Retracted: Relationship Incentives and the Optimistic/Pessimistic Pattern in Analysts' Forecasts

Authors

Errata

This article is corrected by:

  1. Errata: Retraction Statement: Relationship Incentives and the Optimistic/Pessimistic Pattern in Analysts' Forecasts Volume 53, Issue 4, 911–912, Article first published online: 19 May 2015

  • We thank Abbie Smith, an anonymous reviewer, Jean Bedard, Rob Bloomfield, Michael Clement, Pat Hopkins, Bin Ke, Lisa Koonce, Mark Nelson, Jay Thibodeau, Yong Yu, and workshop participants at Bentley College, Cornell University, University of Texas at Austin, and Washington University for their helpful comments, Cornell's Johnson School and Bentley College for financial support, and the financial analysts who participated in our experiments.

ABSTRACT

We examine whether analysts' incentives to maintain good relationships with management contribute to the optimistic/pessimistic within-period time trend in analysts' forecasts. In our experiments, 81 experienced sell-side analysts from two brokerage firms predict earnings based on historical information and management guidance. Analysts' forecasts exhibit an optimistic/pessimistic pattern across the two timing conditions (early and late in the quarter), and the effect is significantly stronger when the analysts have a good relationship with management than when their only incentive is to be accurate. Debriefing results indicate that analysts are aware of this pattern of forecasts, and believe that this benefits their future relationships with management and with brokerage clients. The analysts most frequently cite favored conference call participation and information access when describing benefits from maintaining good relationships with management. Our results suggest the following: The optimistic/pessimistic pattern in forecasts is in part a conscious response to relationship incentives, information access is perceived to be a major benefit of management relationships, and recent regulatory changes may have lessened but have not eliminated this conflict of interest source.

1. Introduction

Both the nature and causes of bias in analysts' forecasts have been the subject of research and regulatory interest since the 1960s. Conclusions about the nature of analysts' forecast bias have changed over time (see Givoly and Lakonishok [1984], Schipper [1991], Richardson, Teoh, and Wysocki [2004], Ke and Yu [2006]), with more recent papers concluding that beginning-of-period forecasts tend toward optimism and end-of-period forecasts tend toward pessimism. Such forecasts have been referred to as reflecting a walk-down or optimistic/pessimistic pattern. While the assumed form of the bias has changed over time, management relationship incentives have been the focus of much of the subsequent research into the causes of systematic bias in analysts' forecasts (Koonce and Mercer [2005]). In this paper, we investigate whether analysts' incentives to maintain good relationships with management contribute to the optimistic/pessimistic pattern in analysts' forecasts.

Schipper [1991] discusses how analysts' responsibilities to investor clients may conflict with their interest in maintaining relationships with managers of the companies they follow. She suggests two components of this conflict: generation of investment banking business and access to management information. Ke and Yu [2006] find that analysts issuing forecasts that are optimistic at the beginning of the period and pessimistic at the end of the period (hereafter, OP forecasts) are more accurate and less likely to be fired. They conclude that this evidence supports the management access incentives hypothesis, reasoning that OP analysts' greater success results from preferential access to information. However, their conclusion that management access incentives cause the OP pattern in forecasts is tentative as they do not have a direct measure of relationship incentives or management access, but instead employ analyst accuracy as a proxy for these constructs. Furthermore, Ke and Yu [2006] find that OP forecasts are associated with a variety of analyst and employer variables, and prior literature indicates that some of these variables (e.g., analysts working for firms with superior resources; see Clement [1999]) are also associated with greater analyst accuracy.

The use of accuracy as a proxy for relationship incentives and the association of OP forecasts with other analyst and employer variables suggest a number of alternative explanations for the association between analyst accuracy and OP forecasts. First, analyst/employer attributes may be driving both accuracy and the OP pattern in forecasts. Second, the opposite causal relation may be true. Management may be attempting to curry favor with more influential analysts (who are also more accurate) by providing them with private information. If this information is downwardly biased (e.g., Hutton [2005]) and analysts do not sufficiently adjust for such bias (e.g., Tan, Libby, and Hunton [2002]), influential analysts would be more accurate and more OP, even in the absence of incentives to please management. Third, slight pessimism in end-of-period forecasts creates a positive surprise at the actual earnings announcement. This may boost trading leading to greater resource availability, which, in turn, may lead to greater analyst accuracy.

The purpose of our experiments is to provide the first tests of the causal relation between management relationship incentives (which may be associated with both information and investment banking business access) and the observed OP tendency in analysts' forecasts. In our experiments, we directly manipulate the analysts' relationship with management (instead of using an indirect proxy) and the timing of their forecasts. Attributes of the company, information environment, and management guidance are held constant, and participants are randomly assigned to incentive and timing treatments. This eliminates the alternative explanations discussed above and controls for analyst characteristics and other potential confounding factors. Manipulating the analysts' relationship with management addresses the difficulty in measuring this variable from archival data and manipulating the timing of the analysts' forecasts controls for the fact that not all analysts revise their forecasts in response to guidance. Our participants are sell-side analysts from brokerage firms with underwriting businesses. These analysts are also more likely to be aware of the importance of management relationship incentives than brokerage-only analysts.

In our debriefing, we also examine whether this OP tendency is intentional by assessing whether analysts are aware of the OP time trend in their forecasts and measuring their perceptions of the benefits of issuing forecasts with such a time trend. Because management relations can be associated with benefits related to information or business access, we also obtain qualitative evidence concerning the specific benefits analysts believe they receive in the current regulatory environment. We conduct our experiments in the post–Regulation Fair Disclosure (FD), Regulation Analyst Certification (AC), and Global Settlement environment, thus offering the first test of the effects of relationship incentives in the new regulatory regime. Doing so also biases against our finding the effects of relationship incentives in a more controlled experimental setting.

In our first experiment, 47 experienced sell-side analysts from a single large investment banking/brokerage firm are randomly assigned to one of two incentive and two timing conditions, and asked to predict second quarter and annual earnings per share in response to management's second quarter range guidance. Results indicate that the participating analysts' forecasts follow an OP pattern. Forecasts issued early in the quarter are above the mean of the range guidance, while those issued late in the quarter are below the mean. More importantly, the forecasts are more OP in the management relationship condition (where the analysts had gradually developed a good professional relationship with the company management) than in the accuracy condition (where their only incentive is to be accurate). These results provide the first direct evidence for Ke and Yu's [2006] premise that incentives to please management cause analysts to issue OP forecasts. Ke and Yu's [2006] finding that analysts who issue OP forecasts are more accurate suggests the possibility that historically accurate analysts are more likely to issue OP forecasts in the presence of relationship incentives because they are more aware of the attendant benefits of maintaining this relationship. Our results on this issue are mixed.

