Linguistic Complexity in Firm Disclosures: Obfuscation or Information?

• DOI: http://doi.org/10.1111/1475-679X.12179
• Published date: 01 March 2018
• Date: 01 March 2018

DOI: 10.1111/1475-679X.12179

Journal of Accounting Research

Vol. 56 No. 1 March 2018

Printed in U.S.A.

Linguistic Complexity in Firm

Disclosures: Obfuscation

or Information?

BRIAN J. BUSHEE,

∗IAN D. GOW,

†AND DANIEL J. TAYLOR

∗

Received 6 January 2014; accepted 26 May 2017

ABSTRACT

Prior research generally interprets complex language in firms’ disclosures as

indicative of managerial obfuscation. However,complex language can also re-

flect the provision of complex information; for example, informative techni-

cal disclosure. As a consequence, linguistic complexity commingles two latent

components—obfuscation and information—that are related to information

asymmetry in opposite directions. We develop a novel empirical approach to

estimate these two latent components within the context of quarterly earnings

conference calls. We validate our estimates of these two latent components by

examining their relation to information asymmetry. Consistent with our pre-

dictions, we find that our estimate of the information component is negatively

associated with information asymmetry while our estimate of the obfusca-

tion component is positively associated with information asymmetry. Our find-

ings suggest that future research on linguistic complexity can construct more

∗The Wharton School, University of Pennsylvania; †University of Melbourne.

Accepted by Christian Leuz. We thank Lauren Cohen, Mirko Heinle, Feng Li,

Tim Loughran, Rabih Moussawi, Luis Palacios, Eddie Riedl, Robert Verrecchia, Yux-

ing Yan, two anonymous reviewers, and seminar participants at Boston University, Cu-

bist Systematic Strategies LLC, Florida International, George Washington, UC-Irvine,

the University of Minnesota Empirical Accounting Conference, University of Toronto,

and University of Zurich for useful discussions, comments, and programming advice.

We gratefully acknowledge the financial support of the Harvard Business School and

of the Wharton School. An online appendix to this paper can be downloaded at

http://research.chicagobooth.edu/arc/journal/onlineappendices.aspx.

85

Copyright C, University of Chicago on behalf of the Accounting Research Center,2017

86 B.J.BUSHEE,I.D.GOW,AND D.J.TAYLOR

powerful tests by separately examining these two latent components of lin-

guistic complexity.

JEL codes: D82; G14; G18; M41; M45

Keywords: disclosure; linguistic complexity; conference calls; information

asymmetry

1. Introduction

An extensive prior literature examines the relation between the linguistic

complexity of corporate disclosures and various firm outcomes. This liter-

ature generally interprets the use of complex language as an intentional

choice by managers to obfuscate in order to increase information process-

ing costs and delay the market reaction to news (e.g., Li [2008]). An alter-

native view is that complex language could be necessary to convey infor-

mation about the firm’s business transactions and operating strategy (e.g.,

Bloomfield [2008]). In reviewing the literature on linguistic complexity,

Loughran and McDonald [2016] conclude, “researchers face the problem

of separating the business and the document. These issues are intertwined

because the document attempts to describe the economic reality of the

business” (p. 1198).

In this paper, we show that linguistic complexity commingles two latent

components that have opposite relations to information asymmetry: an “in-

formation” component and an “obfuscation” component. The informa-

tion component represents linguistic complexity attributable to informa-

tive technical disclosure about the business. The obfuscation component

represents linguistic complexity intended to reduce the informativeness of

the disclosure. As a result of this commingling, the relation between overall

linguistic complexity and information asymmetry is theoretically ambigu-

ous. In contrast, we show that the relations between the two latent compo-

nents of linguistic complexity and information asymmetry are theoretically

unambiguous; that is, the information (obfuscation) component is nega-

tively (positively) associated with information asymmetry.

We develop an empirical approach to estimate the latent components of

linguistic complexity within the context of quarterly earnings conference

calls. The key to our approach is to identify the portion of the linguistic

complexity that represents managers’ appropriate disclosure response to

business complexity in the absence of obfuscation. We exploit the fact that

conference calls allow us to observe the linguistic complexity of an outside

party—security analysts—in the same setting as the manager’s linguistic

complexity. Prior research finds that analysts have incentives to acquire and

convey value-relevant information (Mayew [2008], Twedt and Rees [2012]),

suggesting that they are unlikely to have obfuscation incentives during the

call. As such, we use analysts’ linguistic complexity as a benchmark for es-

timating the two latent components. We then validate our approach by

LINGUISTIC COMPLEXITY IN FIRM DISCLOSURES 87

examining the relation between our estimates of the two latent components

and information asymmetry.

We first follow prior work and examine the association between over-

all managerial linguistic complexity and information asymmetry. Following

an extensive prior literature, we measure linguistic complexity using the

Gunning [1952] Fog index and information asymmetry using the Amihud

[2002] measure of illiquidity. As conference calls include both scripted pre-

sentations and more spontaneous responses to questions, we separately ex-

amine the presentation and the response portions of the call. Consistent

with prior work that suggests that the linguistic complexity of 10-Ks reflects

obfuscation, we find a positive relation between managerial Fog in the pre-

sentation and information asymmetry. However, we find a negative relation

between managerial Fog in the response and information asymmetry. While

these results suggest that the informativeness of linguistic complexity dif-

fers between scripted and spontaneous language, our analysis of the latent

components suggests a more nuanced conclusion.

Specifically, when we decompose linguistic complexity into its two latent

components, we find a negative (positive) relation between the estimated

information (obfuscation) component and information asymmetry. These

findings are similar for both the presentation and response. These results

illustrate that the relation between overall linguistic complexity and infor-

mation asymmetry is ambiguous because complex language is driven by two

components that have opposite relations to information asymmetry. In con-

trast, our decomposition shows that the relation between the components

themselves and information asymmetry is unambiguous.

Finding that the two latent components have opposite relations with in-

formation asymmetry should mitigate concerns that our estimates are bi-

ased in any one direction. For example, if our estimate of the information

component contains some obfuscation, it would bias the negative (positive)

coefficient on the information (obfuscation) component upward (down-

ward) and hence bias against finding opposite directional effects. While any

empirical estimate of an unobserved latent variable will have some amount

of noise and bias, our findings suggest that the noise and bias is sufficiently

mild that our estimates are empirical descriptive of the underlying theoret-

ical constructs.

Our results are robust to a variety of sensitivity tests. First, the results hold

(and in some cases become stronger) when we control for other language

attributes, such as number of words, industry jargon, forward-looking state-

ments, and tone. Second, the results are robust to controls for precall in-

formation asymmetry and to estimating a within-industry, within-firm, and

within-manager analysis, suggesting that time-series variation in these com-

ponents explains time-series variation in information asymmetry. Third, the

results hold under three Monte Carlo placebo tests that use randomly se-

lected nonevent days, alternative conference call days, and language from

another firm’s conference call. Fourth, we further validate our measures

by showing that the information (obfuscation) component is positively