University of Southern California Dissertations and Theses

Essays on the firm and stakeholders relationships: evidence from mergers & acquisitions and labor negotiations

(USC Thesis Other)

Essays on the firm and stakeholders relationships: evidence from mergers & acquisitions and labor negotiations

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Essays on the Firm and Stakeholders Relationships:

Evidence from Mergers & Acquisitions and Labor Negotiations

by

Sunny (Seung Yeon) Yoo

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BUSINESS ADMINISTRATION)

August 2021

Copyright 2021 Sunny (Seung Yeon) Yoo
ii

Acknowledgments
I am grateful to my advisor, Kenneth Ahern, as well as Odilon Camara, Tom Chang,
Patricia Dechow, Gerard Hoberg, Arthur Korteweg, Yaron Levi, Kai Li, Feng Mai, John
Matsusaka, Ekaterina Neretina, Oguzhan Ozbas, João Ramos, Rui Shen, Regina
Wittenberg Moerman, Xinyan Yan, Irene Yi, and Ben Zhang for their thoughtful advice
and in-depth discussions. I also thank Kai Li for providing data on cultural value scores
and Irene Yi for providing data on contract expirations. I am indebted to the seminar
participants at the USC Marshall, the Transatlantic Doctoral Conference, the AFA Ph.D.
poster session, KU Leuven, Nanjing University, Southwestern University of Finance and
Economics, Hong Kong Baptist University, ESCP Business School, and Chinese
University of Hong Kong (Shenzhen) for valuable comments. Last but not least, I would
like to say a heartfelt thank you to my family for their endless love, support, and
inspiration.

iii

Table of Contents

Acknowledgments ii

List Of Tables v

List Of Figures vi

Abstract vii

Chapter 1: Corporate Culture in Mergers and Acquisitions: Evidence from CEO
Letters to Shareholders
1
I. Introduction . . . . . . . . . . . . . . . 1
II. Data Source and Cultural Similarity Measure . . . . . . . . . . . . . . . 9
A. CEO Letter and Similarity Measure . . . . . . . . . . . . . . . 9
B. Similarity Measure using LDA Analysis . . . . . . . . . . . . . . . 12
C. Merger Data . . . . . . . . . . . . . . . 14
D. Other Variables . . . . . . . . . . . . . . . 15
E. Summary Statistics . . . . . . . . . . . . . . . 18
F. LDA Topic Analysis . . . . . . . . . . . . . . . 20
III. Corporate Culture and Merger Volume . . . . . . . . . . . . . . . 25
A. Corporate Culture Measured by Cosine Similarity . . . . . . . . . . . . . . . 25
B. Corporate Culture Measured by LDA Analysis . . . . . . . . . . . . . . . 26
C. Potential Mechanism . . . . . . . . . . . . . . . 27
IV. Cultural Integration and Post-Merger Synergy
Realization
. . . . . . . . . . . . . . . 30
A. Post-Merger Performance . . . . . . . . . . . . . . . 30
B. Post-Merger Divestiture . . . . . . . . . . . . . . . 32
V. Corporate Culture and Merger Gain . . . . . . . . . . . . . . . 32
A. Buyer's Announcement Return . . . . . . . . . . . . . . . 33
B. Combined Announcement Return . . . . . . . . . . . . . . . 34
C. Potential Mechanism . . . . . . . . . . . . . . . 34
VI. Robustness Checks . . . . . . . . . . . . . . . 37
A. LDA Analysis using Different Topic Numbers . . . . . . . . . . . . . . . 38
B. Merger Volume with Li, Mai, Shen, and Yan
(2020) Measure
. . . . . . . . . . . . . . . 38
C. CEO Change and Corporate Culture . . . . . . . . . . . . . . . 39
VII. Conclusion . . . . . . . . . . . . . . . 42

Chapter 2: Do Firms Leave Workers in the Dark Before Wage Negotiations? 76
I. Introduction . . . . . . . . . . . . . . . . 76
iv

II. Information Advantage and Wage Bargaining . . . . . . . . . . . . . . . . 86
III. Institutional Background and Identification Strategy . . . . . . . . . . . . . . . . 88
A. Confidential Treatment Order . . . . . . . . . . . . . . . . 88
B. Identification Strategy . . . . . . . . . . . . . . . . 90
IV. Data . . . . . . . . . . . . . . . . 93
A. Contract Expiration . . . . . . . . . . . . . . . . 93
B. Redacted Disclosure . . . . . . . . . . . . . . . . 94
C. Other Covariates . . . . . . . . . . . . . . . . 96
V. Summary Statistics . . . . . . . . . . . . . . . . 97
A. Collective Bargaining Contract . . . . . . . . . . . . . . . . 98
B. Industry Distribution . . . . . . . . . . . . . . . . 98
C. Time-Series Distribution . . . . . . . . . . . . . . . . 98
D. Frequency and Types of Redacted Contracts . . . . . . . . . . . . . . . . 99
VI. Effect of Labor Negotiations on Firm Disclosure . . . . . . . . . . . . . . . . 100
A. Contract Expiration and Redaction . . . . . . . . . . . . . . . . 100
B. Cross-Sectional Factor Analysis . . . . . . . . . . . . . . . . 103
VII. Effect of Strategic Disclosure on Firm Performance . . . . . . . . . . . . . . . . 109
VII. Robustness Tests and Additional Analyses . . . . . . . . . . . . . . . . 111
A. Substitution Effect between Strategic Disclosure
and Liquidity Management
. . . . . . . . . . . . . . . . 111
B. Endogenous Financial Constraint . . . . . . . . . . . . . . . . 113
C. Cross-Sectional Factor Analysis using
Alternative Proxies
. . . . . . . . . . . . . . . . 114
IX. Conclusion . . . . . . . . . . . . . . . . 115

References 136

v

List Of Tables

Chapter 1: Corporate Culture in Mergers and Acquisitions: Evidence from CEO
Letters to Shareholders

I. Summary Statistics . . . . . . . . . . . 55
II. Bayesian Topic Modeling using Latent Dirichlet allocation
(LDA)
. . . . . . . . . . . 58
III. LDA Topic Culture and CEO & Firm Characteristics . . . . . . . . . . . 62
IV. Similarity in CEO Letters and Likelihood of Being
Targeted
. . . . . . . . . . . 64
V. LDA Topic Similarity and Likelihood of Being Targeted . . . . . . . . . . . 65
VI. Merger Volume and Mechanism Test . . . . . . . . . . . 67
VII. Change of Similarity in CEO Letters and Post-Merger
Performance
. . . . . . . . . . . 68
VIII. Change of Similarity in CEO Letters and Post-Merger
Divestiture
. . . . . . . . . . . 70
IX. Similarity in CEO Letters and Buyer's Announcement
Return
. . . . . . . . . . . 71
X. Buyer's Announcement Return and Mechanism Test . . . . . . . . . . . 72
XI. Change of Similarity and CEO Changes . . . . . . . . . . . 74
XII. Cultural Change around CEO Change . . . . . . . . . . . 75

Chapter 2: Do Firms Leave Workers in the Dark Before Wage Negotiations?
I. Summary Statistics . . . . . . . . . . . 124
II. Descriptive Information on CTO Practice . . . . . . . . . . . 125
III. Contract Expiration and Redaction . . . . . . . . . . . 126
IV. Cross-sectional Analysis . . . . . . . . . . . 127
V. Redaction and Ex-Post Firm Performance . . . . . . . . . . . 132
VI. Strategic Disclosure and Strategic Liquidity Management . . . . . . . . . . . 133
VII. Redaction Tendency and Lagged Financial Constraint . . . . . . . . . . . 134

vi

List Of Figures

Chapter 1: Corporate Culture in Mergers and Acquisitions: Evidence from CEO
Letters to Shareholders

1. Histogram of Number of Words in CEO Letter . . . . . . . . . . . 51
2. LDA Analysis Illustration . . . . . . . . . . . 52
3. Word Cloud . . . . . . . . . . . 53
4. Similarity in CEO Letters and Buyer's Announcement
Return
. . . . . . . . . . . 54

Chapter 2: Do Firms Leave Workers in the Dark Before Wage Negotiations?
1. Changes in Redaction Tendency around Contract
Expiration
. . . . . . . . . . . 121
2. Changes in Number of Exhibits around Contract Expiration . . . . . . . . . . . 122
3. Changes in Types of Redacted Agreements around Contract
Expiration
. . . . . . . . . . . 123