In debriefing, the majority of participants expect their forecasts to be lower at the end of the period than at the beginning of the period. The analysts also believe that short-term company guidance is generally downwardly biased and that issuing forecasts above management's guidance which the company fails to meet or beat damages their relationship with management. They further believe that when companies for which they have issued a buy recommendation beat their most current forecast, it improves their relationship with brokerage customers who have purchased the stock based on their recommendation. All of these findings are consistent with intentional bias on the part of the analysts.

Our second experiment is designed to gather evidence on analysts' perceptions of the specific benefits derived from maintaining good relationships with management. Thirty-four experienced sell-side analysts from a medium-sized investment banking/brokerage firm participate. We manipulate the timing of their forecasts in the same manner as in experiment 1, but all participants receive the management relationship incentive. They perform the same task as in experiment 1 and also answer open-ended questions about the benefits of maintaining good relationships with management. The OP pattern in their forecasts and their answers concerning the expected bias in management guidance and the effect of issuing forecasts above management guidance on relationships with management are also highly similar to experiment 1. Answers to the open-ended question concerning the benefits of maintaining good relationships with management focus on increased access to ask questions during conference calls and enhanced access to informal conversations with management. These results are consistent with other findings related to management discrimination among analysts (Chen and Matsumoto [2006], Mayew[forthcoming]), and provide direct support for Ke and Yu's [2006] conclusion that managers provide better information access to favored analysts.

This paper contributes to the literature on biases in analysts' forecasts, as our experiments provide the first direct causal evidence of the effects of analysts' incentives to please management on the OP tendency, and provide insight into analysts' beliefs that underlie these effects. The results suggest that analysts' incentives to maintain relationships with company management contribute to the within-quarter OP pattern in analysts' forecasts. We control for primary alternative explanations for prior findings by randomly assigning analysts to time and incentive conditions in the first experiment and to time conditions in the second experiment, ensuring that all analysts in each experiment work for the same firm, and measuring experience and historical accuracy. The consistency of our results in the management relationship conditions in both experiments with parts of prior archival results (e.g., Hutton [2005], Baik and Jiang [2006], Cotter, Tuna, and Wysocki [2006], Ke and Yu [2006]) lends credibility to the study's findings. Our ability to eliminate natural confounds in the archival data and test more detailed process explanations also illustrates the benefits of using multiple methods with offsetting strengths and weaknesses (Libby, Bloomfield, and Nelson [2002]).

Gaining a better understanding of the causes of bias in analysts' forecasts may help regulators determine more effective remedies should such bias be judged to be detrimental to investors' interests. The results also suggest that Regulations FD and AC and the Global Settlement may reduce, but do not eliminate, the benefits of building and maintaining good relationships with management (see Cotter, Tuna, and Wysocki [2006], Mohanram and Sunder [2006]). They also indicate that an OP pattern in forecasts is consistent with analysts' incentive to maintain good relationships with brokerage customers, which suggests an avenue for further research.

The rest of the paper proceeds as follows. Section 2 discusses prior research on biases in analysts' forecasts and presents our hypotheses. Sections 3 and 4 describe the method and results of our two experiments. Section 5 provides a brief summary and discussion of implications, limitations, and directions for future research.

2. Background and Hypotheses

2.1 bias in analysts' forecasts

The assumed form of bias in analysts' forecasts has changed over time. The earliest studies show forecasts to be unbiased on average (see Givoly and Lakonishok [1984]). A variety of studies in the late 1980s and 1990s indicate that analysts' forecasts were optimistic on average (e.g., Brown et al. [1987], O'Brien [1988], Affleck-Graves, Davis, and Mendenhall [1990], Easterwood and Nutt [1999]). The most recent studies examining forecasts over time show that average forecast optimism decreased throughout the early 1990s and eventually gave way to average forecast pessimism (Brown [1997, 2001a]). Brown [2001a] links this change with the increasing importance of analyst forecasts as an earnings benchmark.

A separate stream of literature shows a trend from forecast optimism to pessimism within both quarterly and annual fiscal periods. For example, in the Cowen, Groysberg, and Healy [2006] sample of forecasts issued from January 1996 to December 2002, 180-day+ forecasts are positively biased, 91- to 180-day forecasts are unbiased, and 0- to 90-day forecasts are negatively biased. Similarly, Ke and Yu [2006] find that annual forecasts are on average optimistic and quarterly forecasts are pessimistic, and Richardson, Teoh, and Wysocki [2004] document a “walk-down” of both annual and quarterly forecasts that is particularly pronounced in more recent periods.

A number of studies suggest that bias in short-term management guidance contributes to the walk-down pattern. Cotter, Tuna, and Wysocki [2006] find that, compared to a control sample of nonguiding firms, analysts' forecasts for guiding firms are more optimistic before guidance is issued. Nevertheless, analysts' forecasts for guiding firms are significantly less optimistic than the control sample after the guidance is issued. The consensus analysts' forecast is 1.7 times as likely as the control sample to be pessimistic after the guidance is issued. Similarly, in Baik and Jiang's [2006] sample of firms issuing guidance, forecasts are pessimistic before the guidance 42% of the time compared to 61% of the time after the issuance of the guidance. Bartov, Givoly, and Hayn's [2002] 1983–1997 sample shows similar effects, which increase in magnitude during more recent years. Richardson, Teoh, and Wysocki [2004] find that end-of-period forecast pessimism is strongest when the importance to management of short-term share price should be strongest: when firms issue shares or managers sell stock shortly after the earnings announcement. All of these findings suggest that managers prefer optimism in beginning-of-period and pessimism in end-of-period analysts' forecasts, but leave open the question of why analysts appear to cooperate with management and issue forecasts that are consistent with their preferences.

2.2 effects of analysts' incentives

Many studies suggest that analysts issue biased earnings forecasts to please firm management. Two benefits from good relationships with managers have been proposed: prospects for investment banking business and access to management (Schipper [1991]). The effects of investment banking incentives have received the great majority of research attention. Most of the studies focus on the effect of an investment banking relationship with a company (e.g., Hunton and McEwen [1997], Lin and McNichols [1998], Michaely and Womack [1999], O'Brien, McNichols, and Lin [2005]). These studies find that underwriting analysts issue more optimistic forecasts and recommendations than unaffiliated analysts.

