Interactive Reporting Bias Surrounding CEO Turnover^†

† This paper is a substantially revised chapter of my dissertation at Monash University. Paper accepted by Jeroen Suijs.

Abstract

This paper analyzes how CEO turnover affects successive CEOs' financial reporting decisions and the capital market price. I show that when an outgoing CEO (O) in period 1 is succeeded by an incoming CEO (N) in period 2, strategic interaction between O and N leads to interlinked earnings reports. Specifically, when the level of earnings reported by O is lower, N's reporting strategy is more likely to feature a downward reporting bias. Furthermore, by a comparison of the two-CEO setting with a setting with no CEO turnover, I show that with CEO turnover, (i) the period 2 earnings report is more sensitive to the private information of the CEO in control and less sensitive to the period 1 earnings report; (ii) the period 1 earnings report is more sensitive to the private information of the CEO in control; and (iii) the equilibrium stock price has the same sensitivities toward the associated risks, but is less sensitive to the periods 1 and 2 earnings reports. These results provide a novel explanation for managerial under-reporting bias based on strategic interaction between successive CEOs and shed light on the role of CEO turnover in earnings management behavior and capital market responses.

Acknowledgements

I am indebted to Chongwoo Choe (chair), Vai-Lam Mui, and Birendra Rai for their guidance. I am very grateful to an anonymous referee and Jeroen Suijs (associate editor) for numerous insightful suggestions that greatly improved the paper. I would also like to thank Paul Brockman, Chee Cheong, Jong-Seo Choi, Greg Clinch, Shane Dikolli, Luciana Fiorini, Stephen King, Donald Lien, Edward Lin, Rodney Maddock, Pedro Gomis Porqueras, Grant Richardson, Terry Walter, Takeshi Yamada, Xiangkang Yin, and seminar participants at Deakin University, Monash University, the University of Western Australia, the 25th PhD Conference in Economics and Business, and the 2015 AFAANZ Conference for helpful comments. All remaining errors are my own.

Notes

† This paper is a substantially revised chapter of my dissertation at Monash University. Paper accepted by Jeroen Suijs.

1 Choi, Kwak, and Choe (2014) document evidence of interactive earnings reports by successive CEOs. Their empirical analysis explores the relationship between managerial under- or over-reporting behavior and CEO turnover types by distinguishing whether the departure of the outgoing CEO is through a peaceful or forced process and whether the appointment of the new CEO is through internal promotion or external recruitment. It is, however, difficult to directly relate my theoretical results to their empirical findings. Choi et al. (2014) are concerned about the effects of CEO turnover types (internal promotion versus external recruitment; forced turnover versus voluntary turnover) on whether CEOs over-report or under-report (discrete variables). In contrast, this study considers how the presence or absence of CEO turnover affects the extent of earnings management and the capital market responses (continuous variables).

2 Beyer (2008) also considers a leader–follower game between an outside analyst and an inside manager. While the objectives of analysts and managers in Beyer (2008) are to minimize forecast and reporting errors, the objectives of CEOs in this paper are to maximize a weighted average of the reported earnings and the stock price, net of earnings management cost (EMC). Also, a manager always chooses to over-report earnings in Beyer (2008), whereas a manager may over- or under-report earnings in this paper.

3 To illustrate, suppose that CEO compensation packages consist of a reported earnings bonus and a stock grant and that O has publicly issued a poor earnings report, say, . Then, N takes control and privately observes another low level of earnings, say, . N can choose to report 0, , or , to maximize her pay net of the EMC. If the EMC is prohibitively high, then N would choose to report . If not, I argue that it is possible for N to report rather than 0. The cost of reporting is that it reduces N's payoff from the reported earnings bonus and adversely affects the stock price because a low level of reported earnings is a bad signal for true earnings. It can, however, reduce the perceived volatility in earnings since is absolutely closer to than 0 is. This, in turn, reduces the risk premium required by outside investors. If the saving on the risk premium dominates the direct negative impact of lower reported earnings and the loss on the earnings bonus, N would choose to report and a downward reporting bias thus occurs.

4 More discussion is provided in Section 3.5.

5 According to Proposition 1, when all earnings are transitory (i.e. ), . It means that, even if , N's reporting strategy is not fully responsive to true earnings and hence earnings management occurs. Note that this does not contradict with Footnote 9, which refers to the fact that when .

6 See, for example, Fee, Hadlock, and Pierce (2011) for further discussion of exogenous CEO changes. I acknowledge that CEO turnover can be endogenous in some cases, that is, earnings management can affect the likelihood of CEO turnover (cf. Hazarika, Karpoff, and Nahata, 2012). The main intuition, nevertheless, carries over to the more complicated situation.

7 Note that, to focus on the reporting behavior of CEOs, I follow Beyer (2009) and exogenously specify the earnings distribution. The current study thus does not consider real-activity earnings management where managers make choices to determine production endogenously. This paper, unlike Beyer (2009), allows the variance of to be random and assumes it to be a multiple of the variance of . Without this assumption, it would be difficult to derive the equilibrium stock price.

8 That is, the density . The intuition of the results would remain valid if managers privately observed τ before issuing earnings reports. The key is that outside investors are uncertain about the precision. Parameters α and β may depend on the ability of the CEO in control or how well the CEO matches with the firm.

9 If , innovations are transitory and current earnings provide no information about future earnings. Although this would not eliminate earnings management incentives (cf. Footnote 5), it does break down the link between O's and N's earnings reports. I rule out this possibility to focus on the interactive reporting between CEOs.

10 In an earlier version of the paper, I considered another trading stage at the end of period 1 and found that the period 1 stock price would be quadratic in the period 1 earnings report and not normally distributed. The period 2 stock price depends on the period 1 stock price. Without normality, however, the projection theorem would not be applicable and it would then be analytically difficult to derive the period 2 stock price. This study thus only allows trading at the end of period 2 for tractability.

11 Frazzini and Lamont (2007) and Cohen, Dey, Lys, and Sunder (2007) argue that firm-specific volatility surrounding earnings announcements is not diversifiable and this may explain why investors demand a premium for earnings announcements.

12 The assumption that part of O's remuneration is tied to the stock price at the end of period 2 is for analytical tractability (see Footnote 9). What is important here is that O cares about the period 2 stock price. If O is only concerned with what happens in period 1, he would always be driven by his myopia to over-report and the strategic interaction between O and N would evaporate. If that is the case, then there is no point in studying the decision-making problems for O and N jointly. It is acknowledged that O generally cares about both the period 2 stock price and the period 1 stock price (if it exists). Nevertheless, to avoid myopic behavior, the weight on the period 2 price must be sufficiently large. The assumption captures this feature and serves the research purpose.

13 This is different from the signal-jamming literature on earnings manipulation, for example, Stein (1989) and Goldman and Slezak (2006). See Arya, Glover, and Sunder (1998) for an excellent discussion of the Revelation Principle in accounting.

14 Investors may then sue both O and N for misrepresentation of financial information, and the one who biased more earnings faces greater litigation risk (e.g. Laux and Stocken, 2012; Trueman, 1997).

15 See, for example, Shibano (1990) and Evans and Sridhar (1996) for an analysis, and DeFond and Jiambalvo (1991) and Hennes, Leone, and Miller (2008) for empirical support.

16 See Sunder (2002) for a discussion of how common knowledge can be closely related to accounting and capital market research. Note that the magnitudes of and reflect how earnings are affected by shocks and fundamentals. Thus, I choose not to normalize and . Similarly, to capture the extent to which period 1 earnings can potentially predict period 2 earnings, I choose to consider a non-zero ρ in the model (cf. Equation (37(A8) ) in the Appendix).

17 One can alternatively model and to be privately observed by O and N, respectively. There are, however, some technical issues with this approach. First, and shall be positive, so one cannot assume them to be normally distributed. Then, for tractability, one would need to assume and to be binary (e.g. Liang, 2004) or to be lognormally distributed with multiplicative production technology (e.g. Peng and Röell, 2014).

18 The equilibrium concept adopted in this paper is the noisy rational expectations equilibrium. See, for example, Fischer and Verrecchia (2000) and Dye and Sridhar (2004).

19 Nonlinear reporting strategies lead to analytical considerations that are beyond the scope of the current paper. For example, if is a function of , then how can one make a conjecture about the functional form? Standard rational expectations equilibrium analysis does not provide much guidance for this. Moreover, it will become technically difficult to derive the equilibrium stock price because it is hard to calculate conditional expectations with nonlinear reporting functions.

20 As is standard in the rational expectations equilibrium analysis, these reporting and pricing coefficients are not common knowledge at the beginning of the game. While making his or her own decision, a player must conjecture about the coefficients in the other players' decisions. However, in equilibrium, players' conjectures will be true, that is, the s, s and s will become common knowledge in equilibrium. Also, note that, because O's reporting decision is made before N's reporting decision, O's earnings report does not explicitly depend on N's earnings report. Nevertheless, when choosing his earnings report, O anticipates N's reporting strategy and takes it into account (cf. Equation (3(3) )).

21 In that case, based on the observations of and , outside investors' estimates of and are and , respectively. The stock price is hence . Also, and can be shown to be nonlinear functions of and . As such, the model becomes analytically intractable. See the Appendix for the detail.

22 This restriction is non-binding in the case that . Also, numerical illustrations in Sections 3.6 and 4.3 show that the key results qualitatively hold if this restriction is dropped.

23 The equilibrium in Proposition 1 is unique up to cases with linear reporting strategies, quadratic pricing function, and the information integration restriction. Admittedly, equilibria of other forms might possibly exist. It is, however, analytically difficult to solve for those equilibria explicitly.

24 It is assumed that all parameter values are such that and are real numbers.

25 As , I hereinafter use to refer to the equilibrium coefficient on both squared terms in Equation (13(13) ).

26 , Equation (23(23) ), is a complicated function of model parameters. Hence, the discussion is focused on the sensitivities of the pricing function.

27 Recall that , where is the sensitivity of O's reporting function toward his own private information, . In equilibrium, will be a function of exogenous parameters of the model, but it is not affected by its own associated argument, , which is an exogenous random variable of the model. Thus, affects , but not . Moreover, when it comes to the reporting function of N and the market pricing function, it matters how is generated. To see this, note that a change in can be caused by changes in and (due to some changes of model parameters) or by changes in (totally random). In the former case, the equilibrium coefficients of N's reporting function and the market pricing function can be affected. On the contrary, in the latter case, it will not affect the equilibrium coefficients of N's reporting function and the market pricing function.

28 Note that because . Also, I find that when and , which is consistent with the numerical solution in Section 3.6.

29 Note that Equation (17) says that , which is consistent with the numerical solution in Section 3.6.

30 Note that under-reporting by N may also happen even when O does not bias earnings downwards. If the period 1 performance is just really bad and O truthfully reports earnings (i.e. ), then N may still have an incentive to bias earnings downwards in period 2.

31 This is to hold the total compensation across two periods the same for the same values of , , and P between the one-CEO and two-CEO cases. As such, any difference in the equilibrium between these two settings is not driven by differences in exogenously assumed compensation to start with, making things more directly comparable.

32 I could have conjectured . However, this is informationally equivalent to the current conjecture: and , whereby . Because , , and are just constants to be determined in equilibrium, there is no essential difference between these two conjectured forms.

33 It is analytically difficult to compare the equilibrium values of , and in the one-CEO setting with those in the two-CEO setting. Thus, the comparison here focuses on the sensitivities of the market pricing function and managerial reporting functions.

34 This is because CEOs in my model have private information, but there is no difference in their abilities or effort choices.

35 In a slight abuse of notation, the pricing coefficients here represent the ones conjectured by N, not the true ones, and the hats indicate N's conjecture of the market's conjecture about N's reporting rules. I could make a distinction between them, but it would complicate the notation and make no difference in equilibrium. The appropriate meaning should be clear in each case.

36 The second-order conditions (SOCs) for N's and O's decision problems are satisfied if , which is true in equilibrium.

37 I do not wait until the decision problem of O is solved to impose this requirement because equilibrium are determined in period 2 and O would correctly anticipate the outcome.

38 Note that, according to the conjectured equilibrium, and thus the partial derivative of this term with respect to is zero.