The Disciplining Role of Mandatory Disclosure Anna Boisits
Transcription
The Disciplining Role of Mandatory Disclosure Anna Boisits
DISCUSSION PAPER SERIES IN ECONOMICS AND MANAGEMENT The Disciplining Role of Mandatory Disclosure Anna Boisits Discussion Paper No. 12-04 GERMAN ECONOMIC ASSOCIATION OF BUSINESS ADMINISTRATION – GEABA The disciplining role of mandatory disclosure Anna Boisits1 Abstract: The model shows that voluntary disclosure (e.g. earnings forecast) can become informative if a subsequent mandatory disclosure (e.g. earnings announcement) bears information about the truthfulness of the preliminary voluntary disclosure. The sequential reports may differ from one another for two reasons. Firstly, the private information underlying the two reports may change during the respective announcements dates. Secondly, both disclosures may be biased. With regard to the voluntary report, the model shows that bad information is withheld. In case of a voluntary report, the better the private signal, the higher will be the forecasted value and its bias. One result of mandatory disclosure is that managers try to smooth their reports. As a consequence, the bias in mandatory disclosure will be high if the voluntary report which is given in advance is good and the subsequent private accounting information is bad. 1 Center for Accounting Research, University of Graz, Universitätsstraße 15, A-8010. 1 1 Introduction Managers might have incentives for self-serving voluntary disclosure. Though, it is unclear whether these disclosures are credible. Healy and Palepu (2001) mention two mechanisms which enhance credibility, namely third-party intermediaries and financial reporting. The latter makes it possible to verify prior forecasts of earnings or revenues by using actual realizations. However, voluntary disclosure can only be credible if there are adequate penalties for deviations in forecasted and actual numbers. Without costs in the case of biasing private information, the mandatory disclosure of the actual numbers will not have any influence on the truthfulness of the forecasted numbers. The legal system and monitoring activity through the board are important for establishing such penalties. However, there are also other costs which can arise if forecasted and actual numbers deviate from one another, such as litigation risk, psychic costs or reputation costs. Empirical studies have shown that the volatility of the share price on the earnings announcement date is small. Therefore, much of the uncertainty is resolved prior to the actual earnings announcement by management forecast, pre-announcements, or other information sources. This phenomenon is illustrated for example in Beyer et al. (2010). In this study, mandatory disclosure (e.g. earnings announcement and SEC filings) provides only 12 percent of the accounting-based information, whereas voluntary disclosure (management forecast and pre-announcements) amounts to 67 percent. The remaining 21 percent is information which is given by analysts. However, this does not mean that voluntary disclosure makes mandatory disclosure redundant. On one hand, it eliminates the remaining uncertainty, on the other hand, mandatory disclosure has a disciplining role because it confirms the forecasted numbers. For the most part, voluntary disclosure is credible because future mandatory disclosure verifies the previously disclosed information. Voluntarily disclosed information may not fully coincide with future mandatory information for several reasons. Firstly, the private information which underlies these disclosures can change. Secondly, the actual private information may be biased both by the voluntary or mandatory disclosure. What influences both biases? Which bias is higher? The aim of this paper is to examine the interplay between the strategies of voluntary disclosure and subsequent mandatory disclosure, in a situation in which mandatory disclosure is partially capable of evaluating the truth content of the previously disclosed information. One advantage of the joint examination of both disclosure types is that a situation is created in which a manager incorporates future consequences of her voluntary disclosure strategy. Additionally, it is 2 demonstrated that reported values which are given voluntarily have an influence on subsequent mandatory disclosure, because managers try to smooth earnings announcements. The following model is closely related to the model by Einhorn and Ziv (2012) who examine a similar setting in which they illustrate that the voluntary bias function is not constant. However, they ignore the fact that a subsequent mandatory disclosure may contain incomplete information on voluntary disclosure. Yet, this joint examination is important in order to gain insight into the manner in which voluntary and subsequent mandatory disclosures interact with one another. In the present model, the manager receives information concerning the terminal value on two sequential dates, whereby the initial information is noisy information about the second signal. The manager is obliged to disclose her information on the second date (e.g. earnings announcement). However, she can decide whether to disclose a forecast voluntarily (e.g. earnings forecast) on the first date or not. Voluntary disclosure is based on private, noisy information. The private information which underlies a forecast is not verifiable and thus, opens up opportunities for biasing. Since the information is no verifiable, the manager cannot be punished directly for misreporting. The market only observes the forecasted value, not the private signal which underlies it. Mandatory disclosure is also observable, but this report occurs later and thus is based on more precise information. Nevertheless, private signals are correlated and, thus, the difference between the forecasted and the actual mandatory disclosure can be used to assess the credibility of the manager. However, a manager who always reports her actual beliefs can face credibility costs, because the information which underlies these disclosures can change. The model shows that mandatory disclosure restricts the possibility to bias voluntarily, but it does not fully eliminate it. One result is that managers withhold bad information, whereas good private signals are disclosed with a bias. The lowest bias is reached if the information is just good enough to be disclosed. Furthermore, the biasing behavior increases the better the private information is. However, the market discerns the incentive to bias and, thus, corrects for the expected bias. The considered equilibrium is a fully separating equilibrium upon disclosure. In this equilibrium the market can perfectly derive the true signal from the disclosed information. However, reporting the truth or disclosing another information is too expensive, due to the beliefs of the market and the credibility costs which arise because of the difference between mandatory and voluntary disclosure values. 3 The section on comparative statics examines how the mandatory and voluntary disclosure strategies depend on the characteristics of signals, the importance of market prices on disclosure dates for the manager and the credibility and biasing costs. Some results of the model are consistent with empirical findings. The following model, for example, shows that the expected value of the difference between voluntary forecasted value and expected value before the disclosure is positive. It also analyzes the characteristics of the variability of change in stock price on the voluntary and the mandatory announcement date. Most theoretical models of voluntary disclosure are either based on the assumption that voluntary information disclosure must be truthful, or that the information is not verifiable and misreporting is costless, cheap talk. If misreporting bears no costs, any kind of disclosure is ignored when the market expects that the manager favors specific information (“babbling equilibrium”). Some papers deal with credibility in a cheap talk setting. Gigler (1994) argues that a manager can credibly communicate information if there are conflicts between goals concerning the disclosure of specific information. Wagenhofer (2000) examines the disclosure decisions of a firm which wants to sell an operation, while private information can either be verifiable or unverifiable. The following contributes to models which assume that a manager can disclose voluntarily distorted information, whereas misreporting is costly (Korn, 2004, Einhorn and Ziv, 2012 and Beyer and Guttman, 2010). Due to the fact that disclosure is voluntary, the manager has discretion about whether to disclose private information or not. If she opts for disclosure, she can decide whether to bias the information or not. There are also other studies which focus on the interaction between voluntary and mandatory disclosure. Kwon et al. (2009) analyze how the quality of subsequent mandatory accounting reports influences the bias in voluntary disclosure and examine the likelihood of such a report; the higher the quality of mandatory disclosure, the lower the bias in voluntary disclosure. The reason for this is that a higher quality of mandatory disclosure leads to a situation in which the valuation of the firm rather depends on this kind of information than on voluntary information. In this model voluntary information and mandatory information are two different sources which provide information about a firm’s liquidating value. Investors use them in order to evaluate a firm on the announcement date of the mandatory disclosure. In contrast, the following model analyzes the effect of the subsequent mandatory disclosure on the prior voluntary disclosure, whereas the latter is an information source about the former. 