The role of access to management has been the subject of fewer studies. Ke and Yu [2006] suggest that, at least before the issuance of Regulation FD, firm managers provided favored analysts with preferential access to information. They reason further that preferential access to information will allow the favored analysts to make more accurate forecasts and, as a result, be more likely to retain their jobs. Since both analysts' incentives and preferential access to information are unobservable in the archival data, Ke and Yu [2006] support their assertions by demonstrating that analysts who issue forecasts following an OP pattern are more accurate and less likely to be fired. Consistent with Richardson, Teoh, and Wysocki [2004], being OP contributes more to analyst accuracy when the firm is involved in insider selling after the actual earnings announcement. However, they find employment by a firm involved in investment banking does not change the discovered associations. Consistent with this finding, Baik and Jiang [2006] do not find any significant differences between analysts with and without an investment banking relationship in forecast revisions in response to guidance. In addition, Ke and Yu [2006] find that OP analysts are more experienced, on average, and more likely to be affiliated with a large brokerage house and have an All-Star rating.

Ke and Yu [2006] note that, despite their strong results, the associations they document should be interpreted with caution as they may result from other unknown explanations. Ke and Yu [2006] use the number of years an analyst has covered a firm, number of firms an analyst covers, gap between the forecast and the earnings announcement, and brokerage firm fixed effects to control for alternative explanations for their accuracy results. In additional analyses, they also control for last period's accuracy. However, prior studies (e.g., Mikhail, Walther, and Willis [1997], Clement [1999], Jacob, Lys, and Neale [1999], Brown [2001b]) find that a variety of analyst and brokerage characteristics, in addition to past accuracy, are predictive of current period accuracy, rating as an All-Star by Institutional Investor magazine, and influence with investors. These results suggest potential confounds in any archival test of the effects of OP forecasts on subsequent analyst accuracy.

These natural confounds suggest a number of alternative explanations for the association between OP forecasts and accuracy. First, the association may be driven by one or more of the analyst attributes mentioned above. For example, prior studies suggest that public management guidance is informative but downwardly biased, and is issued most often when the consensus forecast is optimistic (e.g., Baik and Jiang [2006]). Furthermore, analysts with some of the above mentioned attributes (e.g., those with superior resources) have been shown to be more likely to revise their forecasts in response to new information and to be more accurate. As a consequence, these responsive analysts' forecasts could be more prone to shift from optimism to pessimism and be more accurate, regardless of whether they are attempting to please management or receive preferential access to information. Second, there is a possibility that the causal direction is misspecified; that is, instead of analysts currying favor with management, managers may be attempting to curry favor with more influential analysts by providing them with more private information. If these private disclosures are informative and downwardly biased, as suggested by Hutton [2005], and analysts do not sufficiently adjust for the downward bias,1 these more influential analysts' forecasts will likely be more accurate and more OP even if they are not attempting to please management. Third, there is a possibility that the OP forecasts are being issued in response to trading incentives. Cowen, Groysberg, and Healy [2006] show that brokerages that perform no underwriting actually produce more optimistic early forecasts than those engaging in both trading and underwriting activities. They conclude that trade-boosting incentives and reputation concerns dominate management relationship incentives in determining analysts' forecast bias (see also Eames, Glover, and Kennedy [2002, 2006]). Slight pessimism in end-of-period forecasts could also provide a basis for analysts to reiterate buy recommendations, further boosting trade. If success at trade boosting leads to superior resources, a relationship between OP forecasts and accuracy could be in evidence.2 The key point is that none of these alternatives require that analysts be reacting to a relationship incentive.

Three regulatory changes, Regulations FD and AC3 and the Global Settlement, made after the time period covered by the Ke and Yu [2006] sample, are designed in part to alter analysts' incentives to please management. Some prior findings suggest that Regulation FD has partially leveled the information playing field among analysts. Mohanram and Sunder [2006] find that analysts from large brokerage houses who are more accurate pre-FD are unable to maintain their superiority post-FD. This suggests a lessening of the benefit derived from pleasing management, and the possibility that the effect of relationship incentives reported in Ke and Yu [2006] will not persist in the current regulatory environment. However, Cotter, Tuna, and Wysocki [2006] find an increasing trend toward “beatable” analysts' forecasts following management guidance even after the effective date of Regulation FD. This suggests increasing pressures to please management. Mayew[forthcoming] finds that after Regulation FD, analysts with better relationships with management are more likely to be allowed to participate in conference calls. This suggests that benefits from good relationships with management that might contribute to the OP phenomenon still exist in the new regulatory regime.

2.3 hypotheses

In our first experiment, we hold constant the timing, sign, and amounts included in management guidance as well as other information available to the analysts, and independently manipulate forecast date (early or late in the quarter) and the analysts' incentive (accuracy or relationship). This allows us to provide a direct test of Ke and Yu's [2006] suggestion that relationship incentives contribute to the OP trend in forecasts and eliminate omitted variables concerns. We test the following hypotheses:

  • H1:  Analysts' forecasts made late in the quarter will be more pessimistic than those made early in the quarter.
  • H2:  Forecasts by analysts with an incentive to maintain an existing relationship with management will exhibit a stronger OP time trend than forecasts by analysts with only an accuracy incentive.

Ke and Yu [2006] hypothesize and find that analysts who issue OP forecasts are more accurate. They reason that this effect occurs because analysts issue OP forecasts to please management and therefore have better access to private management information. In this way, their forecast accuracy is improved. This argument suggests that analysts who have track records of issuing forecasts of higher (vs. lower) accuracy may have better learned the benefits of maintaining existing relationships with management and issuing OP forecasts. This implies that the effects in H2 are likely to be accentuated for analysts with high historical forecast accuracy compared to those with low historical forecast accuracy. We test the following hypothesis in experiment 1 using the participating brokerage firm's data on analysts' historical accuracy:

  • H3:  The greater OP time trend effect for analysts with relationship incentives compared to accuracy incentives will be magnified for analysts with higher (vs. lower) historical forecast accuracy.

We also examine whether the OP bias is intentional, and gather qualitative information to assess the specific benefits (information access vs. investment banking business access) of maintaining good relationships with management. We conduct all of our tests after the implementation of Regulations FD and AC and the Global Settlement. As a consequence, we are testing for the proposed effects in the current regulatory environment. In experiment 1, we gather detailed internal information on analyst experience and historical accuracy to perform additional tests of the association between analyst attributes and the strength of the OP time trend in forecasts. In addition, using the results of our first and second experiments, we provide preliminary evidence on the effects of brokerage size on the OP time trend.

