1

Equity Offerings, Technology Investments, and Employee Skill

Composition

E. Han Kim, Heuijung Kim, Yuan Li, Yao Lu, and Xinzheng Shi†

Abstract

We examine whether and how equity financing affects technology investments, demand for skills,

firm-level employment, and wages. Analyses of job advertisements posted online reveal that seasoned

equity offerings (SEOs) are associated with higher demands for computer skills and non-routine task

skills. To draw causal inferences, we construct an instrument using exogenous shocks on the

eligibility to issue SEOs, treatments of which are based on past irreversible firm behavior. We find

capital infusion through SEOs increases purchases of machines and equipment, innovations, and

employee skill composition. SEOs lead to a net decline in firm-level employment, displacing a greater

number of low-skilled workers than adding high-skilled employees. Average wages increase because

of higher skill composition, but total wages remain unchanged because the higher average wage

applies to a smaller number of employees. These results illustrate channels through which stock

markets affect labor markets.

First Draft: September 5, 2017

Revised: January 18, 2018

Keywords: Capital Skill Complementarity, Equity Issuance, Employment, Investment in Technology,

Innovations, Wages.

JEL Classifications: G32, J24, J31

†E. Han Kim is Everett E. Berg Professor of Finance at the University of Michigan, Ross School of Business,

Ann Arbor, Michigan 48109: ehkim@umich.edu. Heuijung Kim is an instructor at Sungkyunkwan University,

SKK Business School, Seoul, Korea: heuikim@skku.edu. Yuan Li is a doctoral student at University of

Southern California: yuan.li.2019@marshall.usc.edu. Yao Lu is Associate Professor of Finance at Tsinghua

University School of Economics and Management, Beijing, China: luyao@sem.tsinghua.edu.cn. Xinzheng Shi

is Associate Professor of Economics at Tsinghua University School of Economics and Management, Beijing,

China: shixzh@sem.tsinghua.edu.cn. We are grateful to Ben Iverson, Francine Lafontaine, Binying Liu, John

McConnell, Jagadeesh Sivadasan, Stefan Zeume, and participants at 2017 Red Rock and HKUST conferences

and seminars at University of Michigan, SUNY/Buffalo, University of Georgia, UNLV, and Korea University

for helpful suggestions, and Zhang Peng and Yeqing Zhang for excellent research assistance. This project

received generous financial support from Mitsui Life Financial Research Center at the University of Michigan.

Yao Lu acknowledges support from Project 71722001 of National Natural Science Foundation of China.

Xinzheng Shi acknowledges support from Project 71673155 of National Natural Science Foundation of China.

2

1. INTRODUCTION

News stories abound about automation, robots, and artificial intelligence replacing workers. Under the

catchy title “Will robots displace humans as motorized vehicles ousted horses?” The Economist (April

1, 2017) cites evidence from Acemoglu and Restrepo (2017) and warns that robots might replace

humans and depress wages.1 Adopting and advancing technology, whether it involves robots, AI, or

other automation technologies, requires capital, often for large investments with uncertain outcomes.

When such investments require external funding, firms typically access capital markets. So if

technology affects employment, what role does capital markets play in that process?

To investigate the issue, we estimate the impacts that capital infusion through equity offerings

has on firm-level investments in technology, employee skill composition, employment, and wages.

Capital-skill complementarity predicts that technology embodied in capital complements non-routine

abstract tasks and substitutes for routine tasks.2 Because non-routine abstract tasks tend to require

higher skills than routine tasks, the complementary and substitution effects may add high-skilled

workers and displace low-skilled workers, increasing the relative proportion of high-skilled

workers—a higher skill-composition of employees. If these predictions prevail, what happens to

employment and wages at the firm level? Do displaced unskilled workers outnumber newly added

skilled employees? A higher skill composition implies a higher firm average wage due to the skill

premium (Card, 1999). However, if the higher average wage applies to a smaller work force, it is not

clear whether the total wage will increase or decrease.

Equity offerings are similar to other means to raise equity, such as family or venture financing;

they not only raise new equity but also help raise incremental debt by increasing the firm’s equity

base. The finance literature offers numerous studies on how equity offerings affect shareholder value,

1 A Wall Street Journal article, “Firms leave the bean counting to the robots,” (October 23, 2007) also describes

how robots are replacing workers but more even-handedly. 2 See Griliches (1969); Hamermesh (1993); Fallon and Layard (1975); Berman, Bound, and Griliches (1994);

Goldin and Katz (1996, 1998); Doms, Dunne, and Troske (1997); Autor, Katz and Krueger (1998); Machin and

Van Reenen (1998); Krusell, et al. (2000); Caroli and Van Reenen (2001); Bresnahan, Brynjolfsson, and Hitt

(2002); Autor, Levy, and Murnane (2003); Duffy, Papageorgiou, and Perez-Sebastian (2004); Lindquist (2004);

Acemoglu and Finkelstein (2008); Yasar and Paul (2008); Ben-Gad (2008); Lewis (2011); Parro (2013); and

Akerman, Gaarder and Mogstad (2015); Kasahara, Liang, and Rodrigue (2016); Hershbein and Kahn (2016);

Acemoglu and Restrepo (2016, 2017); and Autor and Salomons (2017).

3

financial policies, investments, and agency costs, but mostly from the shareholder perspective.3 This

focus on capital providers omits an important stakeholder—employees, who also can be affected by

equity offerings. Equity offerings can be either seasoned equity offerings (SEOs) or initial public

offerings (IPOs), depending on whether a publicly listed or a private firm makes the offering. We do

not consider IPOs because of the confounding effects associated with private firms becoming public

firms (Bernstein, 2015). SEOs are devoid of the confounding effects since listed firms make SEOs.

Studying how SEOs affect employment and skill composition is challenging because SEOs

are endogenous and data on employee skills are scarce. We address endogeneity issues by

constructing an instrument using the 2006 and 2008 regulatory shocks on the eligibility to issue public

SEOs in China, both of which are exogenous to individual firm decisions. The Chinese experiments

also help solve the data problem. The China Securities Regulatory Commission (the CSRC,

equivalent to the SEC in the U.S.) requires publicly listed firms to disclose yearly the composition of

the workforce by occupation and education in company filings. Since occupation and education are

related to skill, the data allow us to infer how SEOs affect the workforce skill composition.

Additionally, Chinese accounting rules require publicly listed firms to disclose payroll information in

financial statements, providing access to reliable wage data.

Our sample period is 2000 through 2012, spanning the exogenous shocks on the eligibility to

issue SEOs. Because of China’s unique political and economic system, one may be concerned with

the generalizability of results obtained from the data. Appendix 1 reviews the literature on the Chinese

labor market, which suggests that major economic reforms undertaken during the 1980s and 1990s

had transformed the labor market in the 2000s to one resembling those of capitalistic market-oriented

economies. The Chinese stock market also became the second largest in the world during our sample

period in terms of both market cap and total value of shares traded. One limitation of our sample is

that it contains only publicly listed firms. However, publicly listed firms play an important role in the

3 Studies relating equity offerings to shareholder value include Asquith and Mullins (1986); Masulis and Korwar

(1986); Korajczyk, Lucas, and McDonald (1990); Eckbo and Masulis (1995); Bayless and Chaplinsky (1996);

and Eckbo, Masulis, and Norli (2000). Studies relating equity offerings to financial policies include Pagano,

Panetta, and Zingales (1998); DeAngelo, DeAngelo, and Stulz (2010); McLean (2011); and Gustafson and Iliev

(2017). Studies relating equity offerings to corporate investments include Kim and Weisbach (2008) and

Gustafson and Iliev (2017). Studies relating equity offerings to agency costs include Jung, Kim, and Stulz (1996)

and Kim and Purnanandam (2014).

4

Chinese economy; their total outputs in 2010, for example, accounted for 43% of China’s GDP

(Bryson, Forth, and Zhou, 2014).

Our investigation begins by relating SEOs to demand for skills as manifested in job

advertisements by a major online job posting company in China. We find that job advertisements

posted by firms receiving SEO proceeds are significantly more likely to contain words related to (1)

advanced computer skills, (2) basic computer skills, (3) non-routine analytical task skills, and (4) non-

routine interactive task skills. We control for firm-, year-, location-, and job dummies; hence, the

results suggest that capital infusion through SEOs is associated with higher demand for technical and

non-routine task skills even within the same job category. However, the results are about correlations

between endogenous variables without exogenous variations.

To identify causal relations we construct an instrument using the shocks on the eligibility to

issue SEOs. Treatments are based on payout ratios during the most-recent past three years, which

makes it difficult for affected firms (those that became ineligible to issue SEOs) to circumvent the

regulations. Although the shocks limit the ability to issue public SEOs, they do not directly affect how

firms use SEO proceeds for investments and employment. However, eligibility is based on past

dividend payouts and thus treated and untreated firms may be different. To help meet the exclusion

restriction that the instrument is uncorrelated with the error term in the second-stage regression, we

control for determinants of dividend payouts as proposed in prior literature. We also check whether

our results are driven by differences in past dividend payouts. They are not. We test whether outcome

variables of treated and untreated firms are different prior to the first shock and find no difference. We

examine whether firms circumvented the 2006 and 2008 regulations through dividend manipulations

prior to the announcement of the regulations. Such maneuvers, if any, are likely to manifest as a jump

in payout ratios just above the thresholds required by the regulations. Using the McCrary (2008) test,

we find no such discontinuity. In addition, we conduct a battery of robustness tests to virtually all

conceivable alternative ways to construct the instrument. Results are robust.

Using the IV approach, we find that SEOs lead to significantly higher fractions of technicians

and R&D employees, sales and marketing forces, and employees with four-year university Bachelor’s

and post-graduate degrees. The technician category includes engineers and IT staff, who tend to

5

possess technical skills for non-routine tasks. R&D employees include scientists, researchers, and

designers, who work on creative tasks, and employees working on developing new products, non-

routine abstract tasks. Sales and marketing forces tend to perform non-routine interactive tasks. In

sharp contrast, the proportions of production workers and support staff drop significantly. Production

workers consist mostly of blue-collar workers performing assembly line work and other routine

physical tasks requiring low skills and less education. The staff category includes office and non-

office support staff, most of whom perform routine clerical tasks or non-routine manual tasks, which

tend to require a lower education.

SEOs lead to a reduction by about seven percent in firm-level employment. By occupation,

SEOs lead to a 10% increase in technicians and R&D employees, a 13% increase in the sales and

marketing force, a 24% reduction in production workers, and a 44% reduction in support staff. The

displacement effect of the capital infusion on low-skilled workers dominates the positive effect on

high-skilled workers, resulting in a net loss of employment.4

Are these changes the results of SEOs increasing investments in technology? Kim and

Weisbach (2008) and Gustafson and Iliev (2017) report that SEOs increase total capital expenditures,

which include scale-expanding and structural investments unrelated to technology. To provide

evidence focused on technology, we estimate the effects of SEOs on purchases of new machinery and

equipment. We also estimate the effects on an outcome of technology-advancing investments,

innovations as measured by patents. On average, SEOs increase investments in machines and

equipment by 19% and the number of patents by 13%. Moreover, the increase in patents is significant

only for the patent categories considered more technologically innovative.

The higher skill composition of employees should increase average wages because higher

skilled and more educated employees are paid more (Card, 1999; for the Chinese evidence, see Zhang

et al. (2005) and Appendix 4). We are not surprised to find that SEOs lead to significantly higher

average wages. For all non-executive employees, SEOs increase the average wage by 10%. However,

4 The decline in employment at the firm level does not imply lower employment at the economy-wide level

because technology creates new complex tasks, creating new jobs. Autor and Salomons (2017) report that as

aggregate productivity rises, employment at the country level, especially in the tertiary sector, tends to grow,

suggesting that advances in machine capabilities may not reduce total employment.

6

executives’ average wages are unaffected by SEOs, suggesting the effects on employee skill

composition are limited to non-executive employees.

The higher average wages do not lead to higher total wages, which represent the bulk of labor

costs. Average wages increase because of the higher skill composition, but higher wages apply to a

smaller number of employees.

This paper contributes to the emerging labor and finance literature by demonstrating how

stock markets affect labor markets. Our evidence implies that capital infusion through SEOs increases

investments in technology, upgrades employee skill composition, and decreases firm-level

employment. Although average wages increase because of higher skill composition, total wages do

not increase because of the decline in the workforce. To our knowledge, this is the first study on

equity issuance to document these important phenomena.

In addition, we add to the debate on how financial leverage affects wages. Bronars and Deer

(1991); Perotti and Spier (1993); and Michaels, Page, and Whited (2016) argue that a decrease in

financial leverage increases wages by weakening employers’ bargaining position with employees. On

the contrary, Berk, Stanton, and Zechner (2010) and Chemmanur, Cheng, and Zhang (2013) argue a

decrease in financial leverage decreases wages by reducing ex-ante employment risk at the time of

wage negotiation. The data used in these studies are average wages, which depend on the skill

composition of the workforce. Firms with different leverage may have workforces with different skill

compositions; for example, firms with low financial leverage may be less capital constrained,

allowing more investments in technology and human capital, leading to a higher skill composition and

a higher average wage. Thus, the negative relation between leverage and average wages documented

in Michaels et al. (2016) could be due to differences in skill composition of the workforce. Until these

differences are properly accounted for, whether and how leverage affects wages remain open

questions.

This paper builds on the literature on capital-technology-skill complementarity, which offers

a large body of important studies. Identification is difficult because capital infusion, investment in

technology, and skill composition are all firm choices. Three recent studies examine the

complementarity hypothesis with cleaner identification using exogenous shocks. Lewis (2011)

7

employs variation in the growth of the relative supply of low-skilled workers stemming from an

immigration wave, and new immigrants’ tendency to cluster geographically, to identify the

technology and skilled worker complementary relation. Akerman, Gaarder, and Mogstad (2015)

exploit variation in the availability of broadband internet across municipalities in Norway to

demonstrate skill complementarity of broadband internet. These findings are about technology-skill

complementary, and do not directly test the relation with raw capital. Our study provides evidence

directly linking raw capital to technology advances, skill composition, and a net loss in firm-level

employment. These results are similar to those in Acemoglu and Finkelstein (2008), who find that a

decline in the relative price of capital for hospitals leads to higher adoption of new health care

technology, decreases labor input, and upgrades the skill composition of hospital nurses. We add to

their contribution by expanding the scope and generalizability: Capital-technology-skill

complementarity applies well beyond the hospital sector. Less friction in accessing stock markets for

external financing increases investment in technology, innovation, and employee skill composition

and decreases labor input for many other categories of occupations in a wide range of sectors.

The next section reviews the relevant literature to develop empirical predictions. Section 3

analyzes online job posting data. Section 4 provides the main results. Section 5 contains a battery of

robustness tests. Section 6 discusses implications.

2. LITERATURE REVIEW AND PREDICTIONS

In this section, we review the finance and labor economics literature relevant to developing

predictions on the effects that SEOs have on technology investments, skill composition, employment,

and wages. We focus on equity financing because of its close association with innovations. R&D

investments, for example, provide firm-specific intangible knowledge assets. These assets are not

readily re-deployable by other users, making them less attractive as collateral in issuing debt and thus

making equity financing a more viable alternative (Hall and Lerner, 2010). Equity financing also

helps raise incremental debt to adopt new technology via purchasing machines and equipment.

From the various means to raise equity, we choose seasoned equity offerings (SEOs) because

of data availability and the shocks on the eligibility to issue SEOs for identification. We do not

include initial public equity offerings (IPOs) to avoid the confounding effects of private firms going

8

public. Bernstein (2015) argues that IPOs change managerial incentives and/or managerial myopia

stemming from short-term performance pressure from the stock market. Such changes affect corporate

governance, which, in turn, affects employment policies and investment decisions (Bertrand and

Mullainathan, 2003; Atanassov and Kim, 2009). SEOs are devoid of such confounding effects

because already publicly listed firms issue them.5

To predict the effects of SEOs on skill composition and employment, we invoke capital skill

complementarity. Griliches (1969) is the first to document the complementary relation. Goldin and

Katz (1996, 1998) argue whether capital and skilled workers are relative complements or substitutes

depends on the nature of technological change. They consider two distinct stages of manufacturing: (1)

a machine-installation and machine-maintenance segment, and (2) a production or assembly portion.

Capital can be used to acquire machines and assets, and the first stage requires skilled labor. In the

second stage, unskilled labor uses the workable capital created by skilled labor plus raw capital in the

production of the final product or assembly segment of manufacturing. China’s manufacturing and

production processes during our sample period can be characterized by the batch and continuous-

process method of production described in Goldin and Katz (1998). This is important because they

show that capital and skill became relative complements when production processes in the U.S.

shifted from factories to continuous-process and batch methods early in the twentieth century, but not

during the more distant period when the artisanal shop transited to the factory.