4 Einhorn (2005) also focuses on the relation of voluntary and mandatory disclosure. In this model, a manager observes two signals about the firm value before disclosure. The manager has to disclose one signal mandatorily. The author examines the influence of mandatory disclosure, and its underlying parameters on the voluntary disclosure strategy, rather than investigating the credibility of voluntary disclosure. Thus, it is assumed that both disclosures are credible communicable. Bagnoli and Watts (2007) also assume that disclosures must be truthful. In this setting, the private information complements the mandatory disclosure. Thus, the voluntary disclosure of this private information makes mandatory disclosure more informative. In contrast to the afore mentioned approaches, this paper argues that the relationship between mandatory and voluntary disclosure is characterized by the assumption that mandatory disclosure can be used in order to evaluate the accuracy of past voluntary disclosure. Gigler and Hemmer (1998) and Lundholm (2003) also argue that mandatory disclosure has a confirmatory role. Gigler and Hemmer (1998) consider a moral hazard model in which the manager exerts value creating action in several periods, but this action is unobservable. Aside from the moral hazard context, the assumption about the content of voluntary and mandatory information is also different from the following model. It is assumed that private information is superior to the subsequent public signal. The following model considers voluntary disclosure as more timely information, but the private information which underlies it is noisy information about the subsequent mandatory disclosure. They consider an optimal contract in which telling the truth is achieved and the manager has incentives to work hard. Furthermore, they analyze the implication of the frequency of mandated financial reports. An increased frequency leads to less voluntary disclosure and, thus, reduces the informational efficiency. Lundholm (2003) considers the confirmatory role of mandatory disclosure in a repeated game setting. Most of the time, voluntary disclosure will be disclosed truthfully, because there is the possibility of punishment if voluntary information differs from the mandatory one. He uses the strong assumption that the private information coincides with the ex-post mandatory disclosed information. None of these theories consider the interplay between the voluntary reporting strategy and the subsequent mandatory disclosure strategy which is the focus of the following model. It proceeds as follows: first of all, the model will be presented. Secondly, the characteristics of the equilibrium will be illustrated. In addition, a description of the effects of various parameters on 5 disclosure strategies will be provided. Section 4 will analyze the relation between the model and empirical research, before final section will end with a conclusion of the main ideas. 1. The model The aim of this paper is to examine the influence which mandatory disclosure (e.g. earnings announcement) has on voluntary disclosure (e.g. management forecast) which is provided in advance. Voluntary disclosure can become informative if it is verified by a subsequent disclosure and if any kind of deviation is punished. Figure 1 depicts the sequence of events. The firm’s uncertain terminal value is a random variable , which is normally distributed with mean and variance (~, ). This value is realized on date three. On date two, the accounting system reports a noisy signal about the future value , = + ̃, where the noise term ̃ is normally distributed with mean zero and variance . This is the manager’s private information. However, she is obliged by a regulator to issue a report about this information. Yet, by engaging in earnings management, the manager can disclose a value which differs from her private signal. Thus, the market receives an information , which is a biased information about the manager’s private information, = + , while denotes the earnings management. On the first date, the manager receives a noisy signal concerning the subsequent accounting information, ̃ = + ̃ + , where is normally distributed with mean zero and variance . She can decide whether to disclose a voluntary report after receiving the information or not. Because this information is not verifiable the manager can provide any information she wants. The distribution of the two signals and the terminal value is common knowledge. The realized signals are only known to the manager. Insert Fig. 1 about here Both signals provide information about the same terminal value. However, they differ in precision and appearance. It is noteworthy that is a sufficient statistic of ̃ . This setting captures the fact, that the manager will have a conjecture about the terminal value prior to the date of the 6 mandatory disclosure. However, on the date of a given forecast, the private information differs from the private information on the date of the mandatory report. Thus, there are two reasons why voluntarily and mandatorily given reports can vary. Firstly, the private information which underlies the two reports may have changed between the two dates, because manager’s signal is only a noisy estimator of the later value. Secondly, management may bias actual private information on both voluntary and mandatory disclosure. The manager’s goal is to maximize her utility during these two periods, which is given by: (1) = + − − − . The utility function is common knowledge. As can be seen in the equation (1), the manager is interested in the magnitude of the firm’s market prices, and . ( ) is the weight on the market price during the first (second) period. Both weights are assumed to be positive, as a higher price is favorable for both periods. The manager may be interested in high market prices because she wants to sell shares which she owns. Another reason may be that her wage depends on the market price. The manager has two kinds of disutilities. The first one consist of costs which arise from manipulating the accounting earnings for the mandatory report. These costs could be the effort needed in order to conduct the manipulation, psychic costs, negotiation costs with the auditor or litigation costs. The convexity covers the fact that it is increasingly hard to manipulate accounting information. The weight of this cost equals . The second type of disutility are credibility costs which arise ex post if the voluntary information which is given in advance differs from the subsequent mandatory report. If, for example, the manager reports months before the actual earnings announcement that the firm will make a tremendous profit but discloses a loss on the mandatory announcement date, she will lose credibility. Although this difference can result from a change in the information endowment, it is more likely that biasing behavior plays a role if discrepancies are high. The convexity illustrates that the higher the difference the more likely it will be that the manager manipulates her information which leads to an even stronger decrease in credibility. A high discrepancy between the two reported values can create psychic costs, because the manager wants to fulfill the forecasts. If the voluntary report and the subsequent mandatory report strongly differ from one another, the manager could be considered as a fraud. This will affect her reputation as well as the occurrence and the condition of future transactions. As a consequence manager’s utility will be lower if the discrepancy is high. Penalties for deviations in forecasted and actual numbers can also be enforced by the board of directors. It is noteworthy 7 that credibility costs are endogenous. In period two, the manager has already chosen the voluntary report . Thus, by choosing , she has an influence on these specific of costs, because = + . The assumption of credibility costs is important, because without these costs, in the case of diverging numbers, the mandatory disclosure of the actual numbers would never have any kind of influence on the truthfulness of the forecasted numbers. If the manager decides against a voluntary report, credibility costs do not arise. 2 The Equilibrium On the first date, the manager receives the signal . She is not obliged to report this information and the information is not verifiable by other market participants. Therefore, the manager can either disclose any value ℝ or nothing (Ø). Thus, the set of disclosure alternatives (" ∈ $) is given by: $ = %Ø& ∪ ℝ. The disclosure strategy is a function: (: ℝ → $. The market expectation + : ℝ → $. The market observes the voluntary disclosure of the disclosure strategy is depicted by: ( and evaluates the firm given the expected disclosure strategy: : $ → ℝ. If there is no disclosure, the market price will be: Ø ∈ ℝ. If voluntary disclosure occurs the market price will depend on the reported value and on the expected disclosure strategy: . , : $ → ℝ depicts the expected pricing strategy. By assumption, the market is risk-neutral and rational. Thus, the firm is + = "0. evaluated with the expected value conditional on the information available: = -.