3. Experiment 1

3.1 method

3.1.1. Participants Our participants are 47 experienced sell-side financial analysts employed by a major worldwide investment banking, trading, and brokerage firm. Fortune ranks the firm among the top 10 investment firms based on total revenues.4 At the time of the experiment, the analysts are an average of 37 years of age, have been an analyst for 11 years, and have worked for the participating brokerage for 5 years. Forty-four of the participants are chartered financial analysts. None of these demographic variables is significant as a covariate in any of the analyses. The participating firm also provides an accuracy ranking for the 47 participants. The metric used is a multiperiod measure of the absolute deviation of forecast of earnings from actual earnings achieved, divided by the actual earnings achieved.5

3.1.2. Procedure The analysts complete the task during a firm-sponsored training course. After returning from the mid-morning coffee break, they are asked to volunteer to participate in a study about factors that influence analysts' responses to earnings announcements. As an incentive to participate, the researchers provide each analyst with a $50 contribution to the charity of his/her choice. The firm trainer, who is qualified by one of the researchers in how to administer the experimental materials, is unaware of the experimental treatments.

The trainer hands two sealed envelopes to each participant6 and provides them with an Excel spreadsheet in which to record their answers to the study questions.7 Participants are asked to open the first envelope and remove the materials, which include a cover sheet, voluntary consent form, case description, and dependent variable response items. After the participants read the cover sheet and sign the consent form, they read the case materials, respond to the dependent variable items, and place all materials into and seal the first envelope. The analysts next open the second envelope, remove the materials within, respond to a series of debriefing questions, provide some demographic information, and place all open materials back into and seal the second envelope. Importantly, after sealing the first envelope, the trainer ensures that the participants do not reopen the first envelope while the second envelope is open.8

3.1.3. Task and Design To minimize the demands on participant time, our task includes the minimum information necessary to test the phenomenon of interest. Participants read background material about a company called Gamma, Inc., a manufacturer of semiconductor materials. They are provided with an abbreviated earnings history (sales, gross profit, net income, and earnings per share) for each quarter of 2005 and the first quarter of 2006.

They are also given the consensus analyst earnings per share forecast for the second quarter of 2006 ($0.34) and full year of 2006 ($1.44), after the first-quarter 2006 actual results, but before the management guidance has been announced. Next, they read the following earnings guidance statement from Gamma management issued shortly after the first-quarter earnings announcement:

The company expects earnings per share for the second quarter ending June 30, 2006 to be above expectations due to stronger than expected sales. Earnings per share are estimated to be between $0.40 and $0.45 for the quarter.

We employ a good news setting because it avoids two potential alternative explanations for an OP pattern in forecasts. In a bad news setting, an OP time trend could be explained by a desire to adjust a forecast slowly as opposed to all at once. Bad news may also weaken the effect of the management relationship incentive because prior research indicates that many analysts drop coverage of firms with lower earnings growth. We use range guidance because of its prevalence and the fact that it provides the greatest latitude for personal preferences to have an effect on analyst forecasts.

We employ a 2 × 2 between-subjects design. The first manipulation focuses on analysts' incentives. We compare the effects of the management incentive condition to those of a control condition where conflicts of interest are minimized. In our control or “accuracy” condition,9 the analysts read the following statement:

Assume that your only concern is the accuracy of your forecast.

Since one of our research goals is to determine our participants' beliefs about the nature of the benefits received from maintaining relationships with management, we use a general relationship condition to compare to the control.10 In the relationship condition, participants read the following statement:

Assume that over time, you have gradually developed a good professional relationship with Gamma management.

Following the incentive manipulation, participants are asked to indicate how they benefit from being accurate or maintaining a good professional relationship with management. The purpose of this question is to focus the participants' attention on the benefits of their particular incentive condition.11

Forecast timing is manipulated at two levels: early or late in the second quarter of 2006. In the early timing condition participants read the following statement:

Assume that right now, it is April 15, very early in the second quarter. As a consequence, after making the forecast below, you will have the opportunity to revise your forecast again before the issuance of the actual earnings release.

In the late timing condition, they read the following statement:

Assume that right now, it is June 23, very late in the second quarter. As a consequence, after making the forecast below, you will not have the opportunity to revise your forecast again before the issuance of the actual earnings release.

Forecast revision potential is included in the manipulation to make the timing effect more salient, and because the stated association between date and revision potential is ecologically valid. The participants are then asked to indicate the point in time in the second quarter that they make their forecast (1 = beginning of second quarter, 9 = end of second quarter – time of actual earnings release). The purpose of this response item is to focus their attention on the time at which they are making their forecasts.12 Afterward, the analysts provide earnings per share forecasts for the quarter ending June 30, 2006, full year ending December 31, 2006, and full year ending December 31, 2007. Finally, the analysts record their level of confidence in the accuracy of their forecasts (1 = not at all confident, 9 = extremely confident). Appendix A includes the information presented to participants in the management relationship incentive/early timing condition.

3.2 results

3.2.1. Manipulation Checks In the postexperiment debriefing questionnaire, we ask participants to indicate their assumptions concerning the incentive and timing manipulations and the timing of the second-quarter guidance. All analysts correctly recall their incentive (accuracy, relationship) and timing (early, late) conditions, and the timing of the issuance of the second quarter guidance.

3.2.2. Hypothesis Tests To test our hypotheses, we use current-quarter (Q2) earnings per share (EPS) forecasts as the primary dependent variable. We also test the effects of our treatments on the analysts' current-year annual forecasts (fiscal year 2006, FY06), and year-ahead (fiscal year 2007, FY07) forecasts.13 The tests are conducted using a 2 × 2 fixed-factor analysis of variance (ANOVA) with incentive and timing conditions as the independent variables.

H1 predicts that forecasts made late in the quarter will be more pessimistic than those made early in the quarter. Cell means and standard deviations for each dependent variable are presented in table 1. When forecasts are made early in the quarter, participants provide a mean Q2 EPS forecast of $0.434, which is above the mean of the range guidance. The mean forecast made late in the quarter is $0.410, which is below the mean of the range guidance. The main effect of timing is highly significant, as shown in table 2, panel A (F= 55.72, p < 0.0001).14 These findings match the OP pattern and are consistent with empirical findings in prior papers. The effect also holds for FY06 (F= 10.33, p= 0.0025) and FY07 (F= 18.06, p= 0.0001) forecasts, as shown in tables 1 and 2, panels B and C. These results support H1 and provide evidence that proximity to the end of the period exerts downward pressure on analysts' earnings forecasts.

Table 1. 
Experiment 1: Mean (SD) EPS Forecasts by Condition
IncentiveTiming
EarlyLate
  1. This table shows (in dollars) the cell means from experiment 1, in which 47 analysts from a large banking/brokerage firm make EPS forecasts for the second quarter of 2006 (panel A), fiscal year 2006 (panel B), and fiscal year 2007 (panel C). The experiment varies, between analysts, whether forecasts are made early or late in the second quarter (timing), and whether analysts have developed professional relationships with management or are only concerned with the accuracy of their forecasts (incentive).