Acemoglu and Finkelstein (2008) argue that technology is always embodied in capital, and

that it requires a capital outlay. They develop a model and provide evidence of how a shock lowering

the relative price of capital in the U.S. health care sector increases new technology adoption by

hospitals, enhances the skill composition of nurses, and decreases total labor input. The capital skill

complementarity is also evident in cross-country data at both the aggregate and the sectorial levels

(Fallon and Layard, 1975; Duffy, Pagageorigiou, and Perez-Sebastian, 2004; and Parro, 2013).

Capital infusion through SEOs relaxes cash constraints (DeAngelo et al., 2010) and allows

firms to make more technology investments, which, in turn, leads to higher demand for skilled

5 Agrawal and Tambe (forthcoming) report changes in employees’ skill set when firms go private via leverage

buyouts.

9

workers. Our first prediction, therefore, is that SEOs will increase demand for skills.

Testing this prediction requires proxies for skills, which Acemoglu and Autor (2011: p. 1045)

define as “a worker’s endowment of capabilities for performing various tasks,” where a task is

defined as “a unit of work activity that produces output.” Tasks can be classified into three broad

categories: abstract, routine, and manual (Autor, Katz, and Kearney, 2006; Autor and Dorn, 2013).6

Abstract tasks, such as research and legal writing, are referred to as problem solving, creative, and

organizational tasks, which tend to require high skills. Routine tasks, such as picking/sorting,

repetitive assembling, and record keeping, are codifiable manual and cognitive tasks following

explicit procedures, which tend to require low skills. Non-routine manual tasks, such as janitorial

service and driving, are tasks requiring physical adaptability, which also tend to require low skills

(Autor and Handel, 2013). Autor, Levy, and Murnane (2003) provide the conceptual framework and

evidence that technological advances replace routine tasks and therefore substitute for workers in

performing routine tasks, but complement workers in performing non-routine abstract tasks.

Accordingly, we predict that SEOs increase investments in technology, and technology advances in

turn reduce labor input for routine tasks and increase labor input for non-routine abstract tasks.

To test this prediction, we use occupations to proxy for routine vs. non-routine abstract tasks.

Each occupation category may comprise multiple tasks at different levels of intensity, but the

variation is greater across occupations than within an occupation (Autor and Handel, 2013). The

intensity of routine tasks is greater in occupations such as production workers, assemblers, and

support staff than occupations such as engineers, R&D staff, and sales and marketing forces. Thus, we

predict SEOs lead to a reduction in the relative proportion of workers in the routine task-intensive

occupations and an increase in the non-routine abstract task-intensive occupations.

Another useful proxy for skills is the level of education. Hence, we predict that SEOs increase

the relative proportion of employees with Bachelors’ degrees from four-year universities and post-

graduate degrees.

When technological improvement embodied in capital substitutes for low-skilled tasks and

6 Autor, Levy, and Murnane (2003) list five task categories: non-routine analytic tasks, non-routine interactive

tasks, cognitive tasks, manual tasks, and non-routine manual tasks. Studies that are more recent collapse these

five measures to the three task aggregates.

10

complements high-skilled tasks, the effects on employment may not be one-to-one. If displaced low-

skill employees outnumber newly added high-skill employees, SEOs will decrease the number of

employees at the firm.

The predicted higher skill-composition of employees implies a higher average wage. Skill

premiums, namely, skilled employees being paid more than unskilled workers, are widely

documented during the past two decades (Card, 1999) and holds for China (Zhang et al., 2005).

However, the impact on total wages will depend on how SEOs affect the total number of employees.

If there are fewer employees, higher average wages will apply to a smaller number of employees.

Hence, it is not clear how SEOs will affect total wages.

3. DEMAND FOR SKILLS AS MANIFESTED IN ONLINE JOB ADVERTISEMENTS

In this section, we estimate correlations between SEOs and demand for technical and non-

routine task skills using job advertisements posted online. The data is from a major job posting

company in China, Lagou.com (https://www.lagou.com). It started the job posting business in 2013,

so our data covers only 2014 through 2016. Our sample contains 45,585 unique full-time job

advertisements posted by 790 A-share firms listed on Shanghai and Shenzhen Stock Exchanges.7 We

exclude repetition of the same advertisements some firms re-post to attract more attention. Table 1,

Panel A shows the number of new job postings by year. Of the 45,585 job advertisements, 7,791 were

posted by firms when they received proceeds from publicly or privately placed SEOs.

We follow an approach similar to that of Hershbein and Kahn (2016) to construct skill

variables. For each job advertisement, we machine-search for the keywords indicating four types of

skills: (1) advanced computer skills, (2) basic computer skills, (3) non-routine analytical task skills,

and (4) non-routine interactive task skills. Table 1, Panel B lists the English translation of Chinese

keywords used to identify each skill.

7 Stock markets in mainland China offer two types of stocks: A- and B-shares. Originally, the A-share market

was for domestic investors to trade with RMB; the B-share market was for foreign investors to trade with U.S.

dollars. The B-share market was opened to domestic investors in 2001, and qualified foreign institutional

investors were allowed to trade in the A-share market beginning in 2006. A firm can issue both A-shares and B-

shares, and both shares have identical rights. We restrict our sample to the A-share market because the total

market capitalization of the A-share market is about 122 times that of the B-share market as of the end of 2013.

In addition, most firms listed in the B-share market are also listed in the A-share market.

11

All estimations are at the job advertisement level, relating skills mentioned in each job

posting to whether the posting occurred during the year in which a company receives SEO proceeds.8

The dependent variable is either an indicator for the presence of a keyword indicating a specific skill

or the log of one plus the number of key words associated with each skill type to capture the intensity

of the skill requirement. The variable of interest is the SEO indicator, JP_SEO, turned on only in the

year SEO proceeds are received.9 All regressions control for year- and firm dummies to control for

heterogeneity in demand for skills and jobs across time and firm. We also control for location

dummies at the county level because many firms operate in multiple locations and job skill

requirements may vary across location (for example, R&D centers requiring advanced computer and

non-routine analytical skills tend to be located in metropolitan areas, while sales offices tend to be

located in both countryside and metropolitan areas.)

Table 2, Panel A reports results relating advanced computer skills to the SEO indicator.

Columns (1) and (3) show positive and significant coefficients, suggesting that firms receiving SEO

proceeds are more likely to require advanced computer skills.10 Hershbein and Kahn (2016) point out

online job postings tend to target white-collar employees more than blue-collar workers. To control

for job-related omitted variables, we add job dummies in Columns (2) and (4) using job titles

mentioned in the postings. Reestimation results continue to show significant positive coefficients on

the SEO indicator, suggesting that when firms receive SEO proceeds, they are more likely to demand

advanced computer skills for the same type of jobs. Panel B repeats the same exercise on basic

computer skills. Coefficients on the SEO indicator remain positive and significant, indicating the

probability of specifying basic computer skills is higher when firms receive SEO proceeds.

In Table 3, we relate the SEO indicator to non-routine analytical and interactive task skills.

Again, the coefficients are all positive and six of eight are significant. In sum, when firms obtain new

capital through SEOs, they seem to have greater demand for technical and non-routine task skills.

8 We cannot conduct firm level analyses because firms may advertise job openings with other job posting

companies and/or through other recruiting channels. 9 We turn on the indicator only in the year a firm receives SEO proceeds because if a firm fills newly advertised

positions in the year of the posting, the advertisement is unlikely to appear in the following year unless the

newly hired employees leave the firm and because the sample period covers only three years. 10 We lose three observations for OLS regressions because location information is unavailable.

12

These results, though informative, do not establish a causal relation, because they are about

association between two endogenous variables. Furthermore, our analysis is confined to job

advertisement level variation, which indicates only whether given a job posting the probability of

requiring specific skills changes. Because the number of job postings at the firm level may also

change, a more complete analysis requires a firm level analysis.

4. CAUSAL EFFECT ANALYSES WITH PANEL DATA

In this section, we conduct firm-level analyses using exogenous shocks on the eligibility to

issue SEOs to identify the causal effects that SEOs have on investment in technology, innovation,

skill composition, firm-level employment, and firm wages.

4.1. Regulatory Changes on the Eligibility to Issue SEOs

On May 6, 2006, the CSRC issued Decree No.30 requiring that to conduct a public SEO, a

firm’s cumulative distributed profits in cash or stocks during the most recent past three years must be

no less than 20% of the average annual distributable profits realized over the same period. Prior to this

regulation, the eligibility requirement was a positive dividend payment during the past three years.

The CSRC further tightened the requirement on October 9, 2008, when it issued Decree No.57, which

raises the threshold to 30%, counting only cash payments as distributed profits.

These shocks are exogenous to individual firm decisions. Although they limit the ability to

issue public SEOs, they do not directly affect how firms use SEO proceeds for investments and

employment. The catalyst for the 2006 regulation was the Split Share Structure Reform of 2005,

which made non-tradable controlling shares tradable in stock markets beginning 2005. The reform led

to a large increase in the supply of tradeable shares, which the CSRC deemed had an adverse effect on

stock price. The intent of the 2006 regulation was to limit the supply of newly issued shares. When

the Split Share Structure Reform began in April 2005, the CSRC suspended all public equity offerings.

Since the suspension was likely to end eventually, some market participants may have anticipated

some form of regulation on SEOs. To check the extent of public knowledge about the specifics or the

timing of the regulation before the announcement on May 6, 2006, we search for news stories in

Chinese media and find some news reports in early February 2006 that a new regulation would be

13

announced soon.11 The reports turned out to be wrong, however, as the CSRC denied it on February 9,

2006, and did not announce the regulation until three months later.

Two years after the 2006 regulation, the CSRC decided to raise the bar on the eligibility to

issue public SEOs. The intent was to reduce the supply of newly issued shares amid a stock market

crash. The Shanghai Stock Exchange Composite Index reached its peak on October 16, 2007, and

then fell precipitously, dropping more than 50% by June 2008. The CSRC issued a draft of the 2008

regulation on August 22, 2008, followed by an official announcement on October 9, 2008.

4.2. Empirical Design

We construct an instrument using the regulatory shocks. We use the IV approach, instead of a

difference-in-differences (DID) approach, because it provides direct estimates of the impacts of SEOs

on outcome variables, while the DID approach provides estimates of the impacts of the policy

changes.12 The 20% and 30% thresholds in 2006 and 2008 shocks raise the possibility of a regression

discontinuity (RD) design. However, observations in the neighborhood around the thresholds are too

few to conduct meaningful RD analyses.13

4.2.1. Construction of the Instrumental Variable

We construct the instrument for “SEO years,” the period when SEOs are most likely to affect

the outcome variables. SEOs can occur at different points in a year (e.g., February vs. November), and

it takes time for deployment of SEO proceeds to affect outcome variables, especially employment and

employee skill composition. We define SEO years as the year of receiving SEO proceeds and two

years afterward. The SEO process itself also takes time. In our sample, the average time from the

initial announcement of an SEO to the receipt of the proceeds is 337 calendar days. We allow for an

11 Chinese news coverage can be found in http://finance.people.com.cn/GB/1041/4090899.html. 12 Let y = α + β ∗ SEO + ε, where β captures effects of SEOs. We construct an IV from a regulatory shock, and

the relation between SEO and IV is SEO = γ + δ ∗ IV + v . The DID approach estimates y = α + β ∗�γ + δ ∗ IV + v� + ε = α + β ∗ γ + β ∗ δ ∗ IV + β ∗ v + ε . That is, the coefficient we get from the DID

approach is β ∗ δ, not β that we hope to estimate using the IV approach. 13 For the 2006 regulation cutoff, there are no eligible firms conducting SEOs and four ineligible firms not

conducting SEOs in the neighborhood of [19%, 21%]. For wider neighborhoods of [17%, 23%] and [15%, 25%],

there are one and four eligible firms conducting SEOs and 7 and 11 ineligible firms not conducting SEOs,

respectively. For the 2008 regulation cutoff, for the neighborhoods of [29%, 31%], [27%, 33%], and [25%,

35%], the number of eligible firms conducting SEOs is 2, 11, and 22; the number of ineligible firms not

conducting SEOs is 7, 13, and 22. For the neighborhood containing the most observations ([25%, 35%]), the

calculated power of the RD strategy for the estimated effect of SEO on the proportion of production workers by

the IV strategy in the paper (i.e., the coefficient of SEO" in Table 7, Panel A, Column 1) is only 0.052, lower than

the conventional threshold 0.8. Stata code "rdpower" is used for this calculation.

14

extra year because the shocks occurred in May and October of 2006 and 2008, respectively. Thus, if a

firm’s payout ratio over 2003 – 2005 is less than 20%, the firm is treated by the 2006 regulation, and

the instrument, SEOIneligible, is equal to one during 2008 – 2010, the SEO years. The CSRC

specifies the formula to calculate the payout ratio as (Dt-1 + Dt-2 + Dt-3) / [(It-1 + It-2+ It-3) / 3], where D

is the amount of dividends paid and I is the amount of distributable profits.

(http://www.csrc.gov.cn/zjhpublic/zjh/200804/t20080418_14487.htm.) 14 The distributable profit is

measured by net income (the parent’s net income for consolidated financial statements). 15

(http://www.csrc.gov.cn/pub/newsite/gszqjgb/fwzn/201603/t20160329_294910.html). For firms listed

for less than three years, the payout ratio is calculated for the years it has been listed (see the CSRC

internal publication, BaoJianYeWuTongXun (Investment Banking Practice Letters) 2, 2010, p.24).

We also turn on the instrument in 2009, 2010, and 2011 for firms treated by the 2006

regulation in 2007—firms with the payout ratio less than 20% over 2004 – 2006, because it may be

difficult to circumvent the regulation in 2007 by increasing dividends in 2006 alone. The results are

robust to not turning on the instrument for firms affected by the 2006 regulation in 2007 (see Section

5.3.) We assume firms are unaffected by the 2006 regulation in 2008 because firms could have

circumvented the regulation by increasing dividends in 2006 and 2007. We follow the same procedure

for firms treated by the 2008 regulation. SEOIneligible is equal to one in 2010, 2011, and 2012 for

firms with the payout ratio less than 30% over 2005 – 2007, and in 2011 and 2012 for firms with the

payout ratio less than 30% over 2006 – 2008. When a single observation satisfies these conditions

multiple times, it is equal to one only once. Appendix 2 illustrates how the instrument is constructed.

4.2.2. Validity of the Instrument

Because we construct the instrument using past payout ratios, treated and untreated firms may

differ to the extent that dividend payouts reflect firm characteristics. To help meet the exclusion

restriction that the instrument is uncorrelated with the error term in the second-stage regression, we

control for the determinants of dividend payouts offered by the dividend literature (described in the

14 Stock dividends are included in computing the payout ratios for the 2006 regulation but excluded for the 2008

regulation. 15 Some firms show negative payout ratios because the average annual distributable profit over the past three

years is negative. These firms are unaffected by the regulation because the CSRC does not approve public SEOs

for firms with negative profits.

15

next section). Another dividend-related issue is that dividends may reduce misuse of free cash flows

(Jensen, 1986), influencing outcome variables of interest.16 However, the instrument is based on past

dividend payouts, not current dividends. Dividend payouts could be serially correlated due to

persistency in corporate financial policies (Lemmon, Roberts, and Zender, 2008), but firm fixed

effects help control for the persistency. As a precautionary measure, however, we include the current

dividend payout ratio, DIV_PR, as a control variable.

One presumption for the validity of the IV is that if there were no shock, affected and

unaffected firms would have no different time trends in the outcome variables. We examine this issue

in Section 5.1 and find no different pre-trends in outcome variables between treated firms and control

firms prior to the first shock in 2006.

Another source of violation of the exclusion condition is some firms circumventing

regulations by increasing payout ratios prior to the shocks. That is, the instrument could correlate with

the error term in the second-stage regression because firms in greater need of capital for investments

in technology are more likely to manipulate the payout ratios. However, circumventing the regulations

is difficult because otherwise low payout-firms have to anticipate the regulatory changes and increase

payouts to meet the thresholds prior to the regulations. Anticipation is subject to uncertainty, which

makes the benefits from dividend maneuvers uncertain, reducing the present value of the benefits. The

uncertainty is not only about future regulations; there is also the approval uncertainty. SEOs in China

and the size of an SEO require the CSRC’s approval, which adds uncertainty over whether and how

much an SEO can raise capital. The cost of maneuvering dividends in anticipation of the 2008

regulation is likely to be economically significant because it requires paying higher cash dividends,

then grossing up the size of the SEO to make up for the cash used to pay the higher dividends prior to

the SEO. Such maneuvers are costly. Firms wishing to issue SEOs tend to be cash constrained

(DeAngelo et al., 2010). Paying out extra cash dividends may lead to foregoing value-enhancing

16 The regulators tied the eligibility to dividend payouts because the CSRC believed firms paying out more free

cash flows are less likely to waste them and thereby better serve investors. See the press conference on the 2006

regulation (http://www.csrc.gov.cn/pub/newsite/hdjl/zxft/lsonlyft/200710/t20071021_95210.html). For more

details, see Regulation for Issuing Stocks, 2006, China’s Securities Regulatory Commission.