|( On the second date, the accounting system provides the information and, by regulation, the manager has to disclose her information. However, the reported value can differ from the private signal, due to the fact that the manager can conduct earnings management. Thus, the disclosure +: ℝ → strategy is a function: 1: ℝ → ℝ. The market beliefs about the strategy are depicted by: 1 ℝ. The market observes the disclosure . Based on this information and on the expected strategy, the market assesses the firm with :ℝ → ℝ. The manager has expectations about the pricing behavior, , :ℝ → ℝ. In equilibrium, actual strategies coincide with the expected ones, so that: = , , = , , + and 1 = 1 +. (=( 8 2.1 The optimal mandatory disclosure strategy On the first date, the manager will choose a report which maximizes her expected utility, based on his private signal . However, although she does not know the signal at this point, the subsequent mandatory disclosure influences the bias of the voluntary disclosure. Thus, in order to find the optimal voluntary disclosure strategy, the optimal mandatory disclosure strategy has to be considered first. The equilibrium which is used for this second period is one with linear strategies. The reason for using this equilibrium is that it is Pareto optimal.2 The price will be linear in the report : = 3 + 4 and the bias in the report will be linear in signal : = 5 + 6. On the second date, the manager has already chosen either report or a non-disclosure. In addition, the price is already realized as well. Thus, the manager takes these values as given. The reporting strategy in the second period depends on whether she has disclosed a value in the first period or not. Concerning the report , the manager chooses the bias, , to maximize the following utility function: max: = + − − − , (2) while the expected pricing function equals: , = 3; + 4< and the mandatory report is depicted by: = + . Thus, the optimal bias is given by: =∗ = + ?@ A BC DB@ + BC B =EF . C DB@ The market assesses the firm with the expected value given the expected strategy: = 5< + 6<: = -G|H = + + NOEP +N IJ@ KLEKDM + NKIJ@ DIQ@ N . KDM In equilibrium, the actual strategies should coincide with the expected ones: 5< = 5, 6< = 6, 4< = 4 and 3; = 3. Thus, the equilibrium strategies equal: =∗ = (3) BC =EF BC DB@ Q = 3 + 4 = RKI@DI @N − I@ J Q + ?@ IJ@ B@ KIJ@ DIQ@ N IJS ?@ BC DB@ @ KIJ@ DIQ@ N B@@ and J C − KI@DI @ NB T + I@B = J Q @ IJ@ BC DB@ L KIJ@ DIQ@ NB@ . If there is no report (Ø), the manager aims to maximize the following utility function: max: = + − . (4) Based on the linear equilibrium structure, the optimal strategies are: (5) 2 Ø∗ = B ?@ IJ@ @ @ @ KIJ DIQ N Q and = 3 + 4 = RKI@ DI @N − I@ J Q IJS ?@ @ KIJ@ DIQ@ N B@ T + KI@JDI@ N. I@L J Q The proof of this statement is given in the appendix. 9 As can be seen in equation (3) and (5), the market is able to perfectly infer the signal from report . In the case of a voluntary report, the mandatory report increases in the private information and, thus, the equilibrium is separable. Therefore, it is not possible to mislead the market in equilibrium. However, the manager biases her information in equilibrium, whereas the market can infer the true signal. Therefore, on date one, the market price will actually be the expected value of conditioned on the expected signal which corresponds to the realized signal in the equilibrium ( = + KIJ @ I@ FEO @ J DIQ N , this is obtained by inserting the optimal strategy of the manager into the optimal pricing function). However, the manager’s biasing strategies differ depending on the reporting strategy in the first period and in the case of a voluntary report in the first period it depends also on the received signal in period two. If there is no voluntary disclosure in the first period, there will be no credibility costs and, thus, the optimal bias will be constant. Hence, the bias does not depend on signal (see (5)). This result is similar to the result of Fischer and Verrecchia (2000). However, in the case of a voluntary report in the first period, the manager also cares about credibility costs. Thus, the bias in report will also depend on the voluntary reported value and the received information . The bias in the situation of a report in the first period will be higher (lower), compared to the bias in the case of no disclosure, if < ( > ). Both optimal biasing functions share the same fix part. This comes from the influence of the bias on the market price. In the case of a voluntary report, the fix part is adjusted by minimize the sum of the credibility and the biasing costs. BC =EF BC DB@ in order to 2.2 The optimal voluntary disclosure strategy On the first date, the manager receives the signal . Based on this information, she chooses her optimal voluntary disclosure strategy. She can either report any kind of value W or nothing (Ø). Her goal is to maximize the expected utility which is either expressed by the expected value of (2) or (4), taking into account the optimal strategies of the second period, which are given by (3) and (5). If she opts for a voluntary report, she will choose a report which maximizes her expected utility: (6) max= -G= |H = , + - G |H − - G − |H − - G |H. In the case of no disclosure, her utility would be: (7) -GØ |H = , Ø + -G |H − -G |H. 10 The price in the second period depends on the future mandatory disclosure and indirectly on the private information . Given the equilibrium strategy of the market in the second period, it is optimal to bias the information according to the bias function given in (3) or (5). In both cases, this leads to a price which is equal to = + KIJ @ price is equal to - G |H = + IJ@ XGF|YHEO KIJ@ DIQ@ N I@ FEO @ J DIQ N in equilibrium. The expected value of this . This price and its expected value do not depend on the strategy in the first period. If is known, the knowledge of is redundant, because the latter is just noisy information about the former. For this reason, if the market receives the mandatory report it can infer the private information and, thus, the market will not consider the voluntary report or the information derived from the non-disclosure in evaluating the firm in the second period. The expected costs in the case of a report are given by: (8) -G − |H + - G |H = B @DBC − - G|H + B B B while \]| = @ @ KIJ DIQ@ N I^ @ IJ@ DIQ@ DI^ C @ and 4 = KIJ@ I@ BC DB@ @ J DIQ NB@ C DB@ Z ?@ A [ + B @DBC \]|, B B C @ . These costs are minimal if = - G|H, which means that exactly disclosing the expected value of the accounting information , given the information leads to the lowest disutility from reporting. Thus, if the manager’s goal would be to minimize the total expected costs, she would issue a forecast about the future mandatory report which is equivalent to the expected value of the accounting information given her signal . This disclosure behavior is reasonable, because the conditional expected value is the best estimator for the future accounting information . If the manager decides not to issue a report on the first date, the disutility comes only from biasing earnings in the second period. Given the strategy depicted in (5) the expected costs are given by: (9) -.Ø _0 = R ?@ IJ@ T . B@ KIJ@ DIQ@ N As can be seen in (8) and (9), the expected costs in the case of providing a voluntary report are higher, independent of the reporting strategy. Even if the manager discloses her actual expectations about future information, the disutility is higher than if nothing is disclosed in period 11 one. The difference between the costs in (8) for = - G|H and the costs in (9) can be interpreted as the costs of voluntary disclosure3: 5 = R (10) ?@ IJ@ @ NT + B@ KIJ@ DIQ B@ BC BC DB@ \]|. Due to the fact that the manager wants to maximize her utility, she will decide to provide information voluntarily if the utility in (6) is higher than the utility under non-disclosure (7). She will be indifferent if the utility is the same and she will chose no disclosure if - GØ H > max= -G= |H. Lemma 1: There exists a unique threshold value . If the private signal lies above (below) this threshold value, > ( < ), then the manager will decide to voluntarily disclose (withhold) information.4 Bad signals are too expensive to disclose. Verrecchia (1983) and Dye (1985) have obtained a similar result. Withholding information leads to a higher utility for the manager if the signal is bad: s < s. The reason for this is that the expected disutility from issuing a voluntary report is higher than the expected costs under no disclosure, independent of the reporting strategy. 2.3 Truthful reporting The following section illustrates that truthful reporting cannot be an equilibrium. A truthful disclosure would occur if the manager discloses the actual expected future accounting information conditional on her information: = - G|H. This reporting strategy would also minimize the expected costs from reporting in period one. Assuming that investors believe that + = = - G|̂ H) if ̂ > ̂ , the pricing function would the manager reports the actual forecast (( be: = + KIJ@ DI@ N. The optimal strategy for the manager given these beliefs would be: I@ =EO J (11) r ∗ = argmax= R + KIJ@ e 3 4 Q B@ BC BC DB@ I@ =EO T + - G |H − @ J DIQ N − - G|H + B C DB@ Z ?@ A C J C @ [ + B @DBC \]|f = - G|H + I @ DI @ B B . B B C @ ? I@ B DB J Q @ C See Verrecchia (1983). The proof is given in the Appendix. 12 In this case, the actual reporting strategy given the beliefs of the market (r ∗ ) does not correspond + = = - G|̂ H). Thus, reporting the actual forecast is not an with the expected strategy (( equilibrium. 