Panel A: Q2 forecasts
Accuracy0.4270.416
(0.014)(0.012)
Relationship0.4420.403
(0.010)(0.008)
Panel B: FY06 forecasts
Accuracy1.9331.914
(0.140)(0.119)
Relationship1.9711.763
(0.125)(0.098)
Panel C: FY07 forecasts
Accuracy2.1752.100
(0.157)(0.084)
Relationship2.2041.958
(0.154)(0.158)
Table 2. 
Experiment 1: ANOVA Analysis
SourceSSd.f.MS F-statistic p-value
  1. This table presents ANOVA results from experiment 1, in which 47 analysts from a large banking/brokerage firm make EPS forecasts for the second quarter of 2006 (panel A), fiscal year 2006 (panel B), and fiscal year 2007 (panel C). The experiment varies, between analysts, whether forecasts are made early or late in the second quarter (timing), and whether analysts have developed professional relationships with management or are only concerned with the accuracy of their forecasts (incentive). SS and MS refer to the sum of squares and mean square from each source, respectively.

Panel A: Q2 EPS forecasts
Timing (T)0.006910.006955.72<0.0001
Incentive (I)0.000010.0000 0.090.7639
T*I0.002310.002318.51<0.0001
Error0.0054430.0001 
Panel B: FY06 EPS forecasts
Timing (T)0.152510.152510.330.0025
Incentive (I)0.037910.0379 2.560.1166
T*I0.104410.1044 7.070.0110
Error0.6350430.0148 
Panel C: FY07 EPS forecasts
Timing (T)0.301910.301918.060.0001
Incentive (I)0.037110.0371 2.220.1435
T*I0.085610.0856 5.120.0288
Error0.7190430.0167 

H2 predicts that the main effect of timing will be qualified by a significant interaction of timing and incentives, such that forecasts in the relationship condition will exhibit a stronger OP time trend than forecasts in the accuracy condition. We find a significant interaction for Q2 forecasts (F= 18.51, p < 0.0001) in the form suggested by the hypothesis, indicating that participants respond differently to the timing manipulation depending on their incentive condition. The form of the interaction is presented in figure 1. Untabulated simple effects tests demonstrate that participants in the relationship condition provide higher forecasts of Q2 earnings than participants in the accuracy condition when forecasting early in the quarter (F= 10.84, p= 0.0020), but provide lower forecasts than participants in the accuracy condition when forecasting late in the quarter (F= 7.83, p= 0.0077). Participants' forecasts in both the relationship and accuracy conditions exhibit a significant OP trend (F= 70.80, p < 0.0001 and F= 4.89, p= 0.0323, respectively).15 These results support H2 and show that participants' forecasts exhibit a stronger OP time trend in response to management relationship incentives. The interaction is also significant for FY06 (F= 7.07, p= 0.0110) and FY07 (F= 5.12, p= 0.0288) forecasts. Participants in the relationship condition exhibit an OP time trend for FY06 (F= 17.63, p= 0.0001) and FY07 (F= 21.69, p < 0.0001), whereas participants in the accuracy condition do not (both p > 0.1700).16

Figure 1 .—.

Experiment 1: time of forecast × incentive interaction plots for Q2 EPS forecasts.

To assess whether analysts' confidence in the accuracy of their forecasts is lower for early forecasts (because there is more uncertainty in a longer horizon) and for relationship incentives (because they are aware of the bias in the forecasts), we ask participants to provide confidence ratings concerning the accuracy of their forecasts immediately after forecasting EPS. Results indicate that, as expected, confidence is lower in the early than the late timing conditions (F= 16.76, p= 0.0002). Confidence is also lower in the relationship incentive than the accuracy incentive conditions (F= 4.25, p= 0.0453). The second effect suggests that the analysts are aware of the effects of the relationship incentive on the bias in their forecasts, and that the effect is intentional. This issue is addressed in detail in the next section of the paper. There is no significant interaction of timing and incentives.

H3 predicts that the OP effect in H2 will be greater for historically more accurate than less accurate analysts. To test this hypothesis, we run two untabulated ANOVAs including either years of experience or within-firm accuracy ranking as an independent variable (along with interactions). H3 is not supported, as there are no significant (p < 0.05) main or interaction effects of accuracy ranking (or years of experience). A change from significance to marginal significance (p= 0.0554) in the timing by incentive interaction for FY07 is the only qualitative difference between these and our original analyses. The direction and significance of the previously reported simple effects also remain unchanged.

These findings, and the nonsignificance of the demographic variables in the untabulated analysis of covariance analyses support the causal direction suggested, but not directly tested, by Ke and Yu [2006]; that is, relationships with management contribute to the OP pattern in analysts' within period forecasts. However, they are not consistent with the argument that historically accurate analysts are more aware of the need to preserve relationships with management by issuing OP forecasts. Our debriefing results provide additional evidence on this issue.

To determine whether this behavior is intentional, we ask participants the following postexperiment debriefing question:

Assume that on January 15, 2007, management of Company A issued guidance indicating the first-quarter EPS is likely to be between $0.22 and $0.26, and that there has been no further information from management. Assume further that you made a forecast after the guidance was issued and again near the end of the quarter. What would your forecasts most likely be? (Select one letter)

  • a.  Lower at the beginning of the period than the end of the period.
  • b. The same.
  • c.  Lower at the end of the period than the beginning of the period.

The mean (median) response to this measure is 2.74 (3.00) on a scale of 1–3 (where a = 1, b = 2, and c = 3). Thirty-seven of 47 participants indicate that their forecasts will likely be lower at the end of the period, and only two indicate the reverse pattern. A signed-ranks test shows that the median is significantly greater than the midpoint (Wilcoxon statistic = 740.00, p < 0.0001), providing evidence that analysts are aware that they issue lower EPS forecasts later in the accounting period.17 Based on a median split, the historically more accurate analysts provide higher responses to this question, indicating that they believe that they are more likely to issue forecasts following an OP trend (t= 1.74, p= 0.0460, one-tailed).18 While our tests reported above do not support H3, this difference provides support for H3 by suggesting that the more accurate analysts are indeed more aware of the need to preserve relationships with management by issuing OP forecasts.

3.2.3. Debriefing In the postexperiment debriefing questionnaire, we also ask participants two questions about the effect of issuing pessimistic end-of-period forecasts. First, we ask:

If you issue a forecast above management's current guidance, and the company fails to meet or beat your forecast, how do you think this will affect your relationship with the company management if you do this on a regular basis?

