Investors’ reactions to retractions and corrections of management earnings forecasts

Accounting, Organizations and Society 36 (2011) 382–397

Contents lists available at ScienceDirect

Accounting, Organizations and Society journal homepage: www.elsevier.com/locate/aos

Investors’ reactions to retractions and corrections of management earnings forecasts Seet-Koh Tan a,1, Lisa Koonce b,⇑ a b

Division of Accounting, Nanyang Business School, Nanyang Technological University, Singapore 639798, Singapore Department of Accounting, CBA 4M.202, McCombs School of Business, The University of Texas, Austin, TX 78712, United States

a r t i c l e

i n f o

a b s t r a c t Timely voluntary disclosure of information by companies sometimes results in erroneous disclosure that must be later retracted (i.e., withdrawn) and/or corrected (i.e., replaced with a corrected disclosure). Although such retractions and corrections appear to be relatively easy and costless ways to fix the erroneous disclosure, our results generally show that both actions have unexpected effects on investor judgment. The results of four experiments, which are consistent with affect theory from psychology, indicate when a company provides a retraction of a previous erroneous voluntary disclosure, investors’ judgments continue to reflect the implications of the initial erroneous information. That is, investors under-adjust. In contrast, when a company provides a correction (alone or with a prior retraction) with an opposite earnings implication, investors tend to over-adjust. Our results also show that if investors do not form a strong initial affective reaction to the initial erroneous forecast, they are less prone to over-adjustment when the correction is later received. Implications for regulators and standard setters are provided. Ó 2011 Elsevier Ltd. All rights reserved.

Introduction Timely voluntary disclosure of information to investors, creditors and others has been shown to be valued by the market (e.g., Ajinkya & Gift, 1984; Brown, Hillegeist, & Lo, 2004; Healy, Hutton, & Palepu, 1999). Regulators and standard setters also have pushed for more timely release of information to the market (FASB, 2000; SEC, 2001). One potential cost of such timely disclosures is that a reduction in their accuracy may occur (SEC, 2002). Indeed, prior research indicates that the number of companies retracting and/or correcting previous disclosures has risen in recent years (Tan & Tan, 2009). Retraction occurs when a company withdraws a previously issued disclosure, such as an earnings forecast. Correction occurs when a company

⇑ Corresponding author. Tel.: +1 512 471 5576. E-mail addresses: [email protected] (S.-K. Tan), [email protected] (L. Koonce). 1 Tel.: +65 6790 5656. 0361-3682/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.aos.2011.06.004

provides a replacement disclosure, with corrected information.2 These two events are distinct because a retraction can be present without a correction, a correction can be present without a prior retraction, or both can be present. Although such retractions and corrections seem to be relatively easy and costless fixes to erroneous disclosure, a concern exists that investors will nevertheless rely on the initial disclosure in ways that have unintended economic consequences. Prior research has explored the judgment effects associated with retraction of voluntarily disclosed information by companies. Specifically, Tan and Tan (2009) find that retractions of prior voluntary disclosures systematically affect investors’ judgments in ways that arguably are non-

2 Another way to view these two distinct constructs is that retraction is a subtraction of the original information while correction is a replacement of the original information. Further, correction differs from an updated disclosure which is based on new information arising from the passage of time or new events occurring. Rather, correction involves fixing an erroneous original disclosure.

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

normative (also see Tan & Tan, 2008, in auditing).3 Their study documented that erroneous information given to investors had an undue influence on their judgments, even after the information was retracted. That is, their postretraction judgments incorporated the implications of information contained in the erroneous press release. However, their study did not also examine how the correction of the erroneous disclosure affected investor judgment. That is, they did not provide investors with new information to replace the erroneous initial disclosure. Because most retractions eventually lead to correction, the important issue of how investors behave in light of voluntary disclosure correction remains unexplored. In this study, we experimentally investigate how correction of erroneous information influences investors’ judgments. Specifically, we conduct four experiments to investigate how investors react to quantitative earnings forecasts that are issued by company management and then later replaced with corrected forecast. We chose the earnings forecast domain because such forecasts are a valuable tool for a company to communicate its expectations to market participants. Such forecasts have the potential to inform investors, reduce information asymmetry and information risk, and decrease the firm’s cost of capital (Hirst, Koonce, & Venkataraman, 2008). In addition, most forecasts are quantitative in nature, thereby allowing us to investigate our ideas in a context where it is clear that a correction has occurred. We employ a similar research design in each of our four experiments. Specifically, we manipulate whether a company initially issues a high or low quantitative earnings (i.e., EPS) forecast that is later revealed to be erroneous. These initial forecasts are high or low relative to the prior year’s actual earnings. Depending on the experiment, the company’s actions include one or more of the following— retraction only, correction only, and both retraction and correction. Although prior research in accounting has previously studied retraction (Tan & Tan, 2008, Tan & Tan, 2009), its inclusion in our study allows us to draw stronger inferences about the theoretical mechanism at work. An important feature of our research design is that after either a retraction or correction (or both) is provided to study participants, the judgments in the high and low EPS forecast conditions arguably should converge.4 We draw on two distinct lines of research from psychology to develop competing predictions regarding how investors are likely to respond to retractions and corrections of earnings forecasts. Specifically, a cognitive-based theory, belief perseverance, suggests that when investors 3 Although Tan and Tan (2009) refer to their study as one involving ‘‘correction,’’ their study examines ‘‘retraction’’ as defined herein. That is, the participants in their study were told that the outcome in the original press release was to be disregarded. Importantly, they were not provided with an alternative outcome (i.e., correction). 4 Importantly, we do not make predictions about the appropriate levels of the post-retraction and post-correction judgments for our low and high EPS conditions. That is, our design does not allow us to make predictions about whether the low- and high-EPS judgments should be, for example, 6 and 6 or 4 and 4 (on a response scale of 1–11, as described in the design section of experiment one) respectively. Rather, we can only assert that these two judgments should be equal, on average.

383

receive either a retraction or a correction (or both) of a previously issued earnings forecast, their initial beliefs about the earnings potential and investment attractiveness of the company will persevere. That is, they will be unduly influenced by the erroneous information, thus underadjusting for the erroneous earnings forecast. In contrast, affect theory indicates that investors are likely to behave differently. Specifically, when given a retraction, investors will under-adjust as traces of the positive/negative affective response to the erroneous earnings forecast persist and have a lingering effect on judgments. However, when given a correction replacing the initially higher/lower forecast (alone or with a prior retraction), an opposite affective reaction is triggered by the implicit counterfactual comparison (i.e., what earnings could have been before the correction) which, in turn causes investors to over-adjust. Our experiment one results show that when the company action involves a retraction of a previously issued erroneous forecast (retraction only), investors under-adjust in their evaluations of a company. That is, even after the company retracts their initial forecast, those receiving a high earnings forecast still judge the earnings potential and investment attractiveness of the company to be higher than those who initially received a low earnings forecast. When the company action involves a correction of an erroneous forecast—either alone (correction only) or with a prior retraction (both retraction and correction)—our results show a different pattern. Here, we find that investors over-adjust in their evaluations of a company. Although the corrected forecast in our study is identical in both the high and low (initial erroneous) forecast conditions, we observe that those receiving a high earnings forecast judge the earnings potential and investment attractiveness of the company as significantly lower after correction than those who initially received a low earnings forecast. That is, participants in the high and low forecast conditions ‘‘flip’’ (or over-adjust in) their assessments of the company once the correction, with or without a prior retraction, is received. Overall, these results are consistent with affect theory. We conducted three other experiments to test the robustness of these findings. In experiment two, we document that our over-adjustment results are robust even with a more-sophisticated set of study participants. In experiment three, we provide additional evidence ruling out belief perseverance theory by showing that a cognitive task used to increase belief perseverance effects—namely, explanation—has no effect on under- and over-adjustment. Experiment four provides further evidence ruling in affect theory. Specifically, we show that when the investor makes an initial judgment based on the erroneous forecast, the affective response to a subsequent correction is heightened (as compared to when an initial judgment is not made). The initial judgment task causes the investor to form a strong affective response to the initial erroneous (favorable or unfavorable) forecast which, in turn, creates a larger opposite affective response when the subsequent correction is received. Our paper advances the literature in several ways. First, we focus on a company action taken in light of an erroneous voluntary disclosure—namely, correction—that has not