2.4 Biased reporting As demonstrated in the previous section, the manager has an incentive to overstate her actual expectation ( ≥ - G|H). In equilibrium, the market will assess the firm with a value which is lower than the value assumed in the previous section, because the market corrects for the expected bias, < + KIJ@ I@ =EO @ J DIQ N . This lower value reduces the profitability of biasing the forecast and as a result the bias will be lower in equilibrium. In equilibrium, the market is able to correct the report exactly by the actual bias. The voluntary disclosure equilibrium considered in this paper is one in which a full separation upon disclosure occurs. This is the case if two signals do not lead to the same reported value, except the signals which lead to withholding of information. As a result of these separable strategies the market is capable of deducing the private information from the disclosure if the manager provides information. Despite the fact that voluntary and mandatory disclosure is not verifiable, the market is able to infer information from reports. The following proposition summarizes the main characteristics of this equilibrium. Proposition 1: There is an equilibrium which is characterized by a disclosure threshold level: . All private information lower than this threshold level leads to non-disclosure, ( = h for all ≤ . All signals larger than this threshold induce a voluntary report, ( = for all ≥ . The reporting function is strictly monotonically increasing, j Y > 0. If the manager possesses the threshold signal , she is indifferent between a non- disclosure or a voluntary disclosure in which exactly the beliefs about future accounting information is disclosed: KN = -.|0. The bias in excess of the actual, expected value − - G|H increases in and converges to ?C IJ@ BC DB@ 5 . IJ@ DIQ@ B@ BC There is an infinite number of separating voluntary disclosure equilibria in this model setting because there are infinitely many expectations about the bias at the threshold signal. According 5 The proofs are provided in the Appendix. 13 to the D1 criterion by Cho and Kreps (1987) all equilibria which have a positive bias at the threshold value can be eliminated. One single separating equilibrium survives this criterion. This equilibrium is characterized by disclosing the actual expectation about future mandatory information if the manager possesses the threshold information.6 Thus, there will be no bias at the threshold level which exceeds the expected value of the accounting information conditional on that signal. The equilibrium reporting function is an increasing function. Therefore, investors can infer the private information which the manager possesses from this disclosure, in case of a voluntary report. Although the manager biases her information, the market can deduce the true private information. In equilibrium, the expected signal ̂ corresponds to the realized signal . Therefore, on the first date, the market price will be the actual expected value of the terminal value , conditioned on the signal , although is not directly observable. Despite the fact that voluntary information is not verifiable, the market can deduce the information which underlies it from the report. Thus, voluntary disclosure leads to price reactions, although the information underlying the voluntary disclosure is not verifiable ex ante and ex post. There exists only one optimal report for every signal. The reason for that lies in the expected credibility costs and the beliefs of the market. There is a trade-off, because of the credibility and the biasing costs. If a manager with a low signal decides to disclose a high report , there is a great risk that is also low, because and correlate with one another. In order to avoid the credibility cost, the manager can bias the future accounting information but at high manipulation cost. These ex post appearing costs lower the incentive to extremely overstate the actual beliefs about future accounting information. In equilibrium, the market is able to correct for the actual bias. Thus, the manager cannot mislead investors. Nevertheless, the manager will bias her information. The reason for this is that choosing a lower report which is closer to the expected value of the accounting information, would lower the expected credibility cost. However, the market would believe that the voluntary report was given by a manger with a lower signal. Therefore, the market evaluates the firm with a lower price. The reduction in price would be higher than the positive effect of lower credibility costs. Therefore, such a report would not be optimal. The equilibrium voluntary reporting function is given by: 6 This refinement method is also used by Einhorn and Ziv (2012). 14 = - G|H + (12) while - G|H = + ?C IJ@ BC DB@ IJ@ DIQ@ B@ BC IJ@ DIQ@ YEO @ IJ@ DIQ@ DI^ l1 + m n−o E @ @ @Kp@ J qpQ N r@ rC KstsN E @ @ @ uC p@ J ZpJ qpQ qp^ [rC qr@ vw, and m GH is the Lambert W function, with mGH ∈ G−1,0. Insert Fig. 2 about here Figure 2 depicts the voluntary reporting function in equilibrium. The manager will never issue a forecast if the signal is low. In this case, the disutility of a voluntary report is too high, thus reporting is not optimal. Regarding good information ( ≥ ), the manager has an incentive to mislead the market in order to convince the market that the firm value is higher than the actual expected firm value based on private information. It is not possible to be misled in equilibrium. The reporting function for all ≥ lies between - G|H and - G|H + ?C IJ@ BC DB@ IJ@ DIQ@ B@ BC . The report at the threshold value equals the expected value of the accounting information. Proposition 1 asserts that the reporting function is a strictly increasing function in the signal which is the case because: x= xY ? I@ B DB C J C @ = =EXGF|YHKI @ DI@ DI @ NB J Q ^ @ BC > 0 for > - G|H. In addition, for a higher signal , the C J C @ reporting function is closer to - G|H + I @ DI @ B B . Thus, the deviation of the reported ? I@ B DB J Q @ C value from the expected value of the accounting information is an increasing function in . 3 Comparative Statics This section will focus on the consequences of changes in parameter on the disclosure strategies. 3.1 The mandatory disclosure strategy The optimal bias of the mandatory disclosure strategy is given by equation (3), in case of a report in the first period. As can be seen in this equation if the voluntary report is high whereas the accounting information on date two is bad (low ), the earnings management will be high in the 15 second period. In order to lower the credibility costs, higher manipulation costs are accepted. The higher the voluntary report, the higher will be the bias ( y:z∗ y= > 0). Hence, the manager tries to y:z∗ smooth her reports. Conversely, a better signal in the second period leads to a lower bias ( 0). yF > What are the effects on the optimal bias if the manager cares more about the credibility costs? A higher weight on credibility costs means a higher . Differentiating the optimal bias, in equation (3), with respect to shows that the optimal bias in mandatory disclosure increases if > and decreases otherwise ( y:z∗ yBC = B@ BC DB@ − ). The optimal bias consists of two parts, the fix part which depends neither on nor on , ?@ IJ@ B@ KIJ@ DIQ@ N , and the part BC =EF . BC DB@ The first part emerges due to the positive effect which a high report has on the market price. The second part guarantees that the influence of the optimal bias on both credible and biasing costs is optimally apportioned in order to minimize the manager’s disutility from these costs. As increases the fix part remains the same. However, an increase of enlarges the importance of the credibility costs compared to the bias costs. In order to reduce credibility costs, the manager’s willingness for a higher bias increases if > . Conversely, if the weight on the costs for manipulating accounting earnings increases it is optimal to reduce these costs by decreasing the bias. Differentiating the optimal bias with respect to shows that the optimal bias decreases ( otherwise. y:z∗ yB@ < 0) if =∗ > − ?@ IJ@ BC @ B@ KIJ@ DIQ@ N , whereas it increases y:z∗ y?@ The higher the weight on the price in the second period, the higher will be the bias ( > 0). For a higher weight, the incentive to manipulate the second price is higher and thus there will be a higher bias in equilibrium. The weight on the price in the first period, , does not influence the optimal bias in the second period. During the second period, the manager cannot influence the past price, . If the variance of the noise term increases, the bias will be lower (yIz@ < 0). A higher variance y:∗ Q means that the signal is worse which leads to a lower price reaction to the mandatory report . This lowers the incentive to bias information. On the other hand, a higher variance of the terminal 16 y:z∗ yIJ@ value increases the bias ( > 0), because a high risk of the terminal value makes the report more important. Thus, the reaction to the report increases and hence also the incentive to bias. 3.2 The voluntary reporting strategy As shown above a threshold level exists in which all signals higher than the specific value result in voluntary disclosure, whereas bad signals lead to a withholding of information. Thus, the probability of a voluntary disclosure is given by 1 − {KN, where { is the cumulative distribution function of signal (̃ ~K, + + N). If increases the probability of a voluntary report decreases. The threshold value depends on the parameters. For example the higher the weight on the credibility costs, the higher will be the threshold value. Thus, voluntary disclosure is less likely. The reason for this is that if is increases the disutility of disclosing voluntarily will be higher and, thus, reporting is worthwhile only for managers with good signals. An increase of has a direct and an indirect effect on the costs. On one side, an increase in directly increases the expected costs of biasing accounting information in the second