The Structure of Voluntary Disclosure Narratives: Evidence from Tone Dispersion

• DOI: http://doi.org/10.1111/1475-679X.12072
• Date: 01 May 2015
• Author: KRISTIAN D. ALLEE,MATTHEW D. DEANGELIS
• Published date: 01 May 2015

DOI: 10.1111/1475-679X.12072

Journal of Accounting Research

Vol. 53 No. 2 May 2015

Printed in U.S.A.

The Structure of Voluntary

Disclosure Narratives: Evidence

from Tone Dispersion

KRISTIAN D. ALLEE

∗AND MATTHEW D. DEANGELIS

†

Received 6 January 2014; accepted 8 January 2015

ABSTRACT

We examine tone dispersion, or the degree to which tone words are spread

evenly within a narrative, to evaluate whether narrative structure provides in-

sight into managers’ voluntary disclosures and users’ responses to those dis-

closures. We find that tone dispersion is associated with current aggregate and

disaggregated performance and future performance, managers’ financial re-

porting decisions, and managers’ incentives and actions to manage percep-

tions. Furthermore, we find that tone dispersion is associated with analysts’

and investors’ responses to conference call narratives. Our results suggest that

tone dispersion both reflects and affects the information that managers con-

vey through their narratives.

JEL codes: G30; M41; M49

Keywords: codes: voluntary disclosure; conference calls; narrative struc-

ture; tone; analysts

∗University of Wisconsin–Madison; †Georgia State University.

Accepted by Douglas Skinner. Weare grateful for constructive comments by Mary Billings,

Matt Bloomfield, Ted Christensen, Fabio Gaertner, Susannah Gustafson, Alina Lerman,

Ken Merkley, Karen Nelson, Holly Skaife, Francis Smart, Tyler Thomas, Jim Wahlen, Terry

Warfield, an anonymous reviewer for the AAA Annual Meeting, and workshop participants

at Michigan State University, the University of Arkansas, and the University of Wisconsin–

Madison. We also thank an anonymous reviewer and participants at the 2014 Journal of Ac-

counting Research Conference. We appreciate the valuable research assistance of Catherine

Guffey and would like to thank the Wisconsin School of Business, the Eli Broad College of

Business, and the Robinson College of Business for financial support.

241

Copyright C, University of Chicago on behalf of the Accounting Research Center,2015

242 K.D.ALLEE AND M.D.DEANGELIS

1. Introduction

There is now a considerable body of research in accounting and finance

examining the role of linguistic tone in firm disclosures. These studies

often use the “bag-of-words” approach to quantify tone (Henry [2008],

Tetlock [2007], Loughran and McDonald [2011]), ignoring the underly-

ing narrative structure in which tone words are used. However, managers

do not present “loose bits of information” (Spivey [1990]), but rather a

disclosure narrative that conveys and reinforces their interpretation of cur-

rent performance and its implications for future firm performance (Sedor

[2002]). The structure of this narrative reflects managers’ organization of

interrelated ideas and helps users interpret and comprehend their message

(Strunk and White [1979], Kintsch and Yarbrough [1982], Sedor [2002]).

In this paper, we develop and use a parsimonious measure of narrative

structure, tone dispersion, to investigate the role of narrative structure in

financial reporting.

We define tone dispersion as the degree to which tone is spread evenly

throughout a disclosure narrative. Rhetoric and communication theory

(Kintsch and Yarbrough [1982], Spivey [1990]) and related applications in

computational linguistics (Kostoff, Eberhart, and Toothman [1997]) sug-

gest that words are less likely to be topically related as the distance between

them in the narrative increases. Since managers use narratives to “tell a

story” about firm performance (Sedor [2002]), we expect that a more even

distribution of tone (higher tone dispersion) throughout the narrative re-

flects a portrayal of good or bad news as pervasive, while a less even distri-

bution (lower tone dispersion) isolates the news to fewer components of

performance. Given prior research on the valuation implications of news

in earnings components (Ou and Penman [1989], Fairfield, Sweeney, and

Yohn [1996]) and sensitivity of individuals to how information is presented

(Lee, Petroni, and Shen [2006], Bloomfield et al. [2014]), we posit that

tone dispersion is incrementally informative to the level of tone because it

captures the extent to which good or bad news is spread across different

sections of the narrative.