384

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

been previously studied. Although Tan and Tan (2009) examine how investors react to retracted voluntary disclosures, their study does not address the issue of correction (alone or with a prior retraction). Our results, combined with their study, provide a more-complete picture in this area of voluntary disclosures. That is, investors react differently to retracted information than they do to corrected information. Second, we extend insights in the earnings preannouncement domain. Specifically, Tan, Libby, and Hunton (2002) examine how patterns of earnings preannouncements and actual earnings realizations can influence analysts’ earnings forecasts for subsequent time periods. For example, they document that a low earnings preannouncement (relative to the consensus forecast) followed by a higher actual earnings realization leads to higher next-year earnings forecasts than a high earnings preannouncement and the same earnings realization. Although their context was one of strategic behavior around preannouncement and actual earnings surprises (and not retraction and correction as in our study), they observed behavior similar to our over-adjustment results. Our study extends theirs by providing an alternative theoretical explanation for these results and by documenting the circumstances where an opposite reaction, underadjustment, occurs. Finally, we extend psychology research by identifying those situations where the invalidation of information can lead not only to belief perseverance (under-adjustment) effects but also can lead to the opposite effect—that is, over-adjustment. This finding is novel as the existing studies have generally found that while belief perseverance effects may be reduced, they are rarely eliminated. Our study also has important implications for standard setters, regulators, and financial scholars. Both US and international standard setters have commented on the issue of correction. For example, the FASB (2000, p. 65) recognizes that in electronic distribution, there is a ‘‘duty to correct or revise a prior statement which was accurate when made but which has become misleading due to subsequent events.’’ Similarly, IASC (1999, p. 66) recommends as part of a standard for web-based business reporting that ‘‘if errors are found to exist in documents that were placed online, any changes should be clearly indicated on the original documents or linked to the original to prevent misrepresentation.’’ Our results suggest that correction of such erroneous information may not be sufficient, as we find that the initial erroneous information influences how investors respond to the subsequent correction. As a result, there are potentially significant costs that are borne by investors and/or firms when these retraction/correction disclosures occur. In the following section, we more fully develop the theories behind our competing predictions. The next three sections describe the experiments, and the final section concludes the paper.

Theory Regulators have recently called for more timely voluntary disclosures by companies. Moreover, Reg FD also has

Correction Absent

Correction Present

Retraction Absent

1

3

Retraction Present

2

4

Fig. 1. Possible relationships between retractions and corrections. This figure illustrates how retraction and correction are distinct events that may occur alone or in conjunction. See Fig. 2 for examples.

precipitated greater and more immediate disclosures. One type of voluntary disclosure, management earnings forecasts, is quite common. These forecasts provide important information about the expected earnings for a particular firm. They represent one of the key voluntary disclosure mechanisms by which managers establish or alter market earnings expectations, preempt litigation concerns, and influence their reputation for transparent and accurate reporting. Management earnings forecasts are influential as they have been shown to affect stock prices (Pownall, Wasley, & Waymire, 1993), analysts’ forecasts (Baginski & Hassell, 1990), and bid-ask spreads (Coller & Yohn, 1997). One concern about the increased quantity and timeliness of voluntary disclosures, including earnings forecasts, is that the quality of such disclosures may suffer (SEC, 2002). Indeed, Tan and Tan (2009) report that the number of press release retractions and corrections issued by US companies (as distributed by PR Newswire and Business Wire) has been increasing over recent time periods. They document that from 2002 to 2006, the number of press release retractions and corrections has more than doubled, from 949 to 2269. As noted earlier, retraction occurs when a company withdraws a previously issued disclosure, such as an earnings forecast. Correction occurs when a company provides a replacement disclosure, with corrected information.5 As illustrated in Fig. 1, these two events are distinct because a retraction can be present without a subsequent correction (see Cell 2), a correction can occur without a prior retraction (see Cell 3), and both can occur (Cell 4) or not occur (Cell 1). Examples of Cells 2, 3 and 4 are provided in Fig. 2. In light of these retractions and corrections, a natural question is how do investors react to them? To answer this, we look to two lines of research in psychology. We consider each in turn. Belief perseverance theory A long-standing stream of cognitive research in psychology, aptly named belief perseverance theory, shows that individuals are generally unable to ignore information specifically highlighted to be erroneous (i.e., retracted). For 5 Arguably retractions are likely to be less common than corrections, as most retractions are eventually followed by corrections but not all corrections are preceded by retractions. Retractions are nevertheless frequent enough that PR Newswire does track them, using the headline ‘‘K I L L K I L L K I L L.’’

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

385

Panel A: Example of retraction only (cell 2 of Fig. 1) Source: China Executive Education Corp. China Executive Education Corp. Retracts Previously Stated Revenue Guidance for Fiscal Year 2010 HANGZHOU, China, May 28, 2010 (GLOBE NEWSWIRE) -- China Executive Education Corp. (OTCBB:CECX) withdrew its revenue guidance for the full fiscal year ending December 31, 2010. Guidance for the full fiscal year 2010 was previously issued in the Company's earnings release for the first quarter of 2010 issued on May 19, 2010. The previously disclosed revenue guidance for fiscal year 2010 in the range of $35 million to $45 million was entered in erroneously and should not be regarded as financial guidance. The Company may issue revenue guidance as the 2010 fiscal year progresses when adequate information and visibility are available to reasonably base future revenue projections. Panel B: Example of correction only (cell 3 of Fig. 1) C O R R E C T I O N -- Bally Technologies, Inc. 13 February 2008 PR Newswire (U.S.) In the news release, Bally Technologies, Inc. (NYSE: BYI) Announces Record Revenue and Earnings for Second Quarter Fiscal 2008, issued earlier today by Bally Technologies, Inc. over PR Newswire, we are advised by the company that in the Fiscal 2008 Business Update of the text, the first sentence should read, "The Company raised its fiscal 2008 guidance for Diluted EPS to $1.60 to $1.90, from an earlier range of $1.55 to $1.85," rather than, "The Company raised its fiscal 2008 guidance for Diluted EPS to $1.62 to $1.87, from an earlier range of $1.60 to $1.90,” as originally issued inadvertently.

Panel C: Example of both retraction & correction (cell 4 of Fig. 1) Connetics Expects First Quarter Profit on Higher-Than-Expected Revenues … 15 April 2004 PR Newswire (U.S.) PALO ALTO, Calif., April 14 /PRNewswire-FirstCall/ -- Connetics Corporation ... The Company expects sales from its core brands, OLUX(R) and Luxiq(R), to be at the high end of the existing guidance of $19 million to $20 million … /K I L L K I L L K I L L -- Connetics Corporation/ 15 April 2004 We are advised by Connetics Corporation that … readers should disregard the news release, Connetics Expects First Quarter Profit on Higher-Than-Expected Revenues, issued earlier today over PR Newswire, as it contained some erroneous information. Connetics Corporation said a revised release will be issued later today. Revised: Connetics Expects First Quarter Profit on Higher-Than-Expected Revenues … 15 April 2004 PALO ALTO, Calif., April 14 /PRNewswire-FirstCall/ -- Connetics Corporation …The Company expects product sales excluding its new product Soriatane(R) to be within the existing guidance of $19.5 million to $20.5 million … Fig. 2. Examples of retractions and corrections.

example, Anderson, Lepper, and Ross (1980) found that falsely induced beliefs about risk preference and future success as a fire fighter prevailed even after explicitly informing the participants that the relationship was false. In a similar vein, Lepper, Ross, and Lau (1986) led participants to erroneously believe that they had strong or weak

abilities to solve a problem by providing them with either highly effective or completely useless task instructions. They found that these participants’ original beliefs were immune to subsequent invalidation. Belief perseverance effects have been documented in accounting as well. Tan and Tan (2008) find belief perseverance effects when qual-