vii

Abstract
This thesis consists of two essays that study the firm and stakeholders relationships.
Chapter 1 examines the role of corporate culture for mergers and acquisitions. To
quantify corporate culture, I run a textual analysis of the language used in CEOs' annual
letters. This analysis categorizes firms into three different corporate cultures:
collaborative, innovative, and customer-centric. Using the novel measure of corporate
culture, I find that firms with more similar corporate cultures are more likely to merge.
Second, buyers' announcement returns are higher if targets have more similar corporate
cultures. Finally, the cultural integration of two merged firms is positively related to post-
merger performance and is negatively associated with ex post divestiture. In sum, this
paper shows that cultural differences have meaningful impacts on mergers. Chapter 2
examines managers' strategic use of financial disclosure in labor negotiations. Using the
exogenous expiration date of collective bargaining contracts, I find that when wage
negotiations are imminent, firms strategically redact information about material
agreements. Strategic redaction is pronounced when unions cannot accurately predict
firms' prospects, when firms have low growth opportunities, when liquidity is less
constrained, and when the estimated cost of a work stoppage is low. These results suggest
that firms strategically withhold information to balance the costs and benefits of
information asymmetry. Consistent with this interpretation, strategic disclosure is
statistically uncorrelated to ex post performance.
Chapter 1
Corporate Culture in Mergers and Acquisitions:
Evidence from CEO Letters to Shareholders
I. Introduction
This paper proposes a new measure of corporate culture and uses it to study
mergers and acquisitions. Corporate culture is widely believed to be important
for corporate performance and merger success. Corporate culture can be defined
as the shared assumptions, values, and beliefs that help employees understand
which behaviors are appropriate (Zingales (2015)). According to a survey of top
executives, culture is one of the top three factors that affect firm value (Graham,
Grennan, Harvey, and Rajgopal (2016, 2017))
A central challenge for research on corporate culture is that corporate culture
is not directly observable. Theoretically, corporate culture can be understood as
an incomplete contract between the firm and its employees (Gorton and Zentefis
(2020)). Employment contracts may explicitly discipline employees with written
agreements but cannot specify all possible contingencies due to contracting costs.
Corporate culture helps resolve this difficulty by establishing a general rule for
workers’ appropriate actions. It can also help people understand what goals others
1
in the organization are pursuing so that they can work effectively. However, since
this shared assumption is not written in formal documents, outsiders cannot easily
quantify corporate culture.
Corporate culture is likely to be a multidimensional concept, impacting every-
thing from overall strategy to a company’s treatment of employees and customers.
This paper advances the idea that one aspect of culture can be inferred from the
language used in Chief Executive Officers’ (CEOs) letters in their annual reports.
One reason to believe that a CEO’s letter might reveal a company’s culture is be-
cause it captures top management’s view on how to run the company (Graham
et al. (2016)). While the shared experience of group members creates culture, the
leader is arguably the most influential figure in determining corporate culture (Cut-
ler (2004)). She can attempt to reshape the culture by building consensus around
a shared vision among the group members. CEO letters capture the CEO’s view,
namely “tone at the top,” which most significantly influences the current corporate
culture (Guiso, Sapienza, and Zingales (2015)). Another reason is that CEO letters
are a written representation of a CEO who authorizes the document and has legal
responsibility for the contents. A CEO shares her thoughts on the firm’s perfor-
mance as well as the values, attitudes, or mental models of the management team.
Even if the letter contains some legal boilerplate, the CEO can convey meaningful
messages on the culture or her personality in unconscious ways (Aktas, De Bodt,
Bollaert, and Roll (2016)). Third, since CEOs issue letters annually, the letters
capture yearly changes in the top management’s mindset.
2
The primary hypothesis I test is whether cultural similarity can provide an in-
cremental explanation on merger match and merger success. Using the novel mea-
sure of corporate culture based on CEO letters, I explore the implications of cor-
porate culture in M&A transactions among the public U.S. companies during 2004
- 2016. Unsuccessful M&A outcomes are often attributed to the difference in the
cultures of the two combined firms. For example, analysts attributed the alleged
failure of Amazon’s $13.7 billion acquisition of Whole Foods Market in 2018 to
misalignment of Amazon’s rigid culture with Whole Foods Market’s flexible cul-
ture.
1
Despite the importance of the topic to academics and practitioners, there is a
dearth of empirical evidence on the role of corporate culture in the merger market
(Zingales (2015)).
I collect annual reports from four different data sources: Mergent Archives,
ProQuest Historical Annual Reports, D&B Hoovers, and internet resources, in-
cluding the website AnnualReports.com. Since annual reports do not have a con-
sistent form across firms and even within a firm across different years, I manually
locate and compile the CEO letters from each annual report.
I quantify cultural differences between acquirers and targets in two ways. First,
I use cosine similarity of the vector representation of the text in each firm’s CEO
letters. Second, I implement a Bayesian topic analysis, called Latent Dirichlet
Analysis (LDA) model. LDA is a machine learning algorithm and does not re-
quire much researcher-induced priors or bias.
2
The LDA model assumes that a
1
See the news article ”Amazon vs. Whole Foods: When Cultures Collide (Harvard Business Research for Business Leaders,
Michael Blanding, 2018)”
2
LDA analysis has been used in many corporate finance papers, including Israelsen (2014), Hoberg and Lewis (2017), Hanley
and Hoberg (2019), Calomiris and Mamaysky (2019), Lopez-Lira (2019), Bellstam, Bhagat, and Cookson (2019), and Lowry,
3
document is generated from latent distributions over a collection of words, de-
pending on the topics it delivers. LDA helps identify topics by flagging groups of
words that appear in the same context. I find that the topics in CEO letters can
be divided into three distinct groups: collaborative firms, innovative firms, and
customer-centric firms. These three groups suggest three types of cultures.
To achieve an ideal laboratory setting to explore the causal implication of cul-
tural aspects on merger activities, one should have randomly assigned corporate
culture or merger match. In my analysis, however, neither corporate culture nor
merger activity is random. One concern about measuring corporate culture using
CEO letters is that the letters may capture firm-specific attributes, such as product
types or industry characteristics, rather than corporate culture. I try to alleviate
this concern in several ways. First, I control for product relatedness and industry
of the buyer and the target using text-based measures developed by previous lit-
erature (Hoberg and Phillips (2010, 2016); Fr´ esard, Hoberg, and Phillips (2019)).
Second, I exclude words mentioned in the business description of a firm’s 10-K,
which describe the firm’s product or industry.
To validate whether the LDA topic classification captures corporate culture,
I correlate my LDA measures with various firm specifications. The correlation
analysis finds that innovative firms hire younger CEOs, pay higher CEO compen-
sation, and involve in more R&D activities. And customer-centric firms are scored
higher in customer satisfaction scores.
Using my new measure of corporate culture, the first question I ask is whether
Michaely, and V olkova (2020).
4
corporate culture has any explanatory power on merger matches. Business practi-
tioners believe that cultural fit is an important factor when they consider potential
targets (Graham et al. (2016, 2017)). The cultural distance hypothesis argues that
the cost of contracts among two groups is positively related to cultural differences
(Hofstede (1980)).
To test this hypothesis, I match each target to a pseudo-target in the same pri-
mary two-digit SIC industry with similar assets, sales, and market value to allevi-
ate the concern that omitted variables drive the association between culture and the
likelihood of merger. Using this matched sample, I find that two firms are more
likely to merge if their CEO letters are more similar. A one standard deviation
increase in similarity increased the likelihood of a merger by 9.34 percent to 13.56
percent. Given the unconditional probability of being targeted in my matched sam-
ple is 50 percent, the estimated coefficients on cultural similarity is economically
meaningful. The alternative similarity measure based on LDA analysis delivers
consistent results. Thus, the empirical results support the hypothesis that greater
cultural distance reduces the likelihood of merger.
I also investigate mitigating factors which are motivated by incomplete contract
theory. The theory predicts that the target’s labor intensity makes cultural align-
ment a more crucial consideration in M&A decisions. As an incomplete contract,
corporate culture helps the employees choose the best action in unforeseen con-
tingencies. It is more likely to face these contingent situations when the target’s
business involves employees’ judgment. Empirical results confirm this prediction.
5
The positive association of cultural similarity and merger match is stronger when
the target has higher labor intensity. Second, if the potential target has multiple
divisions, it is hard for the buyer’s management team to control all of these distinct
segments. Therefore, the existing corporate culture is likely to be more important.
Consistent with this idea, I find that the interaction between cultural similarity
and the number of segments is positively associated with the likelihood of merger
match.
Next, I investigate the relation between post-merger performance and post-
acculturation. Like pre-acquisition characteristics, post-acquisition integration is
critical in determining post-merger performance (Agrawal, Jaffe, and Mandelker
(1992)). If two cultures are misaligned, the combined firms’ employees may not
coordinate well due to different assumptions, values, and beliefs on the best ways
of conducting business. Mergers between the firms with similar CEO letters will
have better future performance, by improving productivity, securing human cap-
ital, and reducing divestiture probability. To test this prediction, I quantify post-
merger integration by the degree to which the acquiring CEO’s letter becomes sim-
ilar to the target CEO’s pre-merger letter. This integration measure is positively
associated with post-merger performance, and a combined firm with a higher in-
tegration score is less likely to divest the acquisition.
In the final set of analysis, I test whether merger announcement returns are re-
lated to cultural similarity. If two organizations’ cultures are misaligned, it may
prevent the merged entity from realizing synergies. Corporate executives say that
6
they would discount the acquisition premium of a culturally-disparate target by
10 percent to 30 percent (Graham et al. (2016, 2017)). Because culture is an un-
written value shared by insiders, it is not clear a priori how well outsiders can
evaluate it and whether they in fact price it into merger transactions. If the merger
market optimally chooses two firms to combine and these two firms design the
merger contract in an equilibrium, I should not observe any association between
the similarity measure and the announcement return. In the real world, any po-
tential biases, such as information asymmetry or agency conflicts, can generate
systematic associations.
I find that buyers’ announcement returns are positively associated with cultural
similarity between the buyer and the target. This means that the capital market
discounts mergers of culturally disparate firms. The positive association between
cultural similarity and the market return is pronounced when a buyer faces severe
information asymmetry towards a target. If information asymmetry is severe be-
tween the buyer and the target, the buyer cannot precisely comprehend the target’s
culture. While the imprecise evaluation can lead to both overpayment and under-
payment in the merger premium, the target has an incentive to correct the mispric-
ing only when the acquirer underpays the premium. As a result, the information
asymmetry leads to an overpayment to the culturally misaligned target. Since the
similarity measure does not provide an additional explanation on the combined
announcement returns, the contractual parties seem to play a zero-sum game. To
sum up, although an acquirer may choose a target and design the merger contracts
7
after considering the cultural aspects, the acquiring firm cannot fully incorporate
the target’s cultural difference and ends up overpaying the premium, attenuating
its shareholder value.
This paper’s central contribution is to provide a novel approach to measure
time-varying firm-level corporate culture. Prior work uses other proxies to mea-
sure corporate culture, including interviews (Graham et al. (2016, 2017)), Glass-
door employee reviews (Grennan (2019)), earnings calls (Li, Mai, Shen, and Yan
(2020)), corporate social responsibility scores (Bereskin, Byun, Officer, and Oh
(2018)), and nationality (Ahern, Daminelli, and Fracassi (2015)). By offering an
annual snapshot of CEO views on her company, CEO letters can capture an annual
shift in attitude, value, or mental model. The time-varying measurement helps me
identify post-merger acculturation. Also, this paper is one of the first works to
show that corporate culture is related to merger outcomes in a meaningful way.
My work can adds beyond the existing literature in corporate culture, includ-
ing the most closely related paper by Li, Mai, Shen, and Yan (2020). First, we
use different methods in a different context. Li et al. (2020) implement a word
embedding model and analyze what top executives elaborate in their discussion
with analysts in earnings conference calls. In comparison, I exploit CEO letters
to shareholders. There are at least three distinctions between conference calls
and CEO letters: audience, extemporaneousness, and main objective. The pri-
mary goal of conference calls is to help the analysts form accurate predictions
on the firm’s capital market performance. Due to the interactive nature, the dis-
8
cussion during conference calls is steered by analysts to address their questions.
In contrast, CEO letters to shareholders have a broader audience and are scripted
in advance. CEOs provide comprehensive information they want to emphasize,
beyond financial performance, such as ethical values and shared goals. Second,
while Li et al. (2020) also study mergers among other firm policies, I provide more
comprehensive empirical evidence on M&A outcomes, including merger match,
market reaction, and post-merger performance. In general, I view our papers as
complementary. Both papers show that different communication mechanisms can
provide insights into manager type and more generally corporate culture.
This study also speaks to the literature on M&A transactions. Prior research
illustrates factors for successful merger activities or the potential synergy. They
include national culture (Ahern et al. (2015)), information asymmetry (Moeller,
Schlingemann, and Stulz (2007)), and product types (Hoberg and Phillips (2010,
2018); Fr´ esard et al. (2019)). Existing research also shows that CEO characteris-
tics can play a role in the takeover process, such as CEO narcissism (Aktas et al.
(2016)). In this paper, I demonstrate that the similarity in CEO letters can be
another important factor predicting merger success.
II. Data Source and Cultural Similarity Measure
A. CEO Letter and Similarity Measure
The term annual reports and 10-K filings are often assumed to be interchange-
able concepts. However, they are distinct from each other. To illustrate the differ-
9
ence, appendix B provides an example of 10-K filing and annual report of AT&T
Incorporation for the fiscal year 2016. A publicly-traded company files its annual
financial performance to the U.S. Securities and Exchange Commission (SEC) us-
ing 10-K reports. In contrast, annual reports are for a public firm to describe its
operations and financial conditions to its shareholders. Compared to 10-Ks, an-
nual reports provide more visual assistance, using graphics or photos. Although
10-Ks provide the most detailed information on a firm’s financial condition, annual
reports can be a flexible medium for managers to communicate with the sharehold-
ers. They also include some information that 10-Ks do not contain.
3
Under the
proxy rules, firms must send annual reports to their shareholders when they hold
annual meetings to elect directors. If they may choose to use their 10-Ks in lieu of
annual reports, there will be no separate annual report.
There are at least three challenges in retrieving CEO letters of merger firms.
First, while the SEC’s EDGAR database makes 10-Ks publicly available, the SEC
does not store annual reports on EDGAR. Second, it is especially hard to find
annual reports and CEO letters for merger targets. Once they are acquired, they
do not have a separate web presence. Third, even if there are some proprietary
data sources for annual reports, any of these sources do not provide a complete list
of reports. Bearing these difficulties in mind, I try to expand the sample size by
complementing four different data sources: Mergent Archives, ProQuest Histor-
ical Annual Reports, D&B Hoovers, and internet resources, including a website
AnnualReports.com.
3
See the description in the SEC website (”How to Read a 10-K/10-Q”).
10
I measure the corporate culture using CEO letters in annual reports. To be spe-
cific, I assume that the CEO letters in the year (t 1) reveal the target’s and the
buyer’s corporate culture in the merger yeart. Annual reports do not have a con-
sistent form across firms and even within a firm across different years. Therefore,
once the annual reports are collected, I manually locate and compile the CEO let-
ters from each annual report. I then parse the words in the letters and remove num-
bers, punctuation, symbols, and stopwords
4
. Next, the stemming process groups
the words with a similar underlying meaning into a root form. For example, the
words, ‘ran,’ ‘run,’ ‘runs,’ and ‘running,’ are grouped into the stem word ‘run.’ Af-
ter this cleaning process, the CEO letters of the buyers, the targets, and the control
firms display 46,058 unique words. The histogram in figure 1 displays the number
of unique words in CEO letters. The mean (median) number of unique words is
407 (377). Figure 1 is skewed to the right, with the maximum number of words
2,176.
Following Loughran and McDonald (2011), I then calculate the weight of each
stem word using the term frequency-inverse document frequency (TF-IDF) ap-
proach. TF-IDF approach discounts words that frequently appear across differ-
ent documents, assuming those words do not provide unique semantic meaning.
When the words exclusively appear in a specific firm’s CEO letter but do not in
most other letters, it assigns a high weight for those words.
Based on the words used by each firm’s CEO letter, I calculate pairwise cosine
4
Stopwords are a set of commonly used words in any language, which do not convey semantic information. In English, examples
include ‘a,’ ‘an,’ ‘the,’ and ‘they.’
11
similarity scores for the acquirer and the (pseudo-)target in a given merger. For a
given merger transactioni, suppose the buyer’s CEO letter has a word vector ofB
i
and the (pseudo-)target’s CEO letter has a vector ofT
i
, which are both weighted
by TF-IDF approach. The cosine similarity of the paired firms can be calculated as
B
i
T
i
jjB
i
jjjjT
i
jj
=
P
n
j=1
b
ij
t
ij
q
P
n
j=1
b
2
ij
q
P
n
j=1
t
2
ij
, where n represents the number of distinct
words andb (t) is a vector element ofB (T ). The similarity measure ranges from
zero to one: zero similarity implies no overlapping word in the pair. When two
documents have the same word frequencies, the similarity will be one.
One concern in my attempt to measure the corporate culture using CEO letters
is that they also depict other firm specifications, such as product or industry char-
acteristics. To alleviate this concern, most of the empirical specifications exclude
the words mentioned in the firm’s 10-K product description. This process helps
me focus on the words which are exclusively used in the CEO letters. I collect the
words in the 10-K product descriptions from the SEC EDGAR website.
B. Similarity Measure using LDA Analysis
The Bayesian topic analysis, called the LDA model, helps me classify firms into
different groups of the corporate culture. The LDA model is one of topic model-
ing methodologies that has been adopted in the finance literature using textual
analysis (Israelsen (2014); Hoberg and Lewis (2017); Hanley and Hoberg (2019);
Calomiris and Mamaysky (2019); Lopez-Lira (2019); Bellstam et al. (2019); Lowry
et al. (2020)). It assumes that a document is generated from latent distributions
12
over a collection of words, depending on the topics it delivers. Since a researcher
cannot observe this latent distribution of words, LDA infers the distribution using
Bayesian techniques from the observable texts in the collection of CEO letters. It
perceives each document as a mixture of topics. And each word in a document
is attributable to one of the document’s topics. For example, suppose that a CEO
wants to emphasize the cultural value of collaboration among stakeholders in her
letter. Then, the CEO letter should be represented by a topic distribution that
places high weights on certain words, such as teamwork or human capital. Rela-
tively, a topic associated with innovation or high revenue will receive low weights
in this particular CEO letter.
The strength of the LDA model is that a researcher can avoid subjective bias
and priors by minimizing arbitrary parametric choice. This automated process pro-
vides a comparative advantage over the textual approaches that rely on researcher-
selected lists of words. The only input I need to specify is the number of topics.
Based on the rationale described in the following section II.F.1, I fit the model
using three-topic specification.
With the given number of topics, LDA produces the following outcomes. First,
it presents the list of words that are the most representative in each topic. Second,
each document is assigned with the proportional distribution over the topic. Figure
2 illustrates the technical process and the outcomes of LDA analysis.
I use the LDA outputs to measure the cultural similarity of paired firms in two
ways. First, I classify the firms into a topic in which they are assigned with the
13
highest distribution. Then, the dummy variable indicates whether the two merging
firms are assigned to the same topic.
Second, I calculate a pairwise cosine similarity score of the probability distri-
bution over the topics for the acquirer and the target in a given merger. For a given
merger transactioni, suppose the buyer has a vector of probabilitiesBP
i
and the
(pseudo-)target has a vector of probabilities TP
i
. Then, the cosine similarity of
combined firms can be calculated as
BP
i
TP
i
jjBP
i
jjjjTP
i
jj
=
P
n
j=1
bp
ij
tp
ij
r
P
n
j=1
bp
2
ij
q
P
n
j=1
tp
2
ij
,
where n represents the number of topics, and bp (tp) is a vector element of BP
(TP ).
C. Merger Data
The initial sample for merger transactions includes all public mergers from the
SDC Platinum database from 2004 through 2016. For each deal, I collect deal-
specific information, such as contracting parties’ CUSIPs, transaction values, and
state locations. I then match CUSIPs to NCUSIPs of CRSP, then to GVKEY
of COMPUSTAT, consequently. The data filter yields a sample of 803 mergers.
For each merger deal, I find a control firm, which is matched to the target firm,
using two-digit SIC industry code, total assets, sales, and market value. A control
firm is from the COMPUSTAT universe with the same two-digit SIC codes as the
matched target firm, with similar size, in terms of total assets, sales, and market
value at the end of the same fiscal year.
Table I panel B presents the thirty cases of one-to-one matching results. Out of
14
the sample where I can collect the CEO letters for both the target and the pseudo-
target, the table shows the matching results with the highest merger transaction
values. Panel C shows the matched result by comparing the mean value of bal-
anced variables of the target and the pseudo-target, with thet-statistics. It shows
that they are well-balanced and have statistically insignificant differences in every
matching variables.
To investigate the market reaction to the merger announcement, I collect stock
price data from CRSP. I calculate the acquirer’s and the target’s abnormal returns
by subtracting the market return from the firm’s daily return. Cumulative abnormal
return (CAR) is the sum of abnormal return in the three ([-1, 1]) or five ([-2, 2])
days around the announcement dates. For the combined CAR, I use either the
average of the acquirer’s and the target’s CAR or the weighted average using their
market value.
D. Other Variables
Firm-year specific financial data are collected from COMPUSTAT. Based on
previous studies, I construct a set of control variables, which are standard con-
trols in the M&A literature. They include the natural logarithm of total assets,
the natural logarithm of market value, the natural logarithm of sale, and book to
market.
I also consider deal-specific information collected from the SDC Platinum.
First, I control for the relative size of the transaction value to the buyer’s mar-
ket value. Moeller et al. (2007) find that the relative transaction value represents
15
the bidding float. So, it is negatively associated with the announcement return.
Second, I include a dummy to represent the cross-industry transaction.
I further control for the product types and industry groups of the target and the
buyer. This paper assumes that the texts in CEO letters reveal the corporate cul-
ture. And corporate culture provides incremental explanatory factors in merger
activities. One might be concerned that CEO letters mainly convey information
about product types or industries of the firm. To alleviate this concern, I use the
product types and industry groups measured by a firm’s textual representations.
First, I control the product relatedness between the target and the buyer. I consider
the text-based vertical relatedness measure collected from the Fresard-Hoberg-
Phillips data library (Fr´ esard et al. (2019)). They measure the vertical relatedness
between firm-pairs using product descriptions in the 10-K and the Bureau of Eco-
nomic Analysis Input-Output tables. Second, I include industry fixed effects in
every specification using text-based industry groups (Hoberg and Phillips (2016)).
In the tests for CEO and firm characteristics for each LDA type, I use the fol-
lowing variables. For CEO and executive compensation data, I use the COMPUS-
TAT Executive dataset. R&D intensity is measured by research and development
expense, scaled by the total asset. Stoffman, Woeppel, and Yavuz (2019) provide
information on new patent issuance. And customer satisfaction scores are obtained
from the American Customer Satisfaction Index (ACSI) website (www.theacsi.org).
I collect the brand value from the Interbrand website (www.interbrand.com). Since
ACSI and Interbrand do not provide company-specific indicators, I manually match
16
the company name to the COMPUSTAT universe.
For the cross-sectional tests in mechanism analysis, I collect the following data.
First, COMPUSTAT Segments provides the data on the number of business seg-
ments of target firms. Second, the target firm’s labor intensity is calculated as the
number of employment, divided by property, plant, and equipment. The relevant
variables are compiled from COMPUSTAT.
I use three different measures of information asymmetry between the target
and the outside market. The first is the (inverse of) number of analyst forecasts.
The number of analysts following the stock represents the information intermedi-
aries that alleviate information asymmetry. The second measure is the standard
deviation among the forecasts on earnings per share (EPS). Disagreement among
analysts is an indication of the lack of available information about the firm. Last,
I use the forecast error in EPS, measured by the difference between the mean of
forecast EPS and the actual EPS, scaled by the actual EPS. Firms with larger levels
of information asymmetry are expected to have higher forecast errors.
To measure the inverse level of agency conflict between the buyer’s top exec-
utives and the shareholders, I use the management’s share ownership, the stock
option value, and the long-term incentive payments for the executives. I obtain the
executive compensation data from the ExecuComp database.
The post-merger restructuring activities are measured by either divestiture flag
or spin-off flag in the SDC Platinum database. The dummy variable for ex post
restructuring equals one if the acquiring firm has any divestiture flag and spin-off
17
flag in the following year of acquisition, and zero otherwise.
E. Summary Statistics
Table I presents the summary statistics for the final sample. While the similarity
measure is the main variable of interest in this paper, there are three restrictions
to have this observation. First, CUSIPs of both the buyer and the target should be
matched to CRSP and COMPUSTAT. Second, I should locate the annual reports
and CEO letters for both of the combining firms.
Panel A compares the firm specifications of the sample with and without a sim-
ilarity measure. The comparison shows that the merger transaction in the final
sample is conducted between larger firms in terms of total assets, sales, and mar-
ket value. Prior literature on M&A research also notes that due to data availability,
the final sample is skewed towards larger companies (Moeller et al. (2007)). How-
ever, if I compare the relative size of total assets and sale, the relative size of two
contracting parties is not systematically different between the two groups. For the
relative market value, the buyer’s market value is marginally statistically smaller
in the data with similarity measures than the sample without similarity measures.
It implies that the target firms in the final sample are relatively more crucial com-
ponent for the combined entity. This selection bias might inflate my estimation on
the importance of the target’s culture in M&A outcomes.
Panel C shows the firm characteristics of the actual target firms in the merger
activities and the control firms, which are matched based on total assets, sales, and
market value in the same two-digit SIC industry. It shows that the two groups are
18
balanced in the matched covariates by having statistically insignificant differences.
Panel D displays the Pearson correlation coefficients between various pair-wise
similarity measures, including my measures and the measures developed by ex-
isting literature. This paper constructs the measures, including (1) the cosine-
similarity between every word in two CEO letters, (2) the cosine-similarity be-
tween words in two CEO letters, excluding terms mentioned in the 10-K business
description, (3) the cosine-similarity of LDA topic distribution across three topics
in two CEO letters, and (4) a dummy variable indicating whether two CEO letters
have the highest probability mass on the same topic out of three topics. And I com-
pare these measures to (5) the text-based product similarity measure (Hoberg and
Phillips (2010, 2016)), (6) the buyer’s text-based vertical related measure to the
target (Hoberg and Phillips (2016)), and (7) the target’s text-based vertical related
measure to the buyer (Hoberg and Phillips (2016)). When the textual similarity
in CEO letters considers every word, it is positively and statistically significantly
correlated with the text-based product similarity measure. This implies that the
CEO describes the firm’s business and products in her letter. However, once I ex-
clude the words for describing the business, the positive correlation between my
measures and the product similarity measures become statistically insignificant.
Overall, most correlations are modest, suggesting that my pair-wise similarities in
CEO letters can depict incremental aspects of the firm besides industry or product
type.
19
F. LDA Topic Analysis
F.1. LDA Topic Identified
In implementing LDA analysis, the only parameter a researcher needs to choose
is the number of topics. To determine the topic number, I consider the following
two aspects. The first feature is the coherence score. Although there is no definite
consensus on how to choose the optimal number of topics, linguistic literature pro-
poses to consider the coherence score. The coherence score measures the degree
of semantic similarity between high scoring words in the topic. A high coherence
score means that topics are semantically interpretable and are distinct from each
other. Table II panel A shows the coherence scores for the topic numbers ranging
from two to twenty. When the algorithm is assigned with three topic specifica-
tion, it delivers the highest coherence score. Second, I balance interpretability
and accuracy. One should have enough topics to distinguish between the topical
context, meaning accurate prediction. However, too many topics will lose their
interpretability since the same words start appearing in multiple topics.
Considering this trade-off and the coherence measure, I choose the topic num-
ber as three. In section VI, I explore a different set of specifications.
LDA helps me identify a group of words, which appear in the same context. Ta-
ble II panel B illustrates the list of top-20 keywords for each topic. The first topic
has some distinct words, including team, people, and leadership. Some unique
words for the second topic are transform, effici, and grew. For the third topic, I
notice the words, such as custom, product, market, and servic. For convenience,
20
I name each topic as ‘collaborative culture,’ ‘innovative culture,’ and ‘customer-
centric culture,’ henceforth. And, I classify the firms into a topic in which they
are assigned with the highest distribution and name them as collaborative firms,
innovative firms, and customer-centric firms, respectively.
Figure 3 shows the word clouds to illustrate groups of words which firms men-
tion in their CEO letters. It is hard to specify a topical context in figure (a), which
is the word cloud generated by every sample CEO letter. The word cloud in figure
(b) for the first LDA topic, the words, such as team and people, become more out-
standing, compared to figure (a). In figure (c), one can quickly notice the words,
including transform or effici. Figure (d) highlights the words, such as custom or
product.
Table II panel C presents the autocorrelation of corporate cultural value mea-
sures. In row (1), I show the correlation between the cultural type at year t and
the following five years. Rows (2) - (4) present the correlation between the proba-
bilities assigned to collaborative culture, innovative culture, and customer-centric
culture at yeart and the following five years. The correlations between yeart and
the lead years are statistically significant and positive. And they become smaller
as time elapses. The autocorrelation results suggest that corporate culture is sticky
and evolves slowly over time.
Next, I try to show the topics identified by LDA do not capture industry or
product types but illustrate a firm’s cultural aspects. First, the examples of actual
CEO letters can illustrate LDA topics. Appendix C shows an excerpt from the
21
CEO letter of Chesapeake Energy Corporation, which is classified as a collabora-
tive firm. In its CEO letter, it emphasizes the importance of its human capital and
talent. Appendix D is an excerpt from the CEO letter of Eli Lilly and Company,
which is classified as an innovative firm. The CEO highlights the importance of
innovation and invention in his CEO letter. In Appendix E, I present a part of
the CEO letter of Omnicare Corporation. Omnicare Corporation is classified as
a customer-centric firm. The letter mentions the firm’s service mind to retain its
customer group.
Second, panel D in Table II describe the firms’ industrial distribution in each
topic. Panel D explores two-digit SIC classification to describe the industry. None
of the topics is clustered in specific industries. It suggests that the LDA topic
captures some distinctive features within the industry.
In the internet appendix, I test correlations between my measure and two alter-
native measures of corporate culture in Table A1. The purpose of this experiment
is to show that the measures used in this paper are corroborated with other cul-
tural measurements, which are developed by some reliable resources. At the same
time, it will demonstrate that my measures can quantify novel aspects of corporate
culture, which are not captured by other proxies.
F.2. CEO & Firm Characteristics for Each LDA Topic
Table II panel E provides the descriptive statistics of firm specifications for
each group of firms. And panel F compares the groups. The t-statistics imply
that the topic groups have some significant differences in many aspects. The firms
22
assigned for each type are not systematically different in terms of total assets. The
firms with a collaborative culture and with a customer-centric culture have a sim-
ilar size of sales and market value. Among the three groups of firms, innovative
firms make the largest sales. And, their market value is the largest, as well. Book
to market ratio is the highest in the customer-centric firms, followed by collabo-
rative firms and the innovative firms. Innovative firms are the youngest. And the
profitability, measured by ROA, is the highest in the firms with a customer-centric
culture. Overall, the comparison shows that the innovative firms are the youngest
entities with high growth opportunities. And the customer-centric firms have some
contrasting features to the innovative firms, by being the most mature and having
the highest book to market ratio. The collaborative firms are located somewhere
between two groups. Although the comparison does not provide any causal impli-
cations, it reinforces the importance of including financial control variables when
empirical analysis uses the LDA model results.
Table III is to show Pearson correlation coefficients between LDA topic distri-
bution and characteristics of the firms and their CEOs. In the attempt to understand
corporate culture using CEO letters, the underlying assumption is that CEOs are
one of the most crucial players in defining corporate culture. They are not only
the narrators who authorize the document but also the forerunner who can initiate
the cultural change in the organization. In panel A, I try to infer CEOs’ charac-
teristics using their demographics and relative payroll. Innovative culture has a
negative correlation with the CEO age. It corresponds with the prior literature,
23
which finds that younger CEOs are more risk tolerant and are more likely to run
innovative companies (Graham, Harvey, and Puri (2013)). CEOs in innovative
firms are compensated more than the other executives in the organization. It im-
plies that CEOs are more influential figures in the firms focusing on innovation.
This association is aligned with the prior literature, which finds that CEO char-
acteristics are an important determinant of the firm’s innovative strategy (Kaplan,
Klebanov, and Sorensen (2012)).
Panel B analyzes innovation activities, measured by R&D intensity and the to-
tal value of newly issued patents. The topic distribution on innovative culture is
positively correlated with a firm’s innovative activities. The positive associations
between the LDA probabilities assigned to the innovative culture and the innova-
tion activities may imply that when CEO emphasizes innovation in her letter, she
indeed encourages innovative activities.
In panel C, I investigate the correlations between the cultural distribution and
the customer relationship. When I proxy a firm’s customer relationship with the
ACSI score, the probabilities assigned to customer-centric culture are positively
correlated with the customer satisfaction. The brand value does not provide signif-
icant correlations with any cultural probabilities. The positive association between
the LDA probabilities assigned to the customer-centric culture and the ACSI score
may suggests that a firm placing high priority in customer relationship is assigned
to customer-centric culture.
In the internet appendix, I repeat the correlation analysis for the alternative
24
cultural measures for the interested readers. Table A3 investigates Pearson cor-
relations with the scores developed by Li et al. (2020) and the firm specifications
which I explore in this section. In Table A4, I repeat the experiment using the MIT
Sloan measure.
Overall, my LDA measures correlate with external variables, such as CEO char-
acteristics, innovation activities, and customer satisfaction scores. One thing I
want to emphasize is that due to the absence of good instruments, I am unable to
establish causality on the correlations between a firm behavior and its culture. In-
stead, I look at differences in culture and their implications on merger outcomes.
Therefore, even if the labels are not perfect in predicting corporate behavior, it
does not change your results.
III. Corporate Culture and Merger Volume
A. Corporate Culture Measured by Cosine Similarity
The cultural distance hypothesis predicts that the cost of contracts among the
two groups is positively related to cultural differences (Hofstede (1980)). Cultural
differences can also deteriorate productivity or increase employee turnover, lead-
ing to inferior post-merger performance or divestiture. Therefore, the increase in
cultural misalignment will discourage a bidder from acquiring a potential target.
To test this prediction, Table IV presents the linear regression estimates of the
implication of cultural similarity on the level of merger activities. In every column,
I include year fixed effects and industry fixed effects.
25
Columns (1) - (2) explore the similarity in two contracting firms’ CEO letters
when I use every word mentioned in the documents. The similarity measures are
estimated to be positive and statistically significant with and without including
various control variables.
In CEO letters, CEO can deliver her thoughts on the values or attitudes with
which employees should conduct their tasks. However, she can also describe the
company’s industry or products in the letter. To disentangle the message on cul-
tural value from the message on other firm specifications, I exclude the words
which are also used in the 10-K to describe their products and industry. In columns
(3) - (4), the remaining words are used to construct the alternative cosine similarity
scores. The analysis using this alternative measure also delivers the same message
as columns (1) and (2). The coefficient estimates on the main explanatory variable
are positive and statistically significant.
The results provide evidence that cultural similarities had a significant positive
association with merger volume. The implication of cultural similarity is eco-
nomically meaningful, as well. One standard deviation change in the similarity
measure increased the likelihood of being a target company by 9.34% - 13.56%,
depending on the specification. It confirms my theoretical prediction that cultural
alignment can stimulate merger attempts.
B. Corporate Culture Measured by LDA Analysis
To see whether CEO letters depict not only a firm’s business but also their
cultural features, it is crucial to understand what precisely the CEO says about
26
culture. In this section, I employ an LDA analysis to identify CEO letters’ topical
context.
Table V tests the relationship between the LDA topical context and the merger
match. Panel A shows the topical distribution of the control firms and the target
firms. A customer-centric firm is more likely to be a target. But the difference in
distribution between the two groups is not statistically significant. It reaffirms that
the matching process is valid to find a comparable control for a target.
Table V Panel B repeats the analysis of Table IV but uses the similarity measure
constructed from LDA analysis. Columns (1) and (2) evaluate the cosine similarity
of the probability distribution over the LDA topics. And in columns (3) and (4), I
implement a dummy variable, representing whether the buyer and the target have
the highest probability mass on the same topic.
The LDA topic similarity is positively related to the merger likelihood. In terms
of economic significance, one standard deviation increased in the similarity of
LDA topic distribution raises the probability of being targeted by 5.70% - 6.06%.
And if two firms had the same topic with the highest probability, they are more
likely to initiate M&A transactions by 13.08% - 13.68%.
This section finds that when the buyer and the potential target cover similar
topics in CEO letters, they are more likely to conduct M&A transactions.
C. Potential Mechanism
I predict that the cultural alignment between two potential merger firms can im-
prove the synergy gain, leading to a higher merger volume. The ideal experiment
27
to run would be where corporate culture is randomly assigned. However, corpo-
rate culture is not randomly established. One could be concerned that omitted
organizational features, such as industry, profitability, or geographical location,
can drive the results, while those features are proxied by CEO letters. Although I
include various control variables, there is no perfect remedy for this identification
concern.
Table VI addresses some of the concerns in an indirect way. I explore the factors
that potentially amplify or mitigate the role of corporate culture as an incomplete
contract between the acquirer and the target’s employees. The main explanatory
variables are the interaction terms between the similarity measure and the potential
mechanism variables, which are indicated at the top of the table. The variables
that I explore as potential mechanisms include the target’s labor intensity and the
target’s number of business segments.
First, corporate culture helps the employees choose the most preferred action
when there is no written policy on how to solve the contingencies. In an extreme
case where a potential target is fully automated and does not entail any employ-
ees’ subjective judgment, corporate culture does not play many roles. Columns
(1) and (2) examine this prediction by investigating the target’s labor intensity.
The similarity measure has a positive estimate, weakly confirming the evidence in
Table IV. The labor intensity of the potential target has a positive association with
merger likelihood. The positive coefficient on the interactions term implies that
cultural similarity became more critical in predicting merger activity when the po-
28
tential target relied more on human capital than physical capital. At the mean level
of cultural similarity, one standard deviation increase in labor intensity increased
the likelihood of being targeted by 7.27% - 8.63%.
Columns (3) and (4) explore the second factor, the number of the target’s busi-
ness segments. Suppose the potential target has many separate divisions. In that
case, it is hard for the buyer’s management team to discipline all of these distinct
segments, which might have different business models or operational objectives.
Therefore, the existing corporate culture embedded can be crucial. As in columns
(1) and (2), the coefficient of the similarity measure is estimated to be positive as
in the main analysis. The number of business segments of the target is negatively
related to the likelihood, implying that the target’s business complexity may de-
ter merger attempts. The positive and statistically significant point estimates for
the interaction term provide suggestive verification on the prediction. The number
of segments strengthened the positive association between cultural similarity and
the merger volume. At the mean level of cultural similarity, one standard devia-
tion increase in the number of business segments increased the likelihood of being
targeted by 9.91% - 12.55%.
Overall, the mechanism test provides suggestive evidence that corporate cul-
ture can provide an incremental explanation on merger activities as an incomplete
contract.
29
IV. Cultural Integration and Post-Merger Synergy Realization
In prior sections, I show that pre-acquisition cultural characteristics can provide
an incremental explanation for merger volume. This section investigates whether
the post-acquisition process can also be critical in determining post-merger per-
formance and stabilizing the combined entity. This empirical analysis is not free
from endogeneity concern. Any omitted variables might simultaneously affect the
variables, leading to spurious correlations. Or, reverse causality can be an issue
if favorable post-merger performance can help the firm coordinate two distinct
entities and employee groups.
A. Post-Merger Performance
If two corporate cultures are misaligned, two combined firms’ employees may
not coordinate well due to different assumptions, values, and beliefs on the best
ways of conducting business. This can deteriorate productivity or increase em-
ployee turnover, leading to inferior post-merger performance.
Table VII tests this prediction. In panel A, I quantify the post-merger integra-
tion by the degree to which an acquirer’s CEO changes her letter to be similar to
the letter of the target’s CEO. The integration measure is estimated to have a pos-
itive and statistically significant association with post-merger performance, mea-
sured by the change in Tobin’s Q, the change in return on assets, and the change
in SG&A and sales ratio.
The relative firm size between the two merging firms will influence the inte-
30
gration process. Pre-merger firm size may represent the relative importance of the
merged units in the combined firm. It can be optimal to maintain the culture of a
relatively important organization. Even if the cultural integration is successfully
implemented, it is more likely for the relatively substantial buyer to maintain its
existing culture and not to move toward the target’s culture. Considering the im-
pact of relative size, I inflate the integration measure by the natural logarithm of
the relative asset size of the acquirer and the buyer. Panel B implements this alter-
native integration measure. It finds comparable results as in panel A for Tobin’s Q
and return on assets. The results for SG&A and sales ratio becomes insignificant,
but the directional prediction stays the same.
One standard deviation change in the integration measure could increase the
change in Tobin’s Q by 0.05 - 0.06 and the change in return on assets by 0.41 -
0.45. And it could decrease the change in SG&A and sales ratio by 0.12 - 0.13.
Since the outcome variables’ respective mean values are (0.03), (0.36), and
0.14, the results imply that the combined entity could revert performance deteri-
oration into performance improvement by increasing the integration measure by
one standard deviation.
5
To summarize, the empirical analysis supports my prediction on the positive
association between two firms’ cultural similarity and post-merger operation.
5
Untabulated results show that there is no statistically significant association between the pre-merger cultural similarity and the post
performance. One potential explanation is that the merger match is optimally chosen in considering cultural status of two parties.
So among these optimal matches, the one with better cultural alteration could outperform others.
31
B. Post-Merger Divestiture
In the second set of analysis, I use the ex post restructuring decision after the
M&A transactions. The failure to integrate two disparate cultures impedes the
combined firms from realizing synergy and stabilizing the organization. This im-
pediment will increase corporate restructuring activities after the merger.
As in the previous section, the post-merger integration is proxied by how much
the acquiring CEO adapts her voice similar to the target CEO’s voice. Table VIII
regresses the integration measure on the likelihood of post-merger restructuring,
including divestiture and spin-off. Columns (1) and (2) use the integration mea-
sure without considering the relative size of the acquirer and the target. Columns
(3) and (4) exploit the measure after inflating the measure with the relative size.
Across the different specifications, the test generates negative and statistically sig-
nificant point estimates, implying the increase in CEO letter similarity decreased
the restructuring magnitude.
One standard deviation increase in similarity change could decrease the like-
lihood of divestiture by 4.40% - 5.27%. Considering the unconditional mean of
divestiture dummy is 14.30%, it is also economically meaningful.
V. Corporate Culture and Merger Gain
The previous sections show that the bidders select the target considering the
cultural alignment. This section investigates how corporate culture is related to
the merger gain of the buyer and the combined entity. There are two contradicting
32
theoretical predictions.
The first theory predicts no association between the cultural similarity and the
merger gain. If two organizations’ corporate culture is misaligned and hard to
be integrated, it will deter the merged entity from fully realizing the synergies.
Many top executives assert this argument by saying that they would walk away
from a culturally misaligned target. And they would also discount the acquisi-
tion premium of the culturally-disparate target by 10% to 30% (Graham et al.
(2016, 2017)). Since mergers are not randomly assigned, the empirical studies
analyzing merger gains can only use the selected subset. If restructuring decisions
and merger contracts are optimally decided after considering cultural aspects, one
would not observe any relationship in the reduced form analysis (Demsetz and
Lehn (1985)).
The second theory predicts a positive relationship between the cultural align-
ment and the acquirer’s merger gain. Although the acquiring firm attempts to
price the transaction value after considering the target’s underlying culture, it is
challenging to fully understand other organization’s culture. This potential mis-
pricing can either overprice or underprice the value of the target. However, the
target has the incentive to correct the mispricing only when they are underpriced.
As a result, the buyer ends up providing high premium to the culturally-misaligned
culture. Since there are two contradicting predictions, it is an empirical question.
33
A. Buyer’s Announcement Return
First, I investigate the association between cultural alignment and the buyer’s
market reaction. Figure 4 plots the relationship between the three-window (five-
window) cumulative returns and the cultural similarity before the merger. The
scatter plots look noisy, but linear regression lines suggests positive relationships.
It confirms the theory which predicts a positive association between the cultural
similarity and merger gain of the buyer.
In Table IX, the similarity in CEO letters is estimated to have positive estimates.
It is also statistically significant in some specifications. In terms of economic
magnitude, one standard deviation rise in the similarity measure increased the
three-day (five-day) window CAR by 0.83% - 2.11% (1.09% - 1.19%). In dollar
terms, this implies a range of roughly $136.11 million to $346.46 million ($178.18
million to $195.50 million) for the median-size firms.
6
The analysis shows that
buyers’ announcement returns are positively associated with the cultural similarity
measure in CEO letters between the buyer and the target.
B. Combined Announcement Return
In the previous section, I show the positive association between the cultural
alignment and the buyer’s return. I still need to show whether the cultural simi-
larity benefits the combined entity as a whole or only helps the buyer. Since there
are two contradicting predictions, I test the association of corporate culture on the
6
The mean (median) of the buyer’s three-day window CAR is0.97% (0.38%). The mean (median) of the buyer’s five-day
window CAR is1.03% (0.35%).
34
combined announcement return.
Table A5 exploits the combined announcement return. The combined announce-
ment return is either the simple average (panel A) or the weighted average (panel
B) of the announcement return of the target and the buyer.
The empirical analysis delivers the same message in both specifications. The
similarity in CEO letters does not have incremental explanatory power on the com-
bined merger gain. Since the mean and the median value of the combined CAR
is positive,
7
the results show that the merger transactions occur only when they
could create positive synergy and that the cultural similarity is also considered in
this optimal endogenous decision.
C. Potential Mechanism
In the previous section, I find a positive association between the cultural simi-
larity measure and the buyer’s market return. This finding implies that acquiring
firms bid higher (lower) premiums for the culturally disparate (similar) targets.
The following alternative theories can explain these findings. First, agency cost
theory (Jensen (1986)) argues that acquiring firms’ empire-building motives lead
to value-destructive takeover deals. Second, hubris theory (Roll (1986)) attributes
the buyer’s market reaction to its valuation mistakes. The valuation error can be
substantial in the context of the corporate culture. Organizational culture is an
unwritten value shared by insiders, so it is hard to evaluate from outside. That
7
The mean (median) of the three-day window average CAR is 7.97% (5.14%). The mean (median) of the five-day window average
CAR is 8.05% (5.39%). The mean (median) of the three-day window weighted-average CAR is 2.66% (1.51%). The mean
(median) of the five-day window weighted-average CAR is 2.75% (1.47%).
35
is, cultural misalignment can be a factor which raises the manager’s mispricing in
M&A transaction. And the assessment from outside will be more challenging if
there is more information asymmetry between insiders and outsiders, leading to
higher misevaluation.
Table X tests the two theories. In panel A, I explore the hubris theory. In partic-
ular, the valuation error of the acquiring firm will be pronounced when information
asymmetry between the target and the outsiders is severe. Information asymmetry
is estimated by the three proxies, measured by the analyst forecasts following the
target stock: the (inverse of) number of analyst following; the standard deviation in
analyst forecasts on EPS; and the analyst forecast error on EPS. The main variable
of interest is the interaction term between the similarity measure and information
asymmetry level. In columns (1) and (2), the interaction term is estimated to have
negative and statistically significant coefficients. It implies that information trans-
parency decreased the positive association between the similarity measure and the
buyer’s announcement return. At the mean level of cultural similarity, one stan-
dard deviation increases in the analyst followings decreased the three-day window
CAR by 1.17% - 1.30%. Columns (3) - (6) estimate positive coefficients for the
interaction between the similarity measure and information asymmetry proxies.
The cultural similarity has the prediction power when the buyer faced high infor-
mation asymmetry. At the mean level of cultural similarity, one standard deviation
increase in the standard deviation in analyst forecasts (analyst forecast errors) in-
creased the three-day window CAR by 2.28% - 2.35% (0.27% - 0.25%).
36
Panel B investigates agency cost theory. I measure the inverse level of manage-
rial incentive misalignment, using three measures: equity ownership stake, as a
percentage of the total management equity; stock option compensation; and long-
term incentive payments, which is the amount paid to the executive under the com-
pany’s long-term incentive plan. These measures are inversely related to agency
costs. The main variable of interest is the interaction term between the similarity
measure and the inverse measure of agency conflicts. In every specification, the
coefficients are not statistically significant. The findings suggest that agency costs
could not explain the mispricing.
The cultural difference in the target may increase the valuation mistake of the
buyer. If information asymmetry is severe between the buyer and the target, the
buyer cannot precisely comprehend the target’s culture. While the imprecise eval-
uation can lead to both overpayment and underpayment in the merger premium,
the target has an incentive to correct the mispricing only when the acquirer under-
pays the premium. As a result, the information asymmetry leads to an overpay-
ment to the culturally misaligned target. The findings in this section collectively
imply that the acquiring firm cannot fully incorporate the cultural difference of the
target and ends up overpaying the premium, attenuating its shareholder value.
VI. Robustness Checks
37
A. LDA Analysis using Different Topic Numbers
In section III.B, I use three topics in LDA analysis. In this section, I want to
show that the main findings in section III are robust to different parametric choices
on the number of topics. In particular, I fit the LDA model using two different
specifications: four topics and seventy-five topics. First, I consider four topics,
which has the second-highest coherence score, preceded by three topics. Second,
I fit the model using seventy-five topics, following Ball, Hoberg, and Maksimovic
(2015) and Hoberg and Lewis (2017). These previous studies run LDA analysis
to investigate the contents in management discussion and analysis of 10-K filings
using seventy-five topics.
Table A6 repeats the analysis in Table V. Columns (1) and (2) use four topic
specification. And columns (3) and (4) fit the model using seventy-five topics. The
findings deliver the same message as the analysis with three topic estimation. The
similarity measure, using the probability distribution over the LDA topics, have
positive estimates. The economic magnitude is also compatible. One standard
deviation increase in the similarity of LDA distribution over four (seventy-five)
topics raised the probability of being targeted by 4.20% - 6.53% (2.70% - 6.30%).
B. Merger Volume with Li et al. (2020) Measure
In section E, I corroborate my measure of corporate culture with the measure
developed by Li et al. (2020). The cross-validation using correlations implies that
our measures capture unique aspects of culture and complement each other.
38
To show my measure provides incremental explanatory power for merger likeli-
hood, Table A7 repeats Table IV and V, but includes the cosine similarity measure
over Li et al. (2020) scores as an extra explanatory variable. The similarity in Li
et al. (2020) scores is estimated to have positive coefficient with merger likelihood
but is not statistically significant over the various specifications. More impor-
tantly, the coefficient estimates on my cultural similarity measure remain positive.
In some specifications, they lose statistical significance.
Overall, both measures are positively associated with merger volume, implying
potential complementarity of two measures. Due to correlations between two mea-
sures, including Li et al. (2020) scores weakens the explanatory power of my mea-
sure. However, my measure has stronger statistical explanatory power on merger
match than Li et al. (2020) scores.
C. CEO Change and Corporate Culture
This paper measures corporate culture using CEO letters. The underlying as-
sumption is that I can infer the cultural aspects of a firm using the CEO’s public
speech. This attempt starts with the belief that a CEO is one of the most crucial fig-
ures to define the organizational norm and is a forerunner for cultural revision. As
a way to verify this underlying assumption, it would be interesting to investigate
how a CEO plays a role in defining corporate culture. Although it is beyond the
scope of this paper, this section provides some suggestive evidence by exploring
cultural changes around CEO changes.
39
C.1. CEO Change and Post-Merger Performance
First, I explore CEO changes around merger transactions to see whether CEO
replacement can be a mechanism to drive cultural integration. Table XI tests the
association between CEO changes and the post-merger cultural alteration. The
first row explores the indicator variable of the acquiring firm’s CEO change. The
variable in the second row represents whether the target firm’s pre-merger CEO
remains as a combined entity’s executive.
Across different specifications, the first dummy variable does not provide in-
cremental explanation on the cultural integration. The coefficient estimates on the
second row are larger and statistically more significant.
Overall, the analysis implies that the combined firm could be more successful
in integrating two cultures by retaining the CEO of the newly integrated entity.
As a mechanism to blend two cultures, both CEOs of two entities remained in
the merged firm. The incumbent CEOs are associated with a higher post-merger
cultural integration. Although it is beyond the scope of this paper, this finding also
suggests that CEO is a crucial figure in establishing corporate culture.
C.2. Cultural Change around CEO Replacement
This section investigates cultural changes around CEO replacement in general.
I first start with any CEO replacement for the period 2012 - 2016. ExecuComp
yields 1,035 observations. For each CEO replacement episode in yeart, I collect
CEO letters written by the incumbent CEO at (t 1) and the new CEO at t and
40
(t + 1) to construct the cosine similarities of two letters in consecutive years:
Similarity
(t1);t
and Similarity
t;(t+1)
.
First, I compare Similarity
(t1);t
and Similarity
t;(t+1)
. In that way, I can study
how similar corporate culture is in the subsequent year, depending on whether the
firm had the same CEO or not. The mean value of Similarity
(t1);t
is 0:586. And
the mean of Similarity
t;(t+1)
is 0:624. The latter is statistically significantly larger
than the former by havingt-statistics as 3:436. This implies that corporate culture
evolved faster with CEO change.
Next, Table XII investigates potential factors which amplify or mitigate an asso-
ciation between a new CEO and cultural change. The unit of observation is a CEO
replacement event. Due to missing variables, the total number of observations is
450, out of 1,035. I explore Pearson correlations between various firm specifica-
tions and Similarity
(t1);t
. Columns (1) - (6) investigate the idea that CEO styles
or characteristics can be an important determinant in cultural change. Although it
is not statistically significant, the positive coefficient in column (1) suggests that if
a firm had a new CEO in the same gender, the culture does not change much. And
in column (5), the positive estimate is the most significant when a female CEO
replaced another female. Column (6) weakly implies that if a firm recruited a new
CEO from one of the incumbent executives, it experienced less cultural change.
Although a CEO might play a role in infusing different culture, corporate cul-
ture is not solely defined by her. If a firm is well-established, then a CEO can only
impose limited power to initiate cultural change. This is either because the incum-
41
bent culture is rigid or because a CEO does not need to replace well-performing
culture. In columns (7) - (13), Pearson correlation coefficients between the sim-
ilarity measure and the proxies for the firm size and profitability are positive and
statistically significant. They suggest that a new CEO does not deviate from the
existing culture if the size of the firm is big and if the firm performed well.
While the results in this section present some interesting insights, they should
be interpreted as correlations, not causality. Also, this analysis is not to present
any long-term impact from CEO replacement, rather to suggest instant influence.
VII. Conclusion
Business practitioners advocate that corporate culture is one of the key determi-
nants for corporate performance and successful mergers. Despite the importance
of the topic and call for the research evidence, the academic attempts to understand
the role of corporate culture in the M&A market has just been started.
As the shared assumptions, values, and beliefs, corporate cultures help employ-
ees understand which behaviors are appropriate. Since this shared assumption is
not written in formal documents, outsiders can only infer corporate culture from
other sources. In this paper, I test the value of CEO letters as a measure of cul-
ture by examining a number of critical issues related to mergers: (i) Do firms seek
partners with similar cultures? (ii) Does cultural integration benefit the combined
entity? (iii) Does the capital market favor mergers with similar cultures?
The paper contributes to the growing literature studying corporate culture by
42
providing a novel approach to measure time-varying firm-level culture. Benefited
by this measure, this paper is one of the first works that show the corporate culture
is related to merger outcomes in a meaningful way. The findings suggest that cul-
tural difference deters firms’ restructuring activities and decreases the acquirers’
shareholder wealth. It also shows that the post-merger performance of combined
firms is strongly associated with corporate culture integration. Collectively, this
paper demonstrates that the similarity in CEO voice in their letters can be another
important factor determining merger success and synergy realization.
43
Appendix A. Variable Definition
Variable Name Definition
Similarity in CEO
letter
Based on the words used by each firm’s CEO letter, I calculate pairwise cosine similarity
scores for the acquirer and the (pseudo-)target in a given merger. For a given merger
transactioni, suppose the buyer’s CEO letter has a word vector ofB
i
and the
(pseudo-)target’s CEO letter has a vector ofT
i
, which are both weighted by TF-IDF
approach. The cosine similarity of the paired firms can be calculated as
B
i
T
i
jjB
i
jjjjT
i
jj
=
P
n
j=1
b
ij
t
ij
q
P
n
j=1
b
2
ij
q
P
n
j=1
t
2
ij
, wheren represents the number of distinct words
andb (t) is a vector element ofB (T ). The similarity measure ranges from zero to one: zero
similarity implies no overlapping word in the pair. When two documents have the same
word frequencies, the similarity will be one.
Similarity in LDA
topic distribution
I calculate a pairwise cosine similarity score of the probability distribution over the topics
for the acquirer and the target in a given merger. For a given merger transactioni, suppose
the buyer has a vector of probabilitiesBP
i
and the (pseudo-)target has a vector of
probabilitiesTP
i
. Then, the cosine similarity of combined firms can be calculated as
BP
i
TP
i
jjBP
i
jjjjTP
i
jj
=
P
n
j=1
bp
ij
tp
ij
r
P
n
j=1
bp
2
ij
q
P
n
j=1
tp
2
ij
, wheren represents the number of topics, and
bp (tp) is a vector element ofBP (TP ).
Dummy= 1 if
having the same
highest LDA topic
I classify the firms into a topic in which they are assigned with the highest distribution.
Then, the dummy variable indicates whether the two merging firms are assigned to the same
topic.
Change in CEO
letter similarity
before considering
relative size
Change in similarity measures for M&A att represent how much an acquiring firm’s CEO
letter at (t+1) becomes similar to a target firm’s CEO letter at (t1), scaled by the
similarity at (t1)
Change in CEO
letter similarity after
considering relative
size
Similarity change is scaled by the natural logarithm of the relative asset size of the buyer and
the target
Text-based network
industries similarity
score
The text-based product similarity measure is collected from the Hoberg-Phillips Data
Library (Hoberg and Phillips (2010, 2016)).
Vertical upstream
potential relatedness
The text-based vertical relatedness measure is collected from the Fresard-Hoberg-Phillips
data library (Fr´ esard et al. (2019)). They measure the vertical relatedness between firm-pairs
using product descriptions in the 10-K and the Bureau of Economic Analysis Input-Output
tables.
Log (Total Assets) The natural logarithm of total assets. The data is collected from COMPUSTAT.
Log (Sale) The natural logarithm of sales. The data is collected from COMPUSTAT.
Log (Market Value)
The natural logarithm of market value, which is a multiplication of the number of shares and
the fiscal year-end stock price. The data is collected from COMPUSTAT.
44
Variable Name Definition
Book to Market
The ratio of the book value of equity and the market value of equity. The data is collected
from COMPUSTAT.
Firm age The number of years in which the firm appears in COMPUSTAT.
ROA The net income, scaled by the total asset. The data is collected from COMPUSTAT.
Dummy= 1 if
female CEO
For CEO and executive compensation data, I use the COMPUSTAT Executive dataset.
CEO age For CEO and executive compensation data, I use the COMPUSTAT Executive dataset.
CEO compensation
relative to other
executives’
For CEO and executive compensation data, I use the COMPUSTAT Executive dataset.
CEO compensation
divided by sales
For CEO and executive compensation data, I use the COMPUSTAT Executive dataset. Sales
data is from COMPUSTAT.
R&D intensity R&D intensity is measured by research and development expense, scaled by the total asset.
Sum of new patent
value
Stoffman et al. (2019) provide information on new patent issuance.
ACSI score
Customer satisfaction scores are obtained from the American Customer Satisfaction Index
(ACSI) website (www.theacsi.org). Since ACSI does not provide company-specific
indicators, I manually match the company name to the COMPUSTAT universe.
Brand value (in
billions)
I collect the brand value from the Interbrand website (www.interbrand.com). Since
Interbrand does not provide company-specific indicators, I manually match the company
name to the COMPUSTAT universe.
CAR[-1,1] or
CAR[-2,2]
I calculate the acquirer’s and the target’s abnormal returns by subtracting the market return
from the firm’s daily return. Cumulative abnormal return (CAR) is the sum of abnormal
return in the three ([-1, 1]) or five ([-2, 2]) days around the announcement dates.
Number of target’s
business segments
COMPUSTAT Segments provides the data on the number of business segments of target
firms.
Target’s labor
intensity
The target firm’s labor intensity is calculated as the number of employment, divided by
property, plant, and equipment. The relevant variables are compiled from COMPUSTAT.
Log (Number of
analyst following)
The natural logarithm of the number of analyst forecasts. I collect the data on the analyst
forecasts from IBES database.
Standard deviation in
analyst forecasts
The standard deviation among the forecasts on earnings per share (EPS). I collect the data
on the analyst forecasts from IBES database.
Analyst forecast
error
I use the forecast error in EPS, measured by the difference between the mean of forecast
EPS and the actual EPS, scaled by the actual EPS. I collect the data on the analyst forecasts
from IBES database.
Percentage of total
shares owned
The executive stock ownership data is from the ExecuComp database.
45
Variable Name Definition
Option granted The executive option data is from the ExecuComp database.
Long-term incentive
plan payouts
The executive long-term payout plan data is from the ExecuComp database.
Dummy= 1 if
acquiring firm
replaces CEO
The change in executives data is from the ExecuComp database.
Dummy= 1 if target
firm’s CEO stays
The change in executives data is from the ExecuComp database.
46
Appendix B. Example of 10-K Filing and Annual Report
This is the example of 10-K filing and annual report of AT&T Inc. for the fiscal year 2016. Panel A presents
the first two pages of 10-K filings, collected from the SEC EDGAR. Panel B displays the first two pages of
annual report, collected from the company’s website.
Panel A: 10-K filing
FORM 10-K