period. In case of a voluntary report the expected bias is higher than that of withholding information in the first period. This direct effect increases the difference between these costs and makes voluntary report less attractive. On the other side, a higher induces the manager to bias less because it is more expensive. This indirect effect works in favor of voluntary reporting. If is small, the direct effect dominates. Thus, the threshold value increases if gets higher.7 However if is high, decreases, which makes a voluntary disclosure more likely. Voluntary disclosure is also less frequent if the weight on the price in the second period gets higher. A higher weight on the second price causes a stronger incentive to bias the accounting information in the second period. Because the disutility of costs is higher in case of a voluntary report, the difference between the costs of a voluntary report and the costs of withholding will be stronger. Thus, a higher leads to a lower probability of a voluntary report. A higher weight on the price in the first period increases the incentive to report voluntarily. The price in the first period is compared to the cost more important for the manager and, thus, the threshold value decreases and voluntary disclosure becomes more likely. Voluntary disclosure will be less frequent if the signal in period one is less precise (higher variance). On the one hand, a higher uncertainty increases the expected disutility in the case of a voluntary report compared to the 7 The threshold level for θ is given in the Appendix. 17 situation of providing no information. This leads to a higher threshold value. On the other hand, a lower precision induces that the price reaction in period one is lower, whereas this reduction is higher than the price reduction if there is no information. This also reduces the incentive to disclose voluntarily. Thus, a lower precision of the signal in period one reduces the probability of a voluntary disclosure. The more the two signals correlate (which is ceteris paribus the case if is lower) the more likely it will be that the manager discloses voluntarily. The second effect also applies if increases. In this case, the precision of the accounting signal in period two decreases. Because of this, the precision of the signal in period one, which is a signal of this accounting information, decreases as well. The effect on the cost is not that unambiguous. On the one hand, a lower precision of the accounting signal decreases the price reaction of the market and so also the incentive to bias which would lead to lower expected costs and a lower threshold value. However, a lower precision increases the expected uncertainty costs because it increases the conditional variance of the signal given . The voluntary disclosure strategy is given by equation (12). The better the signal , the higher will be the bias in excess of the actual expected accounting information conditioned on the signal (}∗ = − - G|H). This bias approaches ?C IJ@ BC DB@ IJ@ DIQ@ B@ BC because the term in brackets in equation (12) approaches 1 if the signal approaches infinity. The bias in the limit will be higher the higher the weight on the price in the first period or the lower the precision of the terminal value. A lower precision of the accounting information has the exact opposite effect. A higher weight on the credibility cost (a higher ) decreases the limit value. Likewise, if the weight on the bias costs increases the limit is lower. Due to these changes in the limit value, the effect on the bias of high signal values is obvious. However, changes in parameter also influence the curvature of the bias function and the threshold value. Thus, a change in the bias function is ambiguous for signal values slightly above the threshold value. In equilibrium, both the voluntary and the mandatory report are biased. However, which of these two reports is biased to a larger extent? When comparing the ex-ante expected value of the bias of the voluntary report to that of the mandatory report, it is detectible that the expected bias in the voluntary ?@ B@ < ?C BC report l1 + - nm n−o is higher if @ @ @Kp@ J qpQ N r@ rC KstsN E @ @ @ uC pJ ZpJ qp@ Q qp^ [rC qr@ E the following inequality holds: v~ ≥ vw. The term -.m G∙H| ≥ 0 in brackets 18 on the left side is negative and higher than minus one (-.m G∙H| ≥ 0−1,0). Thus, both sides are positive. It can be shown that the difference between the expected voluntary bias and the expected bias in the mandatory disclosure decreases in .8 The manner in which other variables influence this difference is ambiguous. 4 Empirical implications and evidence In line with the presented model also empirical research displays that forecasts induce price reactions. Waymire (1984), for example, analyzes price reactions in consequence of management forecasts. The empirical study shows that forecast deviation is generally positive and that there is a positive relationship between forecast deviation and cumulative abnormal return around the announcement date. Forecast deviation is defined as the normalized difference between the forecasted value from the manager and the analyst forecast prevailing prior to the disclosure of the management. Because a disclosure induces a stock price reaction, the capital market derives information from this voluntary disclosure. Thus, management forecast must be informative. The forecast deviation equals − in the model, in which is the voluntarily given information and stands for prior expectation. The expected value of the forecast deviation is positive (-. − | ≥ 0 > 0) which is consistent with the empirical finding. However, this is not surprising. Firstly, only managers with good information, ≥ , disclose. Secondly, managers bias their actual expectation upwards. The model predicts in common with the empirical study a positive relationship between forecast deviation and change in market price ( − ) on the announcement date. Waymire (1984) computes the correlation between the two variables. However, the relation depicted in the model is not linear. Computing the correlation would suppress the actual relationship. In line with McNichols (1989), the model shows that stock price reactions contain information beyond that of the management report. The model shows that the market is able to correct for the strategic behavior in forecasting. If a voluntary disclosure is observed the market corrects the disclosed value for the expected bias. Empirical studies have shown that the volatility of the share price on the earnings announcement date is small. Thus, most of the information is resolved prior to the actual earnings announcement 8 Proofs are given in the Appendix. 19 by management forecast, for example. Ball and Shivakumar (2008) find empirical evidence that management forecasts provide substantial information. The information is discretionary (which means that it is provided when manager prefers to reduce information asymmetry) and forwardlooking, whereas earnings announcement is not discretionary and backward-looking. The empirical study shows that the “surprise content” of earnings announcements is substantially lower than that of management forecasts. Thus, earnings announcement might not have the role of providing timely, new information. However, this corporate information is important in order to curb management forecasts. One important role of earnings announcement is to discipline prior information. The confirmation role of backward-looking financial outcomes on private forwardlooking information is also examined in Ball et al. (2012). They test the hypothesis that audited financial reporting and previous given voluntary disclosures are complements. They show that market reaction to forecasts increases in the resources committed to audit. Hence, financial statement verification augments the credibility of management forecasts. The present model is able to show under which circumstances price reactions which are caused by voluntary disclosure are higher than reactions on the mandatory disclosure date. For this comparison only firms which disclose voluntarily are considered, ≥ . The expected price reaction for those firms is positive (-. − | ≥ 0 > 0) because only manager with good signal report voluntarily. The volatility of this price reaction is given by: \]. − | ≥ 0 = IJS @ e1 + IJ@ DIQ@ DI^ KYNYEO EKYN − @ @ N KYN KIJ@ DIQ@ DI^ ZEKYN[ @ f. The term in brackets lies between zero and one because of Sampford’s law (1953). Lemma 2: The variance \]. − | ≥ 0 decreases in and in .9 The higher the threshold value the lower will be the variance. The conditional variance is lower if a smaller amount of signals, , leads to a report. The previous section shows that is higher the higher , the higher , the lower and the higher . An increase in has two effects on the conditional variance. On the one hand, it increases the threshold value, with the result that fewer signals lead to a voluntary report. This reduces the conditional variance. On the other hand, a higher makes the voluntary report less informative which leads to a lower market reaction ( is lower). This effect lowers the conditional variance as well. 9 The proof is provided in the Appendix. 