386

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

itative audit evidence is invalidated. These same authors report perseverance effects in investor judgments when qualitative information in voluntary disclosures is retracted (Tan & Tan, 2009). In terms of the mechanism responsible for the continued influence of subsequently invalidated information, Ross, Lepper, and Hubbard (1975) argue that upon exposure to the original uncorrected evidence, decision-makers form judgments and causal explanations for those judgments. These judgments and causal explanations are stored separately from the original evidence in memory (Johnson & Seifert, 1994; Ross & Anderson, 1982). Thus, updates to the evidence do not necessarily lead to revisions in these judgments and their causal accounts. Because the original judgments and causal schemas are stored separately in memory from the original evidence, they continue to influence subsequent judgments despite the invalidation of the evidence. This research also explores various methods of correction, and generally finds that while the belief perseverance effects may be reduced, they rarely are eliminated. For example, Lord, Lepper, and Preston (1984) show that asking individuals to be unbiased does not work. That is, despite good intentions, motivational charges to be fair and unbiased do not undo the effects of belief perseverance. However, attempts to correct the initial erroneous assertion by asking individuals to consider the opposite relationship, or engage in counterexplanation, often reduce the problem. In most cases, though, the initial erroneous beliefs still persevere (Anderson, 1982). In summary, belief perseverance effects (i.e., under-adjustment) appear to be robust, appearing both when retractions and corrections are present. Theories drawing on affect At first blush, theories based on affect appear less applicable than belief perseverance in the domain of retractions and corrections. That is, affect theory does not address constructs such as retraction or correction. However, key ideas of these theories suggest they are applicable to this domain, as explained below. Affect is commonly considered to be an evaluative reaction, such that information (including numerical information) is represented as a positive or negative valence in memory (Kida & Smith, 1995). Thus, when individuals observe quantitative information, such as an earnings forecast, the theory suggests that they will make comparisons among relevant numbers.6 These comparisons, in turn, lead to affective reactions, such as increasing/favorable or decreasing/unfavorable (Kida, Smith, & Maletta, 1998). Although factors such as attention and memory demands may influence the strength of these affective reactions, such 6 One might argue that anchoring and insufficient adjustment (Tversky & Kahneman, 1974) is a relevant theory for our tests. However, our setting involves multiple numbers (i.e., prior year actual earnings, forecasted earnings, and corrected earnings). Hence, it is difficult to reliably argue which one of these is the anchor (particularly between the prior year’s actual earnings and the initial erroneous forecast). Moreover, the fact that there are multiple numbers suggests the theoretical ideas developed in Kida and Smith (1995) are likely to dominate.

reactions are generally more potent than other features of the information set (Kida & Smith, 1995). Thus, they are more easily brought to mind and used in formulating subsequent judgments and decisions. With these ideas in mind, we argue that when an investor receives a retraction of a previously issued erroneous earnings forecast, their affective reaction to that forecast is likely to be central. That is, at the time the initial forecast was received, the investor coded the forecast as either favorable or unfavorable news relative to the prior year’s earnings (Barth, Elliott, & Finn, 1999). Then, when the earnings forecast is subsequently retracted by the company, the favorable or unfavorable affective reaction remains central in the mind of the investor. That is, it is difficult to dismiss from memory and, thus, has an undue influence on their subsequent judgments. Thus, for retractions, affect theory predicts under-adjustment. In the situation where an investor receives a correction of a previously issued forecast, affect-based research suggests that the reaction will be based on yet another comparison—namely, a comparison of what occurred with what might have been (Mellers, Schwartz, Ho, & Ritov, 1997; Shepperd & McNulty, 2002). That is, a bad outcome feels less disappointing when the comparison is worse. A good outcome feels less elating when the comparison is better. An intuitive example of these ideas is illustrated by a study of bronze and silver medalists at the Olympics (Medvec, Madey, & Gilovich, 1995). Bronze medalists showed a surprising tendency to be happier than silver medalists. Why? Bronze medalists apparently focused on the alternative of winning no medal, whereas silver medalists focused on the alternative of winning a gold medal. Indeed, individuals who are objectively better off than others can nevertheless feel worse because of a comparison of what occurred to what might have been. Drawing on this line of reasoning, if a company issues an initial high forecast and then later corrects it to a lower level, the investor compares the corrected forecast to the favorable initial high forecast. Affect theory indicates that the investor will be disappointed that the actual forecast is not the original, higher one (i.e., what might have been). By the same token, an investor who receives an erroneous initial low forecast and then later receives a corrected higher forecast from the company makes the comparison to the initial low forecast (i.e., what might have been). Here, the investor will be pleased that the actual forecast is not the original, lower one. Affect theory indicates that these two investors will have different judgments about the company even if the corrected forecast is identical. Specifically, it suggests over-adjustment as the investor who received the high forecast that was later corrected would judge the company less favorably (i.e., s/he feels worse) than the investor who received the low forecast that was later corrected. In summary, affect theory predicts underadjustment for retraction and over-adjustment for correction. When both retraction and correction are present, it follows that affect theory would predict over-adjustment. In summary, belief perseverance theory predicts underadjustment for all three company actions, i.e., retraction only, correction only, and both retraction and correction. In contrast, affect theory predicts under-adjustment for

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

387

Fig. 3. Initial and corrected forecasts used in experiments. This figure pictorially shows the forecasts used in the low EPS and high EPS forecast conditions, as well as the corrected earnings per share forecast that all participants received.

retraction only and over-adjustment for both correction only and both retraction and correction.7 Experiment one Participants, design, and materials Participants were 89 final-year students majoring in business. Participants had taken an average of 10.5 accounting and/or finance courses. Each participant was paid a small, fixed amount of money for completing the experimental task. Experiment one had a 2  2 between-participants design. Participants were provided with information about a company in the air and petroleum gas production industry. They were told that their objective was to assess the future earnings potential of the company and to make judgments about its investment attractiveness. Accordingly, participants were provided with excerpts of the balance sheet and income statement, including earnings per share (EPS) of $1.23 for the prior year. After reviewing these background materials, participants were given a press release reporting an EPS forecast made by company management for the current year. This forecast, which is our first manipulation, was varied at two levels—low EPS (range of $1.01–$1.11, with a midpoint of $1.06) or high EPS (range of $1.35–$1.45, with a midpoint of $1.40). Participants should recognize the low EPS forecast as a decline 7 These predictions appear similar to those rendered by Hogarth and Einhorn’s (1992) belief-adjustment model (also see Ashton & Ashton, 1988 in accounting). Their model accommodates both predictions of primacy (similar to our under-adjustment) and recency (similar to our overadjustment). However, their model is not appropriate for our experimental setting as it is designed to address situations where there are two pieces of evidence, A and B, that are presented to study participants in either an Athen-B or B-then-A order (Hogarth & Einhorn, 1992, pp. 4–5). In our setting, the order of information is not varied. Participants are given either information item A or B which is later retracted or corrected with the introduction of information item C.

in earnings as compared to the previous year and the high EPS as an increase in earnings (cf. Barth et al., 1999). Fig. 3 provides a pictorial overview of these forecasts. We chose to use range forecasts (as opposed to point forecasts) for several reasons. First, range forecasts are commonly issued by company managements, approximating about one-third of all forecasts issued in the last 10 years (Baginski, Hassell, & Kimbrough, 2004). Second, range forecasts communicate uncertainty in what future earnings will be (Hirst et al., 2008). Such uncertainty is an important element of belief revision studies. Key to our design is that both forecasts represent misinformation as they are subsequently retracted and/or corrected by the company’s management within a short time period.8 The second manipulation conveys the information about the company’s action—both retraction & correction or correction only. These two conditions are illustrated in Fig. 4. In the both retraction & correction condition, participants are initially given the high or low EPS forecast and asked to make two judgments. These judgments are labeled as A in Fig. 4, and they are described in more detail below. Then, they are told by company management, via a press release, that the previous forecast issued earlier in the day should be disregarded (i.e., it was retracted). They respond again to the same two judgment questions (labeled B in Fig. 4). After this information, they are given another press release (issued later in the same day) containing a corrected forecast from company management. The corrected forecast has a range of $1.18–$1.28 and an average of $1.23. This forecast falls between (and does not overlap with either of) the erroneous initial high and low forecasts, thereby providing an unambiguous correction.

8 Although most corrections are made relatively quickly after the initial erroneous disclosure, time gaps nevertheless still exist between the original erroneous disclosure and subsequent company actions. Tan and Tan (2009) provide small-sample evidence that approximately 20% of corrections are made on the same day as the initial erroneous disclosure and approximately 60% are made on the following day.