UNITED STATES
SECURITIES AND EXCHANGE COMMISSION
Washington, D.C. 20549

(Mark One)

x

ANNUAL REPORT PURSUANT TO SECTION 13 OR 15(d)
OF THE SECURITIES EXCHANGE ACT OF 1934

For the fiscal year ended December 31, 2016

OR

o TRANSITION REPORT PURSUANT TO SECTION 13 OR 15(d)
OF THE SECURITIES EXCHANGE ACT OF 1934

For the transition period from to
Commission File Number 1-8610
AT&T INC.
Incorporated under the laws of the State of Delaware
I.R.S. Employer Identification Number 43-1301883
208 S. Akard St., Dallas, Texas, 75202
Telephone Number 210-821-4105
Securities registered pursuant to Section 12(b) of the Act: (See attached Schedule A)
Securities registered pursuant to Section 12(g) of the Act: None.
Indicate by check mark if the registrant is a well-known seasoned issuer, as defined in Rule 405 of the Securities Act. Yes [X] No [ ]
Indicate by check mark if the registrant is not required to file reports pursuant to Section 13 or Section 15(d) of the Act. Yes [ ] No [X]
Indicate by check mark whether the registrant (1) has filed all reports required to be filed by Section 13 or 15(d) of the Securities Exchange Act of 1934 during the preceding 12 months (or for such shorter
period that the registrant was required to file such reports), and (2) has been subject to such filing requirements for the past 90 days. Yes [X] No [ ]
Indicate by check mark whether the registrant has submitted electronically and posted on its corporate Web site, if any, every Interactive Data File required to be submitted and posted pursuant to Rule 405
of Regulation S-T during the preceding 12 months (or for such shorter period that the registrant was required to submit and post such files). Yes [X] No [ ]
Indicate by check mark if disclosure of delinquent filers pursuant to Item 405 of Regulation S-K is not contained herein, and will not be contained, to the best of registrant’s knowledge, in definitive proxy
or information statements incorporated by reference in Part III of this Form 10-K or any amendment to this Form 10-K. [ ]
Indicate by check mark whether the registrant is a large accelerated filer, an accelerated filer, a non-accelerated filer or a smaller reporting company. See definition of “large accelerated filer,” “accelerated
filer” and “smaller reporting company” in Rule 12b-2 of the Exchange Act.

Large accelerated filer [X] Accelerated filer [ ]
Non-accelerated filer [ ] (Do not check if a smaller reporting company) Smaller reporting company [ ]
Indicate by check mark whether the registrant is a shell company (as defined in Rule 12b-2 of the Exchange Act).
Yes [ ] No [X]

Based on the closing price of $43.21 per share on June 30, 2016, the aggregate market value of our voting and non-voting common stock held by non-affiliates was $266 billion.
At February 10, 2017, common shares outstanding were 6,141,570,142.

DOCUMENTS INCORPORATED BY REFERENCE
(1) Portions of AT&T Inc.’s Annual Report to Stockholders for the fiscal year ended December 31, 2016 (Parts I and II).
(2) Portions of AT&T Inc.’s Notice of 2017 Annual Meeting and Proxy Statement dated on or about March 10, 2017 to be filed within the period permitted under General Instruction G(3) (Parts III and
IV).
Panel B: Annual report
A global leader in telecommunications, media & technology
AT&T INC. 2016 ANNUAL REPORT
FINANCIAL HIGHLIGHTS
Cash from operations
(Record)
25% since 2014
$39.3B
$39.3B 2016
$35.9B 2015
$31.3B 2014
Free cash flow is cash from operations minus capital expenditures of $22.4B in 2016, $20.0B in 2015 and $21.4B in 2014
$16.9B
70% since 2014
Free cash flow
$16.9B 2016
$9.9B
$15.9B 2015
2014
Capital spending
>$140B
invested between 2012 and 2016 in our
network, including acquisitions of spectrum
and wireless operations
$22.4B
capital expenditures in 2016 alone
Reflecting DTV acquisition and growth in video and IP services
$163.8B
Consolidated revenues
11.6%
Free cash flow dividend payout ratio
is dividends of $11.8B divided by free
cash flow of $16.9B
70%
Free cash flow dividend payout ratio
47
Appendix C. Example of CEO Letter Classified in LDA Type 1
(Collaborative Culture)
This is an excerpt from the CEO letter of Chesapeake Energy Corporation from the 2006 annual report.
Chesapeake Energy Corporation is classified as a firm with collaborative culture according to LDA topic
analysis.
48
Appendix D. Example of CEO Letter Classified in LDA Type 2 (Innovative
Culture)
This is an excerpt from the CEO letter of Eli Lilly and Company from the 2009 annual report. Eli Lilly and
Company is classified as a firm with innovative culture according to LDA topic analysis.
49
Appendix E. Example of CEO Letter Classified in LDA Type 3
(Customer-Centric Culture)
This is an excerpt from the CEO letter of Omnicare Corporation from the 2008 annual report. Omnicare
Corporation is classified as a firm with customer-centric culture according to LDA topic analysis.
50
Figure 1.
Histogram of Number of Words in CEO Letter
This figure illustrates the frequency distribution of unique words in CEO letters of the buyers, the targets,
and the control target firms.
Number of words
Frequency
0 500 1000 1500 2000
0 100 200 300 400 500 600 700
51
Figure 2.
LDA Analysis Illustration
This figure illustrates the technical process of LDA analysis.
Collection of CEO Letters
LDA Analysis
Cluster of words by topic
Topic 1
Frequency of words
word 1word 2word 3word 4word 5word 6
Distribution of topics
• Topic 1 27%
• Topic 2 25%
• Topic 3 48%
Topic 1
Topic 2
Topic 3
CEO Letter #1
CEO Letter #1
CEO Letter #2
CEO Letter #3
52
Figure 3.
Word Cloud
This figure illustrates the word cloud of unique words in CEO letters of the buyers, the targets, and the
control target firms. Figure (a) includes CEO letters of every sample firms. Figures (b) - (d) include unique
words belonging to topic 1, topic 2, and topic 3.
percent
year
million
network
growth
strong
sharehold
continu
billion
custom
team
grow
help
focus
work
last
product
new
peopl
build
best
compani
deliv
opportun
per
return
market
great
improv
increas
success
leadership
nancial
challeng
now
everi
today
drive
goal
progress
grew
better
share
look
bank
three
valu
record
achiev
way
cant
top
posit
just
excel
two
one
core
innov
commit
invest
signifi
transform leader
earn
near
cultur
strength
revenu
come solid
want
communiti
scal
strengthen
dividend
world
excit
move
ahead
sustain
strateg
much
talent
even
ever
creat
first
confid
foundat
launch
remain
pleas
bring
servic
approach
get
margin
model
across
(a) Words in CEO Letter
percent
team
strong
help
peopl
sharehold
great
last
today
challeng
grow
money
best
just
work
better
leadership
top
build
grew
now
even
goal
everi proud
want
solid
cultur
much
drive
progress
good
get
know
innov
think
look
way
say
come
deliv
core
rose
alway
dilut
return
network
presid
confid
ceo
decad
talent
day
world
disciplin
per
strength
excel
hard
bring
despit
near
billion
communiti
job
ever
foundat
accomplish
join far
fact
move
excit
focus
ago
leader
keep
pleas
averag
digit
dedic
vision
sheet
histori
three
cycl
hope
seen
approach
sustain
retir
role
platform
advantag
start
banker
forward
superior
true
import
(b) Words in LDA Type 1 (Collaborative Culture)
network
percent
financi
signific
fiscal
profit
first
effici
benefi
strong
transform
offic
grew
excit
confid
liberti
billion
ect
phone
leadership
gaap
partnership
tabil
reflect
progress
sharehold
great
excel
effect
dividend
launch
strength
repurchas
smartphon
drive ecommerc
financ
leader
live
difficult
acceler
momentum
today
just
win
strengthen
doubl
opportun
fastest
team
trust
polyon
now
ever
pipelin
colleagu
look
solid
peopl
diseas
scienc
everi near
passion
alon
help
effort
exibl
faster
bone
best
mileston
vision
co2
clear
societi
commit
portfolio
goal
pleas
foundat
saw
impress
driver
bring
core
driven
come
gain
breakthrough
ebitda
(c) Words in LDA Type 2 (Innovative Culture)
year
million
growth
continu
custom
new
product
compani
market
increasinvest
improv
posit
valu
bank
revenu
share
success
servic
oper
focus
perform
manag
earn
cash
also
cost
capit
opportun
provid
industri
program
achiev
expect
pnc
develop
believ
well
asset
time
make
billion
employe
strategi
work
result
vishay
net
acquisit
high
offer
sale
generat
term
financi
technolog
need
futur
can
reduc
mani
end
system
incom
global
base
profit
long
grow
per
plan
record
creat
client
commit
expand
support
made rate
expens
total
remain
serv
signific
includ
return
level
deliv
addit
econom
qualiti
initi
first
store
strateg
board
compar
process
experi
price
(d) Words in LDA Type 3 (Customer-Centric Culture)
53
Figure 4.
Similarity in CEO Letters and Buyer’s Announcement Return
This figure illustrates the scatter plots of the cumulative abnormal return of buyers around the merger an-
nouncement and the similarity measures, accompanied with linear approximations. The unit of observation
is a public merger during the period 2004 - 2016, of which both buyer and target are matched to COMPU-
STAT and CRSP. They-axis is the cumulative abnormal return of buyers around the merger announcement.
Thex-axis is the similarity measures for M&A att. It represent how close an acquiring firm’s CEO letter is
to a target firm’s CEO letter at (t 1), lying in the interval of (0;1). Figure (a) uses cumulative abnormal
return over the three-day window around the merger announcement. Figure (b) uses cumulative abnormal
return over the five-day window around the merger announcement.
-.4 -.2 0 .2
0 .2 .4 .6 .8
Similarity in CEO letter at (t-1)
CAR [-1, 1] Fitted values
(a) CAR over the Three-day Window
-.3 -.2 -.1 0 .1 .2
0 .2 .4 .6 .8
Similarity in CEO letter at (t-1)
CAR [-2, 2] Fitted values
(b) CAR over the Five-day Window
54
Table I
Summary Statistics
This table summarizes firm-year specific covariates in M&A transactions during the period 2004 - 2016.
The unit of observation is a public merger, of which both buyer and target are matched to COMPUSTAT
and CRSP. Panel A compares the observations with and without similarity measures. Similarity measures
for M&A att represent how close an acquiring firm’s CEO letter is to a target firm’s CEO letter at (t1),
lying in the interval of (0;1). Panel B exhibits the examples of matched target and pseudo-target firms,
which I can collect both of the firms’ CEO letters and have the highest merger transaction values. Panel C
compares target firms and control firms, which are matched to target firms, using industry, total assets, sales,
and market value. Columns (1) and (2) represent the mean value of each group. Column (3) shows the mean
difference between two groups with statistical significance indicators. The text-based vertical relatedness
measure is from the Fresard-Hoberg-Phillips data library (Fr´ esard et al. (2019)). Financial data is collected
from COMPUSTAT. Panel D displays Pearson correlation coefficients between pair-wise similarity between
the acquirer and the (pseudo-)target, including (1) the cosine-similarity between every word in two CEO
letters, (2) the cosine-similarity between words in two CEO letters, excluding terms mentioned in the 10-
K business description, (3) the cosine-similarity of LDA topic distribution across three topics in two CEO
letters, (4) a dummy variable indicating whether two CEO letters have the highest probability mass on the
same topic out of three topics, (5) the text-based product similarity measure (Hoberg and Phillips (2010,
2016)), (6) the buyer’s text-based vertical related measure to the target (Hoberg and Phillips (2016)), and
(7) the target’s text-based vertical related measure to the buyer (Hoberg and Phillips (2016)). Significance
levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Panel A: M&A Transactions with and without Similarity Measure
With Data Without Data
Mean Mean Difference
(1) (2) (3)
Buyer’s vertical upstream potential relatedness to target 0:01 0:01 0:00
Buyer’s Log(Total Assets) 9:99 8:74 1:25

Buyer’s Log(Sale) 8:96 7:61 1:35

Buyer’s Log(Market Value) 9:55 8:19 1:36

Buyer’s Book to Market 0:56 0:54 0:02
Target’s vertical upstream potential relatedness to buyer 0:01 0:01 0:00
Target’s Log(Total Assets) 8:47 6:96 1:51

Target’s Log(Sale) 7:48 5:91 1:57

Target’s Log(Market Value) 8:05 6:53 1:52

Target’s Book to Market 0:57 0:58 0:01
Log(Buyer’s Total Assets/Target’s Total Assets) 1:90 2:05 0:16
Log(Buyer’s Sale/Target’s Sale) 1:85 1:99 0:14
Log(Buyer’s Market Value/Target’s Market Value) 1:91 2:10 0:19

Number of observation 297 506
55
Panel B: Example of Matched Target and Pseudo-Target Firm
Deal Year Deal Value ($ bil) Target Target SIC Pseudo-Target Pseudo-Target SIC
2016 79.41 Time Warner Inc 4888 Rogers Communications 4812
2004 66.75 Disney (Walt) Co 4888 At&T Corp 4813
2015 55.64 Time Warner Cable Inc 4841 Liberty Global Plc 4841
2004 38.98 Nextel Communications Inc 4812 Telmex-Telefonos de Mexico 4813
2005 35.4 Burlington Resources Inc 1311 Conocophillips 1311
2014 35.27 Baker Hughes Inc 1381 Canadian Natural Resources 1311
2005 27.86 Guidant Corp 3841 Garmin Ltd 3812
2015 27.54 Norfolk Southern Corp 4011 CSX Corp 4011
2006 26.29 Caremark RX Inc 5912 Medco Health Solutions Inc 5912
2006 25.83 Phelps Dodge Corp 3330 Essar Steel Algoma Inc 3312
2010 23.9 Genzyme Corp 2836 CF Industries Holdings Inc 2870
2011 18.08 Goodrich Corp 3728 Boeing Co 3721
2006 16.61 Caesars Entertainment Corp 7990 MGM Resorts International 7990
2008 15.51 Rohm and Haas Co 2821 Newmarket Corp 2860
2015 15.44 Jarden Corp 3089 Nike Inc 3021
2015 13.57 Starwood Hotels&Resorts World 7011 Hilton Worldwide Holdings 7011
2004 12.29 Public Service Enterprise Group Inc 4931 TC Energy Corp 4922
2005 11.3 Constellation Energy Group Inc 4931 Duke Energy Corp 4931
2004 10.9 Sears Roebuck & Co 5311 Macy’s Inc 5311
2007 9.75 Trane Inc 3585 LAM Research Corp 3559
2016 9.31 Valspar Corp 2851 Glaxosmithkline Plc 2834
2014 9.16 Family Dollar Stores 5331 Macy’s Inc 5311
2007 8.64 Commerce Bancorp Inc 6020 National City Corp 6020
2016 8.2 B/E Aerospace Inc 2531 Knoll Inc 2522
2007 7.95 Navteq Corp 7370 Global Payments Inc 7374
2015 7.94 Southern Co Gas 4924 Pinnacle West Capital Corp 4911
2006 7.42 Keyspan Corp 4931 Uns Energy Corp 4911
2016 6.85 Great Plains Energy Inc 4911 Black Hills Corp 4911
2012 6.45 Freeport McMoRan Oil&Gas 1311 Canadian Natural Resources 1311
2004 6.33 Caesars Entertainment Inc 7990 MGM Resorts International 7990
56
Panel C: Target and Pseudo-Target Firm
Target Firms Pseudo-Target Firms
Mean Mean Difference
(1) (2) (3)
Log (Total Assets) 7:45 7:48 0:02
Log (Sale) 6:44 6:46 0:03
Log (Market Value) 6:97 7:06 0:09
Panel D: Correlation in Pair-Wise Similarity Measures
(1) (2) (3) (4) (5) (6) (7)
(1) Similarity in CEO letter, including every word 1:00
(2) Similarity in CEO letter, excluding words in 10-K business
description
0:22

1:00
(3) Similarity in LDA topic distribution 0:09 0:18

1:00
(4) Dummy = 1 if having the same highest LDA topic 0:02 0:08 0:80

1:00
(5) Text-Based Network Industries similarity score 0:25

0:10 0:06 0:13

1:00
(6) Buyer’s vertical upstream potential relatedness to target 0:00 0:08 0:06 0:07 0:09 1:00
(7) Target’s vertical upstream potential relatedness to buyer 0:02 0:10 0:03 0:04 0:08 0:97

1:00

p< 0:10,

p< 0:05,

p< 0:01
57
Table II
Bayesian Topic Modeling using Latent Dirichlet allocation (LDA)
This table shows the empirical analysis using the LDA approach. For panels A, B, C, and D, the unit of
observation is a CEO letter for buyers, targets, and pseudo-targets. Panel A reports the coherence scores for
LDA analysis for different choice of the number of topics. Panel B shows the top 20 keywords for each of
three topics, constructed by LDA mechanism. Panel C exhibits the autocorrelation of LDA topic assignment
at year t and the following five years. Row (1) is the autocorrelation of the topic type with the highest
probability mass. Row (2) - (4) is the autocorrelation of the probabilities assigned to type 1, type 2, and
type 3, respectively. Panel D presents the industrial distribution over two-digit SIC code of the firms which
are assigned to the topic with the highest probability mass. Panel E summarizes firm specifications of those
assigned firms. Panel F compares the firm specifications of the firms which are assigned to the topic with
the highest probability mass. Column (1) shows the mean difference between the type 1 firms and the type 2
firms with statistical significance indicators. Column (2) shows the mean difference between the type 2 firms
and the type 3 firms with statistical significance indicators. Column (3) shows the mean difference between
the type 3 firms and the type 1 firms with statistical significance indicators. Financial data is collected from
COMPUSTAT. Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Panel A: Topic Number and Coherence Score
Topic Number Coherence Score Ranking
2 0.267 3
3 0.291 1
4 0.280 2
5 0.237 4
6 0.211 6
7 0.210 7
8 0.205 9
9 0.228 5
10 0.205 9
11 0.197 14
12 0.202 11
13 0.193 16
14 0.199 13
15 0.202 11
16 0.197 14
17 0.206 8
18 0.183 18
19 0.188 17
20 0.162 19
58
Panel B: Top 20 Keywords
Type 1 Type 2 Type 3
Topic Description Collaborative Culture Innovative Culture Customer-Centric Culture
(1) (2) (3)
percent network year
strong financi million
team percent growth
sharehold signific continu
peopl fiscal custom
challeng profit product
great first new
build effici compani
help benefit market
today offi increas
grow phone improv
leadership grew invest
last confid valu
work liberti posit
best effect bank
money billion revenu
grew excit success
just transform share
goal strong servic
cultur great oper
Panel C: Autocorrelations of LDA Corporate Culture Type
(1)
Yeart+1 Yeart+2 Yeart+3 Yeart+4 Yeart+5
(1) LDA Type 0.51

0.43

0.43

0.34

0.27

(2) LDA Probability assigned to Type 1 0.65

0.54

0.50

0.49

0.44

(3) LDA Probability assigned to Type 2 0.69

0.61

0.55

0.58

0.54

(4) LDA Probability assigned to Type 3 0.63

0.57

0.52

0.40

0.35

p< 0:10,

p< 0:05,

p< 0:01
59
Panel D: Industry Description for Each Topic
Industry Description
Type 1 Type 2 Type 3 Total
Collaborative
Culture
Innovative Culture
Customer-Centric
Culture
No. % No. % No. % No. %
Agriculture, Forestry, and Fishing (SIC 01 - 09) 1 50.0 1 50.0 0 0 2 100
Mining (SIC 10 - 14) 138 77.1 22 12.3 19 10.6 179 100
Construction (SIC 15 - 17) 19 90.5 0 0 2 9.5 21 100
Manufacturing (SIC 20 - 39) 847 67.4 334 26.6 75 6 1,256 100
Transportation and Public Utilities (SIC 40 - 49) 320 60.3 170 32 41 7.7 531 100
Wholesale Trade (SIC 50 - 51) 52 85.2 6 9.8 3 4.9 61 100
Retail Trade (SIC 52 - 59) 112 61.2 46 25.1 25 13.7 183 100
Finance, Insurance, and Real Estate (SIC 60 - 67) 849 84.1 99 9.8 62 6.1 1,010 100
Services (SIC 70 - 89) 277 66.7 126 30.4 12 2.9 415 100
Nonclassifiable Establishments (SIC 99) 20 64.5 4 12.9 7 22.6 31 100
Total 2,635 71.4 808 21.9 246 6.7 3,689 100
60
Panel E: Firm Specification for Each Topic
Type 1 Type 2 Type 3 Total
Topic Description Collaborative Culture Innovative Culture Customer-Centric Culture
Mean s.d. Mean s.d. Mean s.d. Mean s.d.
Log (Total Assets) 8:56 2:13 8:47 1:94 8:44 2:36 8:53 2:10
Log (Sale) 7:53 2:17 7:85 2:01 7:56 2:35 7:60 2:15
Log (Market Value) 8:06 2:04 8:48 1:90 8:00 2:24 8:15 2:03
Book to Market 0:59 0:70 0:45 0:48 0:61 0:42 0:56 0:65
Age 27:81 18:52 25:44 16:88 33:72 18:34 27:68 18:26
ROA 0:01 0:14 0:03 0:14 0:31 2:94 0:04 0:77
Panel F:t-Statistics Comparing types
(Type 1)-(Type 2) (Type 2)-(Type 3) (Type 3)-(Type 1)
Log (Total Assets) 0.10 0.03 -0.12
Log (Sale) -0.32

0.30

0.03
Log (Market Value) -0.41

0.47

-0.06
Book to Market 0.14

-0.16

0.02
Firm age 2.37

-8.28

5.91

ROA -0.01

-0.28

0.29

61
Table III
LDA Topic Culture and CEO & Firm Characteristics
This table displays Pearson correlation coefficients between LDA topic distribution and various CEO and firm characteristics. LDA analysis assigns each CEO
letter with probabilities on three topics. The variable of interest is indicated at the top of the table. The unit of observation is a CEO letter of buyers, targets,
and pseudo-targets around the years of merger transactions that occurred during the period 2004 - 2016. Panel A shows the correlation coefficients on CEO
characteristics. CEO and executive compensation data are collected from Compustat Executive dataset. CEO’s relative compensation is CEO compensation
divided by the average compensation of executives or by sales amount. Panel B shows the innovation activity, using R&D intensity, and the sum of new patent
value. R&D intensity is measured by research and development expense, scaled by total asset. Patent data is collected by Stoffman et al. (2019). Panel C shows
a firm’s customer relationship, using the American Customer Satisfaction Index (ACSI) score and brand value. ACSI score is collected from www.theacsi.org.
And I collect the brand value from www.interbrand.com. I manually match the company name to the GVKEY universe. Significance levels are indicated: * =
10 percent, ** = 5 percent, *** = 1 percent.
Panel A: CEO Characteristics
(1) (2) (3) (4)
Dummy = 1 if
female CEO
CEO age
CEO compensation
relative to other
executives
compensation
CEO
compensation,
divided by sales
Probabilities assigned to collaborative culture 0:02 0:06