16

investments. If the firm takes on more borrowing to meet the cash needs, financial leverage will

exceed the optimal level.

Costs of maneuvering dividend payouts in anticipation of the 2006 regulation is likely to be

lower because it counts stock dividends towards meeting the dividend requirement. If low payout-

firms anticipated this aspect of the forthcoming regulation, they could have satisfied the dividend

requirement by issuing sufficient stock dividends during 2003 - 2005. Data show otherwise. Stock

dividends were relatively rare in China during that period. Among 600 dividend cases in 2005, for

example, only 41 included stock dividends. Over the 2003-2005 period, 94% of all the dividend cases

did not include any stock dividends.

If, in spite of these considerations, firms somehow manipulated payout ratios prior to the

shocks to meet the eligibility requirements, the past payout ratios are likely to be just above 20% for

2006 and 30% for 2008. They are unlikely to exceed the thresholds by much because the maneuver

forces the firm to payout more than it would have done otherwise. Thus, we use the method proposed

in McCrary (2008) to detect a discontinuity in the past three-year payout ratios at 20% for 2006 and at

30% for 2008. None of the discontinuity estimates is significant.17 Although the McCrary test is only

about the necessary condition, the results support the validity of our instrument. 18

4.2.3. Baseline Specifications

We rely on two specifications throughout the paper to crosscheck robustness of the estimation

results. In the first specification, we control only for year- and firm fixed effects, and four

predetermined variables—firm age and three legal variables. Year-fixed effects control for economy-

wide shocks affecting all firms in the same year, such as a greater supply of university graduates

following a policy change to increase college enrollments in 1999 and the stock market crash

preceding the 2008 regulation, while firm-fixed effects control for time-invariant firm characteristics.

Firm age is proxied by the log of the number of years a firm has been listed, ln(NYEAR_LISTED).

Legal variables include: (1) The minimum wage required in the province or provincial-level city of a

17 We use Stata command “DCdensity,” which automatically chooses bin size and bandwidth. The discontinuity

estimate for 2006 is 0.724, with standard error and P-value of 0.708 and 0.306, respectively. For 2008, the

discontinuity estimate is 0.179, with standard error and P-value of 0.274 and 0.514. 18 Some firms treated by the shocks may raise funds through means other than an SEO, which will bias the

results downward.

17

firm’s headquarters location, ln(MIN_WAGE).19 Minimum wages may affect the skill composition of

employees by imposing a lower limit on what firms can pay unskilled workers. (2) The 2008 Labor

Law of People’s Republic of China on employment and wages. The law is likely to have greater

effects on firms with greater labor intensity. Because the law became effective on January 1, 2008, we

measure the law’s effect, Labor_Law_Effect, by interaction of the industry average ratio of the total

number of employees to all fixed assets in 2007 with an indicator equal to one for 2008 through 2012.

We use industry classifications as defined by the CSRC. (3) Local legal environment, LAWSCORE. A

higher score indicates the firm is located in a region with more developed legal institutions and

stronger law enforcement.20 We include this variable because the law and finance literature (e.g., La

Porta et al., 1998) suggests firms located in countries with stronger investor protection tend to have

better corporate governance and suffer from fewer agency problems, which may affect firms’

investment decisions and labor policies.

The second specification controls for more time-varying firm characteristics. As noted, we

add the determinants of dividend payouts offered by the dividend literature, which began with the

Miller and Modigliani irrelevancy theorem (1961). After much, sometimes-heated, debate over half a

century, a consensus has emerged, at least on which factors are possible determinants of dividend

payouts. A hypothesis lately receiving much empirical support is a life-cycle theory (DeAngelo and

DeAngelo, 2006): In early years, firms’ investment opportunities exceed internally generated capital,

so they retain more earnings and pay few dividends. In later years, internally generated cash exceeds

investment opportunities and firms pay out the excess funds to prevent misuse of the free cash flows.

Empirically, DeAngelo, DeAngelo, and Stulz (2006) report the propensity to pay dividends among

U.S. firms is higher when retained earnings comprise a larger fraction of total equity. Denis and

Osobov (2008) find the same pattern for five other developed economies. Following these studies, we

use the ratio of retained earnings to total equity, RE/TotalEq, as a determinant of dividend payouts.

19 Provinces and provincial-level cities adjust minimum wages every two or three years. In China, there are four

provincial-level cities: Beijing, Shanghai, Tianjin, and Chongqing. 20 The National Economic Research Institute (NERI) constructs the index for each province or provincial-level

region. The index changes over time, reflecting changes in the number of lawyers as a percentage of the

population, the efficiency of the local courts, and the protection of property rights. For a more in-depth

description, see Wang, Wong, and Xia (2008).

18

Other determinants of dividends offered in the literature include: (1) Size and profitability

(e.g., Denis and Osobov, 2008). We proxy firm size by the log of sales, ln(SALES);21 and profitability

by return on assets, ROA. (2) Dividend tax clienteles (e.g., Eades, Hess, and Kim, 1984; Fenn and

Liang, 2001; Allen and Michaely, 2003; and Crane, Michenaud, and Weston, 2016). In China,

individual investors pay a flat 20% tax on dividends (until 2013) and institutional investors pay no tax

on dividends. Tax clientele is therefore proxied by the fraction of shares held by institutional investors,

Inst_OWN. (3) Dividend signaling (e.g., Bhattachaya, 1979; Allen and Michaely, 2003). Demand for

signaling may be greater for firms with greater information asymmetry. We proxy information

asymmetry by outsiders’ costs of acquiring information as measured by stock return volatility,

Tot_Volatility. (4) Executive share ownership (e.g., Brown, Liang, and Weisbenner, 2007). Since the

rationale for this determinant is control and self-interest, we proxy it by the percentage of shares

owned by the largest shareholder, %_LARGST_SH.

We also control for six types of variables related to the outcome variables: (1) Strength of

corporate governance. Governance may affect investment decisions (Jensen, 1986). It may also

influence labor policies, affecting employment and wages (Bertrand and Mullainathan, 2003;

Atanassov and Kim, 2009; Cronqvist et al., 2009; Kim and Ouimet, 2014). Proxies for corporate

governance include LAWSCORE included in the first specification; %_LARGST_SH mentioned above;

the percentage of independent directors on the board, %_IND_DIR; and the percentage of shares held

by local and/or central government, %_STATE_OWN. State share ownership varies substantially over

time and across firms. (2) Sales growth rate, SALES_GR, because growth opportunities affect

employment (Hanka, 1998). (3) An indicator for private equity offerings, D_PRIVATE_PLACE, and

an indicator for overseas SEOs issued by dual-listed firms, Overseas_SEOs. These indicators help

control for the effects of capital infusion through other equity offerings unaffected by the shocks. (4)

Asset tangibility, as measured by property, plants, and equipment over total assets, PPE/TA. High tech

firms tend to have fewer fixed assets and fewer production workers. (5) Financial leverage, Leverage,

21 We measure firm size by sales instead of the number of employees or total assets, because the number of

employees is one of the outcome variables of main interest and total assets automatically increase when firms

receive SEO proceeds. SEOs might also affect sales, but the impacts take time and the control variable is the

concurrent sales revenue.

19

to partial out the leverage channel through which SEOs may affect wages. SEOs reduce leverage

(Pagano, Panetta, and Zingales, 1998; Eckbo, Masulis, and Norli, 2000; Gustafson and Iliev, 2017),

and as mentioned earlier, a number of studies argue leverage affects wages. (6) Percentage of non-

tradable shares, %_NONTRD_SH, to control for the potential confounding effects of the Split Share

Structure Reform.

4.3. Data and Summary Statistics

4.3.1. Sample Construction and Data Sources

The sample period covers 2000 through 2012 to span the regulatory shocks. Underwritten

offerings in China were first allowed in 2000, and some corporate governance variables, such as board

information, are available only after 2000. As before, we construct the sample using all A-share firms

listed on the Shanghai and Shenzhen Stock Exchanges. We exclude financial firms as defined by the

CSRC (e.g., banks, insurance firms, and brokerage firms), firms with total employment less than 100,

and ST (special treatment) and *ST firms, which have had two (ST) or three (*ST) consecutive years

of negative net profit.

The main data source for labor, financial and corporate governance variables is Resset

(http://www.resset.cn/en/). It is similar to Compustat but unlike Compustat, it provides reliable data

on wages and employment. It also provides firm-level panel data on employee occupation and

education. The CSRC does not require a specific format for reporting employee composition, but all

firms report in company filings the number of employees by occupation or job type, and most firms

report the number of employees by education. Resset collects the information and constructs firm-

level data on the number of employees by occupation and education. It also provides verbal

descriptions of each job type coded from the company filings for each firm-year.

We manually clean the occupation data using the verbal descriptions. Firms vary in how they

define occupations due to differences in the nature of business, operation, and organizational structure.

Consequently, the occupation data in Resset show some inconsistencies between occupation variable

names and verbal descriptions of occupation or job type. We also find some jobs classified as “others”

by Resset can be classified into a specific occupation group using the verbal descriptions.

20

The first occupation category, Production, is production workers. It includes mainly blue-

collar workers performing assembly line work, sorting, moving, and other routine physical tasks.

Most firms report this category quite clearly. Many high-tech and non-manufacturing firms have no

employees in this category.

The second category, Staff, stands for support staff. This category is not as clear-cut as the

production worker category. Some firms report the number of employees with a finer breakdown,

such as office support staff and HR staff, while others aggregate them into one category of staff,

which may include both office staff (receptionists, secretaries, customer service providers, and office

administrators) and non-office staff (employees for warehouse maintenance, security, and logistics

support). Some firms report office and non-office staff separately, while others lump them together.

To make the data comparable across firms, we manually check verbal descriptions for each firm-year

and sum the number of employees in all staff positions. Most employees in this group perform routine

clerical or low-skill tasks. However, some (e.g., drivers and janitors) perform non-routine low-skill

manual tasks and some are in administrative positions within the support group (e.g., HR manager,

logistics supervisor, office managers), which require non-routine interactive skills.

The third category, Tech_R&D, includes technicians and R&D employees. Technicians

consist of engineers and IT staff. R&D employees include scientists, researchers, and designers

working on creative tasks, and employees working on developing new products. We group

technicians and R&D staff into one category because only about 20% of our sample firms have a

separate category for R&D employees.

The fourth and fifth categories, S&M and Finance, are the sales and marketing force and

finance staff. These categories are rather straightforward. The sales and marketing force includes

employees in sales, marketing, advertising, and brand management. Finance staff includes

accountants and finance staff involved in investment and asset management.

The last category, Others, includes those reported as “others” by sample firms and job

categories which cannot be put into one of the above five categories. Job descriptions such as

“operating” are ambiguous, making it difficult to put into a specific category, so they are counted as

Others. Some firms report employees in distinctly different occupations, such as sales and technicians

21

or financial accountants and sales, as one category. Since we cannot separate them, we treat them as

Others. We do the same when some firms report the number of managers. We do not make a separate

category for managers because only about 25% of sample firms report the number of managers, which

cannot mean the rest of sample firms do not have managers.

To separate employees by education, we construct two high-education groups: holders of

post-graduate degrees, Grad, and holders of four-year university Bachelor’s degrees and above, BA.

Grad includes all masters and doctorate degrees (e.g., MS, MA, MBA, EMBA, PhD, MD, and JD).

About 50% of sample firms separately report the number of employees with post-graduate degrees.

Others lump four-year university Bachelor’s degrees and above in one category. When firms report

post-graduate degree holders separately from four-year Bachelor’s degree holders, we combine them

to construct BA. Some firms report degree holders from three-year or lower level colleges together

with four-year university degree holders as one group. We do not include them in BA.

For data on SEOs we rely on CSMAR (http://www.gtarsc.com/), because it provides more

detailed SEO information than Resset. We hand-collect minimum wages from provincial government

webpages. To mitigate outlier problems, we winsorize all financial variables at 1% and 99% level and

replace them with the value at 1% or 99%. We normalize all monetary variables to 2000 RMB.

Table 4 lists the sample distribution by year. The sample contains 17,838 firm-year

observations associated with 2,341 unique firms. In total, our sample contains 557 public SEOs. We

do not include privately placed equity offerings because the regulations did not apply to private

offerings. The table shows a surge of public SEOs when underwritten offerings were first allowed in

2000. The small number of SEOs in 2005 and 2006 are due to the suspension of all public equity

offerings during the Split Share Structure Reform. (The suspension began in April 2005 and ended in

May 2006.) SEO activities recovered in 2007 and increased in 2008, but the 2008 regulation seems to

have succeeded in limiting the supply of newly issued shares; the number of SEOs dropped in 2009

and remained relatively low until the end of the sample period.

4.3.2. Descriptive Statistics

Table 5 provides summary statistics for all key variables. Appendix 3 provides variable

definitions and data sources. The SEO indicator, SEO, shows 9% of firm-year observations are in

22

SEO years. The instrument, SEOIneligible, indicates 16% of observations are treated by the

regulatory shocks. The average fractions of production workers, support staff, technicians and R&D

staff, the sales and marketing force, finance staff, and others are 48%, 9%, 17%, 13%, 3%, and 18%,

respectively. 22 The very high percentage of production workers is due to the dominance of the

manufacturing sector among domestic-listed firms and the exclusion of the financial services sector.

About 20% of employees have Bachelor’s degrees and above, and 3% have post-graduate degrees.

The average number of employees is 4,585. The average past three-year payout ratio, P3_PR, is about

three times the average annual dividend payout ratio, DIV_PR, 23 because the denominator in

calculating P3_PR is the average annual distributable profit over the past three years. (See the formula

in Section 4.2.1.) The average wage for all employees, AWAGE, is slightly lower than the average

wage for all non-executive employees, AWAGE_NonExe, which is calculated over 2001-2012 because

firms did not separately disclose payroll information for executives until 2001.

4.4. Skill Composition and Employment

Our job posting data analyses show a positive association between SEOs and demand for

skills. If SEOs increase demand for skills and the demand is met, SEOs will lead to an increase in the

proportion of skilled employees in the work force. In this section, we estimate how SEOs change

employee occupation- and education-composition and level of employment.

The first-stage is estimated by the firm-level conditional (fixed-effects) logistic regression

because the endogenous variable is an indicator. Under the assumption that the instrument has

predictive power over the endogenous variable, IV estimators using the logit model in the first-stage

are asymptotically efficient; i.e., coefficients of the model can be more precisely estimated

(Wooldridge, 2010, p.939). Standard errors of the first-stage regression are clustered at the firm level,

and those of the second-stage regression are corrected by bootstrapping.

Table 6 reports first-stage results. The coefficients on SEOIneligible are negative and highly

significant for both specifications, indicating that the instrument has strong predictive power over the

endogenous variable. F-statistics are not reported because the first-stage regression is conditional logit,

22 The percentages do not sum to 100% because of missing observations. 23 The minimum DIV_PR is zero because no firm in our sample paid dividends in a year of negative profits.

23

a non-linear estimation. When we estimate the first-stage using the OLS with the full set of control

variables, F-statistic is 15.56.

Table 7 reports second-stage results for occupation- and education-composition. Both

specifications show that SEOs significantly increase the proportions of technicians and R&D

employees, and the sales and marketing force. The fractions of employees with Bachelor’s degrees

and above and with post-graduate degrees also significantly increase when the specification contains

the full set of control variables.24 Technicians and R&D employees tend to possess technical and

analytical skills required for non-routine abstract tasks. Sales and marketing forces tend to possess

communication skills required for non-routine interactive tasks. Employees with Bachelor’s and post-

graduate degrees possess higher skills in general.

In contrast, SEOs significantly decrease the fraction of production workers that mostly perform

routine physical tasks. The fraction also declines significantly for support staff, most of whom

perform either routine clerical tasks requiring low skills or non-routine manual tasks requiring low

skills and less education. From these results, we infer that SEOs increase (decrease) the relative

proportion of skilled (unskilled) employees.

Coefficients on control variables are largely consistent with intuition. The proportion of

technicians and R&D staff is positively related to sales growth rate, accumulated earnings, board

independence, and institutional ownership, but is negatively related to asset tangibility and firm size.

The same variables show mostly opposite signs for the proportion of production workers, which is

negatively related to sales growth rate, firm profitability, industry exposure to the labor law, board

independence, and financial leverage, but is positively related to asset tangibility and firm size. The

fraction of employees with Bachelor’s degrees and above is positively related to the minimum wage,

accumulated earnings, ownership concentration, and financial leverage, but is negatively related to

firm size, asset tangibility, and industry exposure to the labor law.