388

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

Both Retraction & Correction Condition High EPS Actual Earnings

Press Release Year 2 EPS forecast Judgment Elicitation

Low EPS

Year 1 actual EPS of $1.23 $1.01-1.11 (average of $1.06)

$1.35-1.45 (average of $1.40)

Perform judgment task

Judgment Elicitation

High EPS

Low EPS

Year 1 actual EPS of $1.23 $1.01-1.11 (average of $1.06)

$1.35-1.45 (average of $1.40)

Perform judgment task (1 of 2)

(1 of 3) Retraction Press Release

Correction only Condition

Retraction: Disregard Year 2 EPS forecast Perform judgment task (2 of 3)

Correction Press Release

Judgment Elicitation

$1.18 - $1.28 (corrected Year 2 EPS forecast) (i.e., average of $1.23)

$1.18 - $1.28 (corrected Year 2 EPS forecast) (i.e., average of $1.23)

Perform judgment task

Perform judgment task

(3 of 3)

(2 of 2)

Fig. 4. Overview of experiment one. This figure provides an overview of the experimental procedure for the two conditions in experiment one—that is, both retraction & correction and correction only. We created a retraction only condition from the both retraction & correction condition by comparing judgments A and B. The letters in the figures and tables that follow refer to the judgments of the corresponding cells above.

As in actual practice, both the retraction & correction press releases are clearly labeled as ‘‘corrections.’’9 The participants then respond again to the same two judgment questions (labeled C in Fig. 4). In the correction only condition, participants are not provided with a retraction. Rather, they are only given a correction. Accordingly, participants are asked to make two sets of judgments, first after receiving the erroneous forecast (labeled D in Fig. 4) and again after receiving the correction (labeled E in Fig. 4). We have two primary dependent variables (cf. Tan & Tan, 2009). The first question asked participants to rate on an eleven-point scale the company’s earnings potential with 1 representing ‘‘very poor’’ and 11 indicating ‘‘very good.’’ The second question asked participants to indicate the attractiveness of the company as an investment, with 1 indicating ‘‘very unattractive’’ and 11 representing ‘‘very attractive.’’10 The final component of the experimental 9 Doing so helps ensure that our study participants did not view the correction as a revision, or updating of management’s beliefs based on new information. Rather, it makes it clear to participants that it is a correction of an erroneous forecast. 10 Because prior research documents an overwhelming tendency to rely on the midpoint when range forecasts are provided (Hirst, Koonce, & Miller, 1999), we did not ask participants to make their own earnings forecast judgments. Although we did not measure the participants’ EPS estimates, we expect that participants’ assessments of the company’s earnings potential over the next 2 years serves as a proxy for their EPS estimates.

materials, that all participants completed, were demographic and manipulation check questions.

Experiment one results To determine whether the forecast manipulation was successful, we asked participants to report whether the initial erroneous forecast was higher, lower, or similar to the prior year’s actual earnings. To check the effectiveness of the forecast correction, we asked participants to indicate whether the corrected forecast was higher or lower than the initial erroneous forecast (or indicate that they did not receive a corrected forecast).11 Both manipulations were successful; correct response rates were 84% and 95% for the forecast and correction questions, respectively. Importantly, chi-square tests indicate that incorrect responses were not clustered in any one condition (all p-values >0.50).12 11 Based on pilot test results, it was difficult to clearly ask participants whether or not they received a retraction. Accordingly, for those participants in the both retraction & correction condition, we asked a slightly different forecast manipulation question—one that also incorporated information about the retraction. 12 For all of the experiments reported in the paper, the results are similar when the participants who failed a manipulation check are excluded from the statistical tests.

389

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397 Table 1 Experiment one descriptive statistics and test results. Judgment D

Judgment E

Panel A: Means (standard deviations) for earnings potential and investment attractiveness judgments Earnings potential judgment High EPS 7.32 (0.89) 6.68 (1.00) 5.68 (1.46) n = 22 n = 22 n = 22 Low EPS 4.96 (1.07) 5.57 (0.87) 6.91 (1.37) n = 21 n = 21 n = 21

Judgment A

Judgment B

Judgment C

7.17 (0.83) n = 23 5.07 (1.75) n = 23

5.65 (1.47) n = 23 6.63 (1.59) n = 23

Investment attractiveness judgment High EPS 7.18 (1.05) n = 22 Low EPS 5.05 (0.97) n = 21

7.17 (1.07) n = 23 4.98 (1.85) n = 22

5.48 (1.70) n = 23 6.46 (2.01) n = 23

6.55 (1.06) n = 22 5.43 (1.16) n = 21

Source

5.32 (1.52) n = 22 6.43 (1.20) n = 21

Earnings potential

Investment attractiveness

df

Statistic

Two-tailed p-value

df

Statistic

Two-tailed p-value

Panel B: Tests of retraction only (judgments A vs B in Fig. 4) Judgment timing (pre- vs post-retraction only) Forecast (high vs low EPS) Judgment timing  forecast

1 1 1

F = 0.00 F = 47.01 F = 18.25

0.95 <0.01 <0.01

1 1 1

F = 0.46 F = 37.67 F = 7.35

0.50 <0.01 <0.01

Simple main effects: (high vs low EPS) Pre-retraction only judgment Post-retraction only judgment

1 1

t = 8.07 t = 3.79

<0.01 <0.01

1 1

t = 6.58 t = 3.44

<0.01 <0.01

Panel C: Tests of both retraction & correction (judgments A vs C in Fig. 4) Judgment timing (pre- or post-retraction & correction) 1 F = 0.51 Forecast (high vs low EPS) 1 F = 3.63 Judgment timing  forecast 1 F = 65.94

0.48 0.06 <0.01

1 1 1

F = 1.36 F = 2.80 F = 61.39

0.25 0.10 <0.01

Simple main effects: (high vs low EPS) Pre-retraction & correction judgment Post-retraction & correction judgment

1 1

t = 6.35 t = 3.28

<0.01 <0.01

1 1

t = 5.78 t = 3.01

<0.01 <0.01

Panel D: Tests of correction only (judgments D vs E in Fig. 4) Judgment timing (pre- or post-correction only) 1 Forecast (high vs low EPS) 1 Judgment timing  forecast 1

F = 0.01 F = 3.14 F = 29.15

0.94 0.08 <0.01

1 1 1

F = 0.15 F = 2.16 F = 32.14

0.70 0.15 <0.01

Simple main effects: (high vs low EPS) Pre-correction only judgment Post-correction only judgment

t = 4.92 t = 2.28

<0.01 0.03

1 1

t = 4.39 t = 1.96

<0.01 0.06

1 1

Recall that under- or over-adjustment would be observed if participants’ judgments in the high EPS and low EPS conditions are different once they have been provided with a press release indicating that they should disregard that forecast (i.e., retraction only), rely on the new, corrected forecast (i.e., correction only), or do both (i.e., both retraction & correction).13 Under-adjustment (over-adjustment) would be indicated if participants’ judgments in the high EPS condition are more (less) favorable than the judgments in the low EPS condition.

Tests of retraction only Although our primary purpose is to study correction, we nevertheless first replicate the prior accounting findings of 13 Recall that all participants were given information about the prior year’s actual earnings per share. Thus, when participants are given a retraction, they should revert back to beliefs based on the prior year’s actual earnings or some common adjustment from that number (Maines & Hand, 1996). Because the prior year’s actual earnings is the same for all participants, the revised judgments should not be different across conditions.

under-adjustment in the presence of retraction (Tan & Tan, 2009) to ensure the robustness of that result. Recall that both belief perseverance theory and affect theory predict that participants would demonstrate under-adjustment when provided with an erroneous initial forecast which they are later told by its source, company management, to disregard. For our tests of retraction only, we compare the initial (A) and second (B) judgments made by participants in the both retraction & correction condition. Table 1 shows the results of these tests, with Panel A showing the means and standard deviations and Panel B documenting the mixed-model analysis of variance (ANOVA) along with follow-up simple main effects. Fig. 5 graphically displays these results. As shown in Table 1, we document a significant statistical interaction between the forecast and judgment timing (i.e., pre- and post-retraction) variables for both the earnings potential and investment attractiveness dependent measures (both p-values <0.01). This interaction reveals that the difference in participants’ judgments between the high and low EPS forecasts was reduced once the forecast was retracted. However, follow-up simple main effect tests show that a

390

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

Tests of retraction only (Judgments A versus B)

Tests of both retraction & correction (Judgments A versus C)