0:10

0:01
Probabilities assigned to innovative culture 0:01 0:07

0:13

0:03
Probabilities assigned to customer-centric culture 0:03 0:00 0:05

0:06

p< 0:10,

p< 0:05,

p< 0:01
62
Panel B: Innovation Activity
(1) (2)
R&D intensity Sum of new patents value
Probabilities assigned to collaborative culture 0:03 0:00
Probabilities assigned to innovative culture 0:06

0:03
Probabilities assigned to customer-centric culture 0:17

0:06

p< 0:10,

p< 0:05,

p< 0:01
Panel C: Customer Relationship
(1) (2)
ACSI score Brand value (in billion)
Probabilities assigned to collaborative culture 0:01 0:06
Probabilities assigned to innovative culture 0:12 0:09
Probabilities assigned to customer-centric culture 0:21

0:07

p< 0:10,

p< 0:05,

p< 0:01
63
Table IV
Similarity in CEO Letters and Likelihood of Being Targeted
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public
merger during the period 2004 - 2016, of which both buyer and target are matched to COMPUSTAT and CRSP. The
outcome variable is a dummy variable to indicate whether the company pair is merged. The control target is constructed
by matching the target firm, using industry, total assets, sales, and market value. Similarity measures for M&A at t
represent how close an acquiring firm’s CEO letter is to a target firm’s CEO letter at(t1), lying in the interval of(0;1).
In columns (1) and (2), the similarity measure incorporates every word from CEO letters. In columns (3) and (4), words
mentioned in the 10-K business description are excluded when the similarity measure is constructed. The text-based
vertical relatedness measure is from the Fresard-Hoberg-Phillips data library (Fr´ esard et al. (2019)). Financial data is
collected from COMPUSTAT. I include industry fixed effects by using text-based industry groups (Hoberg and Phillips
(2016)). Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Similarity including every word
Similarity excluding words in 10-K
business description
(1) (2) (3) (4)
Similarity in CEO letter at (t1) 0:792

0:793

0:682

0:597

(0:223) (0:221) (0:283) (0:298)
Buyer’s vertical upstream potential
relatedness to target
15:644 26:015

(11:633) (14:302)
Buyer’s Log (Total Assets) 0:033 0:046
(0:053) (0:063)
Buyer’s Log (Sale) 0:025 0:045
(0:043) (0:050)
Buyer’s Log (Market Value) 0:071 0:119

(0:052) (0:066)
Buyer’s Book to Market 0:255

0:343

(0:085) (0:094)
Target’s vertical upstream potential
relatedness to buyer
11:312 29:784

(11:679) (14:199)
Target’s Log (Total Assets) 0:041 0:009
(0:038) (0:042)
Target’s Log (Sale) 0:012 0:005
(0:033) (0:037)
Target’s Log (Market Value) 0:013 0:003
(0:037) (0:040)
Target’s Book to Market 0:071 0:058
(0:084) (0:093)
Year Fixed Effect Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes
Observations 464 389 305 291
AdjustedR
2
0.025 0.036 0.027 0.031
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
64
Table V
LDA Topic Similarity and Likelihood of Being Targeted
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public merger during the period 2004 - 2016,
of which both buyer and target are matched to COMPUSTAT and CRSP. The outcome variable is a dummy variable to indicate whether the company pair is
merged. The control target is constructed by matching the target firm, using industry, total assets, sales, and market value. Panel A compares the target and
the matched control in terms of the distribution over the LDA topic. A firm is assigned to one of the LDA topics with the highest probability. In panel B, the
similarity measures for M&A att represent how close an acquiring firm’s CEO letter is to a target firm’s CEO letter at (t1), in terms of topical distribution.
In columns (1) and (2), the main explanatory variable is the cosine similarity of LDA topic distribution across three topics. In columns (3) and (4), the main
explanatory variable is a dummy variable, which represents whether the buyer and the target have the highest probability mass on the same topic out of three
topics. The text-based vertical relatedness measure is from the Fresard-Hoberg-Phillips data library (Fr´ esard et al. (2019)). The financial controls include the
buyer’s and the target’s vertical upstream potential relatedness to the counterparty, the logarithm of total assets, the logarithm of sales, and the logarithm of
market value. The text-based vertical relatedness measure is from the Fresard-Hoberg-Phillips data library (Fr´ esard et al. (2019)). Financial data is collected
from COMPUSTAT. I include industry fixed effects by using text-based industry groups (Hoberg and Phillips (2016)). Significance levels are indicated: * = 10
percent, ** = 5 percent, *** = 1 percent.
Panel A: LDA Topic Distribution across Target and Control-Target Firms
Target LDA Type Pseudo-Target Firm Target Firm Total
No: % % No: % % No: % %
Type 1 (Collaborative Culture) 110 78:6 50:5 108 77:1 49:5 218 77:9 100:0
Type 2 (Innovative Culture) 19 13:6 51:4 18 12:9 48:6 37 13:2 100:0
Type 3 (Customer-Centric Culture) 11 7:9 44:0 14 10:0 56:0 25 8:9 100:0
Total 140 100:0 50:0 140 100:0 50:0 280 100:0 100:0
65
Panel B: LDA Topic Similarity and Likelihood of Being Targeted
Similarity in LDA topic
distribution
Dummy = 1 if having the same
highest LDA topic
(1) (2) (3) (4)
Similarity in LDA topic at (t1) 0:189

0:201

0:137

0:131

(0:099) (0:104) (0:065) (0:068)
Year Fixed Effect Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes
Financial Controls Yes Yes Yes Yes
Observations 305 291 305 291
AdjustedR
2
0.027 0.036 0.030 0.036
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
66
Table VI
Merger Volume and Mechanism Test
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public merger during the period 2004 - 2016,
of which both buyer and target are matched to COMPUSTAT and CRSP. The outcome variable is a dummy variable to indicate whether the company pair is
merged. The control target is constructed by matching the target firm, using industry, total assets, sales, and market value. Similarity measures for M&A att
represent how close an acquiring firm’s CEO letter is to a target firm’s CEO letter at (t1), lying in the interval of (0;1). Potential mechanisms are measured
by three proxies and indicated at the top of the table. Columns (1) and (2) use the target firm’s labor intensity. Labor intensity is calculated as the number of
employment, divided by property, plant, and equipment. The relevant variables are compiled from COMPUSTAT. Columns (3) and (4) use the logarithm of the
number of business segments of target firms, divided by the market value. The number of business segments is compiled from COMPUSTAT Segments. The
financial controls include the buyer’s and the target’s vertical upstream potential relatedness to the counterparty, the logarithm of total assets, the logarithm of
sales, and the logarithm of market value. The text-based vertical relatedness measure is from the Fresard-Hoberg-Phillips data library (Fr´ esard et al. (2019)).
Financial data is collected from COMPUSTAT. I include industry fixed effects by using text-based industry groups (Hoberg and Phillips (2016)). Significance
levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Target’s labor intensity Target’s business segment
(1) (2) (3) (4)
Similarity in CEO letter at (t1) 0:551

0:403 0:381 0:417
(0:295) (0:306) (0:372) (0:419)
Mechanism 0:285 0:324 17:183

21:765
(0:586) (0:666) (9:119) (14:633)
Similarity in CEO letter at (t1)
Mechanism
8:411

9:973

150:714

190:922

(4:313) (5:126) (89:875) (113:423)
Year Fixed Effect Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes
Financial Controls No Yes No Yes
Observations 286 272 220 209
AdjustedR
2
0.023 0.030 0.035 0.066
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
67
Table VII
Change of Similarity in CEO Letters and Post-Merger Performance
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public merger during the period
2004 - 2016, of which both buyer and target are matched to COMPUSTAT and CRSP. The outcome variable is the post-merger performance. In
panel A, change in similarity measures for M&A att represent how much an acquiring firm’s CEO letter at (t+1) becomes similar to a target
firm’s CEO letter at (t 1), scaled by the similarity at (t 1). In panel B, similarity change is scaled by the natural logarithm of the relative
asset size of the buyer and the target. Columns (1) and (2) use the change in Tobin’s Q to measure post-merger performance. Columns (3)
and (4) use the change in return on assets to measure post-merger performance. Columns (5) and (6) use the change in SG&A and sales ratio
to measure post-merger performance. The financial controls include the buyer’s and the target’s vertical upstream potential relatedness to the
counterparty, the logarithm of total assets, the logarithm of sales, and the logarithm of market value. The text-based vertical relatedness measure
is from the Fresard-Hoberg-Phillips data library (Fr´ esard et al. (2019)). Financial data is collected from COMPUSTAT. I include industry fixed
effects by using text-based industry groups (Hoberg and Phillips (2016)). Significance levels are indicated: * = 10 percent, ** = 5 percent, ***
= 1 percent.
Panel A: Before Considering Relative Size
Change in Tobin’s Q Change in ROA Change in SG&A and sales ratio
(1) (2) (3) (4) (5) (6)
Change in CEO letter similarity 0:012

0:013

0:101

0:094

0:029

0:030

(0:001) (0:002) (0:016) (0:034) (0:016) (0:013)
Year Fixed Effect Yes Yes Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes Yes Yes
Financial Controls No Yes No Yes No Yes
Observations 211 210 211 210 130 130
AdjustedR
2
0.299 0.359 0.030 0.079 0.056 0.214
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
68
Panel B: After Considering Relative Size
Change in Tobin’s Q Change in ROA Change in SG&A and sales ratio
(1) (2) (3) (4) (5) (6)
Change in CEO letter similarity 0:006

0:007

0:053

0:049

0:002

0:001
(0:001) (0:001) (0:008) (0:016) (0:001) (0:001)
Year Fixed Effect Yes Yes Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes Yes Yes
Financial Controls No Yes No Yes No Yes
Observations 211 210 211 210 130 130
AdjustedR
2
0.297 0.352 0.030 0.079 0.061 0.118
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
69
Table VIII
Change of Similarity in CEO Letters and Post-Merger Divestiture
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public merger during the
period 2004 - 2016, of which both buyer and target are matched to COMPUSTAT and CRSP. The outcome variable is the dummy to indicate
whether the acquiring firms undergo any divestiture or spin-off in the first year after the M&A transaction. In columns (1) and (2), change
in similarity measures for M&A at t represents how much an acquiring firm’s CEO letter at (t + 1) becomes similar to a target firm’s CEO
letter at (t 1), scaled by the similarity at (t 1). In columns (3) and (4), similarity change is scaled by the natural logarithm of the relative
asset size of the buyer and the target. The text-based vertical relatedness measure is from the Fresard-Hoberg-Phillips data library (Fr´ esard
et al. (2019)). Financial data is collected from COMPUSTAT. I include industry fixed effects by using text-based industry groups (Hoberg and
Phillips (2016)). Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Change in similarity before
considering relative size
Change in similarity after
considering relative size
(1) (2) (3) (4)
Change in CEO letter similarity 0:010

0:012

0:005

0:005

(0:004) (0:005) (0:002) (0:002)
Year Fixed Effect Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes
Financial Controls No Yes No Yes
Observations 211 210 211 210
AdjustedR
2
0.016 -0.005 0.012 0.009
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
70
Table IX
Similarity in CEO Letters and Buyer’s Announcement Return
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public
merger during the period 2004 - 2016, of which both buyer and target are matched to COMPUSTAT and CRSP. The
outcome variable is the cumulative abnormal return of buyers around the merger announcement. Similarity measures for
M&A att represent how close an acquiring firm’s CEO letter is to a target firm’s CEO letter at(t1), lying in the interval
of (0;1). Columns (1) and (2) use cumulative abnormal return over the three-day window around the merger announce-
ment. Columns (3) and (4) use cumulative abnormal return over the five-day window around the merger announcement.
The financial controls include the ratio of transaction value to buyer’s market value, the buyer’s and the target’s vertical
upstream potential relatedness to the counterparty, the logarithm of total assets, the logarithm of sales, and the logarithm
of market value. The text-based vertical relatedness measure is from the Fresard-Hoberg-Phillips data library (Fr´ esard
et al. (2019)). Financial data is collected from COMPUSTAT. I include industry fixed effects by using text-based industry
groups (Hoberg and Phillips (2016)). I include industry fixed effects by using text-based industry groups (Hoberg and
Phillips (2016)). Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
CAR[-1,1] CAR[-2,2]
(1) (2) (3) (4)
Similarity in CEO letter at (t1) 0:055 0:140

0:072

0:079
(0:034) (0:056) (0:037) (0:071)
Financial Controls Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes
Observations 232 75 232 75
AdjustedR
2
0.055 0.189 0.004 0.080
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
71
Table X
Buyer’s Announcement Return and Mechanism Test
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public merger during the period 2004 - 2016, of which both
buyer and target are matched to COMPUSTAT and CRSP. The outcome variable is the cumulative abnormal return of buyers over the three-day window around the merger
announcement. Similarity measures for M&A att represent how close an acquiring firm’s CEO letter is to a target firm’s CEO letter at (t1), lying in the interval of (0;1).
In panel A, the target’s information asymmetry level is measured by three proxies and indicated at the top of the table. Columns (1) and (2) use the natural logarithm of the
number of analyst following. Columns (3) and (4) use the standard deviation in analyst forecasts on EPS. Columns (5) and (6) use the analyst forecasts error in EPS, scaled by
the actual EPS. I collect the data on the analyst forecasts from IBES database. In panel B, the buyer’s agency costs are measured by the inverse of incentive alignment between
the shareholders and the management team, which are measured by three proxies and indicated at the top of the table. Columns (1) and (2) use the percentage of managerial
equity ownership. Columns (3) and (4) use the stock option values granted to executives. Columns (5) and (6) use the long-term incentive payouts for the management
team. I collect the data on the executives’ compensation from the ExecuComp database. The financial controls include the buyer’s and the target’s vertical upstream potential
relatedness to the counterparty, the logarithm of total assets, the logarithm of sales, and the logarithm of market value. The text-based vertical relatedness measure is from the
Fresard-Hoberg-Phillips data library (Fr´ esard et al. (2019)). Financial data is collected from COMPUSTAT. I include industry fixed effects by using text-based industry groups
(Hoberg and Phillips (2016)). Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Panel A: Information Asymmetry
Log (Number of analyst following)
Standard deviation in analyst
forecasts
Analyst forecast error
(1) (2) (3) (4) (5) (6)
Similarity in CEO letter at (t1)
Information asymmetry
0:104

0:115

0:172

0:167

0:022 0:020
(0:052) (0:054) (0:068) (0:079) (0:014) (0:016)
Similarity in CEO letter at (t1) Yes Yes Yes Yes Yes Yes
Mechanism Variable Yes Yes Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes Yes Yes
Financial Controls No Yes No Yes No Yes
Observations 192 191 172 171 187 186
AdjustedR
2
0.062 0.058 0.046 0.026 0.037 0.026
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
72
Panel B: Agency Costs
Percentage of total shares
owned
Options granted black scholes
Long-term incentive plan
payouts
(1) (2) (3) (4) (5) (6)
Similarity in CEO letter at (t1)
Incentive alignment
0:002 0:002 0:003 0:000 0:017 0:026
(0:005) (0:006) (0:011) (0:011) (0:018) (0:016)
Similarity in CEO letter at (t1) Yes Yes Yes Yes Yes Yes
Mechanism Variable Yes Yes Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes Yes Yes
Financial Controls No Yes No Yes No Yes
Observations 189 187 189 187 189 187
AdjustedR
2
0.051 0.124 0.053 0.124 0.061 0.140
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
73
Table XI
Change of Similarity and CEO Changes
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public merger during the period
2004 - 2016, of which both buyer and target are matched to COMPUSTAT and CRSP. The outcome variable is indicated at the top of the table.
In columns (1) and (2), change in similarity measures for M&A att represents how much an acquiring firm’s CEO letter at (t + 1) becomes
similar to a target firm’s CEO letter at (t 1), scaled by the similarity at (t 1). In columns (3) and (4), similarity change is scaled by the
natural logarithm of the relative asset size of the buyer and the target. The dummy variable for CEO replacement equals to one if the acquirer’s
CEO leave the company during the merger year or the following year. The dummy variable related to the target firm’s CEO equals to one if the
target firm’s CEO at (t1) stays in the combined firm as an executive as of (t+1). The financial controls include the buyer’s and the target’s
vertical upstream potential relatedness to the counterparty, the logarithm of total assets, the logarithm of sales, and the logarithm of market
value. The text-based vertical relatedness measure is from the Fresard-Hoberg-Phillips data library (Fr´ esard et al. (2019)). Financial data is
collected from COMPUSTAT. I include industry fixed effects by using text-based industry groups (Hoberg and Phillips (2016)). Significance
levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Change in similarity before
considering relative size
Change in similarity after
considering relative size
(1) (2) (3) (4)
Dummy = 1 if acquiring firm replaces CEO duringt ort+1 0:045 0:099 0:099 0:208
(0:327) (0:364) (0:498) (0:573)
Dummy = 1 if target firm’s CEO stays as oft+1 1:296 1:456 1:528

1:917

(0:923) (0:971) (0:872) (0:945)
Year Fixed Effect Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes
Financial Controls No Yes No Yes
Observations 141 140 141 140
AdjustedR
2
0.113 0.119 0.173 0.157
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
74
Table XII
Cultural Change around CEO Change
This table displays Pearson correlation coefficients between the CEO letter similarity before and after a change in CEO and various CEO
and firm characteristics. The unit of observation is a CEO replacement during the period 2012 - 2016, which are collected from ExecuComp
database. The total number of observations is 450. Similarity measures between before and after the CEO change att represents how close a
firm’s CEO letter att is to its CEO letter at (t 1), lying in the interval of (0;1). CEO data on age and gender are collected from Compustat
Executive dataset. COMPUSTAT database provides the financial information measured at (t 1). Significance levels are indicated: * = 10
percent, ** = 5 percent, *** = 1 percent.
(1) (2) (3) (4) (5) (6)
Dummy
= 1 if
replaced
by same
gender
Dummy
= 1 if
male to
female
Dummy
= 1 if
male to
male
Dummy
= 1 if
female to
male
Dummy
= 1 if
female to
female
Dummy
= 1 if
replaced
by insider
CEO letter similarity before and after CEO replacement 0:009 0:002 0:010 0:014 0:089

0:014
(7) (8) (9) (10) (11) (12) (13)
Total
assets
t1
Market
value
t1
Sales
t1
Tobin’s
Q
t1
ROA
t1
ROE
t1
ROI
t1
CEO letter similarity before and after CEO replacement 0:018 0:083

0:097

0:078

0:139

0:146

0:133

p< 0:10,

p< 0:05,

p< 0:01
75
Internet Appendix to
“Corporate Culture in M&As: Evidence from CEO Letters to Shareholders”
A. LDA Topic Validation Tests
There are some existing measures of corporate culture developing by previous literature. One might
wonder how my measure is distinct from theirs. Table A1 provides correlations between my measure and
two alternative measures of corporate culture. The purpose of this experiment is to show that the measures
used in this paper are corroborated with other cultural measurements, which are developed by some reliable
resources. At the same time, it will demonstrate that my measures can quantify novel aspects of corporate
culture, which are not captured by other proxies.
First, Li et al. (2020) implement a word embedding model and analyze what top executives elaborate
in their discussion with analysts in earnings conference calls. They create five corporate cultural values
of innovation, integrity, quality, respect, and teamwork. There are both commonalities and differences
between my approach and their approach. They focus on the messages delivered by top management, as
I do in this paper. However, their method is distinct from mine by using CEO conversations aimed to
different audiences and purposes. In conference calls, the management team and the analysts discuss the
firm performance. The discussion’s primary goal is to help the analysts form accurate predictions on the
firm’s capital market performance. Also, due to the interactive nature, the discussion is steered by analysts
to address their questions. In comparison, CEO letters to shareholders have a broader audience and are
scripted more carefully. Second, CEO letters allow CEOs to provide comprehensive information they want
to emphasize, beyond financial performance, such as ethical values and shared goals.
Panel A tests the correlations between my LDA measure and the corporate culture measure developed
by Li et al. (2020). Overall, my LDA measure and the measure developed by Li et al. (2020) seem to
provide both confirming and contradicting results. The collaborative culture has positive correlations with
integrity scores and respect scores. And the innovative culture is positively related to their innovation scores.
However, collaborative culture is negatively associated with teamwork scores. And the customer-centric
culture has a negative coefficient with quality scores.
Second, MIT Sloan Management Review provides a cultural measure for a subset of large companies.
8
It scores nine corporate cultural values of agility, collaboration, customer-driven, diversity, execution, inno-
8
https://sloanreview.mit.edu/culture500/
A.1
vation, integrity, performance, and respect, using a data set of 1.4 million employee reviews from Glassdoor.
It defines a list of words for each cultural value and calculates the percentage of a firm’s reviews that dis-
cuss those words. There are several dissimilarities between my measure and MIT Sloan measure. First, my
measure and MIT Sloan’s measure have different underlying assumption on corporate culture. I consider
corporate culture defined by the top-down approach, while MIT Sloan potentially captures the bottom-up
culture. For instance, the MIT Sloan measure might help understand the corporate culture shared by em-
ployees by analyzing their anonymous comments on the firm. Second, the MIT Sloan measure does not
provide time-specific measures about corporate culture, whereas my measure provides yearly measures of
culture. Last, the MIT scores are based on an arbitrary word list, defined by researchers. Therefore, it might
be sensitive to the their subjective choice. By using an unsupervised machine learning algorithm, I can be
free from researcher-induced prejudice.
In panel B, I explore the cultural measures provided by the MIT Sloan Management Review. Like
the results in panel A, my LDA measure confirms the measure developed by the MIT Sloan Management
Review in some cases but also provides distinct features of the corporate culture. The collaborative culture
has positive associations with collaboration scores and integrity scores. The innovative culture is positively
related to agility scores and innovation scores. However, the customer-centric culture does not have a
positive coefficient with customer scores.
Although cross-validating alternative corporate culture measures is beyond the scope of this paper, the
internet appendix provides some analysis for the interested readers. Table A2 correlates Li et al. (2020)
scores and the MIT Sloan measures.
Overall, my LDA measures correlate well with the MIT Sloan measures. In some cases, they do not
seem to correlate with the Li et al. (2020) scores. This might be because we use different methods in a
different context.
A.2
Table A1
LDA Topic Cross-Validation
This table displays Pearson correlation coefficients between LDA topic distribution and various CEO and firm characteristics. LDA analysis assigns each
CEO letter with probabilities on three topics. The variable of interest is indicated at the top of the table. The unit of observation is a CEO letter of buyers,
targets, and pseudo-targets around the years of merger transactions that occurred during the period 2004 - 2016. Panel A investigates the measure developed
by LDA analysis and the cultural value score developed by Li et al. (2020). Li et al. (2020) score the five corporate cultural values of innovation, integrity,
quality, respect, and teamwork, using earnings conference calls. Panel B analyzes the cultural measure for some large companies, deveopled by MIT Sloan
Management Review (https://sloanreview.mit.edu/culture500/). It scores the nine corporate cultural values of agility, collaboration, customer-driven, diversity,
execution, innovation, integrity, performance, and respect, using a data set of 1.4 million employee reviews from Glassdoor. I manually match the corporate
names to GVKEY in COMPUSTAT universe. Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Panel A: Li et al. (2020) Measure
(1) (2) (3) (4) (5)
Integrity score Teamwork score Innovation score Respect score Quality score
Probabilities assigned to collaborative culture 0:16

0:04

0:27

0:08

0:21

Probabilities assigned to innovative culture 0:13

0:09

0:32

0:05

0:26

Probabilities assigned to customer-centric culture 0:05

0:07

0:06

0:06

0:06

p< 0:10,

p< 0:05,

p< 0:01
Panel B: MIT Sloan Culture 500 Measure
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Agility
score
Collabo-
ration
score
Customer
score
Diversity
score
Execution
score
Innova-
tion score
Integrity
score
Perfor-
mance
score
Respect
score
Probabilities assigned to collaborative culture 0:12 0:30

0:23

0:09

0:01 0:16

0:12

0:02 0:05
Probabilities assigned to innovative culture 0:20

0:35

0:23

0:15

0:04 0:21

0:13

0:06

0:03
Probabilities assigned to customer-centric culture 0:13

0:07

0:02 0:10

0:06 0:08

0:01 0:08

0:04

p< 0:10,

p< 0:05,

p< 0:01
A.3
Table A2
Li et al. (2020) Scores and MIT Sloan’s Scores
This table displays Pearson correlation coefficients between Li et al. (2020) score and the
cultural measure for some large companies, deveopled by MIT Sloan Management Review
(https://sloanreview.mit.edu/culture500/). Li et al. (2020) score the five corporate cultural values of in-
novation, integrity, quality, respect, and teamwork, using earnings conference calls. MIT Sloan scores the
nine corporate cultural values of agility, collaboration, customer-driven, diversity, execution, innovation,
integrity, performance, and respect, using a data set of 1.4 million employee reviews from Glassdoor. I man-
ually match the corporate names to GVKEY in COMPUSTAT universe. The unit of observation is a CEO
letter of buyers, targets, and pseudo-targets around the years of merger transactions that occurred during the
period 2004 - 2016. Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
(1) (2) (3) (4) (5)
Li et al.
integrity
score
Li et al.
teamwork
score
Li et al.
innovation
score
Li et al.
respect score
Li et al.
quality score
MIT agility score 0:11

0:03 0:23

0:02 0:20

MIT collaboration score 0:10

0:06 0:06

0:07

0:19

MIT customer score 0:21

0:05 0:10

0:08

0:20

MIT diversity score 0:06

0:03 0:01 0:03 0:10

MIT execution score 0:15

0:07

0:20

0:10

0:07

MIT innovation score 0:03 0:22

0:27

0:03 0:26

MIT integrity score 0:04 0:01 0:15

0:12

0:09

MIT performance score 0:09

0:04 0:10

0:09

0:05
MIT respect score 0:04 0:07

0:07

0:05 0:01

p< 0:10,

p< 0:05,

p< 0:01
A.4
Table A3
Li et al. (2020) Scores and CEO & Firm Characteristics
This table displays Pearson correlation coefficients between Li et al. (2020) score and various CEO and
firm characteristics. Li et al. (2020) score the five corporate cultural values of innovation, integrity, quality,
respect, and teamwork, using earnings conference calls. The unit of observation is a CEO letter of buyers,
targets, and pseudo-targets around the years of merger transactions that occurred during the period 2004 -
2016. Panel A shows the correlation coefficients on CEO characteristics. CEO and executive compensation
data are collected from Compustat Executive dataset. CEO’s relative compensation is CEO compensation di-
vided by the average compensation of executives or by sales amount. Panel B shows the innovation activity,
using R&D intensity, and the sum of new patent value. R&D intensity is measured by research and devel-
opment expense, scaled by total asset. Patent data is collected by Stoffman et al. (2019). Panel C shows a
firm’s customer relationship, using the American Customer Satisfaction Index (ACSI) score and brand value.
ACSI score is collected from www.theacsi.org. And I collect the brand value from www.interbrand.com.
I manually match the company name to the GVKEY universe. Significance levels are indicated: * = 10
percent, ** = 5 percent, *** = 1 percent.
Panel A: CEO Characteristics
(1) (2) (3) (4)
Dummy = 1 if
female CEO
CEO age
CEO compensation
relative to other
executives
compensation
CEO
compensation,
divided by sales
Integrity score 0:03 0:09

0:03 0:01
Teamwork score 0:02 0:05

0:10

0:22

Innovation score 0:11

0:10

0:14

0:04
Respect score 0:01 0:02 0:05

0:06

Quality score 0:02 0:04 0:07

0:02

p< 0:10,

p< 0:05,

p< 0:01
A.5
Panel B: Innovation Activity
(1) (2)
R&D intensity Sum of new patents value
Integrity score 0:09

0:12

Teamwork score 0:41

0:09

Innovation score 0:14

0:05

Respect score 0:04

0:09

Quality score 0:12

0:02

p< 0:10,

p< 0:05,

p< 0:01
Panel C: Customer Relationship
(1) (2)
ACSI score Brand value (in billion)
Integrity score 0:17

0:28

Teamwork score 0:18

0:04
Innovation score 0:04 0:05
Respect score 0:19

0:11
Quality score 0:30

0:12

p< 0:10,

p< 0:05,

p< 0:01
A.6
Table A4
MIT Sloan’s Scores and CEO & Firm Characteristics
This table displays Pearson correlation coefficients between the cultural measure for some large compa-
nies, deveopled by MIT Sloan Management Review (https://sloanreview.mit.edu/culture500/) and various
CEO and firm characteristics. It scores the nine corporate cultural values of agility, collaboration, customer-
driven, diversity, execution, innovation, integrity, performance, and respect, using a data set of 1.4 million
employee reviews from Glassdoor. I manually match the corporate names to GVKEY in COMPUSTAT
universe. The unit of observation is a CEO letter of buyers, targets, and pseudo-targets around the years of
merger transactions that occurred during the period 2004 - 2016. Panel A shows the correlation coefficients
on CEO characteristics. CEO and executive compensation data are collected from Compustat Executive
dataset. CEO’s relative compensation is CEO compensation divided by the average compensation of ex-
ecutives or by sales amount. Panel B shows the innovation activity, using R&D intensity, and the sum of
new patent value. R&D intensity is measured by research and development expense, scaled by total asset.
Patent data is collected by Stoffman et al. (2019). Panel C shows a firm’s customer relationship, using
the American Customer Satisfaction Index (ACSI) score and brand value. ACSI score is collected from
www.theacsi.org. And I collect the brand value from www.interbrand.com. I manually match the company
name to the GVKEY universe. Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1
percent.
Panel A: CEO Characteristics
(1) (2) (3) (4)
Dummy = 1 if
female CEO
CEO age
CEO compensation
relative to other
executives
compensation
CEO
compensation,
divided by sales
Agility score 0:10

0:14

0:09

0:14

Collaboration score 0:02 0:25

0:11

0:17

Customer score 0:11

0:03 0:03 0:25

Diversity score 0:15

0:08 0:09

0:11

Execution score 0:09

0:15

0:02 0:05
Innovation score 0:01 0:06 0:22

0:14

Integrity score 0:01 0:22

0:09

0:04
Performance score 0:12

0:14

0:08

0:01
Respect score 0:03 0:20

0:03 0:05

p< 0:10,

p< 0:05,

p< 0:01
A.7
Panel B: Innovation Activity
(1) (2)
R&D intensity Sum of new patents value
Agility score 0:36