Table 8 reports second-stage results for the level of employment. Both specifications show

SEOs significantly reduce firm-level employment by about 7 - 8%. The remaining columns break

24 Post-graduate degree holder regressions in Tables 7 and 8 contain substantially fewer observations than other

regressions because many firms do not separately report the number of employees with post-graduate degrees.

24

down the number of employees by occupation and education, where the dependent variable is the log

of one plus the number of employees (some firm-years show no employees in some occupation and

education categories.) Both specifications show that the number of technicians and R&D employees

significantly increases and the number of production workers and support staff significantly decreases.

Estimations with the full set of control variables indicate the number of production workers and

support staff decline by 24% and 44%, whereas technicians and R&D staff, the sales and marketing

force, and post-graduate degree holders increase by 10%, 13%, and 10%, respectively.

In sum, capital infusion through SEOs decreases the proportion of employees performing tasks

requiring low skills and increases the proportion of employees in occupations requiring high-skills,

resulting in a higher skill composition. Displaced low-skilled workers outnumber newly added high-

skilled workers, resulting in a net decline in employment.

4.5. Investments in Technology and Innovations

We argue these changes in the skill composition and employment level are the results of

SEOs leading to more investments in technologies. Investments to adopt new technologies require

purchases of new machinery and equipment, while investments to advance technology require

innovative activities, the outcome of which we proxy by the number of patents.

The dependent variable in the first two columns of Table 9 is acquisition costs of newly added

machines and equipment. These data are available from 2003 when the CSRC first required listed

firms to breakdown acquisition costs of newly added fixed assets by type. The estimated coefficient

with the full set of control variables implies that SEOs lead to a 19% increase in investments in

machines and equipment. The last column confirms previous studies on other countries (e.g., Kim and

Weisbach, 2008; Gustafson and Iliev, 2017): Chinese SEOs also increase capital expenditures.

Coefficients of control variables suggest purchases of machines and equipment are positively

related to sales growth rate, accumulated earnings, privately placed equity offerings, 25 firm size,

tangibility of assets, and financial leverage; and negatively related to firm age, industry exposure to

the labor law, and the strength of legal institutions.

25 Investments in machines and equipment are unrelated to equity offerings in foreign stock exchanges, perhaps

because firms deploy the proceeds for foreign subsidiaries’ overseas operations.

25

Investments in innovative activities are proxied by one of their outcomes; the log of the

number of patents granted in year t + 3. 26 The three-year lag allows time to invest SEO proceeds, for

investment to yield innovations, and for innovations to become patents. Although many patents could

turn out to be useless, the number of newly granted patents reflects the effort and resources devoted to

advance technology. Baiten (http://www.baiten.cn/) is the source of our patent data. When we count

the number of patents, we omit patents withdrawn in later years by the State Intellectual Property

Office (SIPO) (http://www.sipo.gov.cn/), the Chinese government agency in charge of patent

administration. It classifies patents into three types: invention patents, utility model patents, and

design patents. According to the Guidelines for Patent Examination 2010 on the SIPO website,

invention (design) patents are considered most (least) innovative among the three, requiring the

longest (shortest) evaluation period with the longest (shortest) protection period.

The first two columns of Table 10 show SEOs increase the number of newly granted patents.

The estimated coefficient with the full set of control variables implies that, on average, SEOs increase

the total number of patents by 13%. The last three columns show the impact is significant only for

patents considered more innovative, invention and utility model patents, with no increase for the least

innovative type, design patents. It appears a significant portion of SEO proceeds is invested to

advance technology, leading to more innovations.

Control variables reveal interesting correlations. In China, firms seem to generate more

patents as they become more established – older and larger with more accumulated earnings.

Institutional ownership is negatively associated with the number of patents, perhaps because

management is less willing to take the risk inherent in innovative activities when their performance

faces greater scrutiny from institutional investors.

4.6. Firm Wages

The higher skill composition following SEOs should increase firm average wages because

skilled workers are paid more, which is also true in China. The China Urban Household Survey shows

26 Another possible proxy for technology-advancing investments is research and development expenditures.

However, we do not have sufficient data on R&D because the Chinese accounting rule did not require disclosure

of R&D expenditures until 2007.

26

Chinese workers with more education are paid more, and technicians are paid substantially more than

production, staff and service, or agricultural workers (see Appendix 4). Table 11 reports the second-

stage results on firm average wages. Unsurprisingly, average wages for all employees increase

significantly following SEOs. The last two columns separate employees into non-executive

employees and executives. Non-executive employees, whose wages increase by 10%, drive the

increase in average wages. Average wages of executives (classified as such in financial statements)

are unaffected by the capital infusion. New capital infusion improves the skill composition of non-

executive employees, increasing their average wages. However, average executive wages do not

increase, suggesting that the effects that SEOs have on skill composition are limited to non-executive

employees.27

Coefficients on control variables are largely consistent with intuition. Average wages are higher

when the minimum wage is higher, when firms are larger and more profitable with more accumulated

earnings, when state share ownership is greater, when ownership is more concentrated, and when

assets consist of more intangible assets.

How do the changes in skill composition and employment affect total wages? Because the total

number of employees declines, the higher average wage does not necessarily imply a higher total

wage. Table 12 reports second-stage results for total wages. Regardless of which specification is used

and how we stratify employee groups, SEOs have no significant impact on total wages.

5. ROBUSTNESS

In this section, we examine (1) pre-trends prior to the first shock; (2) whether past dividend

payout ratios can explain our results; and (3) whether our results are robust to alternative ways to

construct the instrument and to excluding small SEOs.

27 The executive wage results do not reflect the value of equity incentives, which are an important component of

executive compensation in the U.S. In China, wages constitute most of executive compensation, with executive

stock options playing no, or a minor role in the compensation. Brison et al. (2014) reports, “Fewer than 1% of

top executives were granted options in any given year between 2006 and 2010 and, for these few cases, at the

median they were worth 30% of CEO cash compensation and 21% of non-CEO top executive cash

compensation.” Chinese firms were unable to offer stock options until 2006, when equity incentives were

formally introduced in the form of employee stock options and discounted share purchase programs. Stock

options are granted and vested shortly after shareholder approval. They are exercisable according to a fixed

schedule tied to certain performance targets. Discounted share purchase programs allow stock purchases at a

discount but they cannot be sold until a performance target is achieved. These equity incentives are issued to

both non-executive employees and executives.

27

5.1. Pre-Trends

We construct the instrument based on the variation in the impacts of the shocks on the

eligibility to issue SEOs. One presumption for its validity is that if there were no shock, affected and

unaffected firms would have no different time trends in the outcome variables. To test whether this

presumption holds, we conduct a placebo test using the 2000-2005 samples prior to the regulatory

shock in 2006 to simulate the situation with no shock. We do not use post-2006 shock samples

because of the presence of the second shock in 2008. We construct an indicator for firms affected by

the 2006 regulation, Affected. Then we test whether there is any differnce between the outcome

varables of shock-affected and shock-unaffected firms during the years prior to 2006 using 2000 as

the base year. We define five placebo shock indicators, Year01,…, Year05, which are equal to one for

years 2001 through 2005. We then estimate the baseline regression with the full set of control

variables for all key outcome variables with the interactions of Affected and placebo indicators as

variables of interest.

Table 13 reports coefficients on the interaction terms. None of the coefficients is statistically

significant for any of the outcome variables, implying that the outcome variables of affected and

unaffected firms prior to the 2006 shock are not different.28 These results suggest no different time

trends in the outcome variables between affected and unaffected firms had there been no shock.

5.2. Can Past Payout Ratios Explain our Results?

Firms may pay out more of their earnings when management anticipates positive shocks to

cash flows in the future. As the anticipated positive shocks realize, firms make more investments in

technology, increasing demand for skills, which, in turn, leads to higher skill composition and other

results we document. To investigate whether this scenario can explain our results, we add the most

recent past three-year payout ratio, P3_PR, as an explanatory variable and re-estimate regressions.

Some firm-years show negative P3_PR because of negative average annual distributable profit over

the past three years, in which case we replace a negative P3_PR by one and add a dummy for the

negative ratio, P3_PR_D.

28 Placebo shock indicators for investments in machinery and equipment cover only 2004 and 2005 with 2003 as

the base year because data on purchases of new machines and equipment are available only from 2003.

28

Table 14 reports the second-stage results for key outcome variables with the full set of control

variables. (Appendix 5, Column 1 reports the first-stage result.) The coefficients on the predicted SEO

hardly change. The coefficient on P3_PR is positive for the fraction of production workers, the

opposite to the coefficient on the predicted SEO, and is insignificant for most of the other outcome

variables of interest. The alternative scenario cannot explain our results.

5.3. Alternative Ways to Construct the Instrument

Since the key to our identification is the instrument, we re-estimate the baseline regressions

using virtually all conceivable alternative ways to construct it. First, there was a short-lived regulatory

shock on December 7, 2004, when the CSRC issued a set of regulations aiming to standardize the

corporate governance practice of public firms. Included in the regulations was a provision that if a

listed firm did not pay any cash dividend during the past three years, it would not be allowed to issue

a public SEO. (Prior to this requirement, publicly listed firms were eligible to issue SEOs as long as

profits were positive during the past three years.) This provision soon became void when the CSRC

embarked on the Split Share Structure Reform in April 2005 and suspended all public equity offerings

until May 2006 (at which time the CSRC issued the 2006 regulation on the eligibility to issue SEOs.)

Excluding this short-lived shock may bias our estimates if SEOs approved during December 2004 -

April 2005 have different impacts on outcome variables from those of SEOs in the later years. To

check whether our estimates are subject to such bias, we include the 2004 shock in constructing the

instrument. Specifically, we also set SEOIneligible equal to one in 2006 if a firm did not pay any

dividend during 2001 – 2003, and in 2007 if a firm did not pay any dividend during 2001 – 2003 or

during 2002 – 2004, and in 2008 if a firm did not pay any dividend during 2001 – 2003 or during

2002 – 2004. Second, some firms may circumvent the 2006 and 2008 regulations in 2007 and 2009,

respectively, by increasing dividends in 2006 and 2008. To guard against such possibilities, we turn

on the instrument only for firms affected by the 2006 regulation in 2006 and firms affected by the

2008 regulation in 2008. Third, we shorten the time elapsed from the announcement to the receipt of

SEO proceeds from two years to one year. Fourth, we rely only on the 2006 shock because some

firms may have anticipated the 2008 shock. All second-stage results, reported in Table 15, Columns

(2) – (5), are robust.

29

As a final robustness check, we exclude small SEOs in the bottom decile in the size of the

proceeds. Firms conducting these small SEOs are typically small cap firms with highly volatile

performance. The last column in Table 15 shows robust results. Appendix 5 reports all first-stage

results for tests conducted in this section.

6. SUMMARY AND IMPLICATIONS

This paper studies how capital infusion through SEOs affects investment in technology, skill

composition, firm-level employment, and wages. We begin by analyzing job advertisements posted

online, which suggest demand for computer and non-routine task skills increase when firms receive

SEO proceeds. To identify causal effects, we rely on external shocks that cut off access to public

SEOs as a means to raise external capital. We find capital infusion through SEOs increase

investments in technology, innovations, and the proportion of high-skilled employees relative to low-

skilled workers. Importantly, displaced low-skilled workers outnumber newly hired high-skilled

workers, resulting in a net reduction in employment. The higher-skill composition increases average

wages because higher-skilled workers are paid more, but total wages remain unchanged due to the

reduction in total employment. On average, SEOs enable firms to upgrade the skill composition of

employees without increasing total wages.

These findings shed light on how stock markets affect labor markets by altering demand for

high- vs. low-skilled workers. Easier access to capital may not only increase demand for high-skilled

workers but also stimulate their supply, as the demand for and the supply of skills are endogenous to

each other and dynamically move together. If the supply of high-skilled workers increases in response

to increased demand, it may induce greater development of skill complementary technologies, which

may enhance economic growth.

The highly developed, sophisticated, and global financial markets of recent years have

allowed less costly access to external capital, which we show leads to displacement of low skilled and

less-educated workers. Retraining to upgrade skills to meet changing demands requires financial

resources, time, and effort; thus, many low-skilled workers may not be able to leave the shrinking

market for their services, at least not in the short run. The ensuing imbalance between the supply of

and demand for low-skilled and less-educated workers is likely to keep their income low. High-skilled

30

and highly-educated employees, on the other hand, will enjoy increasing demand for their services as

frictions to accessing external capital decline and capital skill complementarity kicks in. The result

might be further widening income inequality in the short-run.

In the end, however, positive spillovers of technology advances to the tertiary sector might

more than offset the negative employment effect on low-skilled workers (Autor and Salomons, 2017).

If and when enough low-skilled workers are properly retrained to perform tasks for tertiary services,

the aggregate employment opportunities might grow as capital markets facilitate development of

complementary technologies and processes, which are necessary to harness the recent technological

advances to yield their full economic benefits.

31

APPENDICES

Appendix 1: Institutional backgrounds on Chinese labor and capital markets

1. Labor Markets and Economic Reforms

China’s labor market has undergone several major changes. In the early years of Communist

China (1952-1978), the state sector dominated employment in the urban area and management did not

have the authority to hire or fire workers without government approval (Lin, Cai, and Li, 1996). Firms

set wages according to a grid determined by the government; wages hardly reflected differences in

productivity (Cai, Park, and Zhao, 2008).

China embarked on economic reforms in 1978, leading to a new, floating wage system by the

mid-1980s. The reforms allowed an enterprise’s total payroll to reflect its performance in the previous

three years. (Prior to this reform, central and local planners had determined the total payroll for each

enterprise (Yueh, 2004)). At the same time, the State Council formally introduced the concept of labor

contracts, giving management the flexibility to adjust employment in response to market competition

(Meng, 2000). However, the labor contract system gave firms the freedom to hire suitable workers,

but the dismissal of workers remained under the government’s tight control.

In 1992 state-owned enterprises (SOEs) were given more autonomy, enabling them to link the

total payroll more closely to firm performance and set their internal wage structures (Li and Zhao,

2003; Yueh, 2004). More reforms followed in 1994-1995, allowing listed SOEs to set their own

wages and encouraging enterprises to consider skills and productivity in addition to occupation and

rank in determining wages (Yueh, 2004). Some SOEs began to lay off workers, as the labor law

issued in 1994 permitted no-fault dismissal of workers in response to changing economic conditions

(Ho, 2006). A major state-sector restructuring followed, closing down or privatizing more than 80%

of SOEs (Hsieh and Song, 2015). When restructuring-affected employees left SOEs, they faced a

more market-driven re-employment process, and the previously inflexible labor market became one in

which supply and demand affected employment and wages. By the mid-2000s, China’s labor market

had become similar to those of other countries based on capitalism; labor is mobile, and enterprises

32

consider market conditions in making employment decisions and in setting wages (Cai, Park, and

Zhao, 2008).

During our sample period, China had well-established legal provisions on hours of work,

payment of wages, and employment. The standard workweek is 40 hours (eight hours per day, five

days per week). Overtime must be paid for any work exceeding standard working hours and cannot

exceed three hours a day or 36 hours per month (Labor Law Article 41). Wages are paid on a monthly

basis, and may not be delayed without reason (Labor Law Article 50). Employees can be fired in the

middle of two fixed-term contracts (or ten years of employment),29 after which contracts must be

made open-ended. Open-ended contracts can be terminated only for cause (Gallagher et al., 2015).

A consequence of these reforms particularly relevant to our study is the increase in returns to

education. Li, et al. (2012) show that the return to an additional year of schooling increased from 2.3

percent in 1988 to about 9 percent in 2000, and the return to college education increased from 7.4

percent in 1988 to 49.2 percent in 2009. These dramatic increases in returns to education are

attributable to the labor reforms and the fast-growing demand for skills (Zhang et al., 2005).

2. Capital Markets and Chinese SEOs

The modernization of Chinese capital markets began when former Premier Rongji Zhu, who

led China to join the World Trade Organization (WTO), spearheaded a series of reforms during his

tenure as vice premier and premier in 1993 – 2003. The reforms included restructuring of state-owned

enterprises (SOEs) and the banking industry.30 A major theme of the reforms was to modernize capital

markets and corporate governance practices of SOEs. The modernization process sped up in 2001

when China officially joined the WTO. In January 2004, the State Council issued a document,

“Opinions on Promoting the Reform, Opening and Steady Growth of Capital Markets,” which sets the

importance of developing capital markets as a high-priority national strategy.31 In response to the

guiding principles from the State Council, the CSRC has implemented a number of new regulations to

29 Contracts are subject to negotiation after the first term. 30 Economist, March 6th, 2003. http://www.economist.com/node/1623179 31 OECD report: Corporate Governance of Listed Companies in China.

https://www.oecd.org/corporate/ca/corporategovernanceprinciples/48444985.pdf

33

modernize stock markets and improve corporate governance.32 According to the World Bank, the

modernization of stock markets, together with the rapid growth of the Chinese economy, have helped

stock markets in mainland China to become the second largest in the world in both market cap and

total value of shares traded in 2009.33

In China, the stock market has been a more important source of external financing than its

corporate bond market, which has been growing at a much slower pace than the stock market.

