Voting over Disclosure Standards

Abstract

This article examines the nature of disclosure standards, under the assumption that (i) standards preferred by more firms are collectively chosen and (ii) privately informed firms prefer standards that increase market perceptions about the value of their assets. A standard is stable if it is preferred by a large enough super-majority of firms over any other standards. Absent any restriction on possible standards, only unanimity would make a standard stable. By contrast, when requiring standards that classify news from best to worst, there is at most a single stable standard, and it must be full disclosure. For a large class of distributions over valuations, the required super-majority is about two-thirds, close to the majority required in many standard-setting boards. Value distributions with heavy tails, such as news that contains extreme risks, require higher super-majorities to be stable. These insights are robust to settings in which the information is used in decision-making.

Acknowledgments

We thank the editor, two reviewers and seminar participants at annual meeting of the American Accounting Association, Notre Dame, Kellogg School, FU Berlin, University of Berne, University of Mannheim and thank Edwige Cheynel, Masako Darrough, Ron Dye, Ivan Marinovic, Ulf Schiller, Dirk Simons and Isabel Wang (the discussant) for helpful suggestions.

Notes

1 Beresford (2001) recounts his experience as a standard-setter responding to comments in the context of standards on business combinations, noting that ‘the alleged economic consequences were beyond the FASB's responsibility. In nearly all cases this satisfied the member of Congress and there was no further activity ’.

2 While our primary interpretation of the model is in terms of a political choice, some of the insights plausibly carry over to consultations made by vote, such as corporate governance issues or say on pay (Ferri & Goex, 2013), that may feature some redistributive effects.

3 For expositional purposes, we assume throughout that parties have equal weight in the standard-setting process; however, this is not a critical assumption for most of the results.

4 The assumption $\underset{\_}{v}>-\mathrm{\infty }$ is unimportant for our analysis and only used to save space. If $\underset{\_}{v}=-\mathrm{\infty }$, one would obtain the same results for the lower tail as for the upper tail using a transformation $g\left(v\right)=f(-v)$ of the original random variable.

5 To keep the model as parsimonious as possible, this definition omits further uses of information, in that the proofs only require prices to be increasing in posterior expectations. We can prove very similar results assuming that the market price is $P\left(E\right(\tilde{v}|\tilde{I}\left)\right)$, where $P(.)$ is increasing and convex, which, in turn, implies that more precise information yields higher prices in expectation. There are other extensions of the model that leave the results entirely unchanged, and which we discuss in Sections 5.4 and 5.5.

6 See also Glode, Opp, and Zhang (2016), Bertomeu and Cianciaruso (2017) and Rappoport (2017) for verifiable disclosure settings in which the sender can report partitions of the message space. One should not necessarily interpret that actual standards are literally written as a function of final cash flows (and the model does not require this). As a matter of practical implementation, regulators may classify events in terms of other tangible characteristics, say $\tilde{\xi }$. When we say that, for example, the signal $A=\{a<\tilde{v}<b\}$ is reported, we mean that firms report some characteristics $\tilde{\xi }$ consistent with $a<\mathbb{E}\left(\tilde{v}|\tilde{\xi }\right)<b$ having occurred. Vice versa, one could state the model in terms of underlying characteristics and then map to financial reporting motives with the transformation $\mathbb{E}\left(\tilde{v}|\tilde{\xi }\right)$.

7 We briefly mention some extensions that are unimportant for any of our results (these have been omitted to simplify the exposition of the model). As in most of the literature in the area, it is unimportant if the final cash flow is noisy in the firm's information $\tilde{v}+ϵ$, as long as we understand $\tilde{v}$ as the expected cash flow. The results are similar if some firms (but not all of them) are uninformed about $\tilde{v}$, if firms draw their information from different distributions, or if the information system is stated more generally as a stochastic report with support over S.

8 With some additional investment in notation, the analysis extends to more general partitions. However, the approach is useful to the extent that it restricts the analysis to empirically reasonable constructs and allows us to state the results with Riemann integrals. It is unimportant for the results if we also allow for closed intervals (the topology is defined on the space S so that A=S is both open and closed).

9 The vote that is implicit in Definition 2.1 may be interpreted as lobbying (any activity in which pressure is exerted on a policy maker). This is a well-documented aspect of accounting standard-setting that is central to our theory. Giner and Arce (2012) investigate lobbying activities in the due process for IFRS 2 (share-based payments). They show that preparers constituted the most active group in terms of lobbying activities. Georgiou (2002) investigates conditions under which firms do not participate in the standard-setting process of the UK's ASB. Finally, Sutton (1984) analyzes the lobbying of those affected by financial accounting standards in the case of accounting standard-setting bodies in the UK and the US.

10 Similar to how a majority vote is used in social sciences, a vote is not meant in a literal sense. Not only for reporting but for many other issues, constituents do not directly cast a vote on an issue. As one example, an elected official makes many decisions that are, once the election has taken place, not subject to an actual referendum. However, in time, those bodies making the decisions are subject to being chosen in a representative process and, thus, must implement choices that are desirable to a large enough majority.

11 Several quick technical remarks are in order. First, all norms are equivalent on the real line, so that the results are unchanged if we use a norm different from the absolute value when defining $d(.,.)$. Second, note that $(\mathrm{\Omega },d)$ is a compact metric space. Third, none of the results requires the distance to be defined in terms of the objective density $f(.)$. Fourth, we may use Riemann integrals given that the sets under consideration are either singletons or countable unions of open sets.

12 This construction is reminiscent of a form of conservative information systems in which information about bad news is less precise (i.e. features a larger region of underlying events) relative to good news, since the modified standard will feature a large fraction of ‘bad’ firms being pooled (Larmande, 2015).

13 The effect of allowing discretionary disclosure in the model would depend on the arrangements for insuring truthful reporting. If regulators enforce the truthfulness of only mandatory disclosures, the additional cost of discretionary disclosure could reinforce pooling in reporting. If all reporting is costlessly truthful, then the full information is reinforced as an equilibrium.

14 This argument is closely related to the lack of a stable redistribution policy. If we assume that a regulator can redistribute wealth among N>3 voters, then no stable outcome exists because N−1 voters would always vote together to expropriate the third. The same holds true for Euclidian preferences where N−1 voters can push toward their preferred bliss point (Plott, 1967). Such transfers are far less simple in our setting because expropriation cannot be done via a direct monetary transfer where a fraction of the voters demand to be paid but, rather, must be achieved via a suitable coarsening of the information.

15 We refer to Pickands (1975) for details. The formal definition is given next: $F(.)$ is in the domain of attraction of an extreme-value distribution function if, for some $c\in {\mathbb{R}}^{+}$, $\underset{u\to +\mathrm{\infty }}{lim}\underset{0<a<\mathrm{\infty }}{inf}\underset{0\le x<\mathrm{\infty }}{sup}\left|\frac{1-F(u+x)}{1-F\left(u\right)}-{\mathrm{e}}^{-{\int }_0^{x/a}(1+ct{)}^{-1}\mathrm{d}t}\right|=0.$

16 This bound is remindful of the model by Caplin and Nalebuff (1988) which results in a similar bound on super-majority. However, the argument is different in their setting in that they model euclidian preferences, that is, such that individual preferences can be represented as a distance from a bliss point. We thank Kai Du for mentioning this reference.

17 In his keynote speech at the American Accounting Association, Sir David Tweedie, then chairman of the IASB, argued that mark-to-market had been completely changed by the FASB in response to the financial crisis of 2007–2008, but the IASB had refused to do so (the complete speech is available on the website at AAA commons 2009). It is plausible that the larger majority required at the IASB would have made difficult large changes away from the status quo that are incompatible with past rules.

18 Note, however, that if such an impairment test is positive, the mark-down to the value of the asset will be based on the discounted present value of the asset.

19 To be fair, our model is not one of the institutional details that enter each step of the standard-setting, as the stability criterion is intended as a broad high-level observation. In an early paper, Amershi, Demski, and Wolfson (1982) offer a model of agenda setting that provides some notion about strategic behavior. More recently, Bertomeu and Cheynel (2013) solve a non-cooperative game in which parties sequentially bargain over which rule to impose.

20 With formal notations, $M\left[x\right]=\{B:B\in \mathcal{M},B\ne A\}\cup \{A_x^{-},\{x\},A_x^{+}\}$.

21 Formally, $f\left(c_1\right)<f\left(c_2\right)$ implies that there exists some $c_3$ such that $f^{\prime }\left(c_3\right)<0$, and we can pick A as a neighborhood of $c_3$.

22 To be a purist, one should not preassume that the limit l necessarily exists and instead use the notations lim inf and lim sup. Replacing all limit signs with these does not present any technical difficulties.