0:05
Collaboration score 0:22

0:11

Customer score 0:36

0:19

Diversity score 0:06 0:22

Execution score 0:31

0:15

Innovation score 0:30

0:16

Integrity score 0:30

0:06
Performance score 0:22

0:15

Respect score 0:30

0:02

p< 0:10,

p< 0:05,

p< 0:01
Panel C: Customer Relationship
(1) (2)
ACSI score Brand value (in billion)
Agility score 0:28

0:11
Collaboration score 0:51

0:46

Customer score 0:67

0:47

Diversity score 0:24

0:13

Execution score 0:43

0:32

Innovation score 0:10 0:44

Integrity score 0:40

0:01
Performance score 0:09 0:18

Respect score 0:34

0:12

p< 0:10,

p< 0:05,

p< 0:01
A.8
Table A5
Similarity in CEO Letters and Combined Announcement Return
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public merger
during the period 2004 - 2016, of which both buyer and target are matched to COMPUSTAT and CRSP. The outcome variable is
the cumulative abnormal return of buyers and targets around the merger announcement. Similarity measures for M&A att represent
how close an acquiring firm’s CEO letter is to a target firm’s CEO letter at (t 1), lying in the interval of (0;1). Columns (1)
and (2) use cumulative abnormal return over the three-day window around the merger announcement. Columns (3) and (4) use
cumulative abnormal return over the five-day window around the merger announcement. Panel A uses the simple average of two
firms’ cumulative abnormal returns. Panel B uses the average of two firms’ cumulative abnormal returns weighted by their market
value. The text-based vertical relatedness measure is from the Fresard-Hoberg-Phillips data library (Fr´ esard et al. (2019)). Financial
data is collected from COMPUSTAT. I include industry fixed effects by using text-based industry groups (Hoberg and Phillips (2016)).
Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Panel A: Average CAR
Combined CAR[-1,1] Combined CAR[-2,2]
(1) (2) (3) (4)
Similarity in CEO letter at (t1) 0:014 0:028 0:011 0:055
(0:033) (0:199) (0:036) (0:209)
Ratio of transaction value to buyer’s
market value
0:014 0:103
(0:097) (0:104)
Dummy=1 if in different text-based
industry
0:043 0:066
(0:055) (0:059)
Buyer’s vertical upstream potential
relatedness to target
1:907 2:039
(7:103) (7:561)
Buyer’s Log (Total Assets) 0:048 0:039
(0:047) (0:047)
Buyer’s Log (Sale) 0:020 0:018
(0:029) (0:031)
Buyer’s Log (Market Value) 0:050 0:030
(0:057) (0:059)
Buyer’s Book to Market 0:112 0:088
(0:069) (0:073)
Target’s vertical upstream potential
relatedness to buyer
3:000 4:469
(7:812) (8:099)
Target’s Log (Total Assets) 0:044 0:047
(0:035) (0:035)
Target’s Log (Sale) 0:007 0:010
(0:021) (0:021)
Target’s Log (Market Value) 0:071

0:061
(0:036) (0:039)
Target’s Book to Market 0:108 0:086
(0:076) (0:081)
Year Fixed Effect Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes
Observations 231 75 231 75
AdjustedR
2
0.004 -0.061 0.006 -0.084
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01 A.9
Panel B: Weighted Average CAR
Combined CAR[-1,1] Combined CAR[-2,2]
(1) (2) (3) (4)
Similarity in CEO letter at (t1) 0:038 0:044 0:041 0:024
(0:029) (0:056) (0:032) (0:076)
Ratio of transaction value to buyer’s
market value
0:018 0:083

(0:038) (0:045)
Dummy=1 if in different text-based
industry
0:030 0:063

(0:018) (0:023)
Buyer’s vertical upstream potential
relatedness to target
2:880 1:326
(2:332) (3:307)
Buyer’s Log (Total Assets) 0:036

0:025
(0:015) (0:018)
Buyer’s Log (Sale) 0:014 0:018
(0:012) (0:015)
Buyer’s Log (Market Value) 0:011 0:014
(0:019) (0:024)
Buyer’s Book to Market 0:024

0:002
(0:013) (0:018)
Target’s vertical upstream potential
relatedness to buyer
1:054 0:320
(2:310) (3:153)
Target’s Log (Total Assets) 0:015

0:018

(0:009) (0:011)
Target’s Log (Sale) 0:006 0:009
(0:007) (0:008)
Target’s Log (Market Value) 0:014 0:022

(0:010) (0:013)
Target’s Book to Market 0:016 0:011
(0:014) (0:019)
Year Fixed Effect Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes
Observations 229 75 229 75
AdjustedR
2
0.061 0.274 0.007 0.176
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
A.10
Table A6
Robustness Check with Different LDA Topic Number
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public
merger during the period 2004 - 2016, of which both buyer and target are matched to COMPUSTAT and CRSP. The
outcome variable is a dummy variable to indicate whether the company pair is merged. The control target is constructed by
matching the target firm, using industry, total assets, sales, and market value. Similarity measures for M&A att represent
how close an acquiring firm’s CEO letter is to a target firm’s CEO letter at (t 1), in terms of topical distribution. In
columns (1) and (2), the main explanatory variable is the cosine similarity of LDA topic distribution across four topics.
In columns (1) and (2), the main explanatory variable is the cosine similarity of LDA topic distribution across seventy-
five topics. The text-based vertical relatedness measure is from the Fresard-Hoberg-Phillips data library (Fr´ esard et al.
(2019)). Financial data is collected from COMPUSTAT. I include industry fixed effects by using text-based industry
groups (Hoberg and Phillips (2016)). Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Similarity in LDA topic
distribution with four topics
Similarity in LDA topic
distribution with seventy five topics
(1) (2) (3) (4)
Similarity in LDA topic at (t1) 0:128 0:199

0:095 0:222

(0:094) (0:099) (0:108) (0:117)
Buyer’s vertical upstream potential
relatedness to target
25:695

25:233

(13:969) (13:892)
Buyer’s Log (Total Assets) 0:049 0:063
(0:064) (0:063)
Buyer’s Log (Sale) 0:067 0:050
(0:048) (0:048)
Buyer’s Log (Market Value) 0:151

0:150

(0:066) (0:066)
Buyer’s Book to Market 0:364

0:359

(0:093) (0:094)
Target’s vertical upstream potential
relatedness to buyer
30:827

29:642

(13:661) (13:682)
Target’s Log (Total Assets) 0:003 0:003
(0:042) (0:042)
Target’s Log (Sale) 0:015 0:008
(0:038) (0:036)
Target’s Log (Market Value) 0:003 0:008
(0:039) (0:040)
Target’s Book to Market 0:067 0:076
(0:085) (0:086)
Year Fixed Effect Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes
Observations 319 297 319 297
AdjustedR
2
0.004 0.046 0.001 0.045
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
A.11
Table A7
Robustness Check Including Li et al. (2020) Measure
This table shows the estimates from linear regressions using various specifications. The unit of observation is a public merger during the period
2004 - 2016, of which both buyer and target are matched to COMPUSTAT and CRSP. The outcome variable is a dummy variable to indicate
whether the company pair is merged. The control target is constructed by matching the target firm, using industry, total assets, sales, and market
value. Similarity measures for M&A att represent how close an acquiring firm’s CEO letter is to a target firm’s CEO letter at (t1), lying in
the interval of (0;1). In column (1), the similarity measure incorporates every word from CEO letters. In column (2), words mentioned in the
10-K business description are excluded when the similarity measure is constructed. In column (3), the main explanatory variable is the cosine
similarity of LDA topic distribution across three topics. In column (4), the main explanatory variable is a dummy variable, which represents
whether the buyer and the target have the highest probability mass on the same topic out of three topics. Similarity in Li et al. (2020) measure
is the cosine similarity of Li et al. (2020) cultural scores over the five corporate cultural values. Li et al. (2020) score the five corporate cultural
values of innovation, integrity, quality, respect, and teamwork, using earnings conference calls. The text-based vertical relatedness measure is
from the Fresard-Hoberg-Phillips data library (Fr´ esard et al. (2019)). Financial data is collected from COMPUSTAT. I include industry fixed
effects by using text-based industry groups (Hoberg and Phillips (2016)). Significance levels are indicated: * = 10 percent, ** = 5 percent, ***
= 1 percent.
Similarity including
every word
Similarity excluding
words in 10-K
business description
Similarity in LDA
topic distribution
Dummy = 1 if
having the same
highest LDA topic
(1) (2) (3) (4)
Similarity in CEO letter at (t1) 0:835

0:501 0:109 0:107
(0:265) (0:342) (0:135) (0:084)
Similarity in Li et al. measure at (t1) 0:097 0:277 0:327 0:331
(0:403) (0:435) (0:444) (0:446)
Year Fixed Effect Yes Yes Yes Yes
Text-Based Industry Fixed Effect Yes Yes Yes Yes
Financial Controls Yes Yes Yes Yes
Observations 243 192 192 192
AdjustedR
2
0.090 0.099 0.095 0.101
Standard errors in parentheses