Legend: High EPS

Tests of correction only (Judgments D versus E)

Low EPS

Fig. 5. Experiment one graphical presentation of results for earnings potential dependent measure (see Fig. 4 for a description of the judgment labels referenced in this figure).

difference in judgments persisted even after the retraction. Specifically, for the earnings potential dependent variable, those who received the initial erroneous high EPS forecast still believed that the company’s earnings potential was greater (mean of 6.68) even after the initial forecast had been retracted as compared to those who received an initial low EPS forecast that was subsequently retracted (mean of 5.57) (t = 3.79, p < 0.01). Similar post-retraction results exist for the investment attractiveness measure (means = 6.55 and 5.43 for high and low EPS, respectively; t = 3.44, p < 0.01).14 Overall, these tests of retraction only are consistent with the predictions of both belief perseverance and affect theory—that is, they reveal under-adjustment. Our tests of correction (alone or with a prior retraction), discussed below, will enable us to discriminate between these two theories. Overall tests of both retraction & correction and correction only Recall that in the presence of a correction, belief perseverance theory indicates that participants will demonstrate under-adjustment, while affect theory predicts over-adjustment in this circumstance. Thus, our correction tests allow us to discriminate between these two theories. Moreover, our research design enables us to study the effect of correction in two ways—one where the correction is preceded by a retraction and the other where it is not preceded by a retraction. Our first analysis involves a 2  2  2 mixed-model ANOVA. The company action variable—both retraction & correction or correction only—and the forecast variable— high or low EPS forecast—are the between-participant variables. The within-participants variable is the timing of the participants’ judgments—that is, immediately after the erroneous forecast or after the correction (i.e., judgments A and D vs C and E in Fig. 4). We conduct this analysis 14 Tests reveal heterogeneity of variances among the four cells, primarily driven by the small standard deviation in the no-retraction/high EPS forecast cell. Therefore, we follow the procedure used by Kachelmeier and Messier (1990) and convert our data to ranks, thereby reducing variance heterogeneity. Using ranks, our results are similar to those in the paper. We also perform this procedure for a small number of other tests in this paper where variance heterogeneity arises, with no change in any inferences.

for both the earnings potential and investment attractiveness dependent variables. Turning first to the results for the earnings potential dependent measure (not tabulated), we see a significant main effect for the forecast variable (F = 6.69, p = 0.01) and a significant forecast  judgment timing interaction (F = 83.68, p < 0.01). No other components of the model were significant, including the forecast  judgment timing  company action three-way interaction (all p-values >0.50). For the investment attractiveness measure (not tabulated), we similarly observe a main effect for the forecast variable (F = 4.64, p = 0.03) and a significant forecast  judgment timing interaction (F = 83.07, p < 0.01), with no other significant terms (all p-values >0.30). Overall, these results indicate that participants reacted similarly whether a retraction & correction were provided or whether just a correction was given. For completeness, we separately analyze these forecast  judgment timing results by company action below.

Tests of both retraction & correction Table 1, Panel C, shows the results for our tests of both retraction & correction (also see Fig. 5 for graphical displays).15 For this 2  2 test, we compare the initial and final judgments for those in the both retraction & correction condition (i.e., compare judgments A and C). For both dependent measures, our results show a significant interaction (both p-values <0.01), suggesting that participants revised their initial (erroneous) judgments after receiving the corrected forecast. Most pertinent to our study, though, their final judgments revealed over-adjustment (i.e., they flipped), consistent with the predictions of affect theory. For the earnings potential dependent measure, those in the high EPS forecast conditions reduced their judgments from a mean of 7.32 to a mean of 5.68, while those in the low EPS forecast conditions increased their earnings 15 Because belief perseverance theory predicts under-adjustment in the presence of either retraction only or correction only, it stands to reason that it predicts under-adjustment when both are present. For affect theory, because the affective reaction driving over-adjustment would occur when both a retraction and correction are present, it follows that over-adjustment will occur in this case.

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

potential judgments from 4.96 to 6.91. These final judgments are significantly different (t = 3.28, p < 0.01). Similar results exist for the investment attractiveness measure.

Tests of correction only For this 2  2 test, we compare the initial and final judgments for those in the correction only condition (i.e., compare judgments D and E). Panel A of Table 1 shows the applicable means and standard deviations and Panel D documents the appropriate ANOVA along with followup simple main effects. Fig. 5 provides a graphical representation of the results. Similar to the both retraction & correction results presented earlier, our results here show a significant interaction for both dependent measures (both p-values <0.01), suggesting that participants revised their initial (erroneous) judgments after receiving the corrected forecast. Once again, their final judgments revealed over-adjustment (i.e., they flipped), consistent with the predictions of affect theory. Those in the high EPS forecast conditions reduced their earnings potential judgments from a mean of 7.17 to a mean of 5.65, while those in the low EPS forecast conditions increased their earnings potential judgments from 5.07 to 6.63. These final judgments are significantly different (t = 2.28, p = 0.03). This crossover pattern in the final judgments indicated that participants over-adjusted for the erroneous initial forecast, once the replacement forecast was provided. Similar patterns are observed for the investment attractiveness dependent measure. Before concluding that these results are fully consistent with affect theory, we perform an additional analysis to rule out a possible alternative explanation for our results. It is possible that participants in the high EPS condition were suspicious of management trying to convey a favorable outlook by initially (and intentionally) forecasting high earnings and then later retracting and/or correcting that (but only after the participant perhaps had made some judgment or taken an action). If so, we would expect participants in the high EPS forecast condition to rate management integrity to be lower than those in the low EPS forecast condition. To test this possibility, we estimated a 2  2 ANOVA for the two forecast conditions and two company actions, using the responses to a management integrity question. This question asked participants to rate management integrity on an 11-point scale with 1 representing ‘‘very low integrity’’ and 11 representing ‘‘very high integrity.’’ The ANOVA results reveal that neither variable, nor their interaction, is significant (all p-values >0.14). The average response was 5.98 (5.81) for the high EPS (low EPS) conditions.16 Hence, the over-adjustment effects obtained in our study is unlikely to be caused by concerns over managerial intent. Overall, the experiment one results are consistent with affect theory. That is, we observe under-adjustment when 16 We also estimated structural equation models using the management integrity judgment as a mediator between the manipulated variables and the main dependent measures with similar inferences—namely, management integrity cannot explain the pattern of results we obtain in the study. All of these tests were replicated for experiments two, three, and four as well.

391

a retraction is present and over-adjustment when a correction is provided. The over-adjustment result occurs both when a correction is provided with and without a preceding retraction.

Experiment two The objectives of the second, third, and fourth experiments are to test the robustness of our findings in experiment one. Turning first to experiment two, we explore whether our choice of study participants in the first experiment influenced these results. Participants, design, and materials The experimental task in this second experiment is identical to the correction only condition from experiment one. However, we use a different population of study participants—namely, Masters in Business Administration (MBA) students. These 40 participants differ from the participants in experiment one as the MBA students have 5.4 years of work experience, on average. Although the nature of our experimental task is such that these additional experiences are unlikely to systematically influence our results (Elliott, Hodge, Kennedy, & Pronk, 2007), we nevertheless test this presumption. Because we focus on correction only in this second experiment, participants initially received either a high EPS or low EPS forecast, assessed the company’s earnings potential and investment attractiveness (i.e., judgment D in Fig. 4), received the corrected forecast, and then re-assessed the two dependent measures (i.e., judgment E). Experiment two results Participants responded to the same two manipulation check questions as in experiment one. The correct response rates were 95% and 100% for the forecast and correction questions, respectively, which suggest a successful manipulation. The results for this second experiment are shown in Table 2 and also Fig. 6. For both the earnings potential and investment attractiveness dependent measures, we observe a significant interaction between the judgment timing and forecast variables (both p-values <0.01). Most interesting for our purposes, though, is that these judgments ‘‘flipped,’’ or over-adjusted, once the earnings forecast correction was received and participants re-assessed their evaluations of the company (both p-values <0.01). That is, after the correction, their earnings potential judgments were less favorable (mean of 5.18) when they received an initial high EPS forecast than when they received an initial low EPS forecast (mean of 6.85) (t = 3.73, p < 0.01). Similar results were obtained for the investment attractiveness measure (means of 4.68 and 6.80, respectively) (t = 3.97, p < 0.01). Even though the corrected forecast was identical for both forecast groups, their revised judgments about the company were much more favorable after an initial low EPS forecast was