On a scale of 1 (harm relationship) to 9 (improve relationship), the mean response is 2.23, which is significantly lower than the midpoint of 5 (t=−22.59, p < 0.0001). This result suggests that the analysts believe that relationships with management may be improved by intentional forecast bias. Second we ask:

If you have issued a buy recommendation on a stock, and it beats your most current forecast of quarterly earnings by a small amount, what effect will this have on your relationship with clients that have purchased the stock?

On a scale from 1 (harm relationship) to 9 (improve relationship), the mean response is 8.83, which is significantly greater than the midpoint of 5 (t= 69.12, p < 0.0001). This result suggests that relationships with clients may also be improved by intentional forecast bias.

Experiment 1 demonstrates the effects of management relationship incentives on the time-trend in analysts' forecasts. Experiment 2 provides more detailed information about the nature of the benefits that accrue to the analysts from maintaining a good relationship with management.

4. Experiment 2

4.1 method

4.1.1. Participants Our participants are 34 experienced sell-side financial analysts employed by a regional investment banking, trading, and brokerage firm. Fortune ranks the firm in the second 10 among investment firms based on total revenues.19 At the time of the experiment, the analysts are an average of 34 years of age, have been an analyst for seven years, and have worked for the participating firm for four years. Twenty-eight of the participants are chartered financial analysts. None of these demographic variables is statistically significant as a covariate in any of the analyses.

4.1.2. Procedure The analysts participate via a computer-based experiment. All experimental materials are placed on an intranet Web server that is owned and operated by the participating firm. The program is written using a combination of Microsoft Access, C++ language, and Java scripts. The analysts can access the experiment from their work computers at any time of day, beginning at 12:00 a.m. on a Monday and ending at 11:59 p.m. on the following Friday.

The week prior to starting the experiment, firm management sends two e-mail messages (one on Wednesday and one on Friday) to 51 analysts working at the headquarters location encouraging them to participate in the upcoming computer-based study. To further encourage participation, the researchers offer to contribute $50 in the name of each participating analyst to the charity of his/her choice. Of the 51 analysts, 34 participate in the experiment, for a response rate of 67%.

Many controls are built into the experiment. For instance, in an attempt to secure the experiment and ensure that only the solicited analysts are participating, once an analyst logs onto the experiment, the software checks the computer number from which the log on request is made. If the computer number does not match one of the preauthorized computer numbers, the experimental software prevents further participation. Also, once an analyst completes the study from his/her computer, no one else can log on to the server from the same computer. Some of the other controls include: prevention of return to prior screens or changing earlier responses, random assignment of participants to treatment conditions, alphanumeric and range checks in all data entry fields, and automatic recording of responses into a database.

4.1.3. Task and Design The task is the same as described for experiment 1, above. Because of participant sample limitations, all participants are assigned to the relationship condition described earlier—assume that over time you have gradually developed a good professional relationship with Gamma management. The early and late forecast conditions are administered as a between-participants manipulation. The participants read the same background information related to Gamma, Inc., management earnings guidance, professional relationship assumption, and the early or late second-quarter timing manipulation used in the first experiment. As in the first experiment, participants then indicate the point in time in the second quarter that they are making their forecast (1 = beginning of second quarter, 9 = end of second quarter – time of actual earnings release).20 Afterward, the analysts provide earnings forecasts for the quarter, full year ending December 31, 2006, and full year ending December 31, 2007. They next indicate their level of confidence in the accuracy of their forecasts (1 = not at all confident, 9 = extremely confident). The analysts then respond to debriefing and demographic items.

4.2 results

4.2.1. Manipulation Checks In the postexperiment debriefing questionnaire, we ask participants to recall the timing manipulation, as well as the timing of the second-quarter guidance. All analysts correctly recall their timing (early, late) conditions and the timing of second-quarter guidance. We also request participants to indicate whether the case materials indicate that they have accuracy or relationship incentives. All correctly answer this question.

4.2.2. Hypothesis Tests To test H1, we use current-quarter (Q2) EPS forecasts as the primary dependent variable. We also test the effects of our treatments on the analysts' current-year (FY06) and year-ahead (FY07) forecasts. The tests are conducted using a 1 × 2 fixed-factor ANOVA with timing condition as the independent variable.

Cell means and standard deviations for each dependent variable are presented in table 3. When forecasts are made early in the quarter, participants provide a mean Q2 EPS forecast of $0.424. The mean forecast made late in the quarter is $0.409. The effect of timing is significant as shown in table 3, panel B (t= 2.09, p= 0.0444). These findings match the OP pattern and are consistent with experiment 1 and empirical findings in prior papers. The effect is marginally significant for FY06 (t= 1.60, p= 0.0585) and becomes insignificant for FY07 (t= 0.80, p= 0.4279). Immediately after forecasting EPS, participants provide confidence ratings concerning the accuracy of their forecasts. Results indicate that confidence is lower in the early than in the late timing conditions (t= 2.14, p= 0.0400). These results support H1 and provide evidence that analysts at a smaller brokerage also appear to exhibit an OP time trend in forecasts.21

Table 3. 
Experiment 2: Mean (SD) EPS Forecasts by Condition and t-statistics
Panel A: Means and standard deviations
ResponseTiming
EarlyLate
Q2 forecasts0.4240.409
(0.019)(0.020)
FY06 forecasts1.5991.537
(0.101)(0.086)
FY07 forecasts2.2402.169
(0.262)(0.257)
Panel B: t-tests
  t-statistic p-value
  1. This table shows (in dollars) the cell means (panel A), as well as t-test results (panel B), from experiment 2, in which 34 analysts from a medium-sized banking/brokerage firm make EPS forecasts for the second quarter of 2006, fiscal year 2006, and fiscal year 2007. The experiment varies, between analysts and whether forecasts are made early or late in the second quarter (timing). All analysts are told that they have developed a professional relationship with management.