p< 0:1,

p< 0:05,

p< 0:01
A.12
Chapter 2
Do Firms Leave Workers in the Dark
Before Wage Negotiations?
I. Introduction
Labor unions are important corporate stakeholders. According to the Bureau of
Labor Statistics, in 2018, more than 16.4 million employees, or 12% of the total
U.S. workforce, were represented by unions. Prior research finds that unions in-
fluence a host of corporate decisions, including governance (Freeman and Kleiner
(1990); Agrawal (2011); Matsusaka et al. (2019)), investment (Fallick and Has-
sett (1999); Yi (2016); Faleye et al. (2006)), and the cost of capital (Falato and
Liang (2016)). The response of managers to labor unions is especially critical
during union negotiations. For example, during negotiations, firms reduce cash
holdings (Klasa et al. (2009)), increase leverage (Matsa (2010)), and implement
value-destroying worker-management alliances (Atanassov and Kim (2009); Kim
and Ouimet (2011, 2014)).
This paper studies another important action by managers when negotiating with
labor unions: the strategic disclosure of information. Strategically withholding
information has both costs and benefits. Withholding information is costly be-
76
cause it increases adverse selection costs of capital (Myers and Majluf (1984)).
However, withholding information can be valuable if it gives managers an in-
formation advantage over labor unions during negotiations. Early information
economics models suggest that firms always fully disclose information in equilib-
rium (Grossman (1981); Milgrom (1981)). However, subsequent research allows
for disclosure costs imposed by bargaining opponents, which generate equilibria
with partial-disclosure (Wagenhofer (1990)). In this partial-disclosure equilib-
rium, firms withhold both favorable and unfavorable information to balance the
beliefs of unions and adverse capital market reaction. Since wage negotiations are
repeated bargaining games, reputation concerns works as a commitment device
for a firm to conceal both good and bad news (Einhorn and Ziv (2008)).
This paper investigates whether labor negotiations cause employers to strategi-
cally manipulate their information environment to enhance their bargaining posi-
tion with labor unions. I also show that theoretically motivated factors influence
this behavior, including information uncertainty, growth opportunities, financial
constraints, and the cost of a strike.
To establish a valid causal link between labor bargaining activities and strategic
disclosure, I exploit within-firm time-series variation in the management’s incen-
tive to withhold information contained in material business agreements. My iden-
tification strategy follows Matsusaka et al. (2019) and Yi (2016) in using collective
labor contract expiration dates, which are plausibly exogenous to a firm’s other
business activities. Once a union becomes a bargaining representative in a certain
77
workplace, a committee of employees and union representatives negotiate con-
tracts over wages, benefits, and working conditions in a collective way (“collective
bargaining”) and handle disputes with management. Ex post negotiating parties
do not time the contract expiration nor initiate the negotiation process (Cramton
and Tracy (1992); Rich and Tracy (2013); Matsusaka et al. (2019)). Therefore,
contract expirations will be uncorrelated with the nature and occurrence of mate-
rial business agreement, which my empirical findings confirm. Hence, changes in
disclosure policy around contract expirations can be attributed to an incentive to
strategically withhold information during labor negotiations.
I use a difference-in-difference analysis to compare the likelihood of strategic
disclosure before imminent labor negotiations to the likelihood of strategic dis-
closure when negotiations are not imminent. The treatment group is composed of
unionized firms which have wage negotiations in the following year. The control
group include unionized firms which do not have expiring wage contracts.
1
To measure strategic disclosure, I use confidential treatment orders (CTOs). A
CTO is an SEC order that approves firms’ requests to redact particular compo-
nents of material business agreements from their public filings. It provides several
advantages for analyzing strategic disclosures. First, a CTO is a clean signal of a
firm’s decision to withhold information. Analysis using CTOs has a comparative
advantage over analysis using voluntary disclosure, such as management forecasts
or conference calls. In voluntary disclosure, managers have authority to choose the
1
In some cases, a firm has overlapping collective bargaining contracts with multiple unions. In these cases, it is harder to find
significant within-firm variations in disclosure policy.
78
content and timing of disclosures. Second, redacting firms are required to note the
omitted parts of their filings, using black-lining or asterisks. Therefore, the public
can observe whether and which components of agreements a firm conceals. Also,
since the SEC rarely denies CTO requests, the public can have almost a complete
set of strategic disclosure decisions intended by management. Third, the infor-
mation approved for a CTO is exempted from the Freedom of Information Act
(FOIA) requests of employees and unions. Under the National Labor Relations
Act Section 8(d) and the FOIA, employers have a legal duty to bargain in good
faith with their employees’ representatives. Employers are obligated to furnish the
union with information which is relevant to the bargaining process or the employ-
ees’ terms or conditions of employment. However, one of the FOIA exemptions is
confidential commercial or financial information.
2
Supported by this exemption,
a CTO provides an effective device to deny FOIA requests and to reinforce the in-
formational barrier. This makes a CTO a powerful instrument for the management
to shield information in labor negotiations.
For my identification strategy to be valid, I rely on the credibility of the fol-
lowing assumption: the value of disclosure exogenously shifts in advance of a
collective bargaining expiration date. In other words, firms facing expiring con-
tracts bear higher costs of sharing information, even as the nature of material
agreements is not systematically different. The following observations support
this claim. First, neither managers nor unions are able to manipulate the timing
2
Freedom of Information Act Guide on exemption 4 is available in the Department of Justice website:
https://www.justice.gov/oip/foia-guide-2004-edition-exemption-4
79
of wage negotiations. Union contracts are rarely renegotiated before their origi-
nal expiration dates (Rich and Tracy (2013); Matsusaka et al. (2019)).
3
Second,
firms are not allowed to time the filings of material contracts. According to the
Sarbanes-Oxley Act (SOX) regulation, a firm shall file its material agreements on
a rapid and current basis. I validate this claim by finding that the nature and the
number of material contracts disclosed in the SEC filings does not significantly
deviate from the mean during the contract expiration years. Third, a firm cannot
arbitrarily initiate nor delay its contracts with third parties. Entering contracts is a
bilateral decision, not unilateral. Hence, while firms have less control over when
to enter the material agreements, they have full control over redactions.
The baseline analysis shows that the firm-years with expiring contracts exhibit
higher frequencies of redactions than those without expiring contracts. In particu-
lar, firm-years with expiring contracts are 2.2 - 3.1 percentage points more likely
to redact material agreements. Considering the overall probability of redaction is
12%, this value is economically substantial.
Next, cross-sectional analysis explores potential modifying factors of strategic
disclosure. These modifying factors illustrate the trade-off between the benefits
and costs of CTOs. First, I predict that the pre-existing level of information asym-
metry will amplify a strategic CTO. Given that a CTO is a bargaining device to
strengthen a firm’s information advantage, the incentive to use this device will
depend on the status quo information environment. If there is limited external
3
Though firms and unions might negotiate with each other even without an imminent collective bargaining expiration, this will,
however, only make it harder to find significant results.
80
information to predict an employer’s future profitability, unions must rely on the
employer’s public disclosure. In this case, a CTO will significantly limit unions’
wage demand, representing an effective bargaining device. To test this prediction,
I run a triple-difference analysis which compares the redaction probability before
expiring contracts for firms with high and low levels of analyst forecast errors. I
confirm my prediction saying that firms with higher external information asym-
metry increase CTOs more than firms with low external information asymmetry.
Second, I predict that firms with low growth opportunities are more likely to
strategically withhold information than firms with high growth opportunities. The-
ory suggests that the greater is the risk of losing employment, the stronger is man-
agement’s bargaining power (Kuhn (1986); Freeman and Kleiner (1999)). And
the costs of losing business are higher for a firm with stronger growth potentials.
Employees of firms with greater growth opportunities can be convinced to set-
tle for lower wages and to wait for a bigger surplus in the future, assuming that
their deferred rents will be better utilized by the employer. Therefore, growth
opportunities strengthen a firm’s bargaining condition and weakens the manage-
ment incentive to implement strategic disclosure. The empirical results confirm
this implication and find that growth opportunities deter a strategic information
withholding.
Next, theory predicts that financially unconstrained firms are more likely to
strategically adopt CTOs before wage negotiations than financially constrained
firms. Organized workers may demand a portion of employers’ excess liquidity,
81
which they interpret as extra surplus (Benmelech et al. (2012)). Prior literature
find that corporate liquidity hurts a firm’s bargaining position since it raises wage
demand (Bronars and Deere (1991); Klasa et al. (2009); Matsa (2010, 2018); My-
ers and Saretto (2016); Chino (2016); Yi (2016)). Therefore, gains from redactions
will be higher in financially unconstrained corporations. The cross-sectional tests
verify my prediction regarding the influence of financial constraints and a strategic
CTO.
Last, I find that firms with low costs of work stoppages are more likely to
conceal information than firms with high costs of work stoppages. Although the
imbalance in information between firms and negotiating unions can strengthen
a firm’s bargaining position, it may lead to a costly work stoppage or holdups
(Reder and Neumann (1980); Mauro (1982); Cramton (1984)). If firms strategi-
cally redact the information before wage negotiation, they will take into account
the fact that it might increase the likelihood of contentious negotiations. I find that
when the cost of a work stoppage is low, strategic redaction is more pronounced.
Next, I perform additional analysis to show that the deviation in disclosure
policy is indeed a strategic choice. First, I test the associations between strate-
gic CTOs and ex-post firm performance. In the partial-disclosure equilibrium,
a firm should conceal both good and bad information to constrain the union on
how to interpret the redacted information. Also, a rational firm will optimally
choose to conceal information about material agreements, up to the point where
the benefit matches the cost. Therefore, the reduced form analysis should not pro-
82
duce any impact on firm performance (Demsetz and Lehn (1985); Duchin et al.
(2010)). Consistent with this prediction, I do not find statistically significant
associations between strategic CTOs and ex-post firm performance. This result
provides suggestive evidence of a trade-off effect of strategic redaction. A firm’s
decision on withholding information should be made in awareness of its poten-
tial consequences. The improved bargaining condition that would be associated
with information advantage should be offset by incremental costs from informa-
tion asymmetry.
Second, I test how different types of managerial actions interact with each other.
Prior literature shows that managers strategically cut liquidity to resist wage de-
mands (Klasa et al. (2009); Matsa (2010); Yi (2016)). This paper shows that if a
firm chooses to raise asset purchases in the year with contract expirations, it does
not increase information asymmetry by redacting material contracts as much as it
would do in the absence of a liquidity cut. Although studying all the interactions
between bargaining devices is beyond the scope of the paper, the exogenous con-
tract expiration dates help me identify the strategic substitutability of two specific
bargaining strategies.
The central contribution of this paper is to provide causal evidence that manage-
ment uses its information advantage as a bargaining device in wage negotiations.
While existing literature shows associations between union strength and firms’ in-
formation environments, these results cannot be interpreted as causal (Liberty and
Zimmerman (1986); DeAngelo and DeAngelo (1991); D’Souza et al. (2000); Hi-
83
lary (2006); Bova (2013); Garc´ ıa Osma et al. (2015); Cheng (2017); Cheng et al.
(2017)). In particular, to investigate the empirical link between the information
environment and union bargaining power, prior studies use various endogenous
factors. If those factors are correlated with other firm fundamentals or omitted
variables, the analysis can lead to spurious correlations. I overcome the poten-
tial endogeneity concern by relating an exogenous shift in bargaining activities
(collective bargaining expiration dates) and a clear measure of strategic disclosure
(CTOs).
My research also contributes to the growing literature on the causal implication
of organized workers on unionized firms. Lee and Mas (2012) argue that union-
ization leads to lower firm value and substantial losses in market value. Prior
literature has offered some explanations on the market value deterioration, such
as principal-agent problem (Faleye et al. (2006); Chyz et al. (2013)), less com-
petitiveness in product markets (Aobdia and Cheng (2018)) and corporate gov-
ernance degeneration (Freeman and Kleiner (1990); Agrawal (2011); Matsusaka
et al. (2019)). I contribute to this stream of research by showing that an increase
in information asymmetry is another potential downside of unionization.
In addition, this paper is one of the first in the literature to show that firms use
their information advantage to gain a bargaining advantage against labor unions.
Existing literature has documented that firms adjust their financial policies to im-
prove their bargaining positions, including taking additional debt and reducing
cash holdings to manage excess liquidity (Klasa et al. (2009); Matsa (2010); Yi
84
(2016)). Strategic disclosure is distinct from financial policy as a bargaining de-
vice. First, it has a broad set of interested parties, which include any potential
information users. Second, by demonstrating strategic disclosure, we can bet-
ter understand the value implication of unionization. Information asymmetry can
be another channel for negative impact on the market value of unionized firms.
Finally, strategic disclosure can have a more immediate impact, while strategic fi-
nancing can take longer to implement and have longer-lasting consequences. For
instance, if a firm adjusts its debt level as a bargaining tool with labor unions, it
must negotiate with its lenders and then readjust the debt level after labor nego-
tiations have ended. In contrast, strategic disclosures are immediate, fully under
the control of management, and easily reversible. This makes strategic disclosure
a more flexible tool for managers and also one that is more easily identified by
researchers, compared to slowly changing debt levels.
Last, I contribute to research on CTOs. Verrecchia and Weber (2006) suggest
that CTOs help firms protect proprietary information. Boone et al. (2016) show
that IPO firms redact proprietary information to shield competitive advantages.
While firms in various life stages routinely use CTOs, their underlying motivations
are rarely explored since it is hard to derive convincing causal implications. I
overcome this problem by investigating contract expirations as an exogenous shift
in information sharing costs. I find union bargaining as an essential determinant
of redaction choice. The SEC recently adopted a new regulation allowing firms
to redact information without filing a request. This amendment makes it more
85
important to understand the underlying motivation of a CTO (SEC (2019)).
II. Information Advantage and Wage Bargaining
Information asymmetry leads to unfavorable financial market outcomes (Myers
and Majluf (1984); Merton (1987); Hoberg and Maksimovic (2014)). Despite the
costs of information asymmetry, full disclosure is not always the best strategy for a
firm. By strategically withholding information, a firm can lower proprietary costs
and preserve its information advantage (Verrecchia (1983); Boone et al. (2016)).
Managers have access to relevant and timely data related firm’s operations and
product market condition. Compared to management, employees and unions are
at an information disadvantage on a firm’s prospects. This information barrier re-
stricts a union’s bargaining activity. Without knowing the firm’s total appropriable
rents, a union cannot offer reasonable counter-offers at the bargaining table. As a
rational player, a union understands that the excessive rent-seeking will increase
the risk of losing business (Freeman and Kleiner (1999)).
This observation implies that a firm faces higher costs of disclosure before wage
negotiations. Without any proprietary costs, firms are forced to disclose full infor-
mation (Grossman (1981); Milgrom (1981)). However, partial-disclosure can be
an optimal policy for a firm when information disclosure induces the bargaining
counterparty to take a strategic action (Wagenhofer (1990)).
In the model describe by Wagenhofer (1990), a firm has two objectives in se-
lecting its disclosure strategy. First, it wishes to avoid the proprietary costs from
86
disclosing information to its bargaining party. In the context of my paper, if a la-
bor union learns favorable information about a firm’s prospective, it will demand
higher rents in wage negotiation. So, the employer has an incentive to disclose
only unfavorable information. Second, a firm is concerned about market reaction.
This second force makes it to disclose favorable information.
In order to sustain the partial disclosure equilibrium, a firm should maintain two
nondisclosure intervals, one with favorable information and another with unfavor-
able information. By pooling favorable and unfavorable information in withheld
information, a firm can balance two opposing forces: one is capital market reac-
tions and another is a union’s rent extraction. This pooling strategy can confuse
the union on how to interpret the concealed information. Since wage negotiations
are repeated bargaining games, reputation concern works as a commitment device
for a firm to conceal both good and bad news (Einhorn and Ziv (2008)).
I demonstrate two pieces of suggestive evidence for the theory of partial-disclosure
equilibrium. First, in the section VII, I test the association between a firm’s strate-
gic disclosure and ex-post firm performance. The test does not find statistically
significant associations. This result suggests that a firm conceals both good and
bad information about future prospects. Second, the baseline analysis shows that
wage negotiations trigger a firm’s information withholding. However, in the untab-
ulated results exploring specific contract types separately, I do not find any strong
associations between wage negotiations and the non-disclosure likelihood of each
contract type. It implies that a firm pools not only how favorable the information
87
is, but also the type of information.
The partial disclosure theory suggests that as proprietary costs of disclosing
favorable information increases, the probability of having partial disclosure equi-
librium increases. Therefore, the main hypothesis I test is that before wage ne-
gotiations, a firm is more likely to pool its good and bad news, as the costs of
information sharing increase. And the size of the proprietary costs will be a func-
tion of many factors, including the status-quo information environment, growth
potentials, financial constraints, and the costs from potential contentions.
III. Institutional Background and Identification Strategy
A. Confidential Treatment Order
According to Regulation S-K, all SEC registrants are required to file material
contracts or agreements with their SEC filings, 8-K, 10-K, 10-Q, or registration
statements. The term “material” represents a level that would influence a reason-
able investor’s investment decision. Management has the discretion to assess the
materiality level, while the SEC has a right to review the compliance.
A CTO enables firms to receive an exemption on this requirement, governed
by Rule 406 under the Securities Act of 1933 and Rule 24b-2 under the Securities
Exchange Act of 1934. Once firms can establish to the SEC that the full non-
redacted disclosure causes competitive harm to their business and investors, they
can request to conceal certain portions of contracts for a designated period. The
confidential period may last for a maximum of ten years.
88
Although the SEC has authority to deny the requests, Meredith B. Cross, the
former Director of Division of Corporation Finance of SEC, admits that this rarely
happens (SEC (2010)). According to Lexis Securities Mosaic database, there are
only nine confidential treatment requests denied out of more than 10,000 requests,
from 1994 to 2015.
4
When managers submit the requests, they know that they
have a high chance of being approved.
5
Once CTO-requests are granted, the redacted filings are exempted from FOIA.
FOIA requires the government agencies, such as the SEC, to fully or partially dis-
close previously unreleased information if any individuals or institutions request.
However, according to FOIA subsection (b) (4), trade secrets and confidential
commercial or financial information are exempted from FOIA obligation. There-
fore, the SEC cannot furnish information with a CTO to the public.
During a CTO-period, firms have the privilege to withhold specific contents of
material contracts. However, they shall note the omitted parts of their public fil-
ings, using black-lining or asterisks. Appendices A, B, and C provide an example
of a CTO filing, the corresponding contract documents, and the 10-K, respectively.
Despite the extensive use of a CTO, our understanding of its underlying mo-
tivation and market reaction is limited. Verrecchia and Weber (2006) is the first
academic research to study CTO practice. It finds that proprietary costs play a role
in redaction decision. Firms in competitive industries choose to withhold propri-
etary information using a CTO. They find associations between CTOs and adverse
4
The reasons of denial include the following: the requested contracts have been publicly disclosed already; other regulatory clauses
require full disclosure; and the registrants fail to provide the information required by Rule 24b-2.
5
Effective of May 2, 2019, the SEC adopted a new regulation to simplify the public company disclosure process. In this amendment,
firms are allowed to redact information without filing a request (SEC (2019)).
89
selection proxies, such as bid-ask spreads, market depth, and share turnover.
To my knowledge, Boone et al. (2016) is the only attempt to explore CTOs
with a corporate finance perspective. In the initial public offering (IPO) context,
they find that IPO firms choose to redact information from their registration state-
ments with around 40% probability. They also show the accompanying costs of
redaction by finding the association between CTOs and underpricing. The costs
of information asymmetry are balanced out by superior performance post-IPO.
B. Identification Strategy
I test the implication of union bargaining power on the information environment
by assessing the causal effect of contract expirations on the likelihood of redacting
business agreements. To make this empirical strategy valid, the identification re-
quires orthogonality between contract negotiations and the nature and occurrence
of material business agreements. The identifying assumption is justified based on
the following observations.
First, ex ante it is unlikely for unions or corporations to manipulate contract
durations to cater to their interests. Matsusaka et al. (2019) find that the newly
negotiated contract usually retains a similar duration as the expiring one for a
given workplace.
Second, ex post negotiating parties do not time the contract expiration nor ini-
tiate the negotiation process. Once collective bargaining contracts are determined,
both unions and firms are obligated to follow the contractual terms and duration.
The vast majority of new contracts are rarely renegotiated before the due dates
90
(Cramton and Tracy (1992); Rich and Tracy (2013); Matsusaka et al. (2019)). In
particular, Matsusaka et al. (2019) document that in 52% of the cases, the length
of the new contract is the same as the old contract, and in 83% it differs by one
year or less.
Third, it is unlikely for employers to have discretion on the occurrence and
the nature of business contracts. Entering contracts is a bilateral decision, not
unilateral. Hence, firms have less control over when to enter the material agree-
ments. To validate this assertion, I provide two tests. First, I test whether the
number of exhibits in material contracts is significantly different during a fiscal
year with imminent contract expirations versus without. Figure 2 shows that the
number of material contracts filed to the SEC does not significantly deviate from
the mean during the firm-years with upcoming contract expirations. Second, I
test whether the nature of business agreements is systematically different with and
without wage negotiations. See section IV to find the detailed data collection pro-
cess and classification criteria. Figure 3 shows that the ratio of the number of
withheld agreements for each type and the total number of redacted agreements is
not statistically different during the contract expiration years.
6
Last, a firm cannot arbitrarily choose which contracts to disclose and when to
file to the SEC. According to Regulation S-K, material contracts represent any
binding agreements that would influence reasonable information users’ decision.
Although the firm determines the materiality, the decision is subject to SEC re-
6
The figure does not include the graphs for the contracts classified as “stockholder agreements.” There is not enough number of
redacted contracts for this type of agreement.
91
view. Also, SOX mandates that the SEC registrants file their material agreements
promptly (Sarbanes (2002)).
Collectively, these stylized facts make collective bargaining expiration dates an
ideal setting to analyze the incentives and consequences of a CTO. They preclude
the possibility that either unions or firms deliberately choose the timing of contract
expirations or the timing and content of material agreement filings. On the other
hands, firms have full control over redactions. Therefore, the empirical design to
test the association between contract expirations and a CTO delivers a valid causal
implication.
Based on the identifying assumption, I exploit within-firm variation in bargain-
ing competition around collective negotiations (Liberty and Zimmerman (1986);
Garc´ ıa Osma et al. (2015); Yi (2016)). The expiration dates of collective bar-
gaining contracts represent a firm’s exposure to bargaining competition with labor
unions. Firms and unions might negotiate with each other even without an im-
minent a collective bargaining expiration. This doesn’t mean, however, that they
actively negotiate all the time. As Frost (2000) claims, the intensity of a union’s
rent-seeking behavior is not constant throughout time but rather dynamic. Consis-
tent with this claim, unions take intensive opportunistic behavior around contract
expiration dates to win favorable wage contracts (Matsusaka et al. (2019)). It is
reasonable to assume that the union becomes more active in collecting relevant
and timely information when expecting intensive negotiations. Hence, it will mo-
tivate firms’ strategic behavior to achieve an advantageous bargaining condition.
92
Since a firm and the union bargain over the common stakes, improving one party’s
bargaining condition always harms another.
This variation in information demand of the union increases the benefit of
a CTO. If I find changes in the use of CTOs around expiring contracts, I can
claim that union bargaining activities cause the corresponding change in CTOs.
A difference-in-difference analysis compares the likelihood of strategic disclosure
before imminent wage negotiations to the likelihood of strategic disclosure when
negotiations are not imminent. The treatment group is composed of unionized
firms which have wage negotiations in the following year. The control group in-
clude unionized firms which do not have expiring wage contracts.
IV. Data
A. Contract Expiration
The National Labor Relations Act (NLRA) requires employers or union repre-
sentatives to send written notice to the other party 60 days before the expiration
date of a proposed termination or modification of a collective bargaining agree-
ment. This notice must also be provided to the Federal Mediation and Concilia-
tion Service (FMCS) within 30 days after notice to the other party.
7
Because the
NLRA requires all terminating or modifiable wage agreements to be submitted to
the FMCS, my sample includes the near universe of wage agreements, not just
disputed agreements.
8
7
https://www.fmcs.gov/services/resolving-labor-management-disputes/collective-bargaining-mediation
8
Table A2 illustrates that the overall unionization ratio is around 3.1% over my sample period. This figure might seem to contradict
the union membership rate published by the Bureau of Labor Statistics (BLS). For example, the BLS states the unionization rate
93
This filed information includes employer names, union names, contract expi-
ration dates, and the number of employees involved in collective bargaining. The
Bureau of National Affairs (BNA) database compiles the information into a dataset
available to the public. Since the dataset does not provide unique company identi-
fiers, such as GVKEY or CIK, the data needs a manual match.
The comprehensive dataset, including matched employer identifiers, is pro-
vided by Irene Yi (Yi (2016); Matsusaka et al. (2019)). She manually matches
the employer names to COMPUSTAT company names and assigns its GVKEY .
To reduce noise, the matching process is limited to unique employer names which
have contracts with more than 500 employees involved. The final sample includes
every contract with these unique names. Also, when the employer name in the
BNA database is on a subsidiary or a plant, the ultimate parent at the point of con-
tract expiration is assigned. If a firm has expiring contracts in the following fiscal
year, I assign a dummy variable which equals to one, and zero otherwise.
B. Redacted Disclosure
To identify whether firms choose to withhold any material contracts or agree-
ments from 10-K filings, I use Lexis Securities Mosaic database. I search 10-K’s
for terms representing CTOs, such as “confidential,” “confidential request,” “con-
fidential treatment,” “CT order,” or “redacted.” 10-K’s found to have those terms
are assigned with an indicator variable to represent a firm’s redaction.
at year 2000 as 13.4%. This discrepancy can be explained by the fact that my sample only covers the entities observed from
COMPUSTAT universe, while the BLS consider every public and private company. Therefore, one can interpret my empirical
analysis as the with-in firm variation of the unionized COMPUSTAT firms with and without imminent wage negotiations.
94
I also use SEC Analytics database to find a comprehensive list of 10-K’s filed in
the SEC EDGAR system. Out of this comprehensive list, if a filing is not identified
in Lexis Securities Mosaic database as having redacted disclosure, I assign zero
for the redaction indicator variable.
Prior literature shows that union bargaining power influences strategic use of
top management’s wage concessions (DeAngelo and DeAngelo (1991)), debt fi-
nancing (Matsa (2010)), and investment policy (Falato and Liang (2016); Yi (2016)).
To reduce potential endogeneity concern, I exclude CTOs granted for employee-,
credit-, or investment-related contracts, in some empirical analysis. To investigate
what kinds of agreements unionized firms redact from their public disclosure, I
read the redacted 10-K’s of unionized firms and examine each exhibit with a CTO.
Then, I classify redacted exhibits into eleven groups, based on the categories intro-
duced by Boone et al. (2016): (i) “sales and purchase related” includes agreement
on the firm’s ordinary business, such as inventory and supply, manufacturing, dis-
tribution, marketing, reseller, vendor, production; (ii) “license or royalty” is for
license or royalty agreements; (iii) “strategic alliance” involves joint ventures,
partnerships, and transition; (iv) “research or consulting” includes research, con-
sulting, or patent agreements; (v) “credit or leasing” is composed of debt contracts,
loans, loan amendments, and guarantees; (vi) “employment related” involves con-
tracts with employees or executives; (vii) “stockholder agreements” are for stock
repurchase or buyback; (viii) “asset investment” is for agreement on investment,
construction, or asset disposal; (ix) “outsourcing” includes outsourcing contracts;
95
(x) “reorganization” is related to merger, acquisition, divestiture or structure reor-
ganization; and (xi) “litigation” stands for legal actions or lawsuit outcomes.
C. Other Covariates
Firm-year specific financial data is collected from COMPUSTAT. I/B/E/S pro-
vides the information on analyst forecasts. Details on debt contracts are obtained
from DealScan. SEC Analytics provides the list of exhibits which firms publish
with SEC filings, such as 10-K, 10-Q or 8-K.
Based on previous studies, I construct a set of control variables which are stan-
dard controls in the financial disclosure literature. First, financial variables include
the natural logarithm of total assets, the natural logarithm of market value, book
to market, and return on assets. Second, to control for proprietary costs related to
market competition, I consider the text-based competition measure of Hoberg and
Phillips (2010, 2016).
9
The measure is negatively associated with product market
competition. Prior literature has shown that the competitive advantage influences
a firm’s redaction decision (Verrecchia and Weber (2006); Boone et al. (2016)).
For cross-sectional analysis, I implement the following proxies to measure in-
formation uncertainty, growth opportunities, financial constraints, and work stop-
page costs, respectively. First, information uncertainty is measured by analyst
forecast errors. Analyst forecast errors are the absolute value of the difference
between a firm’s reported earnings per share and the mean of most recent analyst
forecasts. Second, backward-looking sales growth proxies for growth opportuni-
9
The text-based measure is based on firm pairwise similarity scores from text analysis of firm 10-K product descriptions. It is
available to the public in Hoberg-Phillips data storage: http://hobergphillips.usc.edu
96
ties. It is calculated assale
t
=sale
t1
, wheresale
t
andsale
t1
are sales in years
t and t 1, respectively. Third, Hadlock and Pierce measure (HP index) repre-
sents firms’ financial constraints (Hadlock and Pierce (2010)). It is constructed as
0:737Size + 0:043Size
2
0:040Age, whereSize equals the natural
logarithm of the inflation-adjusted total assets (at from COMPUSTAT) (in 2004
dollars), andAge is the number of years the firm is listed with a non-missing stock
price on COMPUSTAT. TheSize andAge variables are truncated up-to4:5 billion
and 37 years. Last, I use the text-based similarity measure as the expected loss of
contentious wage negotiation. The measure is borrowed from Hoberg and Phillips
(2010, 2016).
Liquidity management is measured by the natural logarithm of loan amount and
asset purchase amount in the following fiscal year. The asset purchase amount is
measured as aqc
t+1
=at
t
, based on COMPUSTAT data items. The loan amount
is measured by the debt which is newly financed in the following fiscal year.
DealScan provides the related information. If more than two facilities are in the
same package, then I use the largest facility in the package to obtain the informa-
tion on initiation date and loan amount. I exclude the deals with primary purpose
of financial restructuring, such as leveraged buyout (LBO), management buyout
(MBO), recapitalization, restructuring, and takeover. I merge DealScan data with
other data using the link file provided by Chava and Roberts (2008), which is
available at Michael Robert’s webpage.
97
V. Summary Statistics
A. Collective Bargaining Contract
Table I panel A summarizes the information regarding collective bargaining
contracts in the sample. Throughout the paper, the main explanatory variable is
to indicate whether a firm has any imminent wage negotiations. It has the mean
value of 0:644. The contracts cover 2;096 employees on average. The average
length of contract is2:468 years, with the standard deviation of2:553 years. Every
unionized firms have about 6:890 expiring contracts over my sample period.
B. Industry Distribution
Table A1 provides the final sample distribution over different industries, which
are defined by two-digit SIC codes. The most common industry is manufacturing,
which has 35.9% of nonunionized observations and 53.4% of unionized obser-
vations. Compared to nonunionized firms, unionized firms are more clustered in
manufacturing and transportation and public utilities industries. Nonunionized
firms are more likely in the finance, insurance and real estate industries. I control
for potential systematic difference in industry by adding firm fixed effects in every
empirical specification.
C. Time-Series Distribution
Table A2 shows the final sample distribution over the period of 1997 to 2013.
The data helps to ensure that year-specific common shocks do not drive the re-
98
sults. Panel A contains the fiscal year distribution of firm-year observations, sep-
arately for unionized firms and nonunionized firms. The observations are evenly
distributed over the sample period.
Panel B provides the yearly distribution of redacted and nonredacted 10-K fil-
ings over the sample period. As the SEC Division of Corporate Finance states in
the Legal Bulletin, CTOs have increased steadily.
10
Panel B in Table I separates the firm-year observations of unionized firms de-
pending on the existence of contract expirations in a certain fiscal year. Contract
expirations slightly decrease over the sample period but are not concentrated in a
particular time window.
Overall, there is no significant clustering in any fiscal years. Although any
year-specific common shock will not drive empirical findings, I further relieve
this concern by controlling for year fixed effects.
D. Frequency and Types of Redacted Contracts
Table II provides general descriptions on redaction activities of the final sam-
ple. Panel A contains summary statistics on redaction practice for nonunionized
firms and unionized firms. The simple mean probability of redaction indicates
that nonunionized firms are more likely to withhold material contracts (17.7%)
than unionized firms (12.1%). Out of every 10-K filings, 17.5% of filings have at
least one material agreement concealed. This figure is close to the findings in Ver-
recchia and Weber (2006), who manually collect the redacted cases and identify a
10
See the SEC Division of Corporation Finance Staff Legal Bulletin No. 1 (with Addendum) “Confidential Treatment Requests”
Action: Publication of CF Staff Legal Bulletin, February 28, 1997.
99
16% redaction rate.
In order to understand what types of contracts firms withhold, I examine nonunion-
ized firms’ 10-K’s and classify redacted contracts into eleven categories. Panel B
reports the frequency distributions of each type of contracts redacted. Since the
unit of observations is exhibit level, the total number of redacted contracts, 1,055,
exceeds the number of redacted filings of unionized firms, 439. Similar to the
findings in Boone et al. (2016), the contracts related to the sales and purchase are
the most notable, followed by credit agreements and strategic alliance contracts.
VI. Effect of Labor Negotiations on Firm Disclosure
A. Contract Expiration and Redaction
The first analysis focuses on the causal impact of contract expirations on CTOs.
Theory provides opposing predictions on whether contract expirations increase
management incentive to hide information from employees to have a favorable
bargaining condition. Information advantage may hinder union’s rent-seeking be-
havior. At the same time, it may increase the costs of information asymmetry.
As a baseline empirical design, I run the following linear regression. Although
the outcome variable is a dummy variable which indicates redaction activity, I use
a linear probability model to control for firm and year fixed effects while avoiding
the incidental parameter problem (Heckman (1987); Greene (2004)).
P(Redaction)
it
=
0
+
1
Expiration
it+1
+
0
X
it
+
i
+
t
+
it
;
100
where i and t represent a firm i and a fiscal year t. The dependent variable is
P(Redaction)
it
, which is a dummy variable to indicate whether firm i chooses
to redact any material contracts or agreements from its 10-K filing in fiscal year
t. The main explanatory variable is Expiration
it+1
. It equals to one if firm i
has collective bargaining contracts expiring in the following fiscal year t + 1. I
also include firm-year specifications as control variablesX
it
, firm fixed effects
i
,
and year fixed effects
t
. The unit of observation is a firm-year, and the panel runs
from 1997 to 2013. Standard deviations are clustered by the firm in all regressions.
Table III presents the baseline results. In columns (1) and (2), the dependent
variable is an indicator variable for CTOs, which equals to one if a unionized com-
pany redacts any material contracts from 10-K in yeart. The coefficient estimates
of expiring contract dummy reject the null hypothesis. When a firm has collective
bargaining contracts expiring in the following year, it shields material business
contract information by requesting CTOs. It implies that bargaining competition
during wage negotiation increases management’s incentive to strategically conceal
information. The point estimate of 3.1 - 3.3 percentage points is statistically sig-
nificant. It is also economically important, considering the overall probability of
redaction in unionized firms is 12.1%.
The identification strategy relies on the assumption of orthogonality between
material contracts and union bargaining power. This assumption might be un-
dermined by the prior literature, which finds the implication of union bargaining
power on various firm policies. Union bargaining power influences the strategic
101
use of top management’s wage concessions (DeAngelo and DeAngelo (1991)),
debt financing (Matsa (2010)), and investment policy (Falato and Liang (2016);
Yi (2016)). Therefore, some types of business agreements might have systemati-
cally different nature and occurrence around contract expirations.
Columns (3) and (4) help to address this concern. I exclude the cases of CTOs
for certain types of agreements, including lending-, employee-, or investment-
related agreements. The dependent variable equals to one if a unionized company
redacts any material contracts other than those types of agreements from 10-K
in year t. I use the same sample observation and the explanatory variable as in
columns (1) and (2). The point estimates are 2.6 percentage points and remain
statistically significant. The economic magnitude is similar as in columns (1) and
(2), given the mean value of outcome variable is 10.5%.
Throughout various specifications, the results indicate that a company increases
the redaction probability by 2.2 - 3.1 percentage points. Considering the overall
probability of redaction is 17.5% for the entire sample and 12.1% for unionized
sample, this value is economically meaningful. If there is a sequence of overlap-
ping contract expirations covering a portion of employees, these point estimates
would be underestimated.
The findings illustrate that union bargaining activities lead to an increase in
proprietary costs of public disclosure. To strategically increase information asym-
metry and strengthen their information advantage over unions, firms use CTOs as
a bargaining device in wage negotiations.
102
B. Cross-Sectional Factor Analysis
In this section, I explore potential modifying factors for strategic disclosure
policy that could affect managerial decision to exploit CTOs. While CTOs can
improve a firm’s information advantage, it may lead to detrimental market reac-
tions, such as an increase in cost of capital. A firm may seek to the balance the
benefits of information advantage against the costs of information asymmetry.
B.1. Information Uncertainty
First, I show how the status-quo information advantage of firm insiders affects
redaction practice as a strategic device. Given that a CTO is a powerful bargain-
ing device that strengthens information barrier, the incentive to use this device
will depend on the status quo information environment. The theory provides two
competing predictions.
One predicts that the existing high-level of informational advantage may reduce
management incentive to implement strategic disclosure policy. If information
barriers between insiders and outsiders are high even in the absence of additional
strategic action, redaction will not provide an additional benefit to enhance infor-
mational strength of firms.
The other predicts that firms may more intensively exploit their exclusive bar-
gaining device. Freeman and Kleiner (1999) find that unionization does not in-
crease the insolvency of firms. It implies that a union is a rational player that does
not push the wages to the point where the firm closes down. If information un-
103
certainty is high and the union is not well-informed with future profitability due
to withheld filings, the union will not ask for high surplus based on the profits of
the past years. It understands that infeasible wage demand will increase the risk
of losing business and jobs. For management, insolvency risk can be a credible
threat to defy wage increase demand. According to this argument, management’s
informational strength or informational uncertainty will increase firms’ strategic
incentive to use a CTO.
In sum, the question of whether the current level of information uncertainty
amplifies or mitigates the strategic CTOs is an empirical question.
To verify which theoretical prediction the empirical tests support, I implement a
triple-difference analysis to compare the group with high information asymmetry
and the group with low information asymmetry. I proxy information uncertainty
using analyst forecast errors. Analyst forecast errors represent sophisticated in-
vestors’ difficulty in forecasting a firm’s future performance (Zhang (2006)). The
main explanatory variable is an interaction term of a dummy to indicate contract
expiration and a dummy to identify higher-than median analyst forecast errors.
Table IV panel A reports the results. Throughout the various specifications,
the triple interaction term preserves positive coefficient estimates. The empirical
evidence supports the theory that status quo information asymmetry strengthens
managements’ incentive to strategically withhold information.
104
B.2. Growth Opportunity
In the next factor analysis, I hypothesize that a firm with low growth opportu-
nities uses CTOs more than a firm with high growth opportunities. As a rational
optimizer, a union concerns both member’s employment and wages. Therefore,
the union will agree on wage concessions when it evaluates the expected benefit
from unfavorable compensation package outweighs the benefit from a favorable
one, by increasing the chance of firm survival (Kuhn (1986); Freeman and Kleiner
(1999)). The benefit of wage concessions is likely to be more prominent when the
firm has a high growth opportunities. In other words, a firm with growth opportu-
nities is exposed to less-severe rent-seeking behavior of a union.
This observation leads to an empirical prediction saying that a CTO will be
used less intensively in a union-management bargaining game when the firm has
higher growth opportunities. Growth opportunities make opportunity costs from
losing business be higher for employees. Unions can be convinced to settle for
lower wages and to wait for a more significant surplus in the future. Therefore,
growth opportunities strengthen the firm’s bargaining condition and weakens the
management incentive to implement strategic disclosure.
Table IV panel B provides some evidence on the empirical prediction. The neg-
ative triple interaction terms imply that firms with high sales growth strategically
redact less, compared to firms with low sales growth. Although the results are
weak in some specifications, they imply that strategic disclosure policy is bene-
ficial for the firms which cannot credibly convince employees to wait for future
105
surplus.
B.3. Financial Constraint
Benmelech et al. (2012) show that financial distress suppresses the rent-seeking
behavior of unions and leads to wage concessions. Financial shortage can be a
credible claim to turn down wage demand when firms are financially constrained.
Therefore, financial constraints strengthen firms’ bargaining position by increas-
ing the risk of losing business (Freeman and Kleiner (1999)). Related to this idea,
firms have incentives to strategically reduce excess liquidity in order to strengthen
their bargaining power in wage negotiation. For example, Klasa et al. (2009)
demonstrate the negative relationship between union power and cash holdings. As
a way to reduce excess liquidity, managers strategically choose to increase debt
financing (Bronars and Deere (1991); Matsa (2010, 2018)) or to increase asset
purchase (Yi (2016)).
The existing evidence collectively indicates that a firm’s incentive to enhance
its bargaining power will have a higher benefit when it has more internal resource.
Motivated by these observations, I hypothesize that the value from redaction will
be higher in financially unconstrained firms. Therefore, expiring contracts will
have stronger prediction power on redaction in financially affluent employers.
Table IV panel C provides weak empirical support for this prediction. The
negative interactions imply that the firms with below-median financial constraints
increase redaction probabilities more than those with above-median financial con-
straints. Although the statistically insignificant coefficients restrict the interpre-
106
tation, the analyses suggest that financially constrained firms are less likely to
manipulate their information environment to preserve information advantage over
unions.
B.4. Costs of Strike
Next, I explore how a threat from potential work stoppages affects strategic
disclosure. Although the imbalance in information between firms and negotiating
unions can strengthen the firms’ bargaining position, it may lead to a costly work
stoppage or holdups. Bargaining theory predicts that incomplete information may
result in a costly delay in settlement (Cramton (1984)). Some anecdotal evidence
proves the theory and shows that the probability of strikes is positively associated
with the information discrepancies on each other (Reder and Neumann (1980);
Mauro (1982)).
If firms strategically redact the information before wage negotiation, they will
take into account the fact that it might increase the likelihood of contentious ne-
gotiations and the cost of a work stoppage. There is a contradicting prediction on
how the potential strike costs would affect the strategic disclosure. On the one
hand, the threat of work stoppage implies a strong bargaining power of the union.
With a comparatively weak bargaining condition, a firm needs to take every pos-
sible bargaining device, including the strategic information sharing. On the other
hand, the increase in information asymmetry might delay wage negotiations. If a
firm has to bear substantial costs due to prolonged strikes, it will discourage the
firm from strategically withholding information.
107
Collectively, empirical analysis is essential to answer the following question:
does the cost of a work stoppage amplify or mitigate the strategic redaction?
The estimated loss from a work stoppage is associated with various factors
which characterize the firms’ productive activity (Reder and Neumann (1980)). In
this literature, I specifically explore how easy the firms’ products can be substi-
tuted in the product market.
Panel D of Table IV weakly confirms the empirical prediction with negative
coefficient estimates for the triple interactions. The firms with above-median sim-
ilarity measure increase redaction probabilities less than those with below-median
similarity measure.
B.5. Summary
The cross-sectional analyses deliver some suggestive evidence that a firm strate-
gically chooses to redact material agreements by considering the size of potential
benefits. The status quo information asymmetry amplifies the strategic disclo-
sure, while growth opportunities, financial constraints, and strike threat mitigate
the strategic behavior. Overall, the coefficient estimates confirm my directional
predictions but in some specifications, do not provide statistically significant re-
sults. One takeaway is that the main result stays significant throughout various
specifications.
108
VII. Effect of Strategic Disclosure on Firm Performance
The results presented above show that contract expirations cause firms to strate-
gically withhold information. A natural question following this analysis is whether
the redaction choice leads to favorable negotiation outcomes for the redacting
firms. To address this question, this section explores the association between the
strategic disclosure and ex post firm performance.
There is an empirical challenge to determine the causal impact of redaction
on ex post firm performance. Although wage negotiations are exogenous events,
disclosure policy is a firm’s endogenous choice. In the partial-disclosure equilib-
rium, a firm should conceal both good and bad information to constrain the union
on how to interpret the redacted information. Also, as a rational player, a firm
determines the right level of information disclosure by assessing the trade-offs be-
tween the benefit of information advantage in wage negotiations and the expected
cost of asymmetric information.
Prior literature shows that if a firm optimizes its strategic variables to maximize
the firm value, empirical tests do not detect any relation between these variables
and firm value (Demsetz and Lehn (1985); Duchin et al. (2010)). Thus, I predict
that reduced-form empirical tests to detect the effect on future performance will
not provide a relation between CTOs and firm performance.
109
To test this speculation, I run the following linear regression model.
FirmPerformance
it+1
=
0
+
1
CTO
it
+
2
Expiration
it+1
+
3
CTO
it
Expiration
it+1
+
i
+
t
+
it
;
where i and t represent a firm i and a fiscal year t. The dependent variable is
FirmPerformance
it+1
, which is measured by return on assets, operating cash
flow scaled by total assets, and operating margin. CTO
it
indicates whether firmi
chooses to redact any material contracts or agreements from its 10-K filing in fiscal
yeart. Expiration
it+1
equals to one if firmi has collective bargaining contracts
expiring in the following fiscal year t + 1. I also include firm fixed effects
i
,
and year fixed effects
t
. The main variable of interest is the interaction term of
the CTO dummy and the contraction expiration dummy, which represents the net
impact of CTO adapted before wage negotiations on firm performance.
Due to the empirical challenge mentioned above, one can interpret a predicted
association between CTOs and ex post performance in two ways. First, it might
imply that redaction alters a firm’s bargaining power in wage negotiations and
leads to a change in operating performance. Second, a manager has private infor-
mation on future performance which motivates the manager to hide information
from employees in a systematically different way.
Since the objective of this paper is to show the strategic value of disclosure as a
bargaining device, the first channel will be the main interest. To alleviate the im-
pact of the second channel, I match redacting firms and non-redacting firms using
110
entropy-balancing (Hainmueller (2012)). The idea is to make redacting and non-
redacting groups look similar in terms of firm-year observables and make CTOs
arguably be randomly assigned. It does not completely remove the endogeneity
concern but might reduce it. To match two groups, I use the same set of firm-year
specifications, including the logarithm of total assets, stockholders’ equity, return
on assets, the logarithm of market value, book to market ratio, and text-based
financial constraint measure. To conserve space, I do not report the balancing
outcomes but find that the matching process is successful.
Table V reports the findings. The interaction terms of CTOs and expiring con-
tracts do not produce any statistically significant coefficients throughout various
performance measures. The findings provide suggestive evidence that manage-
ment strategically implements CTOs by counterbalancing the benefit and the cost.
Although I reduce the endogeneity concern by matching firms with and without
CTOs, the studies with ex post outcomes do not provide causal interpretation but
should be interpreted as correlations.
VIII. Robustness Tests and Additional Analyses
A. Substitution Effect between Strategic Disclosure and Liquidity Management
A union’s rent-seeking behavior can incentivize managers to implement various
strategic policies. In particular, strategic liquidity reduction is well-understood
both in theory and empirical literature. A firm decreases internal resources either
by increasing fixed interest payments (Matsa (2010)) or by purchasing additional
111
assets (Yi (2016)).
While strategic disclosure and liquidity reduction can enhance a firm’s bargain-
ing position, they entail costs: the former increases capital market costs, and the
latter increases financial distress costs. Unless we accurately formulate the cost
and the benefit function of each strategy, it is an empirical question of whether
two bargaining devices substitute or complement each other.
Table VI provides evidence on strategic substitutability between two strategies.
In columns (1) and (2), I test whether debt financing in year t + 1 reduces the
causality of contract expirations on redaction probability. The interaction terms
between contract expirations and debt financing have negative but statistically in-
significant coefficient estimates. Although the results do not confirm any inter-
action between strategic financing and strategic disclosure, this finding does not
contradict the argument made in Matsa (2010). Contract expirations represent
a transitory shock in union bargaining activities. However, Matsa (2010) mea-
sures the bargaining power using firm-level collective bargaining coverage and
state changes in labor laws, which are more permanent. While debt financing
and asset acquisition achieve the same goal, which is to reduce excess liquidity,
the former changes liquidity for a long period. Asset acquisition can be a close
substitute for disclosure policy since both produce short-lived impacts.
The interaction between two bargaining strategies becomes pronounced in columns
(3) and (4). The columns test the substitution between redaction and asset pur-
chase. The interaction terms between contract expirations and asset purchase
112
amounts have statistically significant and negative coefficients. They illustrate
the strategic substitution effect between disclosure policy and investment policy.
Table A4 in the internet appendix tests the strategic substitution between a CTO
and cash holdings. Klasa et al. (2009) finds that firms in more unionized industries
strategically hold less cash to depress union’s rent-seeking behavior. The positive,
yet statistically insignificant, coefficients of the interaction terms suggest the same
message as Table VI. Strategic redaction is more prevalent when a firm fails to
shrink its cash holdings.
The findings indicate that strategic disclosure is a substitute for liquidity man-
agement using asset acquisition. We can explain strategic substitutability using the
following rationale. First, a firm cannot cut wages below a certain point on which
employees start to have better outside options. Therefore, the value of additional
bargaining power is bounded. Second, while additional bargaining power helps a
firm to obtain wage concessions, excessive wage cuts entails reputation costs. If
a firm forces its current employees to settle on wage concessions, it may have a
difficult time to recruit potential workers in the future (Hart (1983)). Collectively,
the total benefit from strong bargaining position is canceled out by reputation loss
in the labor market.
B. Endogenous Financial Constraint
In the main empirical analysis, I use financial constraints as one of the mitiga-
tors for CTOs. The idea is that financial constraints will suppress wage demand
and reduce the benefit of redaction. However, it is also possible that the firms with
113
active union bargaining activities strategically choose to reduce excess liquidity
and increase financial constraints (Klasa et al. (2009); Matsa (2010)).
In order to entirely remove the concern, I need an exogenous shock in union
bargaining power, which exclusively influences the proprietary value of informa-
tion and does not change the value of financial structure. Absent such a shock, I
try to alleviate the concern by using financial constraints in lagged values.
In Table VII, I run the same experiment as Table IV panel C but use the median
value of lagged HP index. The evidence preserves the main message in the cross-
sectional analysis using contemporary HP index. The interaction terms of the
dummy for upcoming wage negotiations and the dummy for high lagged financial
constraints are estimated to have negative and statistically significant coefficients.
C. Cross-Sectional Factor Analysis using Alternative Proxies
In the section VI. B. Cross-Sectional Factor Analysis, I explore potential miti-
gating factors in strategic disclosure, using one proxy for each factor. This section
repeats the analysis but exploits different proxies. Internet appendix reports the
results in Table A3.
In panel A, I use the number of analyst following as the inverse of status quo
information asymmetry. The interaction term provides another weak support on
the intensifying impact of status quo information asymmetry on a strategic CTO.
Panel B tests the mitigating influence of growth opportunities, using book-to-
market ratio. The positive but statistically insignificant coefficient for the interac-
tion term suggests that a firm with a lower growth potential implements a CTO as
114
a bargaining device more intensively.
In panel C, financial constraints, measured by Hadlock and Pierce index, pro-
duce negative interaction coefficients with the expiring contract indicator. It con-
firms the implications of the main analysis: financially constrained firms are less
likely to use disclosure policy as a bargaining tool.
As a proxy for the strike costs, panel D uses the volatility in inventory stock
in previous years. If it is high, it is hard to accurately predict market demand
and to address the market demand throughout any potential work stoppages. The
negative point estimates for the interaction term imply that unpredictable market
demand suppresses a strategic CTO before wage negotiations.
Overall, the coefficient estimates agree to the findings in the section VI. As in
the main analysis in the section VI, they confirm the directional predictions but in
some specifications, do not provide statistically significant results.
IX. Conclusion
Managers are better informed about a firm’s future prospects than other stake-
holders, such as investors and labor suppliers. This inherent information asymme-
try can be harmful because it leads to inefficient resource allocation and a higher
cost of capital. However, it can also enhance a firm’s competitive advantage in
negotiations over wages or pricing contracts.
This paper analyzes the strategic use of public disclosure to improve a firm’s
bargaining condition with unions. To overcome endogeneity concerns and to es-
115
tablish causality, I use collective contract expiration dates. The results provide
causal evidence that firms strategically withhold information to improve bargain-
ing outcomes with labor unions. The analysis on various determinants on CTOs
illustrates trade-offs of withholding confidential information between the benefits
of information advantage over unions and the costs from information asymmetry
in the capital market.
The paper contributes to the growing literature studying how labor forces in-
fluence a firm’s policy. The findings suggest that labor negotiations have a signif-
icant impact on a firm’s information environment. In particular, a firm with low
bargaining power will further distort information asymmetry among management
and investors. The findings also complement the existing studies on the strategic
decision to withhold information using CTOs.
In light of the findings in this paper, there are potential directions for future
research. First, it is important to better establish causality between strategic CTOs
and operating outcomes. Second, future research could investigate wage settle-
ment results to study the direct impacts of strategic information sharing on bar-
gaining outcomes. Last, the cross-sectional findings are centered on the benefits of
redaction, while the strategic decision is made considering a trade-off of the ben-
efit and capital market concerns. To complete the picture, future research could
explore how the capital market reacts to strategic disclosure.
116
Appendix A. Confidential Treatment Example
This is an example of a confidential treatment order form, requested by Progress Power, Inc. and then
approved by the SEC. The redacted contract is Exhibit 10.1 “ENGINEERING, PROCUREMENT AND
CONSTRUCTION AGREEMENT” for “AP1000 NUCLEAR POWER PLANT.” You can locate the docu-
ment here: SEC EDGAR Link. The firm timely disclosed the contract in Form 8-K, with some confidential
information withheld.