Although a regulated bond market for enterprises began in 1996, only very large and stable companies

can issue bonds because of the strict approval process required for issuing bonds. Over the period

2010 through 2012, for example, Chinese listed firms raised 2,147.5 billion RMB through stock

markets (via SEOs and IPOs), while bond markets helped raise only 429.5 billion RMB. Over the

same period, adjusted for differences in stock market capitalization, non-financial Chinese firms

issued SEOs more than three times their U.S. counterparts.34

The Chinese stock market is well suited to study SEOs. The types of SEOs available and the

underwriting procedures in China are similar to those in the U.S. There are three types of SEOs: rights

offerings, underwritten offerings, and private placements to no more than ten qualified investors. As

in the U.S., there are two types of underwriting contracts, best efforts and firm commitments.

In comparison to U.S. SEOs, Chinese SEOs provide a cleaner sample to study how the

proceeds from SEOs affect firm investments, employment, and wages because virtually all Chinese

SEOs are primary shares.35 SEOs in the U.S. often include secondary offerings, sale of shares held by

32 In a Q&A session with media, officials from CSRC explain the background of the regulations in details. See

http://www.china.com.cn/chinese/FI-c/723240.htm (in Chinese). 33http://data.worldbank.org/indicator/CM.MKT.LCAP.CD?end=2016&locations=CN-JP-US-HK-FR-GB-

DE&name_desc=false&page=5&start=2003&view=chart and

https://data.worldbank.org/indicator/CM.MKT.TRAD.CD?end=2016&locations=CN-JP-US-HK-FR-GB-

DE&name_desc=false&page=5&start=2003&view=chart. 34 Over the period 2010 through 2012, the average total Chinese stock market capitalization is 3,949.77 billion

USD and non-financial Chinese listed firms raised 86.09 billion USD through SEOs, 2.18% of total market

capitalization. This is more than three times the ratio for US counterparts. During the same period, the average

market capitalization of US stock market is 17,149.34 billion USD and non-financial US listed firms raised

102.75 billion USD through SEOs, 0.6% of the total market cap. Total stock market capitalization excludes

financial firms. Capital raised through SEOs is taken from SDC Platinum. The market capitalization data are

taken from data on the World Bank website (http://data.worldbank.org/). Capital raised through SEOs includes

only proceeds from primary offerings. 35 There were only three mixed offerings containing secondary offerings of state-owned shares, all of which

occurred in 2001. At that time, the CSRC required that if a firm plans to issue N new shares through an

34

insiders and block holders. Proceeds of secondary offerings do not go to the firm and hence cannot

affect investment and employment decisions. Thus, if one studies the effects of deploying U.S. SEO

proceeds without carefully screening out secondary offerings, the results will contain much noise and

confounding effects.

underwritten offering and has state-owned shares, then the offering must contain 10% of N state-owned shares.

The regulation lasted for only four months, and there have been no mixed offerings since 2001.

35

Appendix 2: Construction of the instrumental variable.

This table illustrates how the instrument, SEOIneligible, is constructed. “Conditions” specify the

past three-year period during which the minimum payout ratio applies to make a firm ineligible

to issue a public SEO. For example, 2003 – 2005 < 20% means that if the payout ratio over 2003

– 2005 is less than 20%, the firm is ineligible to issue a public SEO in 2006. In this table, we

assume it takes two years to complete an SEO. Since SEO years include the year of receiving the

proceeds and two years afterward, we turn on the instrument in 2008, 2009, and 2010 for firms

affected by the 2006 regulation in 2006. We follow the same procedure for firms affected by the

2006 regulation in 2007, and for firms affected by the 2008 regulation in 2008 and 2009.

Year SEOIneligible Conditions

2000 0 NA

2001 0 NA

2002 0 NA

2003 0 NA

2004 0 NA

2005 0 NA

2006 0 NA

2007 0 NA

2008 1 If 2003 – 2005 < 20%

2009 1 If 2004 – 2006 < 20% or 2003 – 2005 < 20%

2010 1 If 2005 – 2007 < 30%, 2004 – 2006 < 20%, or 2003 – 2005 < 20%

2011 1 If 2006 – 2008 < 30%, 2005 – 2007 < 30%, or 2004 – 2006 < 20%

2012 1 If 2006 – 2008 < 30% or 2005 – 2007 < 30%

36

Appendix 3: Variable Definitions and Data Sources.

Variables

Definition

Data

Sources

SEO-related Variables

JP_SEO An indicator equal to one in the year in which a firm

receives SEO (public or private placement) proceeds, and

zero otherwise.

CSMAR

SEO An indicator equal to one in SEO years (the year in

which SEO proceeds are received and two years after),

and zero otherwise. It applies to only public offerings.

CSMAR

SEOIneligible Instrument for SEO years. Appendix 2 illustrates how it

is constructed. Wind

Outcome Variables

Adv_Computer_Dum An indicator for the presence of words indicating

advanced computer skills in a job advertisement. Lagou.com

Ln(Adv_Computer) Log of one plus the number of words indicating

advanced computer skills in a job advertisement. Lagou.com

Basic_Computer_Dum An indicator for the presence of words indicating basic

computer skills in a job advertisement. Lagou.com

Ln(Basic_Computer) Log of one plus the number of words indicating

advanced computer skills in a job advertisement. Lagou.com

Non-routine Analytical

Task Skill_Dum

An indicator for the presence of words indicating non-

routine analytical task skills in a job advertisement. Lagou.com

Ln(Non-routine Analytical

Task Skills)

Log of one plus the number of words indicating non-

routine analytical task skills in a job advertisement. Lagou.com

Non-routine Interactive

Task Skill_Dum

An indicator for the presence of words indicating non-

routine interactive task skills in a job advertisement. Lagou.com

Ln(Non-routine Interactive

Task Skills)

Log of one plus the number of words indicating non-

routine interactive task skills in a job advertisement. Lagou.com

EMP Total number of employees at the firm-level. Unit:100. Resset

Production Number of production workers. Resset

Staff Number of support staff. Resset

Tech_R&D Number of technicians (including engineers and IT staff)

and R&D employees. Resset

S&M Number of employees in sales and marketing. Resset

Finance Number of accounting and finance staff. Resset

Others Number of employees with unidentified occupation. Resset

BA Number of employees with four-year university

Bachelor’s degrees and above. Resset

Grad Number of employees with post-graduate degrees. Resset

AWAGE Average annual cash salary and bonuses to all employees

in 2000 RMB. Unit: 10,000. Resset

AWAGE_NonEXE Average annual cash salary and bonuses to all non-

executive employees in 2000 RMB. Unit: 10,000. Resset

AEXEPAY Average annual cash salary and bonuses to all executives

in 2000 RMB. Unit: 10,000. Resset

Ln(Payroll) Log of total annual cash salary and bonuses to all

employees in 2000 RMB. Unit: 10,000. Resset

Ln(Payroll_NonExe) Log of total annual cash salary and bonuses to all non-

executive employees in 2000 RMB. Unit: 10,000. Resset

Ln(Payroll_Exe) Log of total annual cash salary and bonuses to all

executives in 2000 RMB. Unit: 1,000,000. Resset

37

Outcome Variables

Definition

Data

Sources

Ln(Capx) Log of total capital expenditures in 2000 RMB. Unit:

10,000. Resset

Ln(Inv_Tech_Assets) Log of one plus the cost of newly acquired machines and

equipment in 2000 RMB. Unit: 10,000. CSMAR

Ln(Total_Patent) Log of one plus the total number of patents granted. Baiten

Ln(Invention) Log of one plus the number of invention patents granted. Baiten

Ln(Utility_Model) Log of one plus the number of utility model patents

granted. Baiten

Ln(Design) Log of one plus the number of design patents granted. Baiten

Control Variables

NYEAR_LISTED Number of years a firm has been listed since its IPO. Resset

SALES Total sales in 2000 RMB. Unit: 1,000,000. Resset

Leverage Total liability divided by total assets. Resset

ROA Return on assets: Net income divided by total assets. Resset

PPE/TA Property, plants, and equipment divided by total assets. Resset

SALES_GR Sales growth rate from year t-1 to year t. Resset

%_IND_DIR Percentage of independent directors on the board. Resset

%_STATE_OWN Percentage of shares held by the government. Resset

%_LARGEST_SH Percentage of shares held by the largest shareholder. Resset

%_NONTRD_SH Percentage of non-tradable shares. Resset

Inst_OWN Fraction of shares owned by institutional investors. Resset

Overseas_SEOs

An indicator equal to one if a firm issues an equity

offering in foreign exchanges including the Hong Kong

Stock Exchange.

CSRC

Website

D_PRIVATE_PLACE An indicator equal to one if a firm issues an equity

offering through private placement. CSMAR

RE/TotalEq Retained earnings divided by the book value of equity. CSMAR

DIV_PR Dividend payout ratio, equal to total dividend paid over

net income. Resset

Tot_Volatility Standard deviation of weekly stock returns. CSMAR

MIN_WAGE The minimum monthly wage in the province or

provincial city of the firm’s headquarters location in

2000 Yuan.

Government

Websites

LAWSCORE An index for the strength of legal environment described

in Section 4.2.2. It is updated by the National Economic

Research Institute up to 2009. For years after 2009, we

use the 2009 index.

National

Economic

Research

Institute

Labor_Law_Effect The degree to which the 2008 Labor Law of People’s

Republic of China affects a firm. See Section 4.2.2. CSMAR

Affected An indicator for firms affected by the 2006 regulation. Resset

P3_PR

The payout ratio during the most-recent past three years

as defined by the CSRC. See Section 4.2.1. If it is

negative, we replace it by one.

Resset

P3_PR_D

Indicator equal to one if the payout ratio during the most-

recent past three years as defined by the CSRC is

negative, zero otherwise.

Resset

38

Appendix 4: Average Annual Wages in China by Education and Occupation.

This table reports average annual wage in China by education and occupation. The data is from China Urban Household Survey (2000-2009), which

provides access to nine provinces; Beijing, Liaoning, Zhejiang, Anhui, Hubei, Guangdong, Sichuan, Shaanxi, and Gansu. Annual wage is deflated using

provincial CPI with 2000 as the base year and the unit is Chinese RMB.

Education

Occupation

Year College or

above High School

Middle School or

below Technician

Production

Workers

Staff or Service

Workers

Agricultural

Workers Others

2000 11084.013 8944.776 5139.363

15239.261 9258.860 11053.963 8566.029 7946.278

2001 11976.958 9554.838 5438.288

16852.991 9864.254 11841.001 9827.922 8882.542

2002 15822.367 10409.411 5757.975

18404.414 10912.095 13807.288 9452.208 9661.626

2003 17728.367 11346.542 5975.318

20489.257 12303.120 15216.043 10937.459 11118.318

2004 19451.303 12139.160 6495.877

23086.913 13622.273 16191.782 12360.412 12257.059

2005 21261.428 13013.126 7123.790

25598.902 14743.270 18072.238 15012.060 14361.187

2006 23030.351 14092.422 7931.302

27949.907 16697.195 19682.444 16756.711 15198.924

2007 24665.948 15261.617 8603.666

29624.443 17833.485 21563.516 18206.153 17030.791

2008 27924.529 16415.125 9329.643

32551.162 20094.639 23523.721 19247.500 20093.954

2009 30928.259 18155.407 10323.152

35799.283 22402.561 26124.442 23231.018 20988.433

39

Appendix 5: First-stage Regressions for Robustness Tests.

This table reports first-stage results estimated using conditional logistic regressions at the firm level. Column (1) reports

the first-stage result for Table 14; Columns (2) - (6) for Table 15. Robust standard errors, clustered at the firm level, are

in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

SEO

VARIABLES (1) (2) (3) (4) (5) (6)

SEOIneligible -1.671*** -1.581*** -1.621*** -1.071*** -1.621*** -1.580***

(0.418) (0.372) (0.416) (0.352) (0.416) (0.407)

ln(NYEAR_LISTED) 4.428*** 4.395*** 4.414*** 4.470*** 4.414*** 4.225***

(0.474) (0.467) (0.471) (0.473) (0.471) (0.463)

ln(MIN_WAGE) 1.597** 1.594** 1.581** 1.478** 1.581** 1.472**

(0.692) (0.693) (0.692) (0.684) (0.692) (0.700)

LAWSCORE 0.046 0.040 0.041 0.047 0.041 0.039

(0.071) (0.070) (0.070) (0.069) (0.070) (0.071)

Labor_Law_Effect 0.125 0.118 0.125 0.110 0.125 0.121

(0.086) (0.085) (0.086) (0.082) (0.086) (0.083)

ln(SALES) 1.111*** 1.108*** 1.110*** 1.102*** 1.110*** 1.119***

(0.201) (0.205) (0.200) (0.196) (0.200) (0.193)

RE/TotalEq 0.240 0.454 0.452 0.453 0.452 0.295

(0.428) (0.477) (0.451) (0.457) (0.451) (0.382)

ROA -8.059*** -8.048*** -8.213*** -8.255*** -8.213*** -7.744***

(1.871) (1.928) (1.904) (1.899) (1.904) (1.779)

Inst_OWN 1.763*** 1.767*** 1.768*** 1.729*** 1.768*** 1.693***

(0.503) (0.503) (0.504) (0.501) (0.504) (0.495)

Ln(Tot_Volatility) -0.196 -0.209 -0.202 -0.203 -0.202 -0.182

(0.259) (0.258) (0.258) (0.259) (0.258) (0.256)

%_LARGEST_SH -2.059* -2.066 -1.994 -2.080* -1.994 -1.587

(1.247) (1.272) (1.259) (1.234) (1.259) (1.213)

SALES_GR -0.139 -0.137 -0.145 -0.144 -0.145 -0.179

(0.127) (0.129) (0.127) (0.127) (0.127) (0.123)

DIV_PR 0.082 0.096 0.090 0.083 0.090 0.091

(0.076) (0.076) (0.075) (0.074) (0.075) (0.074)

%_STATE_OWN 0.269 0.290 0.272 0.251 0.272 0.206

(0.534) (0.536) (0.533) (0.528) (0.533) (0.523)

%_IND_DIR -0.679 -0.735 -0.712 -0.735 -0.712 -0.751

(0.651) (0.643) (0.644) (0.641) (0.644) (0.644)

%_NONTRD_SH -1.952*** -1.920*** -1.955*** -1.909*** -1.955*** -1.784***

(0.617) (0.615) (0.615) (0.614) (0.615) (0.604)

D_PRIVATE_PLACE -1.348*** -1.378*** -1.357*** -1.340*** -1.357*** -1.326***

(0.310) (0.314) (0.315) (0.306) (0.315) (0.310)

Overseas_SEOs -13.616*** -13.593*** -13.594*** -14.281*** -13.594*** -13.688***

(0.716) (0.708) (0.708) (0.705) (0.708) (0.702)

Leverage -6.314*** -6.291*** -6.304*** -6.223*** -6.304*** -6.068***

(0.974) (0.973) (0.973) (0.949) (0.973) (0.947)

PPE/TA 0.999 1.041 1.000 1.022 1.000 1.128

(0.994) (0.986) (0.981) (0.971) (0.981) (0.966)

P3_PR 0.124

(0.090)

P3_PR_D -1.054***

(0.401)

Observations 5,261 5,261 5,261 5,261 5,261 5,261

Year Dummies Y Y Y Y Y Y

Pseudo R2 0.435 0.433 0.432 0.429 0.432 0.415

Wald 1023 1001 1004 1056 1004 1006

40

References

Acemoglu, Daron and Amy Finkelstein, 2008, Input and Technology Choices in Regulated Industries:

Evidence from the Health Care Sector, Journal of Political Economy 116, 837-880.

Acemoglu, Daron and David Autor, 2011, Skills, Tasks and Technologies: Implications for

Employment and Earnings, In: David Card and Orley Ashenfelter (Ed.), Handbook of Labor

Economics, Vol. 4b, Elsevier/North-Holland, Amsterdam, Chapter 12, 1043-1171.

Acemoglu, Daron and Pascual Restrepo, 2016, The Race between Man and Machine: Implications of

Technology for Growth, Factor Shares and Employment, NBER Working Paper No. 22252.

Acemoglu, Daron and Pascual Restrepo, 2017, Robots and Jobs: Evidence from US Labor Markets.

Available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2940245.

Akerman, Anders, Ingvil Gaarder and Magne Mogstad, 2015, The Skill Complementarity of

Broadband Internet, Quarterly Journal of Economics 130, 1781-1824.

Allen, Franklin, and Roni Michaely, 2003, Payout Policy, Handbook of the Economics of Finance,

edited by George M. Constantinides, Milton Harris, and René M. Stulz, 1:337–429. Elsevier.