Q2 forecasts2.090.0444
FY06 forecasts1.600.0585
FY07 forecasts0.800.4279

4.2.3. Debriefing To determine whether the forecast trend is intentional for the smaller brokerage firm, participants respond to the same debriefing question used in experiment 1 (forecast lower at beginning of period, end of period, or no difference). The mean response on a scale from 1 (lower at beginning) to 3 (lower at end) to this measure is 2.53. Twenty-two of 34 participants indicate that their forecasts will likely be lower at the end of the period, and only four indicate the reverse pattern. The mean is significantly greater than the midpoint (t= 4.37, p < 0.0001), providing evidence that these analysts are aware that they issue lower EPS forecasts later in the accounting period. As in experiment 1, we also ask the participants two debriefing questions dealing with the impact of OP forecasts on relations with management and with investment clients. As in experiment 1, the participating analysts believe that both relationships improve when they issue forecasts that management meets or beats (means of 3.50 and 8.67 on the nine-point scales, respectively).22

The participants in experiment 2 are allowed to answer the following open-ended question: “Assume that over time, you have gradually developed a good professional relationship with Gamma management. Please indicate how you benefit from having good relationships with company managers.” Twenty of the participants provide specific answers to the question. Of these answers, 11 mention access to information, eight mention participation in conference calls, and two mention additional institutional business.23 Four indicate that a good relationship with management would be valuable in general, three leave the answer blank, two refuse to answer this sensitive question, and five provide uninformative answers. These findings are consistent with Mayew's[forthcoming] finding that supportive analysts have greater conference call participation, and suggest the possibility that material or non-material information may still be available to preferred analysts. The results provide additional support for Ke and Yu's [2006] claim that information access is a key benefit driving the OP pattern in more accurate analysts' forecasts.

5. Summary and Conclusion

In our two experiments, 81 experienced sell-side analysts predict earnings in response to historical data and management guidance. We manipulate the timing of the analysts' forecasts (early and late in the quarter) and the analysts' incentives (management relationship and accuracy). The resulting forecasts exhibit an OP pattern, and the pattern is stronger when the participants have developed a good relationship with management than when their only stated incentive is to be accurate. The analysts are aware of both the general downward bias in management guidance and the general OP temporal pattern in their forecasts. They also believe that this pattern benefits their relationships with both management and investment clients. When asked to indicate how they benefit from a good relationship with management, the answers focus on conference call participation and access to information. Our evidence on whether more accurate analysts within a specific investment bank are more aware of the need to issue OP forecasts is mixed.

Our study provides the first direct evidence of a causal link between relationship incentives and the OP time trend in analysts' forecasts. This contributes to the literature on both analysts' forecasts and management guidance. It confirms Ke and Yu's [2006] principal hypothesis that relationship incentives drive at least part of the OP trend in forecasts, and that differential access to information is an important benefit of good relationships with management. These results are consistent with legislative concerns about sources of analyst conflict of interest that remain even after the issuance of Regulations FD and AC (Cox [2005]). They may also suggest a reason why managers believe they can increase the chances of meeting or beating forecasts by issuing downwardly biased guidance. Successful walk-down of analysts' forecasts requires the cooperation of the analysts, and our results suggest that analysts may not adjust for downward bias in management guidance at the end of the period at least in part because of their incentives to maintain their relationships with management. The participants in our study are not aware of any history of biased guidance on the part of the reporting company. A test of this possibility requires a multiperiod study that examines analysts' responses to a guidance track record under differing incentive conditions.

The implications of our results for the specific causal direction of the other effects suggested by Ke and Yu [2006] are less clear. Our results suggest that OP analysts have greater access to management than those whose forecasts follow other patterns, and analysts recognize the importance of this benefit. While the results are consistent with differential access, it is not clear that differential access is the source of their greater accuracy. It is still possible that more successful analysts are accurate and OP for reasons other than their access to management. For example, beyond any information effects, access to management may build the analysts' reputation with brokerage clients and with managers in the industries covered by the analyst. This may have positive effects on the brokerage and banking businesses of their employers, which may in turn result in increases in resources available to the analysts and increased accuracy in their forecasts. Our large brokerage analysts more strongly believe that their forecasts will be OP—a finding that is consistent with both information and broader reputation explanations. In fact, both forces may be at work. While the sources of our observed brokerage differences are not clear, and may even be the result of administration differences, the results suggest that further research is warranted to determine differences in the incentive functions faced by analysts working for different types of organizations.

Given that the answers to conference call questions are now immediately made public, the exact nature of the benefits from conference call participation is not clear. Our analysts also suggest that the OP pattern can improve their relationships with brokerage customers. This suggests an additional relationship incentive that may have a variety of effects on analysts' work product—a topic that also could be examined in future studies.

The following limitations should be noted when interpreting our results. First, although we find that analysts with accuracy incentives issue forecasts with a smaller OP trend than those with relationship incentives, we cannot determine whether accuracy incentives reduce the OP effect or relationship incentives magnify this effect, or both. A control group would permit this inference to be made. Second, although analysts' qualitative responses indicate that information access rather than investment banking business access is largely the key benefit from good relationships with management, we suspect that few would indicate business access as a benefit even if it were so in the currently regulatory environment (although two respondents actually did so).

Appendix

APPENDIX A

Information Presented in Management Relationship Incentive/Early Timing Condition

BACKGROUND INFORMATION

Gamma, Inc. is a manufacturer of semi-conductor materials listed on the American Stock Exchange. It has a fairly diversified portfolio of customers, with two larger customers accounting for about 20% of its sales.

COMPANY EARNINGS HISTORY

(in thousands except per share data)First QuarterSecond QuarterThird QuarterFourth QuarterYear
FISCAL 2005
Sales …………………$26,171$28,960$30,433$31,316$116,881
Gross profit ……………8,89710,62911,82312,73344,082
Net income ……………1,5772,7332,8383,15410,302
Earnings per share ………0.150.260.270.300.98

FIRST QUARTER 2006 EARNINGS ANNOUNCEMENT

FISCAL 2006First Quarter
Sales …………………$29,955
Gross profit ……………12,823
Net income ……………3,364
Earnings per share ………0.32

Consensus forecasts after the announcement of the first quarter 2006 earnings:

Consensus Second Quarter Forecast for FY 2006$0.34
Consensus 12-month EPS Forecast for FY 2006$1.44

EARNINGS GUIDANCE

Shortly after the first quarter earnings announcement, the management of Gamma issued the following statement:

The company expects earnings per share for the second quarter ending June 30, 2006 to be above expectations due to stronger than expected sales. Earnings per share are estimated to be between $0.40 and $0.45 for the quarter.

Please record your answers on the spreadsheet provided.

ASSUMPTIONS

A1. Assume that over time, you have gradually developed a good professional relationship with Gamma management.

Please indicate how you benefit from having good relationships with company managers.

____________________________________________________________

____________________________________________________________

____________________________________________________________

A2. Assume that right now, it is April 15, very early in the second quarter. As a consequence, after making the forecast below, you will have the opportunity to revise your forecast again before the issuance of the actual earnings release.

Please indicate at what point in time in the second quarter you are making your forecast by selecting a number:

Begining of Second Quarter123456789End of Second Quarter—Time of Actual Earnings Release

REQUIRED

Given the company information on the prior page and assumptions presented above, what is your EPS forecast for the following?