UNITED STATES
SECURITIES AND EXCHANGE COMMISSION

March 18, 2009

ORDER GRANTING CONFIDENTIAL TREATMENT
UNDER THE SECURITIES EXCHANGE ACT OF 1934

Progress Energy, Inc.
File No. 001-15929 - CF# 23240

Florida Power Corporation
d/b/a Progress Energy Florida, Inc.
File No. 001-3274 - CF# 23240
_____________________________

Progress Energy, Inc. and Florida Power Corporation d/b/a Progress Energy
Florida, Inc. submitted an application under Rule 24b-2 requesting confidential treatment
for information they excluded from the Exhibit to a Form 8-K filed on March 2, 2009.

Based on representations by Progress Energy, Inc. and Florida Power Corporation
d/b/a Progress Energy Florida, Inc. that this information qualifies as confidential
commercial or financial information under the Freedom of Information Act, 5 U.S.C.
552(b)(4), the Division of Corporation Finance has determined not to publicly disclose it.
Accordingly, excluded information from the following exhibit(s) will not be released to
the public for the time period(s) specified:

Exhibit 10.1 through March 2, 2019

For the Commission, by the Division of Corporation Finance, pursuant to
delegated authority:

Ellie Bavaria
Special Counsel
117
Appendix B. Redacted Contract Filing
This is an excerpt from contract filed with confidential treatment by Progress Power, Inc., which is about
“ENGINEERING, PROCUREMENT AND CONSTRUCTION AGREEMENT” for “AP1000 NUCLEAR
POWER PLANT.” You can locate the document here: SEC EDGAR Link. The agreement is timely dis-
closed in Form 8-K and then in Form 10-K for the fiscal year end.
EX-10.1 2 g17748exv10w1.htm EX-10.1
Exhibit 10.1
Progress Energy, Inc. and Florida Power Corporation d/b/a Progress Energy Florida, Inc. (“PEF”) have requested
confidential treatment for certain portions of this document pursuant to an application for confidential treatment sent
to the Securities and Exchange Commission. Progress Energy, Inc. and PEF have omitted such portions from this
filing and filed them separately with the Securities and Exchange Commission.
Such omissions are designated as “[***].”
ENGINEERING, PROCUREMENT AND CONSTRUCTION AGREEMENT
BETWEEN
FLORIDA POWER CORPORATION DOING
BUSINESS AS:
PROGRESS ENERGY FLORIDA, INC.
(OWNER)
AND
A CONSORTIUM CONSISTING OF
WESTINGHOUSE ELECTRIC COMPANY LLC
AND
STONE & WEBSTER, INC.
(CONTRACTOR)
FOR AN
AP1000 NUCLEAR POWER PLANT
Progress Energy Contract No. 414310

[…….]

ARTICLE 3 — SCOPE OF WORK AND SCHEDULE
3.1 General.
(a) Contractor will perform the Work identified as Contractor’s responsibility in the Scope of Work
(Exhibit A) and will perform all other obligations and responsibilities of Contractor as set forth in this
Agreement. The Work will be performed in two phases, as more fully described in Sections 3.2 and 3.3 of
this Agreement. Contractor agrees to design, engineer, supply, equip, construct and install a complete and
fully operational Facility, including the Equipment to be incorporated therein and the Services to be
provided in connection therewith.
(b) If there is a dispute as to whether certain work related to the Facility is within the Contractor’s
Scope of Work, then in exigent circumstances Owner shall have the right to require Contractor by written
Notice to begin to perform such work and Contractor will be paid on a Time and Materials Basis for such
work until the DRB makes a determination as to whether such work or a portion thereof is within the
Contractor’s Scope of Work. If there is also no agreement between the Parties on the pricing or the
adjustment to the Contract Price in connection with such work, then either Party may also submit to
Dispute Resolution the determination of the appropriate pricing or Contract Price change, as applicable,
relating to such work. If the DRB determines that such work is within the Contractor’s Scope of Work,
then the DRB shall determine whether such work is priced [***], or on a Time and Materials Basis as set
forth in Exhibit H. If, however, the DRB determines that such work is outside of Contractor’s Scope of
Work, then the DRB shall determine the appropriate adjustment to the Contract Price pursuant to Section
9.4(c).
Page 21
118
Appendix C. Redacted 10-K
This is an excerpt from 10-K filing by Progress Power, Inc., which requested and received the approval for
confidential treatment for its material contract filing. The corresponding contract is about “ENGINEERING,
PROCUREMENT AND CONSTRUCTION AGREEMENT” for “AP1000 NUCLEAR POWER PLANT.”
You can locate the document here: SEC EDGAR Link. The agreement is timely disclosed in Form 8-K and
then in Form 10-K for the fiscal year end.
10-K 1 form10k_2011.htm 2011 FORM 10K
UNITED STATES
SECURITIES AND EXCHANGE COMMISSION
Washington, D.C. 20549
FORM 10-K
[…..]

Exact name of registrants as specified in their charters,
Commission
File Number
state of incorporation, address of principal executive
offices, and telephone number
I.R.S. Employer
Identification Number

[

1-15929 Progress Energy, Inc.
410 South Wilmington Street
Raleigh, North Carolina 27601-1748
Telephone: (919) 546-6111
State of Incorporation: North Carolina
56-2155481

[ … ..]

EXHIBIT INDEX

Number Exhibit Progress Energy,
Inc.
PEC PEF
[…..]
*10d(2) Engineering, Procurement and Construction Agreement, dated as of
December 31, 2008, between Florida Power Corporation d/b/a/ Progress
Energy Florida, Inc., as owner, and a consortium consisting of
Westinghouse Electric Company LLC and Stone & Webster, Inc., as
contractor, for a two-unit AP1000 Nuclear Power Plant (filed as Exhibit
10.1 to Current Report on Form 8-K filed on March 2, 2009). (The
Registrants have requested confidential treatment for certain portions of
this exhibit pursuant to an application for confidential treatment
submitted to the SEC. These portions have been omitted from the above-
referenced Current Report and submitted separately to the SEC.)
X

X

[…..]
119
Appendix D. Order Denying Confidential Treatment Example
This is an example of denied confidential treatment request form, initially requested by Corestream Energy,
Inc. and then denied by the SEC. You can locate the document here: SEC EDGAR Link.

UNITED STATES
SECURITIES AND EXCHANGE COMMISSION
January 24, 2012

ORDER DENYING CONFIDENTIAL TREATMENT
REQUEST UNDER RULE 24b-2
AND
NOTICE OF OPPORTUNITY TO PETITION
FOR REVIEW UNDER THE
SECURITIES EXCHANGE ACT OF 1934

Corestream Energy, Inc.

File No. 0-26383 - CF#26575
_____________________

The Division of Corporation Finance denied your request for confidential treatment of the
information excluded from exhibit 10.1 to the Form 8-K, as filed on March 15, 2011.

We denied your request because we concluded:

• the excluded information is material to those who make investment decisions
concerning your securities; and

• the application failed to include information required to be included by Rule 24b-2,
including an analysis of the applicable exemption from disclosure under the
Commission’s rules and regulations and a specific stated date through which
confidential treatment is sought.

You may request that the Commission review this order by submitting a petition to the
Office of the Secretary within five days, as required by 17 C.F.R. 201.430. Otherwise, we will
make the information for which you requested confidential treatment available to the public.

For the Commission, by the Division of Corporation Finance, pursuant to delegated
authority:

Michael McTiernan
Assistant Director
120
Figure 1.
Changes in Redaction Tendency around Contract Expiration
This figure plots the point estimates and 95% confidence intervals by regressing redaction probability
in year t on indicators for bargaining contracts which expire in different time windows: (t + 2;t + 3],
(t+1;t+2],(t;t+1],(t1;t] and(t2;t1]. Regressions include firm and year fixed effects. Standard
errors are clustered by the firm. The sample consists of firm-year observations during the period 1997 -
2013 for unionized firms which have at least one expiring contracts in the sample period.
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
0 .02 .04 .06 .08
Contract expiration
121
Figure 2.
Changes in Number of Exhibits around Contract Expiration
This figure plots the point estimates and 95% confidence intervals by regressing number of contract dis-
closed in yeart on indicators for bargaining contracts which expire in different time windows: (t+2;t+3],
(t+1;t+2],(t;t+1],(t1;t] and(t2;t1]. The exhibits disclosed in SEC filings are collected from
SEC Analytics. I count the list of exhibits which have titles as “Ex-10. XX” in 10-K, 10-Q and 8-K filings
in fiscal yeart. Regressions include firm and year fixed effects. Standard errors are clustered by the firm.
The sample consists of firm-year observations during the period 1997 - 2013 for unionized firms which
have at least one expiring contracts in the sample period.
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-2 -1 0 1 2
Contract expiration
122
Figure 3.
Changes in Types of Redacted Agreements around Contract Expiration
This figure plots the point estimates and 95% confidence intervals by regressing the ratio of redacted contracts for each
agreement type in year t on indicators for bargaining contracts which expire in different time windows: (t+2;t+3],
(t+1;t+2],(t;t+1],(t1;t] and(t2;t1]. The ratio is the number of redacted agreements for each type and the
total number of redacted agreements. Following Boone et al. (2016), I classify redacted exhibits into eleven groups: (i)
“sales and purchase related”; (ii) “license or royalty”; (iii) “strategic alliance”; (iv) “research or consulting”; (v) “credit
or leasing”; (vi) “employment related”; (vii) “stockholder agreements”; (viii) “asset investment”; (ix) “outsourcing”; (x)
“reorganization”; and (xi) “litigation”. Regressions include firm and year fixed effects. Standard errors are clustered by
the firm. The sample consists of firm-year observations during the period 1997 - 2013 for unionized firms which have at
least one expiring contracts in the sample period.
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-.15 -.1 -.05 0 .05 .1
Contract expiration
(a) Sales and purchase
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-.05 0 .05 .1
Contract expiration
(b) License or royalty
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-.1 -.05 0 .05 .1 .15
Contract expiration
(c) Strategic alliance
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-.05 0 .05
Contract expiration
(d) Research or consulting
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-.04 -.03 -.02 -.01 0 .01
Contract expiration
(e) Credit or leasing
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-.04 -.02 0 .02 .04
Contract expiration
(f) Employment
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-.15 -.1 -.05 0 .05 .1
Contract expiration
(g) Asset investment
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-.01 0 .01 .02 .03 .04
Contract expiration
(h) Outsourcing
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-.1 -.05 0 .05 .1
Contract expiration
(i) Reorganization
Dummy=1 for expiring contracts between (year+2) and (year+3) ((t+2, t+3])
Dummy=1 for expiring contracts between (year+1) and (year+2) ((t+1, t+2])
Dummy=1 if expiring contracts in t+1 fiscal year
Dummy=1 for expiring contracts between (year-1) and (year) ((t-1, t])
Dummy=1 for expiring contracts between (year-2) and (year-1) ((t-2, t-1])
-.01 0 .01 .02 .03
Contract expiration
(j) Litigation
123
Table I
Summary Statistics
This table reports summary statistics for the sample firms during the period 1997 - 2013. Panel A summarize the in-
formation regarding collective bargaining contracts. Panel B presents the fiscal year distribution of unionized firm-year
observations, separately for without and with expiring contracts. The contract expiration data is collected from the Bu-
reau of National Affairs (BNA) by Irene Yi (Yi (2016); Matsusaka et al. (2019)), who manually match employer names to
unique company ID, such as GVKEY and CUSIP. Significance levels are indicated: * = 10 percent, ** = 5 percent, *** =
1 percent.
Panel A: Collective Bargaining Contract
Mean S.D. No.
Dummy=1 if expiring contracts int+1 fiscal year 0.64 0.48 3640
Number of employees under expiring contract (in thousands) 2.10 10.92 2904
Number of years covered by the contract 2.47 2.55 3608
Panel B: Collective Bargaining Expiration by Year
Fiscal Year End
Without Expiring Contract With Expiring Contract Total
No. % No. % No. %
1997 67 5.2 153 6.5 220 6.0
1998 65 5.0 162 6.9 227 6.2
1999 76 5.9 144 6.1 220 6.0
2000 74 5.7 147 6.3 221 6.1
2001 67 5.2 161 6.9 228 6.3
2002 79 6.1 145 6.2 224 6.2
2003 67 5.2 157 6.7 224 6.2
2004 66 5.1 154 6.6 220 6.0
2005 92 7.1 126 5.4 218 6.0
2006 80 6.2 135 5.8 215 5.9
2007 70 5.4 135 5.8 205 5.6
2008 83 6.4 121 5.2 204 5.6
2009 88 6.8 120 5.1 208 5.7
2010 68 5.3 138 5.9 206 5.7
2011 83 6.4 119 5.1 202 5.5
2012 85 6.6 114 4.9 199 5.5
2013 85 6.6 114 4.9 199 5.5
Total 1,295 100.0 2,345 100.0 3,640 100.0
124
Table II
Descriptive Information on CTO Practice
This table presents information on redaction practice of sample firms during the period 1997 - 2013. Panel
A reports the overall redaction tendencies, separately for unionized and nonunionized firms. Firms are
classified as nonunionized firms if they do not have collective bargaining expirations in the sample period.
Unionized firms consist of companies which have at least one contract expiration in the sample period.
The unit of observation is a firm-year 10-K. Panel B presents the frequency distribution for eleven types of
redacted agreements for unionized firms. The unit of observation is an exhibit which stands for a material
contract. The contract category is a modified classification of Boone et al. (2016). The total number of
redacted contracts (=1,055) exceeds the number of redacted 10-K’s (=439) of unionized firms since each
10-K can have more than one contract withheld.
Panel A: Redaction Tendency
Nonunionized Firm Unionized Firm Total
No. % No. % No. %
Non-Redacted 10-K’s 94,429 82.3 3,201 87.9 97,630 82.5
Redacted 10-K’s 20,251 17.7 439 12.1 20,690 17.5
Total 114,680 100.0 3,640 100.0 118,320 100.0
Panel B: Redacted Contract Types
Redacted Exhibit
Contract Type Total Number Average Number Maximum Number
Sales or Purchase Related 574 0.158 17
License or Royalty 73 0.020 6
Strategic Alliance 114 0.031 3
Research or Consulting 11 0.003 3
Credit or Leasing 180 0.049 13
Employee Related 12 0.003 1
Stockholder Agreement 11 0.003 5
Asset Investment 40 0.011 7
Outsourcing 23 0.006 3
Reorganization 12 0.003 2
Litigation 5 0.001 1
Total 1,055
125
Table III
Contract Expiration and Redaction
This table shows the estimates from linear regressions using various specifications. The unit of observation
is a firm-year, and the panel runs 1997 - 2013. The main explanatory variable is a dummy variable to
indicate whether the firm-year has expiring contracts in the following fiscal year t + 1. The reported
numbers are coefficient estimates and their t-statistics (in parentheses). Standard errors clustered by the
firm. All regressions include firm and year fixed effects. The financial controls include the natural logarithm
of total assets, the natural logarithm of market value, book to market, return on assets, and the text-based
competition measure (Hoberg and Phillips (2010, 2016)). In columns (1) and (2), the sample consists of
unionized firms which have at least one contract expiration in the sample period. The dependent variable
is an indicator variable for CTOs on any material contracts in year t. In columns (3) and (4), the sample
consists of unionized firms which have at least one contract expiration in the sample period. The dependent
variable is an indicator variable for CTOs on material contracts other than those related to debt financing,
employees, or investment in year t. The dependent variable is an indicator variable for CTOs on any
material contracts in yeart. Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Dummy=1 if firm
redacted any
material contracts
Dummy=1 if firm
redacted material
contracts other than
lending, employee, or
investment agreements
(1) (2) (3) (4)
Dummy=1 if expiring contracts int+1
fiscal year
0:031

0:033

0:026

0:026

(2:62) (2:67) (2:36) (2:32)
Log (Total Assets) 0:026 0:006
(1:01) (0:24)
Log (Market Value) 0:036

0:034

(2:14) (2:06)
Book to Market 0:001

0:001

(3:75) (3:70)
Return on Assets 0:012 0:028
(0:36) (0:80)
Text-Based Competition Measure 0:069 0:079

(1:52) (1:88)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3640 3468 3640 3468
AdjustedR
2
0.025 0.033 0.018 0.025
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
126
Table IV
Cross-sectional Analysis
This table shows the estimates from triple-difference regressions using various specifications. The unit of
observation is a firm-year, and the panel runs 1997 - 2013. The main explanatory variable is the interaction
terms of two dummy variables. The first dummy indicates whether the firm-year has expiring contracts in
the following fiscal yeart+1. And the second dummy is to identify higher-than median factor variables. The
reported numbers are coefficient estimates and theirt-statistics (in parentheses). Standard errors clustered
by the firm. All regressions include firm and year fixed effects. The financial controls include the natural
logarithm of total assets, the natural logarithm of market value, book to market, return on assets, and the text-
based competition measure (Hoberg and Phillips (2010, 2016)). In columns (1) and (2), the sample consists
of unionized firms which have at least one contract expiration in the sample period. The dependent variable is
an indicator variable for CTOs on any material contracts in yeart. In columns (3) and (4), the sample consists
of unionized firms which have at least one contract expiration in the sample period. The dependent variable
is an indicator variable for CTOs on material contracts other than those related to debt financing, employees,
or investment in yeart. The dependent variable is an indicator variable for CTOs on any material contracts in
yeart. Panel A uses analyst forecast errors as a proxy for information asymmetry. Analyst forecast errors are
the absolute value of the difference between a firm’s reported earnings per share and the mean of most recent
analyst forecasts. Panel B uses backward-looking sales growth as growth opportunities proxy. Backward-
looking sales growth issale
t
=sale
t1
, wheresale
t
andsale
t1
are sales in yearst andt1, respectively.
Panel C uses Hadlock and Pierce measure (HP index) (Hadlock and Pierce (2010)) as a financial constraint
proxy. HP index is constructed as0:737Size+0:043Size
2
0:040Age, whereSize equals the
natural logarithm of the inflation-adjusted total assets (at from COMPUSTAT) (in 2004 dollars), andAge
is the number of years the firm is listed with a non-missing stock price on COMPUSTAT. The Size and
Age variables are truncated up-to 4:5 billion and 37 years. Panel D uses the text-based product similarity
measure, which represents pairwise similarities for given firm’s products with their substitutions (Hoberg
and Phillips (2010, 2016)). All the median values are determined among unionized firm-year observations.
Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
127
Panel A: Information Uncertainty
Dummy=1 if firm
redacted any material
contracts
Dummy=1 if firm
redacted material
contracts other than
lending, employee, or
investment agreements
(1) (2) (3) (4)
Dummy=1 if expiring contracts int+1 fiscal year
0:008 0:013 0:002 0:002
(0:47) (0:74) (0:11) (0:10)
Dummy=1 for high analyst forecast error 0:049

0:046

0:048

0:046

(3:26) (2:98) (3:23) (3:03)
(Dummy=1 for expiring contracts int+1 fiscal
year) (Dummy=1 for high analyst forecast error)
0:033

0:031 0:040

0:038

(1:76) (1:63) (2:28) (2:16)
Log (Total Assets) 0:030 0:008
(1:18) (0:33)
Log (Market Value) 0:038

0:035

(2:16) (2:10)
Book to Market 0:010 0:010
(0:88) (0:91)
Return on Assets 0:011 0:004
(0:33) (0:14)
Text-Based Competition Measure 0:064 0:076

(1:48) (1:97)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3333 3287 3333 3287
AdjustedR
2
0.025 0.036 0.020 0.028
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
128
Panel B: Growth Opportunities
Dummy=1 if firm
redacted any material
contracts
Dummy=1 if firm
redacted material
contracts other than
lending, employee, or
investment agreements
(1) (2) (3) (4)
Dummy=1 if expiring contracts int+1 fiscal year
0:046

0:051

0:048

0:052

(2:97) (3:16) (3:34) (3:46)
Dummy=1 for high backward-looking sales growth
0:013 0:029 0:018 0:033

(0:66) (1:45) (1:04) (1:84)
(Dummy=1 for expiring contracts int+1 fiscal
year) (Dummy=1 for high backward-looking
sales growth)
0:032 0:037

0:046

0:052

(1:54) (1:76) (2:36) (2:61)
Log (Total Assets) 0:026 0:006
(1:00) (0:23)
Log (Market Value) 0:037

0:034

(2:16) (2:04)
Book to Market 0:001

0:001

(3:81) (3:74)
Return on Assets 0:013 0:028
(0:39) (0:79)
Text-Based Competition Measure 0:070 0:080

(1:55) (1:92)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3626 3466 3626 3466
AdjustedR
2
0.025 0.034 0.020 0.027
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
129
Panel C: Financial Constraints
Dummy=1 if firm
redacted any material
contracts
Dummy=1 if firm
redacted material
contracts other than
lending, employee, or
investment agreements
(1) (2) (3) (4)
Dummy=1 if expiring contracts int+1 fiscal year
0:044

0:044

0:049

0:049

(2:22) (2:26) (2:58) (2:61)
Dummy=1 for high HP measures 0:019 0:033 0:007 0:003
(0:58) (0:97) (0:23) (0:11)
(Dummy=1 for expiring contracts int+1 fiscal
year) (Dummy=1 for high HP measures)
0:022 0:021 0:040

0:041

(0:96) (0:91) (1:82) (1:83)
Total Assets 0:000 0:000
(0:21) (0:73)
Log (Market Value) 0:046

0:037

(3:05) (2:68)
Book to Market 0:001

0:001

(4:30) (4:09)
Return on Assets 0:003 0:023
(0:07) (0:65)
Text-Based Competition Measure 0:065 0:079

(1:43) (1:89)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3635 3468 3635 3468
AdjustedR
2
0.025 0.033 0.019 0.027
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
130
Panel D: Cost of Strike
Dummy=1 if firm
redacted any material
contracts
Dummy=1 if firm
redacted material
contracts other than
lending, employee, or
investment agreements
(1) (2) (3) (4)
Dummy=1 if expiring contracts int+1 fiscal year
0:050

0:054

0:036

0:038

(2:84) (3:08) (2:10) (2:28)
Dummy=1 for high text-based similarity measure
0:002 0:002 0:015 0:022
(0:07) (0:07) (0:52) (0:74)
(Dummy=1 for expiring contracts int+1 fiscal
year) (Dummy=1 for high text-based similarity
measure)
0:040

0:042

0:020 0:022
(1:69) (1:79) (0:89) (0:99)
Log (Total Assets) 0:026 0:005
(1:01) (0:22)
Log (Market Value) 0:036

0:034

(2:14) (2:07)
Book to Market 0:001

0:001

(3:84) (3:77)
Return on Assets 0:013 0:028
(0:38) (0:81)
Text-Based Competition Measure 0:087

0:099

(1:87) (2:30)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3475 3468 3475 3468
AdjustedR
2
0.024 0.035 0.018 0.027
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
131
Table V
Redaction andEx-Post Firm Performance
This table shows the estimates from linear regressions using various specifications. The unit of observation
is a firm-year, and the panel runs 1997 - 2013. To balance unionized firms with and without implementing
CTOs, I match redacting firms to non-redacting firms based on the firm-year specifications, including the
logarithm of total assets, stockholders’ equity, return on assets, the logarithm of market value, book to
market ratio, and text-based financial constraint measure. The main explanatory variable is the interaction
terms of two dummy variables. The first dummy is an indicator variable for CTOs on any material contracts
in year t. The second dummy is to identify whether the firm-year has expiring contracts in the following
fiscal year t + 1. The outcome variables are indicated at the top of each column. Return on assets is
calculated as ib
t+1
=at
t+1
. Operating cash flow per asset is equal to (ib
t+1
+ dp
t+1
)=at
t+1
. Operating
margin is determined byib
t+1
=sale
t+1
. The reported numbers are coefficient estimates and theirt-statistics
(in parentheses). Standard errors clustered by the firm. All regressions include firm and year fixed effects.
Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Return on Assets
Operating CF per
Asset
Operating Margin
(1) (2) (3)
Dummy=1 if CTO int fiscal year 0:024

0:021

0:060

(2:14) (2:09) (2:62)
Dummy=1 if expiring contracts int+1
fiscal year
0:022

0:021

0:027

(2:44) (2:44) (2:14)
(Dummy=1 if CTO int fiscal year)
(Dummy if expiring contracts int+1 fiscal
year)
0:000 0:005 0:023
(0:02) (0:44) (0:87)
Firm Fixed Effect Yes Yes Yes
Year Fixed Effect Yes Yes Yes
Entropy Balanced Yes Yes Yes
Observations 2770 2770 2770
AdjustedR
2
0.289 0.267 0.173
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
132
Table VI
Strategic Disclosure and Strategic Liquidity Management
This table shows the estimates from linear regressions using various specifications. The unit of observation is a firm-year,
and the panel runs 1997 - 2013. The dependent variable is an indicator variable for CTOs on any material contracts in year
t. The two main explanatory variables are a dummy variable to indicate whether the firm-year has expiring contracts in the
following fiscal yeart+1 and its interaction terms with the strategic liquidity management devices. Liquidity management
is either debt financing or asset purchases and is indicated at the top of each column. Standard errors clustered by the
firm. All regressions include firm and year fixed effects. The financial controls include the natural logarithm of total
assets, the natural logarithm of market value, book to market, return on assets, and the text-based competition measure
(Hoberg and Phillips (2010, 2016)). Columns (1) and (2) use the natural logarithm of the loan amount as a proxy for
liquidity management. The loan amount is measured by the debt which is newly financed in the following fiscal year
t+1. Columns (3) and (4) use asset purchase amount as a proxy for liquidity management. The asset purchase amount
is measured asaqc
t+1
=at
t
. Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
Liquidity management
using debt financing
Liquidity management
using asset purchase
(1) (2) (3) (4)
Dummy=1 if expiring contracts in
t+1 fiscal year
0:038

0:039

0:038

0:038

(2:96) (2:89) (2:84) (2:73)
Liquidity Management Amount 0:001 0:001 0:149

0:151
(0:67) (0:51) (1:86) (1:43)
(Dummy=1 if expiring contracts int+1 fiscal
year) (Liquidity Management Amount)
0:001 0:001 0:203

0:196

(1:17) (1:14) (2:29) (1:82)
Log (Total Assets) 0:026 0:029
(1:02) (0:94)
Log (Market Value) 0:036

0:027
(2:13) (1:53)
Book to Market 0:001

0:000
(3:72) (0:30)
Return on Assets 0:013 0:022
(0:39) (0:62)
Text-Based Competition Measure 0:070 0:065
(1:55) (1:38)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3640 3468 3258 3109
AdjustedR
2
0.025 0.033 0.026 0.030
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
133
Table VII
Redaction Tendency and Lagged Financial Constraint
This table shows the estimates from triple-difference regressions using various specifications. The unit of
observation is a firm-year, and the panel runs 1997 - 2013. The main explanatory variable is the interaction
terms of two dummy variables. The first dummy indicates whether the firm-year has expiring contracts
in the following fiscal year t+1. And the second dummy is to identify higher-than median Hadlock and
Pierce measure (HP index) (Hadlock and Pierce (2010)) as of t1 year-end. HP index is constructed as
0:737Size+0:043Size
2
0:040Age, whereSize equals the natural logarithm of the inflation-
adjusted total assets (at from COMPUSTAT) (in 2004 dollars), andAge is the number of years the firm is
listed with a non-missing stock price on COMPUSTAT. TheSize andAge variables are truncated up-to4:5
billion and 37 years. The reported numbers are coefficient estimates and theirt-statistics (in parentheses).
Standard errors clustered by the firm. All regressions include firm and year fixed effects. The financial
controls include the natural logarithm of total assets, the natural logarithm of market value, book to market,
return on assets, and the text-based competition measure (Hoberg and Phillips (2010, 2016)). In columns
(1) and (2), the sample consists of unionized firms which have at least one contract expiration in the sample
period. The dependent variable is an indicator variable for CTOs on any material contracts in year t. In
columns (3) and (4), the sample consists of unionized firms which have at least one contract expiration in
the sample period. The dependent variable is an indicator variable for CTOs on material contracts other than
those related to debt financing, employees, or investment in yeart. The dependent variable is an indicator
variable for CTOs on any material contracts in yeart. All the median values are determined among unionized
firm-year observations. Significance levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
134
Dummy=1 if firm
redacted any material
contracts
Dummy=1 if firm
redacted material
contracts other than
lending, employee, or
investment agreements
(1) (2) (3) (4)
Dummy=1 if expiring contracts int+1 fiscal year
0:046

0:046

0:052

0:052

(2:27) (2:32) (2:71) (2:71)
Dummy=1 for high lagged HP measures 0:024 0:040 0:002 0:006
(0:62) (1:03) (0:07) (0:18)
(Dummy=1 for expiring contracts in t+1 fiscal year)
(Dummy=1 for high lagged HP measures)
0:026 0:023 0:046

0:044

(1:13) (0:95) (2:08) (1:97)
Log (Total Assets) 0:030 0:008
(1:15) (0:33)
Log (Market Value) 0:037

0:034

(2:17) (2:09)
Book to Market 0:001

0:001

(3:75) (3:68)
Return on Assets 0:008 0:025
(0:24) (0:72)
Text-Based Competition Measure 0:068 0:079