Agrawal, Ashwini, and Prasanna Tambe, forthcoming, Private Equity and Workers’ Career Paths: The

Role of Technological Change, Review of Financial Studies.

Asquith, Paul and David W. Mullins, 1986, Equity Issues and Offering Dilution, Journal of Financial

Economics 15, 61-89.

Atanassov, Julian and E. Han Kim, 2009, Labor and Corporate Governance: International Evidence

from Restructuring Decisions, Journal of Finance 64, 341-374.

Autor, David H. and David Dorn, 2013, The Growth of Low-Skill Service Jobs and the Polarization

of the US Labor Market, American Economic Review 103, 1553-15597.

Autor, David H. and Michael J. Handel, 2013, Putting Tasks to the Test: Human Capital, Job Tasks,

and Wages, Journal of Labor Economics 31, 59-96.

Autor, David H., Lawrence F. Katz, and Melissa S. Kearney, 2006, The Polarization of the U.S. Labor

Market, American Economics Review 96, 189-194.

Autor, David H., Lawrence F. Katz and Alan B. Krueger, 1998, Computing Inequality: Have

Computers Changed the Labor Market? Quarterly Journal of Economics 113, 1169-1213.

Autor, David. H., Frank Levy and Richard J. Murnane, 2003, The Skill Content of Recent

Technological Change: An Empirical Exploration, Quarterly Journal of Economics 118, 1279-

1333.

Autor, David. H., and Anna Salomons, 2017, Robocalypse Now—Does Productivity Growth Threaten

Employment? Available at: https://ideas.repec.org/h/nbr/nberch/14019.html.

Bayless, Mark and Susan Chaplinsky, 1996, Is There a Window of Opportunity for Seasoned Equity

Issuance? Journal of Finance 51, 253-278.

Ben-Gad, Michael, 2008, Capital-Skill Complementarity and the Immigration Surplus, Review of

Economic Dynamics, 11, 335-365.

Berk, Jonathan B., Richard Stanton and Josef Zechner, 2010, Human Capital, Bankruptcy, and Capital

Structure, Journal of Finance 65, 891-926.

Berman, Eli, John Bound and Zvi Griliches, 1994, Changes in the Demand for Skilled Labor within

U.S. Manufacturing Industries: Evidence from the Annual Survey of Manufacturing, Quarterly

Journal of Economics 109, 367-397.

Bernstein, Shai, 2015, Does Going Public Affect Innovation? Journal of Finance 70, 1365-1403.

Bertrand, Marianne and Sendhi Mullainathan, 2003, Enjoying the Quiet Life? Corporate Governance

and Managerial Preferences, Journal of Political Economy 111, 1043-1075.

Bhattacharya, Sudipto, 1979, Imperfect Information, Dividend Policy, and “the Bird in the Hand

Fallacy", Bell Journal of Economics 10, 259-270.

Bresnahan, Timothy F., Erik Brynjolfsson and Lorin M. Hitt, 2002, Information Technology,

Workplace Organization, and the Demand for Skilled Labor: Firm-Level Evidence, Quarterly

Journal of Economics 117, 339-376.

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.

Brown, Jeffrey R., Nellie Liang, and Scott Weisbenner, 2007, Executive Financial Incentives and

Payout Policy: Firm Responses to the 2003 Dividend Tax Cut, Journal of Finance 62, 1935–65.

41

Bryson, Alex, John Forth, and Minghai Zhou, 2014, Same or Different? The CEO Labour Market in

China’s Public Listed Companies, Economic Journal 124, F90-F108.

Cai, Fang, Albert Park and Yaohul Zhao, 2008, The Chinese Labor Market in the Reform Era in

China’s Great Economic Transformation, In: Loren Brandt and Thomas G. Rawski (Ed.),

China’s Great Economic Tranformation, Cambridge University Press, Chapter 6, 167-214.

Card, David, 1999, The Causal Effect of Education on Earnings, In: David Card and Orley

Ashenfelter (Ed.), Handbook of Labor Economics, Vol. 3a, Elsevier/North-Holland, Amsterdam,

Chapter 30, 1801-1863.

Caroli, Eve and John Van Reenen, 2001, Skill Biased Organizational Change? Evidence from a Panel

of British and French Establishments, Quarterly Journal of Economics 116, 1449-1492.

Chemmanur, Thomas, Yingmei Cheng and Tianming Zhang, 2013, Human Capital, Capital Structure,

and Employee Pay: An Empirical Analysis, Journal of Financial Economics 110, 478-502.

Crane, Alan D., Sebastien Michenaud, and James Weston, 2016, The Effect of Institutional

Ownership on Payout Policy: Evidence from Index Thresholds, Review of Financial Studies 29,

1377-1408.

Cronqvist, Henrik, Fredrik Heyman, Mattias Nilsson, Helena Svaleryd, and Jonas Vlachos, 2009, Do

Entrenched Managers Pay Their Workers More? Journal of Finance 64, 309-339.

DeAngelo, Harry and Linda DeAngelo, 2006, The Irrelevance of the MM Dividend Irrelevance

Theorem, Journal of Financial Economics 79, 293-316.

DeAngelo, Harry, Linda DeAngelo, and René M. Stulz, 2006, Dividend Policy and the

Earned/contributed Capital Mix: A Test of the Life-Cycle Theory, Journal of Financial

Economics 81, 227–54.

DeAngelo, Harry, Linda DeAngelo and René Stulz, 2010, Seasoned Equity Offerings, Market Timing,

and the Corporate Lifecycle, Journal of Financial Economics 95, 275-295.

Denis, David J. and Igor Osobov, 2008, Why Do Firms Pay Dividends? International Evidence on the

Determinants of Dividend Policy, Journal of Financial Economics 89, 62–82.

Doms, Mark, Timothy Dunne and Kenneth R. Troske, 1997, Workers, Wages, and Technology,

Quarterly Journal of Economics 62, 253-290.

Duffy, John, Chris Pagageorgiou and Fidel Perez-Sebastian, 2004, Capital-Skill Complementarity?

Evidence from a Panel of Countries, Review of Economics and Statistics 86, 327-344.

Eades, Kenneth, Patrick Hess, and E. Han Kim, 1984, On Interpreting Security Returns during the Ex-

dividend Period, Journal of Financial Economics 13, 3-34.

Eckbo, B. Espen and Ronald W. Masulis, 1995, Seasoned Equity Offerings: A Survey, In: John R.

Birge and Vadim Linetsky (Ed.), Handbooks in Operations Research and Management Science,

Vol. 9, Elsevier/North-Holland, Amsterdam, Chapter 31, 1017-1072.

Eckbo, B. Espen, Ronald W. Masulis and Øyvind Norli, 2000, Seasoned Public Offerings: Resolution

of the ‘New Issues Puzzle’, Journal of Financial Economics 56, 251-291.

Fallon, P. R. and Richard Layard, 1975, Capital-Skill Complementarity, Income Distribution, and

Output Accounting, Journal of Political Economy 83, 279-302.

Fenn, George W., and Nellie Liang, 2001, Corporate Payout Policy and Managerial Stock Incentives,

Journal of Financial Economics 60, 45-72

Gallagher, Mary, John Giles, Albert Park, and Meiyan Wang, 2015, China’s 2008 Labor Contract

Law: Implementation and Impacts on Workers, Human Relations, 68(2), 197-235.

Goldin, Claudia and Lawrence F. Katz, 1996, Technology, Skill, and the Wage Structure: Insights

from the Past, American Economic Review 86, 252-257.

Goldin, Claudia and Lawrence F. Katz, 1998, The Origins of Technology-Skill Complementarity,

Quarterly Journal of Economics 113, 693-732.

Griliches, Zvi, 1969, Capital-Skill Complementarity, Review of Economics and Statistics 51, 465-468.

Gustafson, Matthew T. and Peter Iliev, 2017, The Effects of Removing Barriers to Equity Issuance,

Journal of Financial Economics 124, 580-598.

Hall, Brownyn H. and Josh Lerner, 2010, The Financing of R&D and Innovation, In: Brownyn H.

Hall and Nathan Rosenberg (Ed.), Handbooks in Economics of Innovation, Vol. 1,

Elsevier/North-Holland, Amsterdam, Chapter 14, 609-639.

Hamermesh, Daniel S., 1993, Labor Demand, Princeton, Princeton University Press.

42

Hanka, Gordon, 1998, Debt and the Terms of Employment, Journal of Financial Economics 48, 245-

282.

Hershbein, Brad and Lisa Kahn, 2016, Do Recessions Accelerate Routine-Biased Technology Change?

Evidence from Vacancy Postings, http://www.nber.org/papers/w22762.

Ho, Virginia, 2006, Labor Law in China’s Reform Era: The Evolving Legal Framework for Labor

Rights. In: Mary Gallagher, Ching Kwan Lee, and Albert Park (Ed.), The Labor of Reform in

China, Unpublished Edited Vol.

Hsieh, Chang-Tai and Zheng (Michael) Song, 2015, Grasp the Large, Let Go of the Small: The

Transformation of the State Sector in China, Brookings Papers on Economic Activity Spring,

295-346.

Jensen, Michael C., 1986, Agency Costs of Free Cash Flow, Corporate Finance, and Takeovers,

American Economic Review 76, 323-329.

Jung, Kooyul, Yong-Cheol Kim and René M. Stulz, 1996, Timing, Investment Opportunities,

Managerial Discretion, and the Security Issue Decision, Journal of Financial Economics 42,

159-185.

Kasahara, Hiroyuki, Yawen Liang and Joel Rodrigue, 2016, Does Importing Intermediates Increase

the Demand for Skilled Workers? Plant-Level Evidence from Indonesia, Journal of

International Economics 102, 242-261.

Kim, E. Han and Paige Ouimet, 2014, Broad-Based Employee Stock Ownership: Motives and

Outcomes, Journal of Finance 69, 1273-1319.

Kim, E. Han and Amiyatosh Purnanandam, 2014, Seasoned Equity Offerings, Corporate Governance,

and Investments, Review of Finance 18, 1023-1057.

Kim, Woojin and Michael S. Weisbach, 2008, Motivations for Public Equity Offers: An International

Perspective, Journal of Financial Economics 87, 281-307.

Korajczyk, Robert A., Deborah Lucas and Robert L. McDonald, 1990, Understanding Stock Price

Behavior around the Time of Equity Issues, In: Hubbard, R. Glenn (Ed.), Asymmetric

Information, Corporate Finance and Investment, University of Chicago Press, Chicago.

Krusell, Per, Lee E. Ohanian, José-Victor Ríos-Rull and Givanni L. Violante, 2000, Capital-Skill

Complementarity and Inequality: A Macroeconomic Analysis, Econometrica 68, 1029-1053.

La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer and Robert W. Vishny, 1998, Law and

Finance, Journal of Political Economy 106, 1113-1155.

Lemmon, Michael L., Michael R. Roberts, and Jaime F. Zender, 2008, Back to the Beginning:

Persistence and the Cross-Section of Corporate Capital Structure, Journal of Finance 63, 1575-

1608.

Lewis, Ethan, 2011, Immigration, Skill Mix, and Capital-Skill Complementarity, Quarterly Journal of

Economics 126, 1029-1069.

Li, Hongbin, Lei Li, Binzhen Wu, and Yanyan Xiong, 2012, The End of Cheap Chinese Labor,

Journal of Economic Perspectives, 26(4), 57-74.

Li, Shi and Yaohui Zhao, 2003, The Decline of In-kind Wage Payments in Urban China, Journal of

Chinese Economics and Business Studies 1, 245-258.

Lin, Justin, Fang Cai, and Zhou Li, 1996, The China Miracle: Development Strategy and Economic

Reform, Hong Kong, Chinese University of Hong Kong Press.

Lindquist, Matthew J., 2004, Capital-Skill Complementarity and Inequality over the Business Cycle,

Review of Economic Dynamics 7, 519-540.

Machin, Stephen and John Van Reenen, 1998, Technology and Changes in Skill Structure: Evidence

from Seven OECD Countries, Quarterly Journal of Economics 113, 1215-1244.

Masulis, Ronald W. and Ashok N. Korwar, 1986, Seasoned Equity Offerings: An Empirical

Investigation, Journal of Financial Economics 15, 91-118.

McCrary, Justin, 2008, Manipulation of the Running Variable in the Regression Discontinuity Design:

A Density Test, Journal of Econometrics, 142, 698-714.

McLean, R. David, 2011, Share Issuance and Cash Savings, Journal of Financial Economics 99, 693-

715.

Meng, Xin, 2000, Labor Market Reform in China. Cambridge and New York: Cambridge University

Press.

43

Michaels, Ryan, T. Beau Page and Toni Whited, 2016, Labor and Capital Dynamics under Financing

Frictions. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2850790.

Miller, Merton H., and Franco Modigliani, 1961, Dividend Policy, Growth, and the Valuation of

Shares, The Journal of Business 34, 411–33.

Pagano, Marco, Fabio Panetta and Luigi Zingales, 1998, Why Do Companies Go Public? An

Empirical Analysis, Journal of Finance 53, 27-64.

Parro, Fernando, 2013, Capital-Skill Complementarity and the Skill Premium in a Quantitative Model

of Trade, American Economic Journal: Macroeconomics 5, 72-117.

Perotti, Enrico C. and Kathlyn E. Spier, 1993, Capital Structure as a Bargaining Tool: The Role of

Leverage in Contract Renegotiation, American Economic Review 83, 1131-1141.

Wang, Qian, T. J. Wong and Lijun Xia, 2008, State Ownership, the Institutional Environment, and

Auditor Choice: Evidence from China, Journal of Accounting and Economics 46. 112-134.

Wooldridge, Jeffrey, 2010, Econometric Analysis of Cross Section and Panel Data, Cambridge, the

MIT Press.

Yasar, Mahmut and Catherine J. Morrison Paul, 2008, Capital-Skill Complementarity, Productivity

and Wages: Evidence from Plant-Level Data for a Developing Country, Labor Economics 15,

1-17.

Yueh, Linda, 2004, Wage Reforms in China during the 1990s, Asian Economic Journal 18, 149-164.

Zhang, Junsen, Yaohui Zhao, Albert Park, and Xiaoqing Song, 2005, Economic Returns to Schooling

in Urban China, 1988 to 2001, Journal of Comparative Economics 33, 730-752.

44

Table 1: Online Job Posting Data and Words Related to Skills.

This table provides information obtained from job posting data. Panel A reports the number of full-

time job advertisements posted by firms listed in Shanghai and Shenzhen Stock Exchanges over 2014-

2016 in Lagou.com (https://www.lagou.com). Panel A, Column (1) shows the number of all new full-

time job advertisements by year, and Column (2) shows the number of new full-time job

advertisements in the year firms issued seasoned equity offerings (including underwritten offerings,

rights offerings, and private placements). Repetition of the same advertisements is excluded. Panel B

provides the list of key words used to identify requirements for different types of skills. The key words

are English translation of Chinese words mentioned in job advertisements.

Panel A: Sample Distribution

Year Number of Unique Job Advertisements JP_SEO=1

(1) (2)

2014 5,702 1,410

2015 15,041 3,591

2016 24,842 2,790

Total 45,585 7,791

Panel B: Key Words Used to Identify Different Skill Requirements

Skills Key Words

Advanced

computer

Programming, Java, SQL, Python, developing, server, artificial intelligence,

big data, machine learning, html, and software

Basic computer Diannao (an unofficial name of computer), PPT, presentation slides, Excel,

spreadsheets, Microsoft Office, Windows, and Word.

Non-routine

analytical task

skills

Research, analysis, problem solving, analytical critical thinking, math,

statistics, learning, thinking, changing, improving, professional writing, and

reporting.

Non-routine

interactive task

skills

Communication, cooperation, negotiation, services, clients, persuading,

selling, management, monitoring, supervisory, leadership, mentoring,

guidance, and making a deal.

45

Table 2: SEOs and Computer Skills.

This table relates SEOs to computer skills mentioned in online job postings. Panels A and B report the

results for advanced and basic computer skills, respectively. Key words used to identify advanced and

basic computer skills are listed in Table 1, Panel B. The dependent variable in Columns (1) and (2) is an

indicator equal to one if any of the key words related to advanced and basic computer skills, respectively,

appear in a job description. The dependent variable in Columns (3) and (4) is the log of one plus the

number of words appearing in a job description. Columns (1) and (2) are estimated by logit regressions;

Columns (3) and (4), the OLS regressions. The sample period covers 2014 – 2016. Regressions in Columns

(1) and (3) control for year-, firm-, and location dummies, and regressions in Columns (2) and (4) add job

dummies. Standard errors (in parentheses) are clustered at the firm-job pair level in all regressions.

Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

Panel A: Advanced Computer Skills

Adv_Computer_Dum Ln(Adv_Computer)

VARIABLES (1) (2) (3) (4)

JP_SEO 0.183*** 0.153** 0.059*** 0.042***

(0.060) (0.063) (0.017) (0.012)

Constant 2.033*** 1.122** 1.855*** 1.215***

(0.779) (0.481) (0.293) (0.115)

Year Dummies Y Y Y Y

Firm Dummies Y Y Y Y

Location Dummies Y Y Y Y

Job Dummies N Y N Y

Observations 44,767 44,767 45,582 45,582

Pseudo-R-squared 0.078 0.297 Adjusted R-squared 0.101 0.398

Panel B: Basic Computer Skills

VARIABLES Basic_Computer_Dum Ln(Basic_Computer)

JP_SEO 0.370** 0.381*** 0.008** 0.008**

(0.148) (0.146) (0.004) (0.003)

Constant -4.514*** -3.880*** 0.018 0.048**

(1.196) (1.183) (0.015) (0.019)

Year Dummies Y Y Y Y

Firm Dummies Y Y Y Y

Location Dummies Y Y Y Y

Job Dummies N Y N Y

Observations 40,218 40,218 45,582 45,582

Pseudo-R-squared 0.085 0.120 Adjusted R-squared 0.032 0.045

46

Table 3: SEOs and Non-routine Task Skills.

This table relates SEOs to non-routine task skills mentioned in online job postings. Panels A and B report the

results for non-routine analytical and interactive task skills, respectively. Key words used to identify non-

routine task skills are listed in Table 1, Panel B. The dependent variable in Columns (1) and (2) is an indicator

equal to one if any of the key words related to non-routine analytical amd interactive task skills appear in a

job description. The dependent variable in Columns (3) and (4) is log of one plus the number of words

appearing in a job description. Columns (1) and (2) are estimated by logit regressions; Columns (3) and (4),

the OLS regressions. The sample period covers 2014 – 2016. Regressions in Columns (1) and (3) control for

year-, firm-, and location dummies, and regressions in Columns (2) and (4) add job dummies. Standard errors

(in parentheses) are clustered at the firm-job pair level in all regressions. Coefficients marked with *, **, and

*** are significant at 10%, 5%, and 1%, respectively.

Panel A: Non-routine Analytic Task Skills

Non-routine Analytical Task

Skills_Dum

Ln(Non-routine Analytical

Task Skills)

VARIABLES (1) (2) (3) (4)

JP_SEO 0.166*** 0.169*** 0.022* 0.022

(0.057) (0.058) (0.013) (0.013)

Constant 0.578 0.373 0.524*** 0.419**

(0.365) (0.380) (0.125) (0.164)

Year Dummies Y Y Y Y

Firm Dummies Y Y Y Y

Location Dummies Y Y Y Y

Job Dummies N Y N Y

Observations 44,992 44,992 45,582 45,582

Pseudo-R-squared 0.072 0.090 Adjusted R-squared 0.082 0.115

Panel B: Non-routine Interactive Task Skills

VARIABLES

Non-routine Interactive Task

Skills_Dum

Ln(Non-routine Interactive

Task Skills)

JP_SEO 0.110* 0.143** 0.015 0.024**

(0.061) (0.063) (0.011) (0.011)

Constant 2.490*** 3.030*** 0.959*** 1.215***

(0.545) (0.608) (0.158) (0.205)

Year Dummies Y Y Y Y

Firm Dummies Y Y Y Y

Location Dummies Y Y Y Y

Job Dummies N Y N Y

Observations 44,214 44,214 45,582 45,582

Pseudo-R-squared 0.052 0.096 Adjusted R-squared 0.069 0.156

47

Table 4: Sample for Panel Data Analyses.

This table reports the number of firms in the total sample and in the seasoned equity offering sample for the

panel data analyses. The sample includes Chinese firms listed on Shanghai and Shenzhen Stock Exchanges

from 2000 to 2012. Financial firms, firms with fewer than 100 employees, ST (special treatment), and *ST

firms are excluded. Firms are classified as ST or *ST if they have two (ST) or three (*ST) consecutive years

of negative net profits. Column (1) shows the number of firms in the full sample by year. Column (2) shows

the number of public offerings (underwritten offerings and rights offerings) by offering year.

Year Full Number of SEOs

(1) (2)

2000 885 154

2001 951 131

2002 1,002 44

2003 1,059 38

2004 1,153 32

2005 1,172 7

2006 1,204 7

2007 1,323 28

2008 1,395 43

2009 1,485 18

2010 1,830 20

2011 2,120 23

2012 2,259 12

Total 17,838 557

48

Table 5: Summary Statistics of Variables Used in the Panel Data Analyses.

This table reports summary statistics for variables used in the panel data regressions. Appendix 3 provides

variable definitions and data sources.

Mean Std. Dev. Min Max

Key Variables (1) (2) (3) (4)

SEO 0.088 0.283 0.000 1.000

SEOIneligible 0.155 0.362 0.000 1.000

%_Production 0.478 0.284 0.000 0.995

%_Staff 0.092 0.108 0.000 0.998

%_Tech_R&D 0.172 0.156 0.000 0.987

%_S&M 0.128 0.160 0.000 0.917

%_Finance 0.033 0.034 0.000 0.672

%_Others 0.183 0.266 0.000 1.000

%_Grad 0.031 0.043 0.000 0.237

%_BA 0.203 0.181 0.000 0.959

Ln(EMP) 2.899 1.205 0.000 8.618

Ln(Production) 6.253 1.916 0.000 12.728

Ln(Staff) 4.414 1.592 0.000 11.353

Lu(Tech_R&D) 5.478 1.229 0.000 12.204

Ln(S&M) 4.876 1.411 0.000 11.456

Ln(Finance) 3.802 1.017 0.000 9.578

Ln(Others) 3.789 2.957 0.000 12.330

Ln(Grad) 3.337 1.401 0.000 10.112

Ln(BA) 5.531 1.320 0.000 11.937

AWAGE 6.799 10.351 0.013 285.293

AWAGE_NonExe 6.933 11.333 0.011 493.721

AEXEPAY 20.192 20.009 0.360 506.227

Payroll 296.153 1908.561 0.039 108031.000

Payroll_NonExe 306.816 1964.043 0.019 108015.900

Payroll_Exe 2.876 3.534 0.022 111.370

Ln(Inv_Tech_Assets) 2.278 2.119 0.000 11.422

Ln(Capx) 4.199 1.884 -6.896 12.420

Ln(Total_Patent) 0.518 1.081 0.000 4.394

Ln(Invention) 0.315 0.767 0.000 3.497

Ln(Utility_Model) 0.306 0.793 0.000 3.664

Ln(Design) 0.122 0.497 0.000 2.944

NYEAR_LISTED 7.011 5.013 0.000 22.000

SALES 4517.473 39862.920 0.003 2085363.000

ROA 0.038 0.088 -1.752 2.933

LEVERAGE 0.456 0.201 0.047 0.889

49

Table 5: Summary Statistics of Variables Used in the Panel Data Analyses. (Continued)

Other Variables Mean Std.Dev. Min Max

PPE/TA 0.320 0.201 0.000 0.975

SALES_GR 0.228 0.497 -0.609 3.379

RE/TotalEq 0.058 0.602 -3.922 1.829

Inst_OWN 0.128 0.170 0.000 0.995

Ln(Tot_Volatility) -2.871 0.319 -6.039 -1.408

%_IND_DIR 0.305 0.127 0.000 0.833

%_STATE_OWN 0.215 0.252 0.000 0.886

%_LARGEST_SH 0.390 0.163 0.022 0.894

%_NONTRD_SH 0.212 0.296 0.000 0.913

D_PRIVATE_PLACE 0.046 0.210 0.000 1.000

Overseas_SEOs 0.001 0.029 0.000 1.000

DIV_PR 0.259 0.306 0.000 1.500

P3_PR 0.767 0.827 0.000 4.085

P3_PR_D 0.027 0.161 0.000 1.000

Ln(MIN_WAGE) 6.404 0.351 5.340 6.990

LAWSCORE 7.784 3.916 0.000 16.610

Labor_Law_Effect 3.689 3.850 0.000 13.312

50

Table 6: First-stage Regressions for Main Results. This table reports first-stage estimation results using conditional logistic regressions at the firm level. Column (1) reports the first stage result for Panel A of Tables 7 and 8, Column (1) of Tables 9, 10, 11, and 12. Column (2) reports the first stage result for Panel B of Tables 7 and 8, Columns (2)-(3) of Table 9, Columns (2)-(5) of Table 10, Columns (2)-(4) of Table 11, and Columns (2)-(4) of Table 12. Robust standard errors clustered at the firm level are in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

SEO

VARIABLES (1) (2)

SEOIneligible -1.036*** -1.621***

(0.318) (0.416)

ln(NYEAR_LISTED) 4.471*** 4.414***

(0.431) (0.471) ln(MIN_WAGE) 1.011 1.581**

(0.633) (0.692) LAWSCORE 0.036 0.041

(0.058) (0.070) Labor_Law_Effect 0.101 0.125

(0.069) (0.086) ln(SALES) 1.110***

(0.200) RE/TotalEq 0.452

(0.451) ROA -8.213***

(1.904) Inst_OWN 1.768***

(0.504) Ln(Tot_Volatility) -0.202

(0.258) %_LARGEST_SH -1.994

(1.259) SALES_GR -0.145

(0.127) DIV_PR 0.090

(0.075) %_STATE_OWN 0.272

(0.533) %_IND_DIR -0.712

(0.644) %_NONTRD_SH -1.955***

(0.615) D_PRIVATE_PLACE -1.357***

(0.315) Overseas_SEOs -13.594***

(0.708) Leverage -6.304***

(0.973) PPE/TA 1.000

(0.981) Year Dummies Y Y Observations 5,438 5,261

Pseudo R2 0.355 0.432 Wald 553.4 1004

51

Table 7: SEO Impact on Employee Composition by Occupation and Education.

This table reports the second-stage estimation of the impact that SEOs have on the employee composition by occupation and education. The dependent variable is the percentage of production workers in Column (1), support staff in Column (2), technicians and R&D employees in Column (3), sales and marketing forces in Column (4), finance staff in Column (5), employees in uncategorized occupations in Column (6), employees with post-graduate degrees in Column (7), and employees with four-year university Bachelor’s degrees and above in Column (8). All regressions include firm- and year fixed effects. Regressions in Panel A control for only firm age and legal variables. Regressions in Panel B include the full set of time-varying firm characteristic variables. Appendix 3 provides variable definitions and data sources. The sample period covers 2000 – 2012. Bootstrapped standard errors are reported in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

Panel A

%_Production %_Staff %_Tech_R&D %_S&M %_Finance %_Others %_Grad %_BA

VARIABLES (1) (2) (3) (4) (5) (6) (7) (8)

SEO� -0.041*** -0.023*** 0.053*** 0.018** -0.001 0.013 0.004 0.005

(0.016) (0.008) (0.012) (0.008) (0.002) (0.022) (0.003) (0.010)

ln(NYEAR_LISTED) 0.007 -0.000 -0.021*** -0.011** -0.002 0.016* -0.002* -0.002

(0.007) (0.003) (0.005) (0.005) (0.001) (0.009) (0.001) (0.005)

ln(MIN_WAGE) -0.004 0.007 -0.006 -0.002 0.001 0.015 0.007** 0.034***

(0.020) (0.010) (0.011) (0.009) (0.002) (0.024) (0.003) (0.012)

LAWSCORE 0.001 0.000 0.002 0.001 0.000 -0.005*** 0.000 0.002*

(0.002) (0.001) (0.001) (0.001) (0.000) (0.002) (0.000) (0.001)

Labor_Law_Effect -0.008*** 0.002** 0.002** -0.002* 0.001*** 0.005*** -0.001*** -0.004***

(0.001) (0.001) (0.001) (0.001) (0.000) (0.002) (0.000) (0.001)

Constant 0.521*** 0.020 0.207*** 0.147*** 0.025* 0.108 -0.010 -0.044

(0.121) (0.059) (0.070) (0.057) (0.013) (0.147) (0.019) (0.072)

Firm & Year FE Y Y Y Y Y Y Y Y

Observations 17,443 17,443 14,314 10,857 13,719 17,443 8,313 11,942

52

Panel B

%_Production %_Staff %_Tech_R&D %_S&M %_Finance %_Others %_Grad %_BA

VARIABLES (1) (2) (3) (4) (5) (6) (7) (8)

SEO� -0.037** -0.014*** 0.026*** 0.026*** 0.002 0.012 0.006*** 0.026***

(0.015) (0.005) (0.009) (0.009) (0.002) (0.019) (0.002) (0.008)

ln(NYEAR_LISTED) -0.002 -0.001 -0.007 -0.010** -0.001 0.020** -0.003* -0.006

(0.008) (0.003) (0.004) (0.005) (0.001) (0.009) (0.001) (0.005)

ln(MIN_WAGE) -0.001 0.007 -0.002 -0.003 0.001 0.015 0.005 0.032***

(0.017) (0.009) (0.010) (0.017) (0.002) (0.026) (0.003) (0.012)

LAWSCORE 0.002 -0.000 0.001* 0.000 0.000 -0.006*** -0.000 0.001

(0.002) (0.001) (0.001) (0.001) (0.000) (0.002) (0.000) (0.001)

Labor_Law_Effect -0.006*** 0.002** 0.001* -0.003*** 0.001** 0.005*** -0.001*** -0.005***

(0.001) (0.001) (0.001) (0.001) (0.000) (0.002) (0.000) (0.001)

ln(SALES) 0.011*** -0.009*** -0.007*** -0.000 -0.003*** -0.002 -0.002* -0.008***

(0.004) (0.002) (0.002) (0.002) (0.000) (0.005) (0.001) (0.003)

RE/TotalEq -0.009* 0.004* 0.010*** 0.004 0.001 -0.004 0.003*** 0.014***

(0.005) (0.002) (0.003) (0.003) (0.001) (0.006) (0.001) (0.003)

ROA -0.085*** -0.005 0.015 0.073*** 0.014** 0.047 -0.000 0.009

(0.026) (0.015) (0.015) (0.019) (0.006) (0.037) (0.008) (0.019)

Inst_OWN 0.011 0.011* 0.012** -0.004 -0.000 -0.025* 0.002 0.004

(0.012) (0.006) (0.005) (0.005) (0.002) (0.013) (0.002) (0.007)

Ln(Tot_Volatility) -0.008 -0.001 -0.002 0.005 -0.001 0.009 0.002 0.000

(0.008) (0.004) (0.004) (0.004) (0.001) (0.009) (0.001) (0.004)

%_LARGEST_SH -0.040* 0.030*** 0.009 0.018 0.014*** 0.016 0.006 0.053***

(0.021) (0.010) (0.014) (0.019) (0.004) (0.029) (0.004) (0.019)

SALES_GR -0.010** 0.004** 0.005** -0.000 0.001** 0.008* 0.000 -0.001

(0.004) (0.002) (0.002) (0.002) (0.001) (0.005) (0.001) (0.002)

DIV_PR 0.001 0.000 -0.001 -0.000 0.000 0.000 -0.000 -0.000

(0.002) (0.001) (0.002) (0.001) (0.000) (0.002) (0.001) (0.001)

%_STATE_OWN -0.002 -0.011* -0.004 -0.022*** -0.001 0.012 -0.003 0.002

(0.013) (0.006) (0.007) (0.006) (0.001) (0.013) (0.002) (0.007)

%_IND_DIR -0.052*** 0.014 0.028*** 0.032*** -0.002 0.007 -0.002 -0.011

(0.018) (0.010) (0.010) (0.009) (0.003) (0.021) (0.003) (0.013)

%_NONTRD_SH -0.010 0.004 -0.001 -0.000 0.001 0.009 0.005* -0.012

(0.013) (0.007) (0.008) (0.007) (0.002) (0.018) (0.003) (0.008)

D_PRIVATE_PLACE 0.001 -0.002 0.000 0.005 -0.001 -0.001 0.001 0.007

(0.007) (0.004) (0.004) (0.007) (0.001) (0.009) (0.001) (0.004)

Overseas_SEOs 0.049 -0.007 -0.009 -0.002 -0.007 -0.038 -0.004 -0.004

(0.034) (0.027) (0.027) (0.028) (0.008) (0.029) (0.005) (0.026)

Leverage -0.107*** 0.021** 0.012 0.024* 0.013*** 0.054** 0.008** 0.035***

(0.019) (0.010) (0.013) (0.014) (0.002) (0.023) (0.003) (0.011)

PPE/TA 0.169*** -0.009 -0.029*** -0.050*** -0.024*** -0.119*** -0.016*** -0.082***

(0.017) (0.010) (0.009) (0.015) (0.002) (0.021) (0.004) (0.011)

Constant 0.436*** 0.050 0.210*** 0.163* 0.031** 0.141 0.016 0.019

(0.112) (0.056) (0.058) (0.096) (0.016) (0.154) (0.020) (0.073)

