mean/median earnings forecast as the market’s expectation of earnings) is that disagreement among analysts reflects general disagreement among investors.
Based on the notion that investor disagreement is one factor that triggers trading in stock markets, forecast dispersion is generally used to study trading volume around information events such as earnings announcements. Barron (1995) builds on prior research that had documented a positive relation between trading volume and the level of prior dispersion in analysts’ forecasts as well as changes in analyst forecast dispersion (e. g. , Ziebart 1990).
He suggests that even if there is no change in the level of dispersion, trading may result because analysts change their relative positions from one forecast period to the next. He refers to this reordering of analyst beliefs as “belief jumbling. ” Using I/B/E/S annual forecast data from 1984-1990 he finds that belief jumbling is positively related to monthly trading volume even after controlling for other variables that have been posited to be related to trading volume such as the absolute price change (Abarbanell et al. 1995]). Consistent with prior research, he also finds that both the level of prior dispersion in analysts’ forecasts and changes in analyst forecast dispersion are positively related to monthly trading volume. Mikhail et al. (2004) find that stock recommendations of superior analysts (measured based on profitability of prior year recommendations) continue to be more profitable than that of inferior analysts. However, even stock recommendations of historically inferior analysts are profitable (see their tables 3 and 4). 9 25
While Barron (1995) focuses on the general relation between monthly trading volume and monthly disagreement measures, Bamber et al. (1997) extend the analysis to earnings announcement periods. They restrict their sample to firm-quarters (I/B/E/S forecasts; 19841994) where at least five analysts issue annual earnings forecasts in the 45 days preceding interim announcements and also revise their annual earnings forecasts in the 30 days following the announcement. Bamber et al. find that all three measures of investor disagreement identified in Barron (1995) are also positively elated to abnormal trading volume around earnings announcements. Prior research suggests that earnings announcements generally resolve preannouncement uncertainty; therefore, trading around earnings announcements should be related to prior dispersion in beliefs (i. e. , forecast dispersion prior to the announcement). Bamber et al. show that earnings announcements may actually trigger further trading through two other mechanisms: belief jumbling and increases in forecast dispersion from before to after the announcement. Barron et al. 1998) analytically show that the mean forecast error, together with forecast dispersion and number of forecasts, can be used to estimate analysts’ total uncertainty as well as their consensus (common uncertainty relative to total uncertainty). Specifically, they show that higher dispersion implies higher total uncertainty but lower consensus. More importantly, they show that forecast dispersion is a measure of analysts’ idiosyncratic uncertainty and therefore does not fully capture total earnings uncertainty. Total uncertainty is a combination of the common uncertainty shared by all analysts and forecast dispersion.
They point out that decreases in forecast dispersion (after for example, earnings announcements), may not signal a decrease in overall uncertainty but rather a decrease in uncertainty related to the idiosyncratic component of analyst forecasts. 26 Barron et al. (2002) use the Barron et al. (1998) measures of consensus and uncertainty and examine the dispersion in analysts’ annual earnings forecasts before and after interim earnings announcements to test whether earnings announcements spark gathering of private information by analysts (as observed through their forecast revision activity).
Prior research suggested that release of public information such as earnings announcements may decrease the need for private information gathering. Barron et al. (2002) find that consensus among analysts actually decreases in the days following the earnings announcement consistent with analysts embedding more private information into their forecast revisions (for example through their individualistic interpretation of the earnings news). They also show that the decrease in consensus is related to the number of analysts revising heir earnings forecasts after an announcement which could explain the increased revision activity documented by Stickel (1989) following interim earnings announcements. Barron et al. (2005) further extend this line of research and find that trading volume around earnings announcements is related to the extent of private information gathering around earnings announcements. Diether et al. (2002) use I/B/E/S data and find that stocks with high (low) earnings forecast dispersion earn negative (positive) returns in the subsequent month.
The difference in returns between stocks in the highest and lowest quintile of forecast dispersion is 9. 48% (annualized). They also find that the forecast dispersion effect is strongest for small stocks, even though return differences are observed in larger stocks as well. They interpret their results as consistent with Miller’s (1977) prediction that in the presence of short sale constraints, investor disagreement will result in share prices reflecting the most optimistic valuations. This overvaluation of high dispersion stocks leads to negative returns in subsequent periods.