R1. Quarter Ending June 30, 2006:    $___________

R2. Full Year Ending December 31, 2006: $___________

R3. Full Year Ending December 31 2007: $___________

R4. How confident are you in the accuracy of the above forecasts? (Select one)

Not at all confident Extremely Confident
0%10%20%30%40%50%60%70%80%90%100%

Footnotes

  • 1

    Learning from feedback in this type of environment has been found to be difficult in prior psychology and accounting studies.

  • 2

    Trading incentives alone do not explain the interaction of the OP/accuracy relationship with insider selling reported by Ke and Yu [2006].

  • 3

    Regulation AC, effective April 13, 2003, requires analysts to issue a statement certifying that the views expressed in the research report accurately reflect the research analyst's personal beliefs. This could increase the possible penalties for purposeful bias in an attempt to curry favor with management.

  • 4

    See http://money.cnn.com/magazines/fortune/fortune500/industries/Securities/1.html. Based on our confidentiality agreement with the participating firms, we cannot provide further identifying information.

  • 5

    Brokerage management did not disclose the time period over which the forecast errors were averaged for the participants.

  • 6

    The experimental materials are stacked in random order by the researchers and the trainer hands out the materials from the top to the bottom of the stack.

  • 7

    The Excel spreadsheet is designed such that the cursor automatically starts at the first data entry cell. When items require a numeric response, participants cannot move on to the next cell until they have entered a valid response into the cell at which the cursor is currently positioned. Once participants enter a response into a cell, a macro automatically moves the cursor to the next data entry cell and password-protects the prior cell. This way, participants cannot go back and change any answers. As well, before leaving a given cell, the participants have to respond to the following warning: “Are you sure? (Yes/No).” This gives them a chance to stay on the current cell until they are sure that they have correctly entered their intended response.

  • 8

    To verify whether the trainer followed the procedure, in the week following the experiment, we contacted two of the participants directly and asked them to describe the procedures followed by the experimenter. Both participants confirmed that the experimenter followed the prescribed procedures.

  • 9

    We also considered a “no relationship” condition as a control. But this leaves open the possibility that participants will still be quite anxious to please management in an attempt to develop a future relationship with the firm managers, and thus reduces the power of our tests. Our results do not address whether analysts with good relationships or those who wish to develop future relationships have a greater incentive to please management.

  • 10

    Because of the regulatory changes described in section 2, an investment banking relationship condition or a condition referring to disclosure of private information will not pass human subjects or participating firm review.

  • 11

    The first participating firm did not allow its analysts to record their answers to any open-ended questions.

  • 12

    They rate their point in time as significantly earlier in the early condition than in the late condition (mean = 1.71 and 8.17, t= 31.50, p < 0.0001). After participants enter their responses to this “attention” item, the spreadsheet automatically moves the cursor to the next data entry cell and fills the “attention” item cell with black color. This control is included so that when participants later respond to manipulation check items dealing with forecast timing (early or late), they cannot look back on the spreadsheet to recall how they responded to the “attention” item.

  • 13

    The need for consistency between the quarterly and annual forecasts should produce similar effects in the annual forecasts. However, we expect the effects of our treatments to be smaller because of the lack of proximity of the forecasts to the actual earnings announcement as well as the greater variation expected in longer-term forecasts.

  • 14

    All p-values are two-tailed except where indicated.

  • 15

    This significant OP trend in the control (accuracy only) condition for the Q2 forecasts could have arisen from a variety of effects, including a carryover from the real world incentives that the analysts face or the belief that accuracy can be enhanced by maintaining good relationships with management, among others. This effect occurs only for the Q2 forecasts. For the FY06 and FY07 forecasts, we do not detect any significant OP trend in the accuracy incentives condition.

  • 16

    Strictly speaking, the annual earnings forecasts (especially those for FY07) do not fall within the domain of OP forecasts to the extent that the OP effect relates to quarterly earnings forecasts within a financial year. We suspect that the FY06 second-quarter earnings results extend to the annual earnings forecasts because the FY06 annual earnings forecasts incorporate the FY06 second-quarter earnings, while the annual FY07 earnings forecasts result from applying a similar growth rate to the higher early or lower late FY06 earnings forecasts.

  • 17

    Parametric tests provide similar results (t= 9.63, p < 0.0001). Debriefing results do not vary by treatment.

  • 18

    We perform a chi-square test with debriefing response (answer c vs. others) and median-split accuracy as the column and row factors. The chi-square statistic is 1.87; p= 0.086, one-tailed.

  • 19

    Based on our confidentiality agreement with the participating firms, we cannot provide further identifying information.

  • 20

    They rate their point in time on this “attention” item as significantly earlier in the early condition than in the late condition (mean = 1.75 and 8.39, t= 27.96, p < 0.0001).

  • 21

    We also compare the magnitude of the OP time trend between the larger brokerage analysts in experiment 1 and the smaller brokerage analysts in experiment 2. Again, we use the Q2 forecasts as the primary variable of interest, but also examine the FY06 and FY07 forecasts. As shown in table 4, a 2 × 2 ANOVA with forecast timing and brokerage as between-participants factors produces a main effect of timing (F= 36.24, p < 0.0001) for Q2 forecasts. There is a significant timing by brokerage interaction (F= 7.55, p < 0.0081) indicating a stronger OP time trend on the part of the larger brokerage analysts. This interaction remains significant for FY06 (F= 7.26, p= 0.0094), but not for FY07 (F= 2.28, p= 0.1370). These results suggest that analysts from both brokerages produce more pessimistic forecasts at the end of the period. The differences in administration across the two experiments and nonrandom selection of analysts from the brokerages make further inferences difficult.

  • This table presents ANOVA results concerning the OP pattern in forecasts, compared between brokerages. Twenty-four analysts from the management relationship condition in experiment 1 are compared with the 34 analysts from experiment 2. Forecasts include the second quarter of 2006 (panel A), fiscal year 2006 (panel B), and fiscal year 2007 (panel C). The between-participant factors are forecast timing (early, late) and brokerage (large, medium-sized). SS and MS refer to the sum of squares and mean square from each source, respectively.

  • 22

    Consistent with their forecasts, the smaller brokerage analysts are marginally less likely than the large brokerage analysts to believe that their forecasts would follow an OP pattern (t= 1.50, p= 0.0700, one-tailed). The smaller brokerage analysts also expect less harm than the larger brokerage analysts to their relationship with management resulting from issuing a forecast that the company fails to meet or beat (t=−5.17, p < 0.001).

  • 23

    One of the participants mentioned both information and conference call participation.

Ancillary