20 If there was a disclosure in the first period, is known and already inherent in the price. The expected price reaction is zero (- G − |H = 0) because all firms which have disclosed voluntarily are obliged to provide a subsequent mandatory report. The volatility of the price ^ reaction on the mandatory announcement date is given by: \]G − |H = I@ DIJ@ DI@ I@ DI @. J IS Q ^ Lemma 3: The variance \]G − |H decreases in and increases in and . I@ J Q The higher the poorer is as a signal for . Thus, the variability on the mandatory announcement date will be higher for high because contains more new information. The main effect which lead to a decreasing conditional variance, \]G − |H, in the case of an increasing variance of the signal , , is the lower weight on the signal compared to the Q J weight on the mean, , in the firm price = I@ DI @ + I@ DI @ . The opposite is true for a higher variance of the terminal value, . I@ J Q I@ J Q If is a good signal about the future accounting information (which is the case if the two signals highly correlate with one another, low ) then the price reaction on the date of voluntary disclosure will on average vary more than on the day of the mandatory disclosure. The intermediate value theorem proves that there is a threshold value for : 0 < < + . If is small, < , then the variance on the voluntary disclosure date, \]. − | ≥ 0, is higher than on the mandatory disclosure date, \]G − |H. If, for example, the date of the forecast is close to that of the mandatory disclosure then there will be a high correlation between the two private signals. Therefore, the price reaction will be high on the voluntary disclosure date whereas it will be low on the mandatory announcement date. 5 Conclusion This paper shows that voluntary disclosure can become informative if a subsequent mandatory disclosure bears information about the truthfulness of the preliminary voluntary disclosure. The voluntary report and the subsequent mandatory report may deviate because of two reasons. Firstly, the private information which underlies these two reports may change during the two announcement dates. Secondly, management may bias actual private information. Both voluntary and mandatory disclosure can be biased. Although neither voluntary nor mandatory disclosure is verifiable, the biasing and credibility costs create a situation in which the market can 21 deduce information from the reports. In equilibrium, the manager will provide a forecast if her private information is above a threshold level. Thus, bad information is withheld because it is too costly to disclose. Due to the non-verifiability of the report, the manager has an incentive to overstate her actual expectations in order to affect the market price. However, manipulating the voluntary forecast so that it exceeds the actual expectation has an effect on future credibility and indirectly on biasing costs. To avoid credibility costs the manager can bias the future accounting information but at the risk of high manipulation costs. Given this trade-off of costs and the beliefs about the pricing behavior, there is only one optimal voluntary report for every signal. In equilibrium the market is able to correct for the actual biasing strategy and, thus, the market price will be the expected value of the firm given the private information. The model shows that both the mandatory as well as the voluntary report will be biased. The magnitude of the biases depends on the uncertainty of the terminal value, the precisions of the signals and the realized signal values. In addition, the frequency of the occurrence of voluntary disclosure also depends on the aforementioned variables. For example, the more the two signals correlate (which is the case if is lower) the more likely it is that the manager discloses voluntarily. The bias in the mandatory report also depends on voluntarily disclosed information if the manager provides a voluntary report in the first period because, in this case, the manager is trying to smooth the announced earnings values. 6 Appendix Mandatory disclosure strategy The mandatory disclosure equilibrium considered in this paper is separable.10 In such an equilibrium managers with different information provide different reports. Therefore, a manager with the signal should not have a higher utility if she would imitate a + Δ manager (with Δ > 0) by reporting + Δ instead of disclosing : (13) , ≥ + Δ, . The price in the first period is not dependent on the message because it is the past market price. Likewise, the strategy of the voluntary report is already selected and, thus, is taken as 10 The approach is similar to the approach of finding the voluntary disclosure strategy which is discussed more precisely in the next section. 22 given. In equilibrium, the price in the second period is given by KN = + IJ@ FEO IJ@ DIQ@ , if the market observes , because investors believe that a message is sent by a manager with information . If, however, a report of + Δ is observed, the market price would be K + ΔN = + . The costs are given by: − + if the manager IJ@ FDFEO IJ@ DIQ@ opts for a voluntary report in the first period whereas they are: if the manager decides to withhold information in the first period. In the following, the existence of a preceding voluntary report is assumed. The approach of finding the optimal mandatory reporting function when information is withheld in the first period is similar. The minimal costs are realized if the manager reports: = B BC C DB@ +B B@ C DB@ . However, this is not an equilibrium strategy, because the manager has an incentive to overstate the true accounting information. Thus, in equilibrium, > BC BC DB@ + B@ BC DB@ . By inserting both the price in the second period and the cost function into the aforementioned inequality (13) and rearranging terms, the following inequality can be obtained: (14) + ≥ B B B B ? I@ F − @ − C + @ @@ J e− − @ − C f≡Θ − KIJ DIQ NBC DB@ F ≤ F BC DB@ BC DB@ BC DB@ BC DB@ L The second solution area ( F ≤ Θ) can be neglected because for positive ∆, ∆ has to be L positive as well. According to the rule of l’Hospital, the limit when Δ goes to zero of the right side of the inequality in (14) equals lim∆F→ Θ = ?@ IJ@ KIJ@ DIQ@ NBC DB@ ZLE rC r =E @ F[ rC qr@ rC qr@ . Furthermore, in order to be a separating equilibrium, a manager with the information + Δ must have a higher utility by disclosing + Δ than by imitating a -type and disclosing . Analog to the approach above it can be shown that this is the case if L F lies between two values while one of these is negative and, thus, can be neglected. The limit of the positive value when Δ goes to zero is equal to the limit calculated above. Thus, the following inequalities must hold: ?@ IJ@ r r @ @ KIJ DIQ NBC DB@ ZLE C =E @ F[ rC qr@ rC qr@ ≤ equation equals: (15) xL xF = xL xF ≤ ?@ IJ@ r r @ @ KIJ DIQ NBC DB@ ZLE C =E @ F[ rC qr@ rC qr@ ?@ IJ@ r r @ @ KIJ DIQ NBC DB@ ZLE C =E @ F[ rC qr@ rC qr@ . Hence, the differential . 23 Solving the equation leads to a mandatory reporting function of: (16) = BC BC DB@ if +B BC BC DB@ B@ C DB@ + B@ BC DB@ + ?@ IJ@ KIJ@ DIQ@ NB@ + KI@@DIJ@ NB > > B ? I@ J Q 1 + m BC C DB@ @ +B EKIJ@ DIQ@ NBC DB@ B@ C DB@ ?@ IJ@ o @ @ t@Kp@ J qpQ Nr@ FD u@ p@ J rC qr@ , , where m G∙H is the Lambert W function11 which lies between zero and -1 since the argument is negative. And for + B@ BC DB@ ?@ IJ@ KIJ@ DIQ@ NB@ (17) = BC BC DB@ < the mandatory message is given by: + B@ BC DB@ + ?@ IJ@ KIJ@ DIQ@ NB@ 1 + m KIJ@ DIQ@ NBC DB@ ?@ IJ@ o @ @ t@Kp@ J qpQ Nr@ FD r u@ p@ qr C @ J BC BC DB@ + . The third solution of the differential equation is: = B BC C DB@ (18) +B B@ C DB@ + KI@@DIJ@ NB . ? I@ J Q @ The reporting function in (16) can be ruled out, because there exists no constant of integration so that B BC C DB@ +B B@ C DB@ + KI@@DIJ@ NB > > B ? I@ J Q @ BC C DB@ +B B@ C DB@ for all . For low the function is not defined. This is not the case for equation (17), there infinitely many equilibria exist because there is an infinite number of beliefs about the “constant of integration”. In addition, the reporting function in (18) is a potential equilibrium strategy. Because each of these equilibria are separating, the market is able to deduce the actual signal underlying the report. Thus, the market is indifferent towards the different equilibria. However, this is not true for the manager, since the bias of all equilibrium disclosure strategies given in (17) (independent of the value of the constant of integration) is higher than the bias given if (18) would be the mandatory equilibrium strategy. Hence, the Pareto optimal equilibrium is the one where the disclosure strategy is given by (18), which is a linear strategy in the signal . Voluntary reporting function The equilibrium is separable upon disclosure if a manager who possesses the signal does not have a higher expected utility if she imitates a + Δ manager (with Δ > 0) by reporting + Δ: The Lambert W function is the inverse function of = m = mo . Important characteristics of this relation are that m ≥ −1 and that m < 0 for < 0, m > 0 for > 0 and m0 = 0. 11 24 -G=Y _H ≥ -G=YDY _H. (19) In a separating equilibrium, the market assesses a firm with = + IJ@ YEO @ IJ@ DIQ@ DI^ if a report is observed. If the market observes a report of + Δ the firm is valued with: = + IJ@ YDYEO @ IJ@ DIQ@ DI^ in the first period. The expected disutility is given by equation (8). The last two terms of equation (8) are independent of the reporting strategy only the first term differs, which equals B@ BC BC DB@ − - G|H if the -type discloses and B@ BC BC DB@ + ∆ − - G|H if the -type discloses + Δ. The change in the report is the difference between the report of a manager with the signal + Δ and the report from a -type manager (Δ = + Δ − ). By inserting the terms into inequality (19), the following is obtained through rearangment: Δ + 2. − - G|H0Δ − @C KI B DB@ ?C IJ@ Y @ @ J DIQ DI^ NB@ BC (20) ≥0 This inequality holds if: (21) Δ ≥ + B DB ? IJ@ Y − . − - G|H0 . − - G|H0 + KI@C @@ C@ NB ≤ − J DIQ DI^ @ BC J The second solution area, Δ ≤ −. − - G|H0 − . − - G|H0 + KI@CDI@@DIC@ NB neglected because for positive ∆, ∆ has to be positive as well. Dividing (21) by Δ > 0 leads to: (22) ∆= ∆Y ≥Γ≡ @ J Using the rule of l’Hospital, the limit equals: lim∆Y→ Γ = zero. J Q ^ @ BC , can be r qr u p@ s E.