We examine tone dispersion in earnings conference calls, an increas-

ingly important and value-relevant channel of corporate communication

(Frankel, Johnson, and Skinner [1999], Bushee, Matsumoto, and Miller

[2003]). Prior research finds that analysts and investors make decisions

regarding their earnings estimates based on what managers choose to

discuss (Black et al. [2013]), as well as how they say it, documenting signif-

icant responses to word choice (Price et al. [2012]) and vocal cues (Mayew

and Venkatachalam [2012]). Matsumoto, Pronk, and Roelofsen [2011] also

find that the question-and-answer (Q&A) portion of the conference call

elicits significant incremental information from managers and analysts.

Since the Q&A occurs immediately after managers’ prepared remarks, ex-

amining conference calls provides an opportunity to investigate the associ-

ation between managers’ narrative structure and analysts’ questions.

THE STRUCTURE OF VOLUNTARY DISCLOSURE NARRATIVES 243

We use a measure of linguistic dispersion from the computational linguis-

tics literature, average reduced frequency (ARF), to measure the degree to

which tone words are evenly distributed throughout the prepared remarks

section of conference call transcripts. We find significant variation in tone

dispersion across firms but little time-series variation within firms, indicat-

ing a significant firm- or manager-specific component to tone structure. In

examining the relationship between the level of tone and its dispersion, we

find that only a small portion of this variation is mechanically related to the

level of tone. Together, these findings suggest that managers deliberately

structure tone as part of their overall narrative.

We further investigate whether narrative structure complements other

aspects of managers’ financial reporting. We provide evidence of associ-

ations between tone dispersion and three characteristics of firm disclo-

sures. First, we find that tone dispersion is significantly associated with firm

performance (e.g., changes in performance margins and missing analyst’s

expectations). Second, consistent with prior evidence that managers use

multiple strategies to manage users’ expectations, including disclosure tim-

ing (Kasznik and Lev [1995], Libby and Tan [1999]) and pro forma earn-

ings (McVay [2006], Curtis, McVay, and Whipple [2013]), we find that

tone dispersion is associated with managers’ financial reporting choices

(e.g., earnings guidance news in the conference call, the magnitude of

special items and discussions of pro forma earnings). Third, we find evi-

dence of strategic motivations for tone dispersion. We find that tone dis-

persion is associated with managers’ incentives to manage perceptions

when firm performance is unusually high or low (Graham, Harvey, and

Rajgopal [2005]), classification shifting (McVay [2006]), and the readabil-

ity of disclosure narratives (Li [2008]). Overall, our findings suggest that

tone dispersion both reflects and affects the information that managers

convey through their narratives.

We next examine whether tone dispersion is associated with analysts’ and

investors’ responses. We document that analysts ask more positive (nega-

tive) questions and the market responds more positively (negatively) when

positive (negative) tone is more dispersed in the prepared remarks section

of the conference call. Additionally, we document that the market responds

strongly to the tone of analyst questions. Since the tone of analyst questions

is also strongly associated with tone dispersion, this finding suggests that

the association between tone dispersion and market reaction is partially

mediated by analyst response.

This study makes several contributions. First, our study unifies two

strands of accounting and finance literature examining word choice and

linguistic structure in firm disclosures. Whereas prior research uses word

counts to measure information content and word and sentence structure

to measure the costs of analyzing that information (e.g., Li [2008], Lehavy,

Li, and Merkley [2011], Lee [2012]), we examine linguistic structure across

the entirety of a disclosure narrative. In addition, despite concern that the

so-called “bag-of-words” approach to textual analysis ignores the context in