Consistent with the return evidence, Diether et al. also find that analyst optimism is strongest 27 when disagreement among analysts (forecast dispersion) is high. They conclude that their evidence is consistent with forecast dispersion being a proxy for investor disagreement and not as a proxy for risk. Forecast dispersion is negatively related to returns; if dispersion is a proxy for risk, dispersion should be positively related to future returns. Johnson (2004) suggests that the finding in Diether et al. s consistent with a standard asset pricing model where forecast dispersion proxies for uncertainty about an upcoming signal of the value of the underlying asset (i. e. , current earnings). He argues that this effect should be most evident in highly leveraged firms where the (option) value of equity should increase with uncertainty, leading to lower returns in the future. Results based on IBES forecast data are consistent with the predictions; the negative relation between forecast dispersion and future returns documented by Diether et al. xists only for firms with risky debt. Alternatively, Chen and Jiambalvo (2005) argue that the findings of Diether et al. may be driven by post earnings announcement drift that has been well documented in the accounting literature. Higher forecast dispersion is typically associated with poor earnings performance, which is followed by negative price drifts. Findings from the forecast dispersion studies described above suggest more interesting avenues for future research.
For example, in their model of analyst uncertainty, Barron et al. (1998) assume that the precision of private information is the same across all analysts. This assumption appears too restrictive when we consider the empirical evidence regarding the distribution of analyst forecasts for a firm. It will be interesting to examine implications for analyst uncertainty and market trading if this assumption is relaxed and the precision of private information is allowed to vary across analysts. 0 Second, more research is necessary in tying the 10 In a recent working paper, Gu (2005) relaxes this assumption and provides generalized measures of the analysts’ common and private information based on observable forecasts. 28 Barron et al. (1998) uncertainty measures to disclosure practices of firms. For example, do firms that have a reputation of providing higher quality disclosures have higher precision of common information and higher level of consensus as defined by Barron et al.? 3. Information content of analysts’ research output This section addresses questions related to informativeness of output from analysts’ research. The output variables include earnings forecasts, target price forecasts, stock recommendations, and other information rationalizing the forecasts and recommendations. We first discuss research related to the information content of earnings surprise measured with reference to analysts’ quarterly earnings forecasts. Then section 3. 3. 2 discusses research related to the informativeness of analysts’ earnings forecast revisions.
Section 3. 3. 3 discusses research related to the combination of all four output variables, and section 3. 3. 4 discusses research relying on analysts’ long-term earnings growth forecasts. 3. 3. 1 Information content of analysts’ earnings forecast errors The association between returns and analysts’ earnings forecast errors depends on the degree to which the forecast proxies for the market’s expectations at the beginning of the return accumulation period, the value relevance of the earnings variable, and the model translating current earnings into value.
An interesting puzzle emerged with evidence in O’Brien (1988) suggesting that, although more accurate than time-series model earnings predictions, analysts’ quarterly earnings forecasts are not necessarily better proxies for the market’s earnings expectations. 11 As noted by Brown (1993) this raises the question as to why accuracy and association approaches to comparing alternative earnings forecasts are not “two sides of the same 11 Although outside the scope of this paper, Foster (1977) similarly notes the curious result that simpler time series models of quarterly earnings (e. g. seasonal random walk) produce forecasts that outperform more descriptive models (e. g. , seasonally differenced first order autoregressive) in producing forecasts that conform more closely to the market’s earnings expectations. Bernard and Thomas (1989, 1990) and others build on this evidence by demonstrating profitable trading strategies that take advantage of this apparent inefficiency. 29 coin. ” Wiedman (1996) builds on Brown et al. (BRS; 1987) to address this issue. BRS argue that the analyst’s forecasting advantage (relative to time-series models) should increase with the dimensionality of the information set (i. . , the number of available information signals beyond the time-series of historical earnings), and decrease with both the variance of the signals as unbiased estimators of the target earnings variable (i. e. , signal imprecision) and the correlation (or common information) among the signals. Using a more recent sample (1988-91 versus 1977-82) of over 19,000 firm-quarters of I/B/E/S earnings forecasts, Wiedman confirms the BRS results that analysts’ forecasting advantage increases with firm size (the proxy for dimensionality) and decreases with the dispersion in analysts’ forecasts (the proxy for signal variance).
Also similar to BRS, Wiedman fails to find the hypothesized positive relation between superior analyst accuracy and number of lines of business (proxying for lack of correlation among signals). In fact, she observes a significantly negative effect on both the accuracy and market expectations proxy dimensions of analysts’ forecasting advantage. The interpretation of the apparently negative effect of lines of business on analyst’s forecasting advantage awaits further research.
Nevertheless, Wiedman’s results support the n