=EXGF|YH0D.=EXGF|YH0 D @C @@ C@ J Zp qp qp [r@ rC ∆Y B DB ? I@ Y Q ^ . ?C IJ@ BC DB@ @ NB B =EXGF|YHKIJ@ DIQ@ DI^ @ C if ∆ goes to Furthermore, in order to be a separating equilibrium, the utility of a + Δ-type must be larger than the utility of imitating a -type: (23) -G=YDY _ + ΔH ≥ -G=Y _ + ΔH. The solution of this equation is: 25 −. − - G| + ΔH0 − . − - G| + ΔH0 + ∆ + Δ K + + N −. − - G| + ΔH0 + . − - G| + ΔH0 + ∆ =Ψ≤ ∆ = ≤Ψ ∆ + Δ K + + N The limit of the upper boundary can be calculated by using the rule of l’Hospital: = lim∆Y→ Ψ ?C IJ@ BC DB@ @ NB B =EXGF|YHKIJ@ DIQ@ DI^ @ C . Due to that: equation equals: x= xY (24) ? I@ B DB whereas: Κ = KI@CDIJ@ DIC @ NB@ J Q ^ @ BC =EXGF|YH ≤ x= xY ≤ =EXGF|YH , the differential = =EXGF|YH. . The manager has an incentive to overstate her actual forecast and thus, ≥ - G|H, which is why equation (24) is positive. The differential equation (24) can be solved by transforming the variables: = 2K − $ + N, where $ + = - G|H, $ = @ J KI function with respect to leads to: (25) KI@ DIQ@ N @ @ J DIQ DI^ N x xY =2 x= xY and = KI@ @ I^ O @ @ J DIQ DI^ N . Differentiating this − 2$. By inserting (24) into (25) the following separable differential equation is obtained: x xY (26) If 2Κ − 2Az = 0, we know from (25) that for leads to = £ ¤ = x= xY E . = $ follows. Inserting x= xY = $ into (24) and solving + $ + . If 2Κ − 2Az ≠ 0 then (26) is a separable differential equation with the following solution: (27) ¦ ln|¦ − $| − ¦ − $ = + ¨ −2$ , whereas C is the constant of integration. For ¦ − $ > 0, which occurs if $ + ≤ < $ + the reporting function is given by: £ ¤ + 26 (28) = - G|H + I@DI@ B ?C IJ@ BC DB@ J 1 + m − @ BC Q © tK@ª@ Nsq« ¬ , where m GHis the Lambert W function, with m GH ∈ G−1,0 because < ¤ + $ + . For ¦ − $ < 0, which is the case if > (29) = - G|H + £ ¤ £ + $ + the reporting function is given by: ?C IJ@ BC DB@ IJ@ DIQ@ B@ BC 1 + m © tK@ª@ Nsq« ¬ , where m GHis the Lambert W function, with m GH > 0 because > 0 in this case. Threshold value The expected utility for a manager with a signal who discloses the report equals in equilibrium: (30) while \]| = - G= |H = R + −e B@ BC BC DB@ IJ@ YEO @ IJ@ DIQ@ DI^ − -G|H + @ @ KIJ DIQ@ N I^ @ IJ@ DIQ@ DI^ , 4 = KIJ@ I@ BC DB@ @ J DIQ NB@ T + R + BC DB@ ?@ A Z [ + IJ@ XGF|YHEO KIJ@ DIQ@ N B@ BC BC DB@ T \]|f, and is given by eihter = £ ¤ + $ + , equation (28) or (29). Equation (30) is obtained by inserting (8), the expected price in the second period and the equilibrium price in the case of a report into (6). The constant of integration ¨ in (28) and (29) is determined by the expectation of the manipulation in excess of the expected value -.|0 at the threshold value, K̂ N − -.|̂ 0. This bias in excess of the expected value is assumed to be = K̂ N − -.|̂ 0 (this value will be discussed at a later point). In equilibrium, the expected has to be the same as the actual one. If this is the case, then the reporting function coincides with the expected reporting function. If > reporting function is given by 29 (28). ?C IJ@ BC DB@ IJ@ DIQ@ B@ BC ( < ?C IJ@ BC DB@ IJ@ DIQ@ B@ BC ), then the If investors assume that all signals which are lower than ̂ lead to a non-disclosure then observing no disclosure would induce the following price in the first period: (31) Ø = -.| ≤ ̂ 0 = − KŶ N KŶ N , 27 whereby ∙ is the density function of the signal and { ∙ is the cumulative distribution function. In equilibrium, the expected price coincides with the actual price: , Ø = Ø, thus ̂ = . The expected utility from non-disclosure is given by: (32) - GØ|H = R − T + R + N KŶ N KŶ IJ@ XGF|YHEO KIJ@ DIQ@ N T − R ?@ IJ@ T . B@ KIJ@ DIQ@ N Equation (32) is obtained by inserting (9) and the expected price in the second period and the price in the case of no disclosure (equation (31)) into (7). In equilibrium, the manager voluntarily reports her information if: - G= |H ≥ - GØ|H and she will disclose nothing if : -G= |H ≤ -GØ|H. denotes a signal for which the manager is indifferent: -.= |0 = -.Ø|0. In order to prove that there is only one threshold value, the intermediate value theorem is used by establishing a function ® j : (33) ® j = - G= | j H − - GØ | j H = R + − B@ BC BC DB@ − R ?@ IJ@ IJ@ KY¯ EON @ IJ@ DIQ@ DI^ T − B@ KIJ@ DIQ@ N T − Z − B@ BC BC DB@ \]| j , [ KY¯ N Y¯ whereby - G= | j H equals the expected utility of equation (30) at the signal j with j = - G| j H + and - GØ | j H which is given by equation (32) with ̂ = = j . The function ® j must be equal to zero if the utility of reporting is equal to the utility of withholding the information (® j = ®KN = 0). If ® j is a strictly monotone function and there exists a signal jj for which ® jj > 0 and a jjj for which ® jjj < 0 then the threshold value is unique. The first derivative equals: y°KY¯ N (34) whereby ± j = Y¯ EO yY ¯ @ IJ@ DIQ@ DI^ = ?C IJ@ @ − IJ@ DIQ@ DI^ ?C IJ@ Z± j + ¯ KY± ¯ N @ Y± IJ@ DIQ@ DI^ [, KY± ¯ N Y± ¯ . The first term of equation ((34) is positive. Sampford’s (1953) inequality implies that: 0 < j ¯ Ẑ + KY± ¯ N Y± [= KY± ¯ N Y± ¯ } E} Z−² + } [ < 1 follows, where E} ± j = −². Thus ® j is a strictly monotonically increasing function ( y°KY¯ N yY¯ > 0). Next, it must be shown that ® j can be positive and negative. Calculating the limits (if it was necessary, the rule of L’Hospital was used) shows that: 28 (35)limY¯ →³ ® j = ∞ − − − R B@ BC BC DB@ (36) limY¯ →E³ { j j − = limY¯ →E³ limY¯ →E³ EKY¯ EON t¶Ks¯ N s¯ tµ E @ eEp@ qp@ qp@ f ¯ J Q ^ Z¶Ks N[ N j = 0 (37) limY¯ →E³ (38) = limY¯ →E³ @ KY¯ NKY¯ EONDKIJ@ DIQ@ DI^ NKY¯ N @ NY¯ KIJ@ DIQ@ DI^ KY¯ N C Ks¯ tµN = limY¯ →E³ EKY¯ EON t¶Ks¯ N s¯ tµ E @ eEp@ qp@ qp@ f ¯ J Q ^ Z¶Ks N[ = limY¯ →E³ KI@ C @ Due to (35) and (38) and the fact that which ®KN = 0. y°KY¯ N yY¯ @ NT − B@ KIJ@ DIQ limY¯ →E³ ® j = 0 − B @DBC − RB B B ?@ IJ@ E KY¯ N C @ Ks¯ tµN B@ BC BC DB@ = limY¯ →E³ C ¶Ks¯ N J KY¯ N Q = Y →E³ =limY¯ →E³ KY¯ N EY¯ Y¯ EO T − B @DBC \]| j <0 KI@ DI@ N ?@ IJ@ @ EKY¯ EON = ¯lim − 2K + + @ @ ¯ J DIQ DI^ NY @ \]| j = ∞ > 0, B B C =0 @ > 0, there exists only one threshold value, for The bias in excess of the expected value of the accounting information As stated above, there is an infinite number of voluntary disclosure functions depending on the beliefs about . Each of these equilibria is characterized by a threshold value which depends on whereby all signals higher (lower) than lead to a voluntary disclosure (withholding of information). All of these equilibria require out-of-equilibria beliefs so that no type has an incentive to deviate from the equilibrium strategy. These out-of-equilibrium beliefs will determine the market price if a report which is lower than -.|0 + is observed. In equilibrium, a manager with the signal must be indifferent between disclosing KN = -.|0 + and disclosing nothing. But what happens if the manager with signal decides to disclose: -.|0 + − · with > · > 0? How high will the market price be in the first period? If ≥ ,Y , then disclosing -.|0 + − · is just as good as disclosing -.|0 + or disclosing nothing, whereby ,Y is defined by: (39) ,Y K-.|0 + − ·N − 5 − B @DBC − · = R − KYNT B B C @ = -.|0 − 5 − KYN B@ BC BC DB@ , 29 where 5 = R ?@ IJ@ @ NT + B@ KIJ@ DIQ B@ BC BC DB@ \]|. The last equality follows from the fact that a manager with the signal is indifferent between disclosing KN = -.|0 + and disclosing nothing. It is noteworthy that the expected disutility is smaller if the manager deviates ( − · < ), thus, -.|0 > ,Y K-.|0 + − ·N. For all out-of-equilibrium beliefs which lead to a higher market price if a report of -.|0 + − · is observed K-.|0 + − ·N > ,Y K-.|0 + − ·N, the manager with the signal would deviate from her disclosure strategy. Assume a manager with a signal lower than , for example − ¸ (¸ > 0) would report -.|0 + − ·=-.| − ¸0 + $¸ + − ·. She will be indifferent if: (40) ,YE¹ K-.|0 + − ·N − 5 − B@ BC BC DB@ $¸ + − · = R − KYN T. KYN Due to that $¸ + − · > − · the price in period one where a − ¸ type is indifferent between disclosing -.|0 + − · and nothing is higher than ,Y K-.|0 + − ·N, the price where a type would be indifferent. A type − ¸ prefers to deviate for all prices which are at least as high as ,YE¹ . For such a price reaction she either strictly prefers reporting -.|0 + − · or is indifferent. If ≥ ,YE¹ a type strictly prefers to deviate from the equilibrium strategy. Thus, according to the D1 criteria of Cho and Kreps (1987), the market should not believe that the report came from a type − ¸ because a type prefers the deviation stronger. Thus, the market should believe that the report was sent from a manger with a signal which is at least as high as . However, if the market believes this, the report -.|0 + − · should lead to a price in the first period which is at least -.|0, K-.|0 + − ·N ≥ -.|0. If this is the case, then a manager with signal would deviate from the equilibrium strategy (KN = -.|0 + ), because the disutility would be lower whereas the price is at least as high as that of reporting KN = -.|0 + . Thus, all equilibria with positive can be deleted by using this refinement method. The only surviving equilibrium is one in which = 0. If a report j ≤ -.