(1:48) (1:88)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3629 3467 3629 3467
AdjustedR
2
0.026 0.035 0.020 0.027
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
135
Internet Appendix to
“Do Firms Leave Workers in the Dark Before Wage Negotiations?”
Table A1
Industry Distribution of Sample Firms
This table reports the industry distribution of sample firms during the period 1997 - 2013, separately for
unionized and nonunionized firms. Each column represents the number of firm-year observations belonging
to the corresponding SIC industry group. Firms are classified as nonunionized firms if they do not have
collective bargaining expirations in the sample period. Unionized firms consist of companies which have at
least one contract expiration in the sample period.
Industry Description
Nonunionized Firm Unionized Firm Total
No. % No. % No. %
Agriculture, Forestry, and Fishing (SIC 01 - 09) 419 0.4 0 0.0 419 0.4
Mining (SIC 10 - 14) 5,734 5.0 29 0.8 5,763 4.9
Construction (SIC 15 - 17) 1,085 0.9 65 1.8 1,150 1.0
Manufacturing (SIC 20 - 39) 41,119 35.9 1,944 53.4 43,063 36.4
Transportation and Public Utilities (SIC 40 - 49) 9,912 8.6 1,007 27.7 10,919 9.2
Wholesale Trade (SIC 50 - 51) 3,446 3.0 63 1.7 3,509 3.0
Retail Trade (SIC 52 - 59) 5,745 5.0 241 6.6 5,986 5.1
Finance, Insurance, and Real Estate (SIC 60 - 67) 24,213 21.1 39 1.1 24,252 20.5
Services (SIC 70 - 89) 20,590 18.0 201 5.5 20,791 17.6
Nonclassifiable Establishments (SIC 99) 2,417 2.1 51 1.4 2,468 2.1
Total 114,680 100.0 3,640 100.0 118,320 100.0
A.1
Table A2
Fiscal Year Distribution of Sample Firms
This table reports the fiscal year distribution of sample firms during the period 1997 - 2013. Each column
represents the number of firm-year observations belonging to the corresponding fiscal year. Panel A presents
the fiscal year distribution of firm-year observations, separately for unionized and nonunionized firms. Firms
are classified as nonunionized firms if they do not have collective bargaining expirations in the sample
period. Unionized firms consist of companies which have at least one contract expiration in the sample
period. Panel B presents the fiscal year distribution of firm-year observations, separately for 10-K’s without
and with any redacted material agreements. 10-K filings are defined to have redacted exhibits if they have
text-strings which represent a CTO, such as “confidential,” “confidential request,” “confidential treatment,”
“CT order,” or “redacted.”
Panel A: Sample Firms
Fiscal Year End
Nonunionized Firm Unionized Firm Total
No. % No. % No. %
1997 8,274 7.2 220 6.0 8,494 7.2
1998 8,176 7.1 227 6.2 8,403 7.1
1999 8,447 7.4 220 6.0 8,667 7.3
2000 8,321 7.3 221 6.1 8,542 7.2
2001 7,876 6.9 228 6.3 8,104 6.8
2002 7,344 6.4 224 6.2 7,568 6.4
2003 6,963 6.1 224 6.2 7,187 6.1
2004 6,740 5.9 220 6.0 6,960 5.9
2005 6,501 5.7 218 6.0 6,719 5.7
2006 6,319 5.5 215 5.9 6,534 5.5
2007 6,176 5.4 205 5.6 6,381 5.4
2008 5,808 5.1 204 5.6 6,012 5.1
2009 5,615 4.9 208 5.7 5,823 4.9
2010 5,515 4.8 206 5.7 5,721 4.8
2011 5,569 4.9 202 5.5 5,771 4.9
2012 5,521 4.8 199 5.5 5,720 4.8
2013 5,515 4.8 199 5.5 5,714 4.8
Total 114,680 100.0 3,640 100.0 118,320 100.0
A.2
Panel B: Confidential Treatment
Fiscal Year End
Without Redacted Exhibit With Redacted Exhibit Total
No. % No. % No. %
1997 7,422 7.6 1,072 5.2 8,494 7.2
1998 7,322 7.5 1,081 5.2 8,403 7.1
1999 7,449 7.6 1,218 5.9 8,667 7.3
2000 7,251 7.4 1,291 6.2 8,542 7.2
2001 6,891 7.1 1,213 5.9 8,104 6.8
2002 6,420 6.6 1,148 5.5 7,568 6.4
2003 6,035 6.2 1,152 5.6 7,187 6.1
2004 5,745 5.9 1,215 5.9 6,960 5.9
2005 5,482 5.6 1,237 6.0 6,719 5.7
2006 5,280 5.4 1,254 6.1 6,534 5.5
2007 5,104 5.2 1,277 6.2 6,381 5.4
2008 4,794 4.9 1,218 5.9 6,012 5.1
2009 4,605 4.7 1,218 5.9 5,823 4.9
2010 4,488 4.6 1,233 6.0 5,721 4.8
2011 4,510 4.6 1,261 6.1 5,771 4.9
2012 4,456 4.6 1,264 6.1 5,720 4.8
2013 4,376 4.5 1,338 6.5 5,714 4.8
Total 97,630 100.0 20,690 100.0 118,320 100.0
A.3
Table A3
Robustness Test: Cross-sectional Analysis
This table shows the estimates from triple-difference regressions using various specifications. The unit of
observation is a firm-year, and the panel runs 1997 - 2013. The main explanatory variable is the interaction
terms of two dummy variables. The first dummy indicates whether the firm-year has expiring contracts in
the following fiscal yeart+1. And the second dummy is to identify higher-than median factor variables. The
reported numbers are coefficient estimates and theirt-statistics (in parentheses). Standard errors clustered
by the firm. All regressions include firm and year fixed effects. The financial controls include the natural
logarithm of total assets, the natural logarithm of market value, book to market, return on assets, and the
text-based competition measure (Hoberg and Phillips (2010, 2016)). Panel A uses the number of following
analyst as a proxy for the inverse of information asymmetry. Analyst following is the number of analysts
producing forecasts on a firm’s earnings per share. Panel B uses book-to-market ratio as the inverse of
growth opportunities proxy. Book-to-market ratio is ceq=(prcc f csho, based on COMPUSTAT data
items. Panel C uses Whited and Wu index (Whited and Wu (2006); Hennessy and Whited (2007)) as a
financial constraint proxy. Whited and Wu index is constructed as0:091[(ib+dp)=at]0:062[indicator
set to one ifdvc+dvp is positive, and zero otherwise]+0:021[dltt=at]0:044[log(at)]+0:102[average
industry sales growth, for three-digit SIC industry]0:035[sales growth]. Panel D uses the volatility in
inventory stock as the inverse of strike costs proxy. The volatility in inventory stock is the standard deviation
of inventory to total assets ratio for the last five years. All the median values are determined among unionized
firm-year observations. In columns (1) and (2), the sample consists of unionized firms which have at least
one contract expiration in the sample period. The dependent variable is an indicator variable for CTOs on
any material contracts in yeart. In columns (3) and (4), the sample consists of unionized firms which have
at least one contract expiration in the sample period. The dependent variable is an indicator variable for
CTOs on material contracts other than those related to debt financing, employees, or investment in yeart.
The dependent variable is an indicator variable for CTOs on any material contracts in yeart. Significance
levels are indicated: * = 10 percent, ** = 5 percent, *** = 1 percent.
A.4
Panel A: Information Uncertainty
Dummy=1 if firm
redacted any material
contracts
Dummy=1 if firm
redacted material
contracts other than
lending, employee, or
investment agreements
(1) (2) (3) (4)
Dummy=1 if expiring contracts int+1 fiscal year
0:032

0:037

0:025

0:028

(1:95) (2:20) (1:68) (1:81)
Dummy=1 if high number of following analysts 0:009 0:009 0:015 0:003
(0:33) (0:33) (0:60) (0:10)
(Dummy=1 for expiring contracts int+1 fiscal
year) (Dummy=1 for high number of following
analysts)
0:013 0:015 0:011 0:012
(0:54) (0:64) (0:47) (0:51)
Log (Total Assets) 0:033 0:010
(1:32) (0:44)
Log (Market Value) 0:037

0:032

(2:09) (1:93)
Book to Market 0:007 0:007
(0:75) (0:79)
Return on Assets 0:010 0:006
(0:29) (0:19)
Text-Based Competition Measure 0:056 0:068

(1:25) (1:68)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3346 3298 3346 3298
AdjustedR
2
0.022 0.032 0.017 0.024
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
A.5
Panel B: Growth Opportunities
Dummy=1 if firm
redacted any material
contracts
Dummy=1 if firm
redacted material
contracts other than
lending, employee, or
investment agreements
(1) (2) (3) (4)
Dummy=1 if expiring contracts int+1 fiscal year
0:027

0:027

0:017 0:016
(1:74) (1:66) (1:18) (1:09)
Dummy=1 for high book-to-market ratio
0:012 0:032 0:007 0:026
(0:54) (1:22) (0:30) (1:04)
(Dummy=1 for expiring contracts int+1 fiscal
year) (Dummy=1 for high book-to-market ratio)
0:002 0:012 0:011 0:019
(0:11) (0:54) (0:58) (0:94)
Log (Total Assets) 0:019 0:002
(0:70) (0:07)
Log (Market Value) 0:041

0:036

(2:21) (2:04)
Book to Market 0:001

0:001

(3:84) (3:78)
Return on Assets 0:008 0:025
(0:23) (0:73)
Text-Based Competition Measure 0:071 0:080

(1:55) (1:89)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3548 3468 3548 3468
AdjustedR
2
0.022 0.034 0.016 0.025
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
A.6
Panel C: Financial Constraints
Dummy=1 if firm
redacted any material
contracts
Dummy=1 if firm
redacted material
contracts other than
lending, employee, or
investment agreements
(1) (2) (3) (4)
Dummy=1 if expiring contracts int+1 fiscal year
0:048

0:051

0:048

0:048

(2:80) (2:92) (2:87) (2:98)
Dummy=1 for high WW measures 0:053

0:037 0:062

0:053

(2:40) (1:57) (3:03) (2:50)
(Dummy=1 for expiring contracts int+1 fiscal
year) (Dummy=1 for high WW measures)
0:032 0:033 0:039

0:040

(1:51) (1:46) (1:96) (1:92)
Log (Total Assets) 0:024 0:001
(0:91) (0:05)
Log (Market Value) 0:036

0:033

(2:12) (2:04)
Book to Market 0:001

0:001

(3:70) (3:66)
Return on Assets 0:009 0:022
(0:28) (0:63)
Text-Based Competition Measure 0:070 0:081

(1:55) (1:95)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3612 3452 3612 3452
AdjustedR
2
0.025 0.034 0.020 0.027
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
A.7
Panel D: Cost of Strike
Dummy=1 if firm
redacted any material
contracts
Dummy=1 if firm
redacted material
contracts other than
lending, employee, or
investment agreements
(1) (2) (3) (4)
Dummy=1 if expiring contracts int+1 fiscal year
0:046

0:043

0:044

0:041

(2:44) (2:23) (2:56) (2:30)
Dummy=1 if high inventory stock volatility 0:008 0:002 0:021 0:013
(0:31) (0:07) (0:85) (0:50)
(Dummy=1 for expiring contracts int+1 fiscal
year) (Dummy=1 for high inventory stock
volatility)
0:018 0:011 0:026 0:021
(0:77) (0:45) (1:17) (0:88)
Total Assets 0:000 0:000
(0:13) (0:86)
Log (Market Value) 0:041

0:031

(2:63) (2:18)
Book to Market 0:001

0:001

(4:40) (3:96)
Return on Assets 0:002 0:026
(0:06) (0:73)
Text-Based Competition Measure 0:046 0:057
(0:99) (1:33)
Firm Fixed Effect Yes Yes Yes Yes
Year Fixed Effect Yes Yes Yes Yes
Observations 3356 3210 3356 3210
AdjustedR
2
0.025 0.030 0.018 0.023
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
A.8
Table A4
Strategic Disclosure and Strategic Liquidity Management
This table shows the estimates from linear regressions using various specifications. The unit of observation
is a firm-year, and the panel runs 1997 - 2013. The dependent variable is an indicator variable for CTOs
on any material contracts in year t. The two main explanatory variables are a dummy variable to indicate
whether the firm-year has expiring contracts in the following fiscal year t + 1 and its interaction terms
with the strategic liquidity management devices. Liquidity management is proxied by the inverse of cash
and short-term investments divided by total assets is indicated at the top of each column. Standard errors
clustered by the firm. All regressions include firm and year fixed effects. The financial controls include the
natural logarithm of total assets, the natural logarithm of market value, book to market, return on assets, and
the text-based competition measure (Hoberg and Phillips (2010, 2016)). Significance levels are indicated: *
= 10 percent, ** = 5 percent, *** = 1 percent.
Cash holdings scaled by total asset
(1) (2)
Dummy=1 if expiring contracts in
t+1 fiscal year
0:025 0:023
(1:62) (1:53)
Cash holdings scaled by total asset 0:002 0:110
(0:01) (0:85)
(Dummy=1 if expiring contracts int+1 fiscal
year) (Cash holdings scaled by total asset)
0:094 0:122
(0:78) (1:04)
Log (Total Assets) 0:027
(1:01)
Log (Market Value) 0:036

(2:10)
Book to Market 0:001

(3:72)
Return on Assets 0:012
(0:35)
Text-Based Competition Measure 0:069
(1:51)
Firm Fixed Effect Yes Yes
Year Fixed Effect Yes Yes
Observations 3633 3466
AdjustedR
2
0.024 0.033
t statistics in parentheses

p< 0:1,

p< 0:05,

p< 0:01
A.9
136

References

Agrawal, Anup, Jeffrey F Jaffe, and Gershon N Mandelker, 1992, The post-merger
performance of acquiring firms: a re-examination of an anomaly, The Journal of
Finance 47, 1605 –1621.

Agrawal, Ashwini K, 2011, Corporate governance objectives of labor union shareholders:
Evidence from proxy voting, Review of Financial Studies 25, 187 –226.

Ahern, Kenneth R, Daniele Daminelli, and Cesare Fracassi, 2015, Lost in translation? the
effect of cultural values on mergers around the world, Journal of Financial
Economics 117, 165 –189.

Aktas, Nihat, Eric De Bodt, Helen Bollaert, and Richard Roll, 2016, CEO narcissism and
the takeover process: From private initiation to deal completion, Journal of Financial
and Quantitative Analysis 51, 113 –137.

Aobdia, Daniel, and Lin Cheng, 2018, Unionization, product market competition, and
strategic disclosure, Journal of Accounting and Economics 65, 331 –357.

Atanassov, Julian, and E Han Kim, 2009, Labor and corporate governance: International
evidence from restructuring decisions, Journal of Finance 64, 341 –374.

Ball, Christopher, Gerard Hoberg, and Vojislav Maksimovic, 2015, Disclosure, business
change and earnings quality, Available at SSRN 2260371.

Bellstam, Gustaf, Sanjai Bhagat, and J Anthony Cookson, 2019, A text-based analysis of
corporate innovation, Available at SSRN 2803232.

Benmelech, Efraim, Nittai K Bergman, and Ricardo J Enriquez, 2012, Negotiating with
labor under financial distress, Review of Corporate Finance Studies 1, 28 –67.

Bereskin, Fred, Seong K Byun, Micah S Officer, and Jong-Min Oh, 2018, The effect of
cultural similarity on mergers and acquisitions: Evidence from corporate social
responsibility, Journal of Financial and Quantitative Analysis 53, 1995 –2039.

Boone, Audra L, Ioannis V Floros, and Shane A Johnson, 2016, Redacting proprietary
information at the initial public offering, Journal of Financial Economics 120, 102 –123.

Bova, Francesco, 2013, Labor unions and management's incentive to signal a negative
outlook, Contemporary Accounting Research 30, 14–41.
137

Bronars, Stephen G, and Donald R Deere, 1991, The threat of unionization, the use of
debt, and the preservation of shareholder wealth, Quarterly Journal of Economics
106, 231 –254.

Calomiris, Charles W, and Harry Mamaysky, 2019, How news and its context drive risk
and returns around the world, Journal of Financial Economics 133, 299 –336.

Chava, Sudheer, and Michael R Roberts, 2008, How does financing impact investment?
The role of debt covenants, Journal of Finance 63, 2085 –2121.

Cheng, Lin, 2017, Organized labor and debt contracting: Firm-level evidence from
collective bargaining, The Accounting Review 92, 57 –85.

Cheng, Lin, Santanu Mitra, and Hakjoon Song, 2017, Organized labor and audit fees,
Accounting Horizons 31, 93 –108.

Chino, Atsushi, 2016, Do labor unions affect firm payout policy?: Operating leverage and
rent extraction effects, Journal of Corporate Finance 41, 156 –178.

Chyz, James A, Winnie Siu Ching Leung, Oliver Zhen Li, and Oliver Meng Rui, 2013,
Labor unions and tax aggressiveness, Journal of Financial Economics 108, 675 –698.

Cramton, Peter C, 1984, Bargaining with incomplete information: An infinite-horizon
model with two-sided uncertainty, Review of Economic Studies 51, 579 –593.

Cramton, Peter C, and Joseph S Tracy, 1992, Strikes and holdouts in wage bargaining:
Theory and data, American Economic Review 100–121.

Cutler, Stephen M, 2004, Tone at the top: Getting it right, Second Annual General
Counsel Roundtable, Washington 3.

DeAngelo, Harry, and Linda DeAngelo, 1991, Union negotiations and corporate policy:
A study of labor concessions in the domestic steel industry during the 1980s, Journal
of Financial Economics 30, 3–43.

Demsetz, Harold, and Kenneth Lehn, 1985, The structure of corporate ownership: Causes
and consequences, Journal of Political Economy 93, 1155 –1177.

D'Souza, Julia, John Jacob, and K Ramesh, 2000, The use of accounting flexibility to
reduce labor renegotiation costs and manage earnings, Journal of Accounting and
Economics 30, 187 –208.

138

Duchin, Ran, John G. Matsusaka, and Oguzhan Ozbas, 2010, When are outside directors
effective?, Journal of Financial Economics 96.2, 195 –214.

Einhorn, Eti, and Amir Ziv, 2008, Intertemporal dynamics of corporate voluntary
disclosures, Journal of Accounting Research 46, 567 –589.

Falato, Antonio, and Nellie Liang, 2016, Do creditor rights increase employment risk?
Evidence from loan covenants, Journal of Finance 71, 2545 –2590.

Faleye, Olubunmi, Vikas Mehrotra, and Randall Morck, 2006, When labor has a voice in
corporate governance, Journal of Financial and Quantitative Analysis 41, 489 –510.

Fallick, Bruce C, and Kevin A Hassett, 1999, Investment and union certification, Journal
of Labor Economics 17, 570 –582.

Freeman, Richard B, and Morris M Kleiner, 1990, The impact of new unionization on
wages and working conditions, Journal of Labor Economics 8, S8 –S25.

Freeman, Richard B, and Morris M Kleiner, 1999, Do unions make enterprises
insolvent?, ILR Review 52, 510–527.

Fresard, Laurent, Gerard Hoberg, and Gordon M Phillips, 2019, Innovation activities and
integration through vertical acquisitions, The Review of Financial Studies.

Frost, Ann C, 2000, Explaining variation in workplace restructuring: The role of local
union capabilities, ILR Review 53, 559 –578.

Garcia Osma, Beatriz, Araceli Mora, and Ana M Sabater, 2015, Strategic accounting
choice around firm-level labor negotiations, Journal of Accounting, Auditing &
Finance 30, 246 –277.

Gorton, Gary B, and Alexander K Zentefis, 2020, Corporate culture as a theory of the
firm, NBER Working Paper 27353.

Graham, John R, Jillian Grennan, Campbell R Harvey, and Shivaram Rajgopal, 2016,
Corporate culture: The interview evidence, Duke I&E Research Paper 16–70.

Graham, John R, Jillian Grennan, Campbell R Harvey, and Shivaram Rajgopal, 2017,
Corporate culture: Evidence from the field, National Bureau of Economic Research.

Graham, John R, Campbell R Harvey, and Manju Puri, 2013, Managerial attitudes and
corporate actions, Journal of Financial Economics 109, 103 –121.

139

Greene, William, 2004, Fixed effects and bias due to the incidental parameters problem
in the Tobit model, Econometric Reviews 23, 125 –147.

Grennan, Jillian, 2019, A corporate culture channel: How increased shareholder
governance reduces firm value, Available at SSRN 2345384.

Grossman, Sanford J, 1981, The informational role of warranties and private disclosure
about product quality, Journal of Law and Economics 24, 461 –483.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales, 2015, Corporate culture, societal
culture, and institutions, American Economic Review 105, 336 –39.

Hadlock, Charles J, and Joshua R Pierce, 2010, New evidence on measuring financial
constraints: Moving beyond the KZ index, Review of Financial Studies 23, 1909 –
1940.

Hainmueller, Jens, 2012, Entropy balancing for causal effects: A multivariate
reweighting method to produce balanced samples in observational studies, Political
Analysis 20, 25 –46.

Hanley, Kathleen Weiss, and Gerard Hoberg, 2019, Dynamic interpretation of emerging
risks in the financial sector, The Review of Financial Studies 32, 4543 –4603.

Hart, Oliver D, 1983, Optimal labour contracts under asymmetric information: An
introduction, Review of Economic Studies 50, 3 –35.

Heckman, James J, 1987, The incidental parameters problem and the problem of initial
conditions in estimating a discrete time-discrete data stochastic process and some
Monte Carlo evidence (University of Chicago Center for Mathematical studies in
Business and Economics).

Hennessy, Christopher A, and Toni M Whited, 2007, How costly is external financing?
Evidence from a structural estimation, Journal of Finance 62, 1705 –1745.

Hilary, Gilles, 2006, Organized labor and information asymmetry in the financial
markets, Review of Accounting Studies 11, 525 –548.

Hoberg, Gerard, and Craig Lewis, 2017, Do fraudulent firms produce abnormal
disclosure?, Journal of Corporate Finance 43, 58 –85.

Hoberg, Gerard, and Vojislav Maksimovic, 2014, Redefining financial constraints: A
text-based analysis, Review of Financial Studies 28, 1312 –1352.

140

Hoberg, Gerard, and Gordon Phillips, 2010, Product market synergies and competition in
mergers and acquisitions: A text-based analysis, Review of Financial Studies 23,
3773 –3811.

Hoberg, Gerard, and Gordon Phillips, 2016, Text-based network industries and
endogenous product differentiation, Journal of Political Economy 124, 1423 –1465.

Hoberg, Gerard, and Gordon M Phillips, 2018, Product integration and merger success,
Tuck School of Business Working Paper 17 –21.

Hofstede, Geert, 1980, Culture and organizations, International Studies of Management
& Organization 10, 15 –41.

Israelsen, Ryan D, 2014, Tell it like it is: Disclosed risks and factor portfolios, Available
at SSRN 2504522.

Jensen, Michael C, 1986, Agency costs of free cash flow, corporate finance, and
takeovers, American Economic Review 76, 323 –329.

Kaplan, Steven N, Mark M Klebanov, and Morten Sorensen, 2012, Which CEO
characteristics and abilities matter?, The Journal of Finance 67, 973 –1007.

Kim, E Han, and Paige Ouimet, 2011, Employee stock ownership plans: Employee
compensation and firm value, Available at SSRN 1823745.

Kim, E Han, and Paige Ouimet, 2014, Broad-based employee stock ownership: Motives
and outcomes, Journal of Finance 69, 1273 –1319.

Klasa, Sandy, William F Maxwell, and Hernan Ortiz-Molina, 2009, The strategic use of
corporate cash holdings in collective bargaining with labor unions, Journal of
Financial Economics 92, 421 –442.

Kuhn, Peter, 1986, Wages, effort, and incentive compatibility in life-cycle employment
contracts, Journal of Labor Economics 4, 28 –49.

Lee, David S, and Alexandre Mas, 2012, Long-run impacts of unions on firms: New
evidence from financial markets, 1961 –1999, Quarterly Journal of Economics 127,
333 –378.

Li, Kai, Feng Mai, Rui Shen, and Xinyan Yan, 2020, Measuring corporate culture using
machine learning, The Review of Financial Studies.

141

Liberty, Susan E, and Jerold L Zimmerman, 1986, Labor union contract negotiations and
accounting choices, The Accounting Review 692 –712.

Lopez-Lira, Alejandro, 2019, Risk factors that matter: Textual analysis of risk disclosures
for the cross-section of returns, Available at SSRN 3313663.

Loughran, Tim, and Bill McDonald, 2011, When is a liability not a liability? Textual
analysis, dictionaries, and 10-Ks, The Journal of Finance 66, 35–65.

Lowry, Michelle, Roni Michaely, and Ekaterina Volkova, 2020, Information revealed
through the regulatory process: Interactions between the sec and companies ahead of
their IPO, The Review of Financial Studies.

Matsa, David A, 2010, Capital structure as a strategic variable: Evidence from collective
bargaining, Journal of Finance 65, 1197 –1232.

Matsa, David A, 2018, Capital structure and a firm's workforce, Annual Review of
Financial Economics 10, 387 –412.

Matsusaka, John G, Oguzhan Ozbas, and Irene Yi, 2019, Opportunistic proposals by
union shareholders, Review of Financial Studies 32, 3215 –3265.

Mauro, Martin J, 1982, Strikes as a result of imperfect information, ILR Review 35, 522 –
538.

Merton, Robert C, 1987, A simple model of capital market equilibrium with incomplete
information, Journal of Finance 42, 483 –510.

Milgrom, Paul R, 1981, Good news and bad news: Representation theorems and
applications, The Bell Journal of Economics 380 –391.

Moeller, Sara B, Frederik P Schlingemann, and Rene M Stulz, 2007, How do diversity of
opinion and information asymmetry affect acquirer returns?, The Review of Financial
Studies 20, 2047 –2078.

Myers, Brett W, and Alessio Saretto, 2016, Does capital structure affect the behavior of
nonfinancial stakeholders? An empirical investigation into leverage and union strikes,
Management Science 62, 3235 –3253.

Myers, Stewart C, and Nicholas S Majluf, 1984, Corporate financing and investment
decisions when firms have information that investors do not have, Journal of
Financial Economics 13, 187 –221.

142

Reder, Melvin W, and George R Neumann, 1980, Conflict and contract: The case of
strikes, Journal of Political Economy 88, 867 –886.

Rich, Robert, and Joseph Tracy, 2013, Early contract renegotiation: An analysis of us
labor contracts, 1970 –1995, Journal of Labor Economics 31, 825 –842.

Roll, Richard, 1986, The hubris hypothesis of corporate takeovers, Journal of Business
197 –216.

Sarbanes, Paul, 2002, Sarbanes-Oxley act of 2002, in The Public Company Accounting
Reform and Investor Protection Act. Washington DC: US Congress.

SEC, 2010, Assessment of corporation finance's confidential treatment processes and
procedures, Report No. 479.

SEC, 2019, Fast act modernization and simplification of regulation S-K, Release No. 33-
10618.

Stoffman, Noah, Michael Woeppel, and M Deniz Yavuz, 2019, Small innovators: No
risk, no return, Kelley School of Business Research Paper.

Verrecchia, Robert E, 1983, Discretionary disclosure, Journal of Accounting and
Economics 5, 179–194.

Verrecchia, Robert E, and Joseph Weber, 2006, Redacted disclosure, Journal of
Accounting Research 44, 791 –814.

Wagenhofer, Alfred, 1990, Voluntary disclosure with a strategic opponent, Journal of
accounting and economics 12, 341 –363.

Whited, Toni M, and Guojun Wu, 2006, Financial constraints risk, Review of Financial
Studies 19, 531 –559.

Williamson, Oliver E, 1979, Transaction-cost economics: The governance of contractual
relations, The Journal of Law and Economics 22, 233–261.

Yi, Irene, 2016, Slashing liquidity through asset purchases: Evidence from collective
bargaining, Available at SSRN 2862397.

Zhang, X. Frank, 2006, Information uncertainty and stock returns, Journal of Finance 61,
105 –137.

143

Zingales, Luigi, 2015, The Cultural revolution" in finance, Journal of Financial
Economics 117, 1 –4.

Abstract (if available)

Abstract This thesis consists of two essays that study the firm and stakeholders relationships. Chapter 1 examines the role of corporate culture for mergers and acquisitions. To quantify corporate culture, I run a textual analysis of the language used in CEOs' annual letters. This analysis categorizes firms into three different corporate cultures: collaborative, innovative, and customer-centric. Using the novel measure of corporate culture, I find that firms with more similar corporate cultures are more likely to merge. Second, buyers' announcement returns are higher if targets have more similar corporate cultures. Finally, the cultural integration of two merged firms is positively related to post-merger performance and is negatively associated with ex post divestiture. In sum, this paper shows that cultural differences have meaningful impacts on mergers. Chapter 2 examines managers' strategic use of financial disclosure in labor negotiations. Using the exogenous expiration date of collective bargaining contracts, I find that when wage negotiations are imminent, firms strategically redact information about material agreements. Strategic redaction is pronounced when unions cannot accurately predict firms' prospects, when firms have low growth opportunities, when liquidity is less constrained, and when the estimated cost of a work stoppage is low. These results suggest that firms strategically withhold information to balance the costs and benefits of information asymmetry. Consistent with this interpretation, strategic disclosure is statistically uncorrelated to ex post performance.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Slashing liquidity through asset purchases: evidence from collective bargaining

PDF

How large are the coinsurance benefits of mergers?

PDF

Understanding the disclosure practices of firms affected by a natural disaster: the case of hurricanes

PDF

Insider editing on Wikipedia

PDF

How do acquirers govern the deal-making process? Three essays on U.S. mergers and acquisitions 1994 – 2017

PDF

Investment behavior of mutual fund managers

PDF

Essays in corporate finance

PDF

Executive compensation: the trend toward one size fits all

PDF

Essays on information and financial economics

PDF

Shareholder litigation as a disciplining device: evidence from firms' financial policies

PDF

Implied equity duration: international evidence

PDF

Regulatory team development in post-merger integration: a survey of views from medical product companies

PDF

Three essays on young entrepreneurial firms

PDF

Changing fundamental analysis in the new economy: the case of DuPont analysis and STEM firms

PDF

The changing policy environment and banks' financial decisions

PDF

Essays in financial intermediation

PDF

Three essays on macro and labor finance

PDF

Essays in corporate finance

PDF

Three essays on board structure

PDF

Automated contracts and the lawyers who don't review them: adoption and use of machine learning technology

Asset Metadata

Creator Yoo, Sunny (Seung Yeon) (author)

Core Title Essays on the firm and stakeholders relationships: evidence from mergers & acquisitions and labor negotiations

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Business Administration

Degree Conferral Date 2021-08

Publication Date 07/25/2021

Defense Date 05/10/2021

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag collective bargaining,confidential treatment order,corporate culture,labor negotiations,mergers & acquisitions,OAI-PMH Harvest

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Ahern, Kenneth (committee chair), Dechow, Patricia (committee member), Hoberg, Gerard (committee member), Matsusaka, John (committee member)

Creator Email yeoniyoo83@gmail.com,yoo399@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC15622075

Unique identifier UC15622075

Legacy Identifier etd-YooSunnySe-9874

Document Type Dissertation

Format application/pdf (imt)

Rights Yoo, Sunny (Seung Yeon)

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu