Core Earnings Uncertainty, Dividend Change Announcements and the Reduction of Covariance Component Risks

• Published date: 01 November 2015
• Date: 01 November 2015
• DOI: http://doi.org/10.1111/jbfa.12129

Journal of Business Finance & Accounting

Journal of Business Finance & Accounting, 42(9) & (10), 1075–1120, November/December 2015, 0306-686X

doi: 10.1111/jbfa.12129

Core Earnings Uncertainty, Dividend

Change Announcements

and the Reduction of Covariance

Component Risks

STEPHEN J. DEMPSEY,DAVID M. HARRISON AND HAINAN SHENG∗

Abstract: We present evidence of two systematic market risk implications associated with

core earnings news implicit in dividend change announcements: (1) a decline in firm-market

correlation intensity, consistent with reduced investor reliance on overall market movements

to value shares, and (2) a downward shift in standard deviation of returns, consistent with

increased core earnings information precision. Decoupling these two covariance component

risk effects is important because they can offset one another at the firm level, masking unique

market influences on total systematic risk. Each is influenced by the information environment

in different ways and each is shown to incrementally explain returns in a manner consistent with

the capital asset pricing model (CAPM).

Keywords: systematic risk, CAPM, covariance, information risk, dividend changes, correlation,

variance, standard deviation, core earnings

1. INTRODUCTION

While changes in cash flow point estimates are in theory priced quickly, changes

in uncertainty concerning cash flow levels logically resonate throughout the entire

returns series until resolution takes place, when cash flow realizations occur. It follows

that the ‘durability’ of a second-moment shift is dependent upon the extent to

which new information has implications for the firm’s complete intertemporal cash

flow distribution. Information that helps resolve uncertainty about sustainable ‘core’

earnings would therefore have a more secular impact on second-moment returns than

information about transitory earnings alone. Furthermore, other things equal, such

second-moment effects would hold equivalently for positive and negative core earnings

signals.

∗The first author is from the Grossman School of Business, University of Vermont. The second author is

from the University of Central Florida. The third author is from the Rawls College of Business, Texas Tech

University. Wewish to thank an anonymous reviewer and Andrew Stark (editor) for helpful comments and

suggestions on earlier versions of this paper.

Address for correspondence: Stephen Dempsey, University of Vermont, 55 Colchester Ave,Burlington, VT

05405, United States.

e-mail: stephen.dempsey@uvm.edu.

C

2015 John Wiley & Sons Ltd 1075

1076 DEMPSEY, HARRISON AND SHENG

We use dividend change announcements as an empirically established core earnings

signal and examine daily returns data surrounding these announcements to distin-

guish their ability to separately affect (1) the correlation component of covariance

and (2) the standard deviation component of covariance. The mathematically distinct

influences of the two components are clear from the following single-factor market

model beta decomposition:

βj,M=ρj,Mσjσ−1

M(1)

where, respectively, ρand σdenote correlation and standard deviation, and subscripts

jand Mdenote firm returns and overall market returns. Although the extant

information economics literature appears to focus exclusively on variance as the

attribute of interest (or its inverse, information precision), correlation is potentially

the more harmful of the two influences on systematic risk because it is, by definition,

entirely non-diversifiable.

Irrespective of the sign and size of the dividend change, we find (1) a significant

decline in firm-market return correlation intensity, and (2) a significant downward

shift in the temporal standard deviation of daily returns. Each effect persists through-

out the postannouncement return time-series we analyze and each is pervasive over

the more than four decades of dividend changes we investigate. Although eliminating

observations with confounding earnings announcements (t=−5, ..., +5) greatly

reduces the sample size, the ameliorated second-moment effects we document are

inferentially identical to those for the full sample. We therefore conclude that the

covariance component shifts we observe are due more to core earnings information

embedded in dividend change announcements than what is contained in routine

quarterly earnings reports. Importantly, the correlation and standard deviation shifts

behave quite differently and have different economic effects. For example, nearly

half of the changes we detect in one component are associated with a shift in the

other component that runs in the opposite direction. While opposing shifts have the

potential to diminish their multiplicative impact on covariance, we nonetheless find

strong cross-sectional systematic risk reduction following both dividend increase and

dividend decrease announcements.

(i) Motivation for Separately Studying Correlation Shifts and Standard Deviation

Shifts

Ex ante correlation in the capital asset pricing model (CAPM) reflects dependency

between firm cash flows and overall economy cash flows. Whether such dependence is

real or assessed, it is investors’ expectations that govern realized returns. Consequently,

if the arrival of significant firm-specific information subsumes investors’ more na¨

ıve

reliance on general price movements to value shares, this would serve to diminish

the empirical intensity of the firm-to-market correlation. The economic intuition is

straightforward: market makers without information advantages to incent trade would

have no reason to price shares at values other than those implied by a mechanistic

pricing model (overall market returns making up at least one key factor). Such

heuristics underlie nearly all program trading. At the other extreme, if investors are

omniscient about a particular stock (i.e., information is infinitely precise), this would

render the overall market irrelevant as a risk factor and the assessed firm-to-market

C

2015 John Wiley & Sons Ltd

DIVIDEND CHANGES, REDUCTION OF COVARIANCE COMPONENT RISKS 1077

correlation would be zero. For all levels of informed trading between absolute

ignorance and complete clairvoyance, information-conditioned prices depart from –

i.e., are less reliant upon – market-based priors.1

In a classical CAPM setting every portfolio of any size, including N=1, sits squarely

on the ex ante security market line. As Roll (1988) argues, the ex post implication

should be high characteristic line R2’s, approaching unity once hindsight influences

are accounted for (e.g., industry influences and firm-specific news releases). Instead,

he finds R2’s are generally low (around 35% with monthly data and 20% with

daily data), which, according to the more optimistic of his two interpretations, is

consistent with private information being imparted into stock prices in a competitive,

punctuated fashion.2Consequently, much of what appears on the chart as white noise

at the individual firm level is a likely outcome of private information-based trading.

The more tightly realized returns collect around the characteristic line, the higher

the R2coefficient, and the higher the indicated percentage of systematic risk to total

risk. Inasmuch as R2is the squared correlation coefficient, it follows that information

arrival causes correlation to drop in empirical intensity. According to our correlation

hypothesis to the extent information innovation lessens investors’ more na¨

ıve reliance

on overall market movements to value the firm, there is a suggested reduction in

systematic risk.

We predict reduced correlation intensity regardless of the nature of firm-specific

news, good or bad. The directional change in standard deviation, however, is likely

to be a function of both the quality and the content of information. With respect to

information quality, Lambert et al. (2007) analytically demonstrate in a CAPM setting

that improved disclosure about the firm leads to an unambiguous decrease in its

assessed cash flow (and return) variance. The conditional covariance of returns with

the market is also shown to be reduced, proportional to the amount of signal noise

relative to total assessed cash flow variance. By contrast, as distinguished from the

precision of information, the fundamental message conveyed by information clearly

has the ability to increase, decrease or leave unchanged investor uncertainty about

future cash flows. Thus, better information may communicate more rather than less

underlying stochastic cash flow uncertainty (e.g., Johnstone, 2015a, 2015b).3In the

1 To embellish a classic analogy, suppose one seeks to shop for a used car online. Without any specific

information beyond the make, model, year, and mileage, the shopper’spriors leave no reason to presume a

price other than the average market ‘blue book’ value. As more is learned about the particular specifications

of the car that has caught the shopper’s interest, this refines value to incorporate those features as well.

Such information alone is insufficient to prompt trade, however, due to the ‘lemons’ problem (Akerlof,

1970). More specific fundamental information is required to motivate a trading interest because the risk

in the transaction cannot be diversified away. The prospective buyer therefore hires a mechanic to get

under the hood. On the basis of this investigation, significant specific information about the car’s true

condition is ascertained. While such probing might lead to the conclusion that the car is indeed just average

(and therefore the ‘blue book’ market value is appropriate), as deeper fundamental knowledge is gained

it becomes mathematically less likely that the average does in fact apply. This is because the area over any

point on a continuous distribution is zero, including the expected value. Consequently, for any risky asset,

the probability that the unconditional ex ante expectation describes true value becomes infinitesimally small

as additional information is gathered to refine the estimate.

2 Roll’s pessimistic interpretation is that the noise is due to market ‘frenzy.’ Durnev et al. (2003) present

evidence in support of the optimistic interpretation. Specifically, they find firms with low R2’s exhibit a

higher association between current returns and future earnings than firms with high R2’s, which, they

conclude, is evidence in favor of market efficiency.

3 Lambert et al.’s analysis (2007, pp. 395–6) utilizes a classic Bayesian model in which new information

unequivocally reduces the assessed variance of jointly-normal posterior distributions. Johnstone (2015a,

2015b) relaxes this assumption to accommodate the reasonableness of higher rather than lower variance

C

2015 John Wiley & Sons Ltd