|0 is observed, in such an equilibrium, the market will evaluate the firm with j = j because the strongest incentives for this report has a firm of 30 type j whereas j = - G| j H. Thus, for all < it is not valuable to report voluntarily, because the utility of withholding is higher. If = 0 in equilibrium, the constant of integration in equation (28) equals: ¨= (41) º» £E E¤@ − . The manager with the signal discloses her actual beliefs about the future accounting information and, thus,: m − © tK@ª@ NKsq«N ¬ = −1 as can be seen in equation (28). This occurs if the constant of integration is given by equation (41). By inserting (41) into (28), the reporting function (12) is obtained. Disclosure strategies The differential equation (24) guarantees that, if a manager decides to issue a report, a type j will report j and a type jj will report jj , for all j , jj .,∞. Deviating from the equilibrium reporting strategy, , would lead to a lower utility. However, is it possible that a better type jj > j decides for a non-disclosure while a type j decides to disclose voluntarily? Firstly, it can be shown that the expected utility is increases in the signal . The higher the signal the higher will be the expected utility of reporting. The expected utility is given by equation (30). Differentiating this equation with respect to leads to: (42) yX.¼zs _Y0 yY J Q = I@ DIJ@ DI@ + I@ DIJ@ DI@ − B @DBC 2 − - G|H RxY − I@ DI @ DI @ T J I@ Q = ^ IJ@ @ @ IJ DIQ@ DI^ J I@ Q + B @ B BC C DB@ ^ B B C @ 2 − - G|H x= IJ@ DIQ@ @ @ IJ DIQ@ DI^ > 0. I@ DI@ J Q ^ Because ≥ - G|H equation (42) is positive. Even without the positive effect of the expected price in the second period, the utility of a better type is higher, (43) yX.¼zs _Y0EXG½@ |YH yY - =KY¯ N ¾ j − - G | j H > -.=Y _0 − -. |0, ≥ 0. Thus, if j > . Since a manager with the threshold information is indifferent between disclosing and withholding information (-.=Y _0 = -.Ø |0). The term on the right side of the inequality is 31 equal to -.Ø|0 − -. |0. Based on this equality and the inequality in (43) the following inequality can be calculated: - =KY¯ N ¾ j > -.Ø |0 − -. |0 + - G | j H = - GØ | j H. (44) Therefore, a manager of type j is better off if she issues a report. If jj > j , then - =KY¯¯ N ¾ jj − - G | jj H > - =KY¯ N ¾ j − - G | j H > - GØ | j H − - G | j H and, thus, also - =KY¯¯ N ¾ jj > - GØ | jj H. Thus, if j discloses information, a firm with better information will issue a report as well. Frequency of voluntary disclosure The threshold value is calculated by setting the function in (33) equal to zero (®KN = 0). Thus, is defined by this implicit equation. Due to of this equation, the derivative of the threshold level, with respect to a parameter, can be determined by using the implicit function theorem. It is the ratio of the derivative of ®∙ with respect to the parameter and the derivative of this function with respect to multiplied with minus one. As already known from above, y°KY¯ N yY¯ > 0. In order to calculate whether increases or decreases depending on a change of a parameter, the partial derivative of ® ∙ with respect to the parameter has to be calculated: y° yBC (45) = −R y° yB@ (46) y° y?C (48) (49) (50) y° @ yI^ = = ?@ IJ@ y° = Á−± − @ NT B@¿ J Q KIJ@ DIQ IJ@ KYEON @ N¿⁄@ KIJ@ DIQ@ DI^ yIQ@ = R B@@ C DB@ − B @ @ KIJ DIQ@ N I^ R @ BC@ C DB@ = − I@DI@J@CB @@B + @ IJ@ DIQ@ DI^ ?C IJ@ y° B@ KIJ@ DIQ y?@ (47) @ NT − B ?@ IJ@ ?C IJ@ KYN = + ± N KY± N KY± KYN ISB ? @ N¿⁄@ KIJ@ DIQ@ DI^ @ @ C DB@ KY± KY± N KY± N @ IJ@ DIQ@ DI^ < 0, T, ê + ZY± [ KY± T − N B@ BC KY± N @ IJ@ DIQ@ DI^ R± + N T < 0, @ @ KIJ DIQ@ N I^ R IJ@ KY± N Á−̂ − @ IJ@ DIQ@ DI^ KY± N + ̂ KY± N KY±N f > 0, @ KIJ@ DIQ@ N BC DB@ KIJ@ DIQ@ DI@ N@ R± + KY±N KY± N T ^ < 0, 32 − S B@ BC I^ @ N@ BC DB@ KIJ@ DIQ@ DI^ The derivative in (46) is negative if θ < + BC ?@@ IJS ¿ B@@ KIJ@ DIQ@ N , @ IJ@ DIQ@ ¿ I^ BC B@¿ @ S @ @ IJ IJ DIQ DI^ BC DB@ @ . The partial derivative, with respect to the weight on the second price, is negative. The derivative, with respect to , is positive due to Sampford’s law. Equation (49) is negative for all ± . This is obvious for ± > 0. Because KY± N KY± N ê + ZY±[ KY± N f 0,1, the upper bound of the third term in square brackets equals ± . Thus, the addition of the first term and this term is always negative and all the other terms are negative as well. If ± < 0 additional conditions are needed. The expression in the square brackets can be transformed to: (51) whereby à = −± > 0, W = à − EÅ Å Ä − à Z−à + [ < 0, Ä Ä , à = −à and 1 − { à = { −Ã. The last two equalities follow from the characteristics of the normal distribution function. Because the Mills’ ratio W lies between 2⁄KÆà + 4 + ÃN < W < 4⁄KÆà + 8 + 3ÃN 12, the term in (51) must be negative for all à > 0. Equation (50) is at least negative if ¿ S B ¿ KI @ DI @ N I^ Q @ J @ @ N IS B DB KIJ@ DIQ@ DI^ @ J C > , which can be derived from the last two terms of equation (50). If this inequality does not hold, it can be positive as well. The derivative with respect to is not unambiguously which is why it is neglected. Proof of the difference between the expected biases The ex-ante expected difference between the two variables is given by: (52) -.=∗ | ≥ 0 − -.}∗ | ≥ 0 = K1 + -.m G∙H| ≥ 0N R E?C IJ@ KIJ@ DIQ@ NBC T+ ?@ IJ@ KIJ@ DIQ@ NB@ . The inequality in the text can be derived from (52). Differentiating (52) with respect to leads to: 12 See Baricz (2008). 33 yX.:z∗ |YÊY0EX.:Ë∗ |YÊY0 y?@ (53) =l w l @ C @JNB w + @ J @ NB > 0. yY ÍÏ y? KIÍÎÍ KIJ DIQ @ Ñ@ ÌÍÍÎÍ ÌÍ ÍÏC J DIQ yX.G∙H|YÊY0 yY E? I@ Ò Ð I@ Ð This derivative is positive. yX.G∙H|YÊY0 yY (54) = ÓY m G∙H Ô ³ R Y KYN EKYN EKYN − YEO Is@ TÕ " < 0 Since the function in curly brackets in (54) has only one root and is positive on the left side and negative on the right side and since the integral of this term equals zero, weighting it with a negative and increasing function (mG∙H) must now add to a negative number. Proof of Lemma 2 The conditional @ @ N KYN KIJ@ DIQ@ DI^ ZEKYN[ @ variance f= IJS @ @ IJ DIQ@ DI^ is given is given by: y} yY = } @ IJ@ DIQ@ DI^ y×Ø=.½C EO|YÊY0 @ yI^ . =− yM yY = The derivative y×Ø=.½C EO|YÊY0 IJS @ @ IJ DIQ@ DI^ IJS @ @ IJ DIQ@ DI^ e1 + KYNYEO EKYN − 1 − 5. Based on Sampford (1953) we know that: 0 < 5 = Z−² + E}[ < 1 and y} > 0 while ² = E} } \]. − | ≥ 0 = by: IJS @ @ IJ DIQ@ DI^ R− YEO @ IJ@ DIQ@ DI^ yM y} y} yY . Thus, the derivative with respect to T < 0. This is negative because with respect to is yM y} > 0 and given by: J − l1 − 5 − w < 0. Ù yY y} y} Ù yI Ñ^@ KIJ@DIQ@ DI@ N@ ÌÍÍÍÎÍÍÍÏ yM y} yY Ò Ò Ò IS ^ Ò yM } 34 Figure 1 Timeline 35 Figure 2 Biasing function 36 7 References Bagnoli, M. and S. Watts (2007), Financial reporting and supplemental voluntary disclosures, Journal of Accounting Research, Vol, 45 (5), 885-913. Ball, R. and L. Shivakumar (2008), How much new information is there in earnings?, Journal of Accounting Research, Vol. 46 (5), 975-1016. Ball, R., S. Jayaraman and L. Shivakumar (2012), Audited financial reporting and voluntary disclosure as complements: A test of the Confirmation Hypothesis, Journal of Accounting and economics, Vol. 53 (1-2), 136-166. Baricz, Á. (2008), Mills’ ratio: monotonicity patterns and functional inequalities, Journal of mathematical analysis and applications, Vol. 340 (2), 1362-1370. Beyer, A., A. Cohen., Th. Lys and B. Walther (2010), The financial reporting environment: review of the recent literature, Journal of Accounting and Economics, Vol. 50 (2-3), 296-343. Beyer, A. and I. Guttman (2010), Voluntary disclosure, manipulation and real effects, Working Paper. Cho, I.-K. and D. M. Kreps (1987), Signaling games and stable equilibria, The Quarterly Journal of Economics, Vol. 102 (2), 179-221. Dye, R. (1985), Disclosure of nonproprietary information, Journal of Accounting Research, Vol. 23 (1), 123-145. Einhorn, E. (2005), The nature of the interaction between mandatory and voluntary disclosures, Journal of Accounting Research, Vol. 43, 593-621. Einhorn, E. and A. Ziv (2012), Biased voluntary disclosure, Review of Accounting Studies, Vol. 17 (2), 420-442. Fischer, P. and R. Verrecchia (2000), Reporting bias, The Accounting Review, Vol. 75 (2), 229245. 37 Gigler, F. B. (1994), Self-enforcing voluntary disclosures, Journal of Accounting Research, Vol. 32, 224-240. Gigler, F. and Th. Hemmer (1998), On the frequency, quality, and informational role of mandatory financial reports, Journal of Accounting Research, Vol. 36. 117-147. Healy, P. and K. Palepu (2001), Information asymmetry, corporate disclosure, and the capital markets: A review of empirical disclosure literature, Journal of Accounting and Economics, Vol. 31 (1-3), 405-440. Korn, E. (2004), Volunatry disclosure of partially verifiable information, Schmalenbach Business Review, Vol. 56. 139-163. Kwon, Y., D. Newman and Y. Zang (2009), The effect of accounting report quality on the bias in and the likelihood of management disclosure, Working Paper. Lundholm, R. (2003), Historical accounting and the endogenous credibility of current disclosures, Journal of Accounting Auditing and Finance, Vol. 18 (1), 207-229. McNichols, M. (1989), Evidence of the informational asymmetries from management earnings forecasts and stock returns, The Accounting Review, Vol. 64, 1-27. Pownall, G. and G. Waymire (1989), Voluntary disclosure credibility and securities prices: evidence from management earnings forecasts, Journal of Accounting Research, Vol. 27, 227246. Sampford, M. R. (1953), Some inequalities on mill’s ratio and related functions, The Annals of mathematical statistics, Vol. 24 (1), 130-132. Verrecchia, R. (1983), Discretionary disclosure, Journal of Accounting and Economics, Vol. 5, 179-194. Wagenhofer, A. (2000), Disclosure of proprietary information in the course of an acquisition, Accounting and Business Research, Vol. 31 (1), 57-69. Waymire, G. (1984), Additional evidence on the information content of management earnings forecasts, Journal of Accounting Research, Vol. 22, 703-718. 38