Asymmetric impact of earnings news on investor uncertainty

• Author: Zihang Peng,Demetris Christodoulou,David Johnstone
• DOI: http://doi.org/10.1111/jbfa.12428
• Date: 01 January 2020
• Published date: 01 January 2020

DOI: 10.1111/jbfa.12428

Asymmetric impact of earnings news on

investor uncertainty

Zihang Peng1David Johnstone2Demetris Christodoulou3

1Accounting, University of New South Wales

2Accounting, University of Wollongong

3Accounting, University of Sydney

Correspondence

DemetrisChristodoulou, Accounting, University

ofSydney, NSW 2026, Sydney,Australia.

Email:d.christodoulou@econ.usyd.edu.au

Abstract

We describe a model that predicts an asymmetric impact of dis-

closure on investor uncertainty. We show that good news tends to

resolvemore uncertainty than bad news, and that uncertainty can be

revisedupwards if the investors’ prior belief is sufficiently strong and

the signal is sufficiently bad. This result is in contrast to classical dis-

closure models, where new information always resolves uncertainty

and the change in uncertainty depends only on the relative preci-

sion of the news. Using option-implied volatility as a proxyfor uncer-

tainty, we find strong support for our predictions. We also show

that our results are robust to competing explanations,notably to the

leverage effect and volatility feedback, as well as to the jump risk

induced in anticipation of the earnings announcements.

KEYWORDS

disclosure, earnings, implied volatility, investoruncertainty

1INTRODUCTION

We study the impact of disclosure on investor uncertainty in a generalized Bayesian setting. We describe a model

that caters for circumstances whereby new information, especially unfavourable information, can shakeinvestor con-

fidence and add to uncertainty about the firm’s stochastic cash flow. Previous models, such as Lambert, Leuz, and

Verrecchia (2007), carry the inherent limitation that any new information, regardless of whether it is favourable or

unfavourable,must by the model’s construction leave investors with reduced uncertainty.

Toavoid that constraint on belief revision, while still using the same convenient Bayesian mathematics, we assume

that the firm has two possible latent states (regimes) underlying its economic fundamentals, ‘Normal’ and ‘Distress’,

where Distress is generally much less probable a priori. Investors then face two layers of uncertainty: (i) they do not

know which state the firm is in (between-state uncertainty), and (ii) they do not know what the firm’s cash flow will be

even when they know which state it is in (within-state uncertainty).

The appeal of this model is that new information affects investors’ beliefs in two ways,simultaneously. First, it gives

an indication of which state the firm is in (unfavourable information points towards the bad state, but possibly not

strongly). Second, it allows investors to revise the probability distribution of the firm’s cash flow conditional on the

firm being in one of the states. A concession to reality in this model is that bad news can raise significant doubts and

leave investors very unclear about which state the firm is in. The firm’s cash flow is then perceived as a draw from

J Bus Fin Acc. 2020;47:3–26. wileyonlinelibrary.com/journal/jbfa c

2020 John Wiley & Sons Ltd 3

4PENG ET AL.

a mixture distribution of two conditional distributions, neither of which might have very high posterior probability

(their posterior probabilities might both be 0.5), potentially leaving the cash flow much more uncertain than before

the relevant information was received.1Giventhe arrival of sufficiently stronger information, investors will eventually

become virtually certain about the firm’s state and its cash flow. Certainty can thus be reached, but typically not

monotonically.

The ‘usual’ Bayesian model, used throughout the accounting information literature, implies that (i) investors’

uncertainty is affected by the precision of the information, but not by ‘what is says’, whether favourable or

unfavourable, and (ii) uncertainty decreases monotonically with any new information. That workhorse model is

elegant and tractable for theoretical work, but is unrealistic empirically. If the market has broadly favourable prior

expectations of the firm but learns that the firm’s costs have increased and sales have decreased, such clearly

relevant accounting information should not necessarily, if ever, leave investors more certain that the firm is in

fact doing well. Johnstone (2018) provides a detailed critique of the usual Bayesian model in accounting research,

including a technical explanation of why its assumptions imply that any new information must always decrease

uncertainty.2

Incontrast, our model describes the impact of news on uncertainty as the net effect from the change in the between-

state uncertainty and the change in the within-state uncertainty. The model allows for a natural asymmetry where

good news resolves more uncertainty than bad news; however, it also allows for the possibility that disclosure can

heighten uncertainty when investors’ prior beliefs are sufficiently tight and the news is sufficiently bad. The upwards

revision in uncertainty is attributed to the increase in between-state uncertainty, which offsets the resolution of the

within-state uncertainty.These results are in stark contrast to the usual Bayesian model.

Closest to our study, Dye and Hughes (2018) describe a voluntary disclosure model that also implies a possibility

of an increase in uncertainty in response to new information. Specifically, they show that, when the manger’s infor-

mation endowment is uncertain, the perceived variance of the firm’s cash flows must go up when the manager makes

no voluntary disclosure. In their model, the absence of disclosure causes investors to guess whether the manager is

genuinely uninformed or the manager is withholding unfavorableinformation, thus giving rise to a mixture distribution

for investors’ posterior beliefs. As a result, uncertainty increases when no disclosure is made even though investors

can extractinformation from the absence of disclosure. This updating mechanism is in line with our model in the sense

that ambiguity regarding the latent state that generates the observed information can increase investor uncertainty.

While the uncertainty-increasing result in Dye and Hughes (2018) applies to a strategic disclosure setting in a risk-

averse market, our model can explain information-driven uncertainty increases in a non-strategic setting regardless

of investor preferences. In fact, Johnstone (2016) explains that the possibility of such uncertainty-increasing effects

of information is more general because the resolution of uncertainty holds only on average but not necessarily in

every case.3Our study extends Johnstone (2016) by systematically identifying a sufficient condition for the uncer-

tainty increase to arise.

Our model describes a causal asymmetric effect of disclosure on uncertainty as a consequence of just Bayesian

updating. This is a distinct effect to volatility feedback embraced in the asset pricing literature (e.g. Campbell &

Hentschel, 1992). For example, Veronesi (1999) proposes a dynamic regime-shift model that implies a negative

association between stock returns and volatility through the discount rate, suggesting that bad news predicts higher

1Our model falls into the Neururer,Papadakis, and Riedl (2016, p. 401) category of “Bayesian Learning with Increased Posterior Uncertainty” models , which the

authorsdescribe as being the most realistic.

2Larson and Resutek (2017) draw a distinction between cash flow uncertainty and uncertainty about the quality of the firm’s accounting information. We

adoptthe general Bayesian decision assumption that the market’s uncertainty about the firm’s cash flow is based on the set of all available information, which

includesinformation about the quality of the accruals and other accounting signals that are included in that set.

3Ourmodel subsumes the classical understanding that disclosure always resolves uncertainty on average (e.g. Verrecchia, 1983; Lambert, Leuz, & Verrecchia,

2007),where between-state uncertainty is notconsidered and only within-state uncertainty takes effect. Specifically, the classical disclosure models suggest

that uncertainty resolution increases in the precision of the disclosure irrespective of the content of the disclosure, and that information can never leave

investorsmore uncertain.