392

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

Table 2 Experiment two descriptive statistics and test results. Judgment D

Judgment E

Panel A: Means (standard deviations) for earnings potential and investment attractiveness judgments Earnings potential judgment High EPS 7.48 (1.33) n = 20 Low EPS 4.78 (1.34) n = 20 Investment attractiveness judgment High EPS

5.18 (1.32) n = 20 6.85 (1.66) n = 20

6.98 (2.22) n = 20 4.65 (1.30) n = 20

Low EPS

Source

4.68 (1.57) n = 20 6.80 (1.55) n = 20

Earnings potential df

Investment attractiveness

Statistic

Two-tailed p-value

df

Statistic

Two-tailed p-value

Panel B: Tests of correction only (judgments D vs E in Fig. 4) Judgment timing (pre- or post-correction only) 1 Forecast (high vs low EPS) 1 Judgment timing  forecast 1

F = 0.15 F = 2.21 F = 57.80

0.70 0.15 <0.01

1 1 1

F = 0.05 F = 0.05 F = 47.70

0.82 0.82 <0.01

Simple main effects: (high vs low EPS) Pre-correction only judgment Post-correction only judgment

t = 6.01 t = 3.73

<0.01 <0.01

1 1

t = 4.34 t = 3.97

<0.01 <0.01

1 1

Tests of correction only (Judgments D versus E)

Legend: High EPS

Low EPS

Fig. 6. Experiment two graphical presentation of results for earnings potential dependent measure (see Fig. 4 for a description of the judgment labels referenced in this figure).

corrected than when an initial high EPS forecast was corrected. In sum, these results fully replicate those from experiment one and are consistent with affect theory. That is, we observed over-adjustment once again, suggesting that our participant choice in experiment one is not a concern. Experiments three and four The third and fourth experiments provide additional evidence testing the potential applicability of belief perseverance theory and affect theory. In particular, in experiment three, our tests are designed to further rule out the applicability of belief perseverance theory and, in experiment four, they are designed to further rule in affect theory.

In experiment three, we ask study participants to explain the initial forecast received from the company management. This explanation task precedes any retraction and/or correction that is subsequently received. Explanation has been found to bolster belief perseverance effects (i.e., under-adjustment), as it results in more elaborate and explicit generation of inferences and causal accounts based on the erroneous information (Anderson et al., 1980). Thus, if belief perseverance theory is applicable, we should see an increase in the amount of under-adjustment (after retraction) and a reduction in the amount of over-adjustment (after both retraction & correction or correction only). In contrast, affect theory indicates that the explanation task, which is cognitive in nature, should have no effect on adjustment. In experiment four, we omit the initial set of judgment questions that require the study participants to evaluate the earnings potential and investment attractiveness of the company based on the erroneous initial forecast. By not asking these questions, participants’ initial affective reactions to the favorable or unfavorable initial forecast should be weaker, as they have spent less time making the comparison and generating the affective reaction to the initial (erroneous) forecast. As a result, when the corrected forecast is subsequently received, those who did not make an initial set of judgments are less likely to make the ‘‘what could have been’’ comparison suggested by affect theory (Kida & Smith, 1995). Stated differently, they are less likely to have a strong opposite affective reaction to the correction. Thus, these participants should be less prone to over-adjustment when the correction is received. Belief perseverance research, in contrast, reveals that the presence or absence of an initial judgment has no effect on the amount of adjustment after a retraction and/or correction (Anderson et al., 1980).

393

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397 Table 3 Experiment three descriptive statistics and test results. Judgment D

Judgment E

Panel A: Means (standard deviations) for earnings potential and investment attractiveness judgments Earnings potential judgment High EPS 7.10 (1.14) 6.14 (1.32) 4.86 (1.82) n = 21 n = 21 n = 21 Low EPS 5.29 (1.31) 5.29 (1.27) 6.52 (1.57) n = 21 n = 21 n = 21

Judgment A

Judgment B

Judgment C

7.14 (1.01) n = 21 5.27 (1.35) n = 22

5.46 (1.82) n = 22 7.00 (1.02) n = 22

Investment attractiveness judgment High EPS 6.48 (1.75) n = 21 Low EPS 4.81 (1.44) n = 21

7.10 (0.94) n = 21 5.05 (1.50) n = 22

5.55 (1.44) n = 22 6.86 (1.32) n = 22

6.14 (1.46) n = 21 5.14 (1.39) n = 21

Source

5.00 (1.90) n = 21 6.43 (1.78) n = 21

Earnings potential

Investment attractiveness

df

Statistic

Two-tailed p-value

df

Statistic

Two-tailed p-value

Panel B: Tests of retraction only (judgments A vs B in Fig. 4) Judgment timing (pre- vs post-retraction only) Forecast (high vs low EPS) Judgment timing  forecast

1 1 1

F = 5.37 F = 16.32 F = 5.37

0.03 <0.01 0.03

1 1 1

F = 0.00 F = 9.28 F = 4.12

1.00 <0.01 0.05

Simple main effects: (high vs low EPS) Pre-retraction only judgment Post-retraction only judgment

1 1

t = 4.65 t = 2.20

<0.01 0.03

1 1

t = 3.56 t = 2.14

<0.01 0.04

Panel C: Tests of both retraction & correction (judgments A vs C in Fig. 4) Judgment timing (pre- or post-retraction & correction) 1 F = 3.24 Forecast (high vs low EPS) 1 F = 0.04 Judgment timing  forecast 1 F = 39.16

0.08 0.85 <0.01

1 1 1

F = 0.05 F = 0.08 F = 25.44

0.82 0.79 <0.01

Simple main effects: (high vs low EPS) Pre-retraction & correction judgment Post-retraction & correction judgment

1 1

t = 3.95 t = 3.64

<0.01 <0.01

1 1

t = 3.13 t = 2.69

<0.01 0.01

Panel D: Tests of correction only (judgments D vs E in Fig. 4) Judgment timing (pre- or post-correction only) Forecast (high vs low EPS) Judgment timing  forecast

1 1 1

F = 0.00 F = 0.36 F = 31.59

0.95 0.55 <0.01

1 1 1

F = 0.29 F = 1.40 F = 41.41

0.60 0.24 <0.01

Simple main effects: (high vs low EPS) Pre-correction only judgment Post-correction only judgment

1 1

t = 4.56 t = 3.81

<0.01 <0.01

1 1

t = 5.06 t = 3.31

<0.01 <0.01

Experiment three Participants, design, and materials Experiment three, conducted at the same time as experiment one, involved 86 students, drawn from the same population. The task was identical to that used in experiment one, but with the addition of the explanation task. After being given the high or low forecast of earnings for the coming year, participants were asked to explain the factors that they believed were behind the earnings forecast just received. After providing this explanation, participants were randomly assigned to the both retraction & correction condition or the correction only condition. Experiment three results We asked participants the same manipulation check questions described previously in experiment one. Both manipulations were successful; correct response rates were 92% and 97% for the forecast and correction questions, respectively. Importantly, chi-square tests indicate that incorrect responses were not clustered in any one condition (both p-values >0.10). Further, participants in all

four conditions provided, on average, just over one factor (mean of 1.02) in their explanations. This number does not differ among the four conditions (p > 0.10). Separate chi-square tests for the high EPS and low EPS forecast conditions indicate that there are no significant differences in the types of factors cited by participants in the two conditions (both p-values >0.10).17 The results for this third experiment are shown in Table 3 and also Fig. 7. Overall, our results in experiment three are similar to those in experiment one. For ease of reference, we use the same judgment letter references shown in Fig. 4. We also focus here on our earnings potential judgments as the investment attractiveness measures reveal similar results. For our test of retraction only, we observe that retraction of the initial forecast reduces the difference

17 A content analysis of these explanations reveals comments about factors such as good cost management (42%) and improved revenue and customer base (37%) in the high EPS conditions. The explanations of those in the low EPS forecast conditions indicated more negative remarks such as higher costs and expenses (60%) and increased competition and worsening economic conditions (18%).

394

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

Tests of retraction only (Judgments A versus B)

Tests of both retraction & correction (Judgments A versus C)

Legend: High EPS

Tests of correction only (Judgments D versus E)

Low EPS

Fig. 7. Experiment three graphical presentation of results for earnings potential dependent measure (see Fig. 4 for a description of the judgment labels referenced in this figure).

in investors’ earnings potential judgments between the high EPS (A = 7.10 vs B = 6.14) and low EPS forecast conditions (A = 5.29 vs B = 5.29) but does not eliminate it (t = 2.20, p = 0.03 for judgments B). This result indicates under-adjustment, which is similar to that observed in experiment one. Turning next to our test of both retraction & correction, we observe over-adjustment, again similar to that found in experiment one. The retraction of the initial forecast and introduction of the corrected forecast causes the difference in judgments between the high EPS (A = 7.10 vs C = 4.86) and low EPS forecast (A = 5.29 vs C = 6.52) conditions to ‘‘flip.’’ That is, those receiving the initial erroneous high EPS forecast judge the company worse than those who initially received the low EPS forecast (t = 3.64, p < 0.01 for judgments C). Finally, our test of correction only also shows over-adjustment, similar to experiment one (and also similar to experiment two with the MBA’s). When participants are given a corrected forecast (but not a retraction), their judgments about the company ‘‘flip,’’ or over-adjust (high EPS: D = 7.14 vs E = 5.46; low EPS: D = 5.27 vs E = 7.00) (t = 3.81, p < 0.01 for judgments E).18 Although these experiment three results are strongly suggestive of the notion that belief perseverance theory is not applicable as the results’ pattern is consistent with affect theory, one additional test will help in validating that conclusion. Specifically, we test whether the explanation task in experiment three affects the size of the judgment revisions. That is, does explanation influence the magnitude of the under- or over-adjustments we observed in the first experiment? If the explanation task does influence the size of the participants’ under- and over-adjustments, then it provides some argument for belief perseverance and against affect theory. To conduct these tests, we compare the applicable postretraction or post-correction (alone or with a prior retraction) judgments from this experiment to the comparable 18 The experimental design used in experiments one and three allow us to not only perform within-participants tests (as reported in the paper) but also select between-participants tests. For example, to test the effect of retraction on a between-participants’ basis, we can compare judgment D (see Fig. 4) to judgment C. For both experiments one and three, these between-participants’ tests reveal similar inferences as the within-participants’ tests we report in the paper.

judgments in experiment one (that did not involve explanation). Recall that all aspects of experiments one and three were identical except for the inclusion of the explanation task in experiment three. Further, both experiments were conducted at the same time, with random assignment of treatment conditions to participants. In these 2  2  2  2 mixed-model ANOVA’s (i.e., forecast  company action  judgment timing  presence vs absence of explanation), we are looking for an interaction involving the explanation variable. Separately analyzing our two dependent measures—earnings potential and investment attractiveness—reveals no significant interactions involving the explanation variable (all p-values >0.15, results not tabulated). Thus, these results indicate that the explanation task did not systematically affect the size of the under- or overadjustments we observed, further arguing against belief perseverance theory and in favor of affect theory.19 Experiment four Participants, design, and materials Experiment four, conducted at the same time as experiment one, involved 45 students, drawn from the same population. Because correction is the primary focus of our paper, we only tested one company action—namely, correction only—in this final experiment. Key to experiment four, though, is that we omitted the initial set of judgments (i.e., those immediately after the erroneous forecast was provided) in our attempt to reduce the strength of the initial affective reaction. Otherwise, the experimental task was identical to that used in experiment one. Referring to Fig. 4, our experiment tests the correction only condition, with the exception that judgment D is omitted. The idea is that by not asking for these judgments, participants are less likely to form a strong affective reaction to the initial erroneous forecast, thereby reducing the affective reaction that occurs once a correction is provided. Thus, this manipulation should temper the degree of over-adjustment and directly address the applicability of affect theory. 19 It is possible that this conclusion may differ depending on the content and/or length of the written explanation. Future research could explore this possibility.

395

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397 Table 4 Experiment four descriptive statistics and test results. Correction only with initial judgmenta

Correction only with no initial judgment

Panel A: Means (standard deviations) for earnings potential and investment attractiveness judgments Earnings potential judgment High EPS 5.65 (1.47) n = 23 Low EPS 6.63 (1.59) n = 23 Investment attractiveness judgment High EPS Low EPS

Source

5.48 (1.70) n = 23 6.46 (2.01) n = 23

6.20 (1.36) n = 23 5.96 (1.53) n = 22

Earnings potential df

a

6.46 (1.36) n = 23 5.95 (1.50) n = 22

Investment attractiveness Two-tailed p-value

df

Statistic

Two-tailed p-value

Panel B: Tests of presence/absence of initial judgment on correction only Initial judgment (present vs absent) 1 F = 0.05 Forecast (high vs low EPS) 1 F = 0.63 Initial judgment  forecast 1 F = 5.68

0.82 0.43 0.02

1 1 1

F = 0.11 F = 1.15 F = 3.05

0.74 0.29 0.08

Simple main effects: (high vs low EPS) Initial judgment present Initial judgment absent

0.03 0.26

1 1

t = 1.99 t = 0.49

0.05 0.63

1 1

Statistic

t = 2.24 t = 1.14

These correction only with initial judgment data originate from experiment one (see Table 1, judgment E).

Experiment four results We asked participants the same manipulation check questions described previously in experiment one. Both manipulations were successful; correct response rates were 87% and 98% for the forecast and correction questions, respectively. As before, incorrect responses are not clustered by condition (both p-values >0.40). Our results show that the over-adjustment result that we documented in experiment one is fully eliminated once a replacement forecast is provided. Specifically, participants in the high EPS forecast condition judged the earnings potential to be statistically similar (mean of 6.46) to that judged by those in the low EPS forecast condition (mean of 5.95) (F = 1.39, p = 0.24). A similar result was observed for the investment attractiveness dependent measure (means of 6.20 and 5.96, respectively, F = 0.31, p = 0.58). These results suggest that when investors do not form a strong prior belief about the erroneous forecast, they experience less affective reactions and thus, are not prone to over-adjustment. Although these experiment four results are strongly suggestive of the notion that affect theory is operating, one additional test will help in validating that conclusion. Specifically, we test whether presence or absence of an initial judgment task—which captures the strength of the initial affective reaction—affects the size of the judgment revisions that were observed once a correction was received. To conduct this test, we compare the post-correction judgments from the experiment one correction only condition where an initial judgment was made (judgment E in Fig. 4) to the comparable judgments here in experiment four (where an initial judgment was not made). Again, recall that the only difference between these two conditions is the presence or absence of an initial set of judgments

made immediately after the erroneous forecast was received from management. Further, both experiments one and four were conducted at the same time with random assignment of treatment conditions to participants. Results are shown in Table 4 and illustrated in Fig. 8. In the 2  2 between-participant tests, we are expecting a significant interaction between high EPS and low EPS forecasts and presence/absence of initial judgment. As expected, results indicate a significant forecast by presence/ absence of initial judgment interaction (both p-values 60.08). Follow-up simple main effects reveal that when an initial judgment is made, we observe over-adjustment (both p-values 60.05). In contrast, when an intial judgment

Tests of correction only

Legend: High EPS

Low EPS

Fig. 8. Experiment four graphical presentation of results for earnings potential dependent measure (the correction only with initial judgment data corresponds to judgment E in Fig. 4 and was previously reported in experiment one (see Table 1). The correction only with no initial judgment data was collected at the same time as experiment one; it corresponds to the correction only condition shown in Fig. 4 with one exception—namely, an intial judgment (i.e., judgment D) was not elicited).

396

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397

is not made, neither over- or under-adjustment occurs (both p-values P0.26). These results are consistent with the ideas in affect theory. That is, when an initial judgment is not made, the affective reaction to a corrected forecast is minimized, thereby reducing over-adjustment.

Conclusion In this paper, we conduct multiple experiments to investigate how investors react to retractions and corrections. We develop competing theoretical predictions based on belief perseverance theory and affect theory. The results of four experiments are consistent with affect theory from psychology. Specifically, when a company provides a retraction of a previous erroneous voluntary disclosure, investors’ judgments continue to reflect the implications of the initial erroneous information. That is, investors under-adjust. Affect theory indicates that under-adjustment is likely as the positive or negative affective reaction from the initial erroneous disclosure lingers in the mind of the investor. In contrast, when a company provides a correction of a previous erroneous disclosure, affect theory makes an opposite prediction. That is, it indicates that over-adjustment is likely, as the investor will make a comparison of what actually occurred (i.e., the corrected information) with what might have been (i.e., the initial erroneous disclosure). This comparison leads to overadjustment because of the opposite affective reaction created by this comparison. Interestingly, our experimental tests of correction indicate that over-adjustment occurs but only when ‘‘what might have been’’ is salient in the mind of the investor. That is, the strength of the initial affective reaction, based on the erroneous forecast, is key to observing the subsequent affective reaction and, thus, whether over-adjustment occurs. Based on the insights of our study, we believe additional research opportunities are possible. One such area pertains to the long-standing line of research that documents under-adjustment (i.e., belief perseverance) in various domains. We show not only under-adjustment but also over-adjustment. Potentially important differences exist between the studies documenting belief perseverance effects and our study. For example, our use of numerical information in the original, erroneous press release and in the correction may have been conducive to the formation of comparisons, thus leading to the affective reactions. Most of the prior belief perseverance research examines qualitative information. Research could profitably explore those circumstances where belief perseverance theory vs affect is likely to be applicable. Yet another area for future research pertains to potential remedies for the under- and over-adjustment behavior we document herein. Such research is likely to involve new methods and/or theories, as many of the traditional cognitive remedies for what appears to be non-normative behavior are not applicable for the affect-based biases we document. Arkes (1991) suggests that if the source of the judgment error is affect-based, then only certain remedies are likely to work. In particular, he indicates that incentives to be accurate will not work nor will tactics such as

‘‘considering the opposite.’’ Rather, reframing the problem or changing the reference point are more apt to be effective. These ideas await future research. Acknowledgments This research is supported by a research grant from the Singapore Ministry of Education and also the Deloitte Foundation. We thank Rob Bloomfield (editor), an anonymous reviewer, Wei Chen, Mary Fischer, Tracie Majors, Premila Gowri Shankar, James Smith, Todd Thornock, and Elaine Ying Wang for their helpful comments on an earlier version of this paper. References Ajinkya, B., & Gift, M. (1984). Corporate managers’ earnings forecasts and symmetrical adjustment of market expectations. Journal of Accounting Research, 22(2), 425–444. Anderson, C. A. (1982). Inoculation and counter-explanation: Debiasing techniques in the perseverance of social theories. Social Cognition, 1(2), 126–139. Anderson, C. A., Lepper, M. R., & Ross, L. (1980). Perseverance of social theories: The role of explanation in the persistence of discredited information. Journal of Personality and Social Psychology, 39(6), 1037–1049. Arkes, H. (1991). Costs and benefits of judgment errors: Implications for debiasing. Psychological Bulletin, 110(3), 486–498. Ashton, A. H., & Ashton, R. H. (1988). Sequential belief revision in auditing. The Accounting Review, 63(4), 623–641. Baginski, S. P., & Hassell, J. M. (1990). The market interpretation of management earnings forecasts as a predictor of subsequent financial analyst forecast revision. The Accounting Review, 65(1), 175–190. Baginski, S. P., Hassell, J. M., & Kimbrough, M. D. (2004). Why do managers explain their earnings forecasts? Journal of Accounting Research, 42(1), 1–29. Barth, M., Elliott, J., & Finn, M. (1999). Market rewards associated with patterns of increasing earnings. Journal of Accounting Research, 37(2), 387–413. Brown, S., Hillegeist, S., & Lo, K. (2004). Conference calls and information asymmetry. Journal of Accounting and Economics, 37(3), 343–366. Coller, M., & Yohn, T. L. (1997). Management forecasts and information asymmetry: An examination of bid-ask spreads. Journal of Accounting Research, 35(2), 181–191. Elliott, W. B., Hodge, F., Kennedy, J., & Pronk, M. (2007). Are M.B.A. students a good proxy for nonprofessional investors? The Accounting Review, 82(1), 139–168. Financial Accounting Standards Board (FASB) (2000). Business reporting research project: Electronic distribution of business reporting information. Norwalk, CT: FASB. Healy, P., Hutton, A., & Palepu, K. (1999). Stock performance and intermediation changes surrounding sustained increases in disclosure. Contemporary Accounting Research, 16(3), 485–520. Hirst, D. E., Koonce, L., & Miller, J. (1999). The joint effect of management’s prior forecast accuracy and the form of its financial forecasts on investor judgment. Journal of Accounting Research, 37(Supplement), 101–124. Hirst, D. E., Koonce, L., & Venkataraman, S. (2008). Management earnings forecasts: A review and framework. Accounting Horizons, 22(3), 315–338. Hogarth, R. M., & Einhorn, H. J. (1992). Order effects in belief updating: The belief-adjustment model. Cognitive Psychology, 24(1), 1–55. International Accounting Standards Committee (IASC) (1999). Business reporting on the internet. London: IASC. Johnson, H. M., & Seifert, C. M. (1994). Sources of the continued influence effect: When misinformation in memory affects later inferences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(6), 1420–1436. Kachelmeier, S. J., & Messier, W. F. (1990). An investigation of the influence of a nonstatistical decision aid on auditor sample size decisions. The Accounting Review, 65(1), 209–226. Kida, T., & Smith, J. (1995). The encoding and retrieval of numerical data for decision making in accounting contexts: Model development. Accounting, Organizations and Society, 20(7-8), 585–610.

S.-K. Tan, L. Koonce / Accounting, Organizations and Society 36 (2011) 382–397 Kida, T., Smith, J., & Maletta, M. (1998). The effects of encoded memory traces for numerical data on accounting decision making. Accounting, Organizations and Society, 23(5-6), 451–466. Lepper, M. R., Ross, L., & Lau, R. (1986). Persistence of inaccurate beliefs about the self: Perseverance effects in the classroom. Journal of Personality and Social Psychology, 50(3), 482–491. Lord, C. G., Lepper, M. R., & Preston, E. (1984). Considering the opposite: A corrective strategy for social judgment. Journal of Personality and Social Psychology, 47(6), 1231–1243. Maines, L. A., & Hand, J. R. M. (1996). Individuals’ perceptions and misperceptions of the time series properties of quarterly earnings. The Accounting Review, 71(3), 317–336. Medvec, V. H., Madey, S., & Gilovich, T. (1995). When less is more: Counterfactual thinking and satisfaction among Olympic medal winners. Journal of Personality and Social Psychology, 69(4), 603–610. Mellers, B. A., Schwartz, A., Ho, K., & Ritov, I. (1997). Decision affect theory: Emotional reactions to the outcome of risky options. Psychological Science, 8(6), 423–429. Pownall, G., Wasley, C., & Waymire, G. (1993). The stock price effects of alternative types of management earnings forecasts. The Accounting Review, 68(4), 896–912. Ross, L., & Anderson, C. A. (1982). Shortcomings in the attribution process: On the origins and maintenance of erroneous social assessments. In D. Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment under Uncertainty: Heuristics and Biases (pp. 129–152). New York: Cambridge University Press.

397

Ross, L., Lepper, M. R., & Hubbard, M. (1975). Perseverance in selfperception and social perception: Biased attributional processes in the debriefing paradigm. Journal of Personality and Social Psychology, 32(5), 880–892. Securities and Exchange Commission (SEC) (2001). Strengthening financial markets: Do investors have the information they need? Washington, DC: SEC. Securities and Exchange Commission (SEC) (2002). Final rule: Acceleration of periodic report filing dates and disclosure concerning website access to reports. Washington, DC: SEC. Shepperd, J. A., & McNulty, J. K. (2002). The affective consequences of expected and unexpected outcomes. Psychological Science, 13(1), 85–88. Tan, H-T., Libby, R., & Hunton, J. E. (2002). Analysts’ reactions to earnings preannouncement strategies. Journal of Accounting Research, 40(1), 223–246. Tan, S-K., & Tan, H-T. (2008). Effects of exposure to subsequently invalidated evidence on judgments of audit workpaper preparers and reviewers. Contemporary Accounting Research, 25(3), 921– 946. Tan, H-T., & Tan, S-K. (2009). Investors’ reactions to management disclosure corrections: Does presentation format matter? Contemporary Accounting Research, 26(2), 605–626. Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131.