Firm & Year FE Y Y Y Y Y Y Y Y

Observations 17,011 17,011 13,957 10,612 13,378 17,011 8,122 11,673

53

Table 8: SEO Impact on Firm-Level Employment by Occupation and Education. This table reports the second-stage estimation of the impact that SEOs have on the number of employees at the firm level. The dependent variable is the number of all employees in Column (1), production workers in Column (2), support staff in Column (3), technicians and R&D employees in Column (4), sales and marketing forces in Column (5), finance staff in Column (6), employees in uncategorized occupations in Column (7), employees with post-graduate degrees in Column (8), and employees with Bachelor’s degrees and above in Column (9). All dependent variables are logged, and all regressions include firm- and year fixed effects. Regressions in Panel A control for only firm age and legal variables. Regressions in Panel B include the full set of time-varying firm characteristic variables. Appendix 3 provides variable definitions and data sources. The sample period covers 2000 – 2012. Bootstrapped standard errors are reported in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively. Panel A

Ln(EMP) Ln(Production) Ln(Staff) Ln(Tech_R&D) Ln(S&M) Ln(Finance) Ln(Others) Ln(Grad) Ln(BA)

VARIABLES (1) (2) (3) (4) (5) (6) (7) (8) (9)

SEO� -0.079** -0.259** -0.617*** 0.197** 0.048 -0.065 0.010 -0.011 -0.039

(0.040) (0.113) (0.121) (0.079) (0.086) (0.077) (0.254) (0.083) (0.068)

ln(NYEAR_LISTED) 0.258*** 0.266*** 0.305*** 0.156*** 0.166*** 0.173*** 0.484*** 0.195*** 0.196***

(0.021) (0.049) (0.052) (0.034) (0.051) (0.028) (0.107) (0.046) (0.029)

ln(MIN_WAGE) -0.108* -0.109 -0.101 -0.106 0.043 -0.051 0.196 0.321** 0.112

(0.058) (0.123) (0.145) (0.084) (0.085) (0.062) (0.287) (0.126) (0.090)

LAWSCORE -0.022*** -0.008 -0.002 -0.009 0.010 -0.015*** -0.056*** -0.027*** -0.026***

(0.005) (0.009) (0.011) (0.007) (0.012) (0.005) (0.021) (0.009) (0.010)

Labor_Law_Effect -0.004 -0.259** -0.617*** 0.197** 0.048 -0.065 0.010 -0.011 -0.039

(0.005) (0.113) (0.121) (0.079) (0.086) (0.077) (0.254) (0.083) (0.068)

Constant 3.307*** 6.703*** 4.276*** 5.910*** 4.367*** 3.837*** 2.246 0.567 4.257***

(0.348) (0.721) (0.889) (0.494) (0.517) (0.364) (1.704) (0.767) (0.542)

Firm & Year FE Y Y Y Y Y Y Y Y Y

Observations 17,443 17,443 17,443 14,314 10,857 13,719 17,443 8,313 11,942

54

Panel B

Ln(EMP) Ln(Production) Ln(Staff) Ln(Tech_R&D) Ln(S&M) Ln(Finance) Ln(Others) Ln(Grad) Ln(BA)

VARIABLES (1) (2) (3) (4) (5) (6) (7) (8) (9)

SEO� -0.070** -0.239** -0.444*** 0.096* 0.131* -0.003 0.048 0.103* 0.042

(0.034) (0.101) (0.106) (0.055) (0.069) (0.045) (0.197) (0.057) (0.060)

ln(NYEAR_LISTED) 0.099*** 0.119** 0.130*** 0.068** 0.043 0.053*** 0.329*** 0.004 0.016

(0.018) (0.051) (0.050) (0.031) (0.040) (0.019) (0.083) (0.038) (0.032)

ln(MIN_WAGE) -0.255*** -0.214* -0.224* -0.230*** -0.113 -0.156*** 0.080 0.233** -0.038

(0.055) (0.111) (0.130) (0.079) (0.108) (0.055) (0.276) (0.113) (0.076)

LAWSCORE -0.013*** 0.003 0.002 -0.007 0.011** -0.014*** -0.055*** -0.025** -0.014*

(0.005) (0.010) (0.009) (0.007) (0.005) (0.004) (0.021) (0.010) (0.009)

Labor_Law_Effect -0.002 -0.042*** 0.008 0.009* -0.024*** 0.004 0.058*** -0.034*** -0.017***

(0.003) (0.008) (0.009) (0.005) (0.007) (0.004) (0.015) (0.005) (0.005)

ln(SALES) 0.447*** 0.375*** 0.262*** 0.402*** 0.417*** 0.325*** 0.340*** 0.386*** 0.413***

(0.013) (0.024) (0.025) (0.017) (0.025) (0.013) (0.042) (0.023) (0.016)

RE/TotalEq 0.014 -0.013 0.058** 0.089*** 0.053* 0.043*** -0.074 0.132*** 0.149***

(0.012) (0.027) (0.024) (0.021) (0.031) (0.016) (0.062) (0.029) (0.020)

ROA -0.593*** -0.862*** -0.648*** -0.620*** -0.160 -0.281*** -0.498* -0.758*** -0.743***

(0.116) (0.199) (0.170) (0.164) (0.166) (0.101) (0.297) (0.201) (0.176)

Inst_OWN -0.001 0.063 0.015 0.050 0.032 -0.006 -0.123 -0.085 -0.009

(0.023) (0.075) (0.065) (0.046) (0.069) (0.031) (0.148) (0.053) (0.044)

Ln(Tot_Volatility) -0.035 -0.053 -0.015 -0.033 0.011 -0.058** -0.002 -0.014 -0.046

(0.022) (0.055) (0.050) (0.031) (0.044) (0.023) (0.112) (0.039) (0.033)

%_LARGEST_SH 0.020 -0.146 0.290** 0.040 0.139 0.135 0.097 -0.072 0.112

(0.078) (0.146) (0.144) (0.112) (0.147) (0.086) (0.277) (0.137) (0.092)

SALES_GR -0.104*** -0.127*** -0.050** -0.071*** -0.109*** -0.066*** -0.044 -0.048** -0.079***

(0.015) (0.025) (0.020) (0.015) (0.020) (0.011) (0.038) (0.021) (0.017)

DIV_PR 0.005 0.008 0.007 0.002 -0.002 0.003 -0.004 -0.006 0.005

(0.011) (0.012) (0.021) (0.017) (0.008) (0.006) (0.024) (0.015) (0.008)

%_STATE_OWN 0.112*** 0.019 0.072 0.087* 0.008 0.065* 0.223 0.018 0.153***

(0.031) (0.076) (0.073) (0.048) (0.050) (0.034) (0.151) (0.053) (0.045)

%_IND_DIR 0.015 -0.202* -0.031 0.202*** 0.199** 0.069 0.070 0.041 0.041

(0.052) (0.105) (0.131) (0.077) (0.082) (0.045) (0.215) (0.109) (0.080)

%_NONTRD_SH 0.004 -0.011 -0.099 -0.070 -0.049 -0.024 -0.002 -0.017 -0.033

(0.040) (0.101) (0.108) (0.059) (0.059) (0.042) (0.188) (0.074) (0.059)

D_PRIVATE_PLACE 0.097*** 0.117** 0.045 0.126*** 0.152*** 0.102*** 0.034 0.104*** 0.101***

(0.023) (0.050) (0.045) (0.026) (0.049) (0.023) (0.102) (0.038) (0.025)

Overseas_SEOs 0.148 0.302 -0.211 0.108 0.271 0.024 -0.290 0.132* 0.073

(0.108) (0.499) (0.518) (0.172) (0.201) (0.093) (0.642) (0.068) (0.103)

Leverage 0.205*** -0.266*** 0.258*** 0.261*** 0.341** 0.468*** 0.572** 0.284*** 0.381***

(0.056) (0.099) (0.099) (0.086) (0.140) (0.051) (0.267) (0.102) (0.100)

PPE/TA 0.515*** 0.954*** 0.464*** 0.271*** -0.227 -0.083 -0.589** -0.091 0.178**

(0.057) (0.118) (0.110) (0.070) (0.140) (0.059) (0.233) (0.103) (0.080)

Constant 1.216*** 4.924*** 3.184*** 4.001*** 2.810*** 2.205*** 0.889 -1.161* 2.349***

(0.345) (0.669) (0.799) (0.494) (0.643) (0.352) (1.672) (0.671) (0.458)

Firm & Year FE Y Y Y Y Y Y Y Y Y

Observations 17,011 17,011 17,011 13,957 10,612 13,378 17,011 8,122 11,673

55

Table 9: SEO Impact on Investments in Technology and Capital Expenditures. This table reports the second-stage estimation of the impact that SEOs have on investments in technology and total capital expenditures. The dependent variable is the log of one plus the value of newly purchased machines and equipment in Columns (1) and (2), and the log of capital expenditures in Column (3). Appendix 3 provides variable definitions and data sources. The sample period covers 2003 – 2012 for Columns (1) and (2), and 2000-2012 for Column (3). All regressions include firm- and year fixed effects. Bootstrapped standard errors are reported in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

Ln(Inv_Tech_Assets) Ln(Capx)

VARIABLES (1) (2) (3)

SEO� 0.512*** 0.188** 0.238***

(0.128) (0.094) (0.091)

Ln(NYEAR_LISTED) 0.026 -0.163*** -0.417***

(0.066) (0.059) (0.040)

Ln(MIN_WAGE) 0.196 0.005 0.028

(0.180) (0.165) (0.090)

LAWSCORE -0.064*** -0.035** -0.007

(0.018) (0.016) (0.008)

Labor_Law_Effect -0.031*** -0.030*** -0.044***

(0.010) (0.009) (0.008)

Ln(SALES) 0.482*** 0.653***

(0.034) (0.019)

RE/TotalEq 0.156*** 0.451***

(0.033) (0.036)

ROA -0.221 1.085***

(0.185) (0.335)

Inst_OWN 0.142 0.389***

(0.094) (0.063)

Ln(Tot_Volatility) -0.012 -0.157***

(0.057) (0.037)

%_LARGEST_SH 0.012 0.147

(0.229) (0.149)

SALES_GR 0.102*** -0.055**

(0.029) (0.023)

DIV_PR 0.002 0.009

(0.033) (0.021)

%_STATE_OWN 0.086 0.085*

(0.086) (0.050)

%_IND_DIR -0.072 0.197*

(0.159) (0.102)

%_NONTRD_SH -0.104 -0.540***

(0.138) (0.089)

D_PRIVATE_PLACE 0.188*** 0.372***

(0.053) (0.036)

Overseas_SEOs -0.035 -0.034

(0.287) (0.148)

Leverage 0.869*** 0.878***

(0.124) (0.125)

PPE/TA 2.252*** 3.005***

(0.133) (0.111)

Constant 1.205 -1.769* -1.768***

(1.096) (1.002) (0.546)

Firm & Year FE Y Y Y

Observations 14,834 14,469 17,114

56

Table 10: SEO Impact on Innovations. This table reports the second-stage estimation of the impact of SEOs on patents granted. The dependent variable is the log of one plus the total number of all patents granted in t+3 in Columns (1) and (2), invention patents, utility model patents, and design patents granted in t+3 in Columns (3) - (5), respectively. Appendix 3 provides variable definitions and data sources. The sample period covers 2000 – 2012. All regressions include firm- and year fixed effects. Bootstrapped standard errors are reported in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

Ln(Total_Patent)t+3 Ln(Invention)t+3

Ln(Utility

Model)t+3 Ln(Design)t+3

VARIABLES (1) (2) (3) (4) (5)

SEO� 0.389*** 0.129** 0.121** 0.113** 0.024

(0.109) (0.065) (0.061) (0.052) (0.045)

Ln(NYEAR_LISTED) 0.152*** 0.185*** 0.124*** 0.149*** -0.002

(0.036) (0.030) (0.024) (0.025) (0.021)

Ln(MIN_WAGE) 0.110 0.116 0.016 0.084 0.068

(0.067) (0.071) (0.044) (0.061) (0.043)

LAWSCORE -0.008 -0.007 -0.002 -0.008* -0.003

(0.005) (0.006) (0.004) (0.004) (0.003)

Labor_Law_Effect -0.006** -0.005 -0.001 -0.003 -0.002

(0.003) (0.004) (0.003) (0.004) (0.001)

Ln(SALES) 0.024** 0.026*** 0.024*** -0.003

(0.011) (0.008) (0.007) (0.005)

RE/TotalEq 0.027** 0.016* 0.024*** 0.004

(0.011) (0.009) (0.009) (0.006)

ROA -0.016 0.068 -0.102 0.013

(0.092) (0.061) (0.069) (0.045)

Inst_OWN -0.102** -0.114*** -0.096** 0.012

(0.048) (0.038) (0.040) (0.026)

Ln(Tot_Volatility) 0.038 0.053** 0.020 -0.022

(0.031) (0.023) (0.027) (0.018)

%_LARGEST_SH -0.118 -0.078 -0.097* -0.001

(0.084) (0.060) (0.051) (0.040)

SALES_GR 0.004 -0.003 -0.002 0.001

(0.008) (0.007) (0.007) (0.004)

DIV_PR 0.003 -0.004 0.002 -0.001

(0.013) (0.010) (0.012) (0.008)

%_STATE_OWN 0.033 0.000 0.028 0.039

(0.045) (0.037) (0.033) (0.026)

%_IND_DIR -0.107 -0.158*** -0.029 -0.032

(0.078) (0.058) (0.066) (0.044)

%_NONTRD_SH -0.072* -0.060* -0.051* -0.017

(0.043) (0.036) (0.029) (0.024)

D_PRIVATE_PLACE 0.019 -0.018 0.025 -0.006

(0.033) (0.025) (0.035) (0.018)

Overseas_SEOs -0.346 -0.130 -0.087 -0.201

(0.378) (0.270) (0.246) (0.287)

Leverage 0.040 0.059 0.024 -0.016

(0.060) (0.051) (0.051) (0.034)

PPE/TA -0.042 0.035 -0.085** -0.052*

(0.056) (0.041) (0.037) (0.028)

Constant -0.401 -0.379 0.058 -0.379 -0.317

(0.398) (0.460) (0.297) (0.380) (0.281)

Firm & Year FE Y Y Y Y Y

Observations 11,138 10,808 10,808 10,808 10,808

57

Table 11: SEO Impact on Firm Average Wages. This table reports the second-stage estimation of the impact of SEOs on firm average wages. The dependent variable is the log of average wage of all employees in Columns (1) and (2); the log of average wage of non-executive employees in Column (3); the log of average wage of executives in Column (4). Appendix 3 provides variable definitions and data sources. The sample period covers 2000 – 2012 for Columns (1) – (2) and 2001 – 2012 for Columns (3) – (4). All regressions include firm- and year fixed effects. Bootstrapped standard errors are reported in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

Ln(AWAGE) Ln(AWAGE_NonExe) Ln(AEXEPAY)

VARIABLES (1) (2) (3) (4)

SEO� 0.092** 0.078** 0.099*** 0.003

(0.041) (0.039) (0.035) (0.040)

Ln(NYEAR_LISTED) -0.009 -0.007 -0.015 -0.052***

(0.016) (0.019) (0.016) (0.019)

Ln(MIN_WAGE) 0.332*** 0.294*** 0.281*** 0.164***

(0.045) (0.048) (0.047) (0.048)

LAWSCORE -0.010* -0.009* -0.012** -0.029***

(0.005) (0.005) (0.005) (0.005)

Labor_Law_Effect 0.007** 0.001 0.000 0.013***

(0.003) (0.003) (0.003) (0.003)

Ln(SALES) 0.099*** 0.104*** 0.163***

(0.010) (0.011) (0.008)

RE/TotalEq 0.045*** 0.036*** 0.102***

(0.015) (0.013) (0.012)

ROA 0.219*** 0.185* 0.546***

(0.083) (0.101) (0.135)

Inst_OWN 0.008 0.007 0.168***

(0.027) (0.027) (0.022)

Ln(Tot_Volatility) 0.040* 0.040*** -0.019

(0.021) (0.015) (0.018)

%_LARGEST_SH 0.175*** 0.199*** -0.052

(0.062) (0.065) (0.061)

SALES_GR -0.002 -0.001 -0.038***

(0.010) (0.010) (0.009)

DIV_PR 0.001 0.001 0.005

(0.012) (0.013) (0.005)

%_STATE_OWN 0.095*** 0.085*** -0.011

(0.024) (0.029) (0.026)

%_IND_DIR 0.002 -0.014 -0.036

(0.049) (0.050) (0.045)

%_NONTRD_SH -0.014 0.009 -0.021

(0.037) (0.038) (0.041)

D_PRIVATE_PLACE -0.020 -0.016 -0.011

(0.016) (0.017) (0.016)

Overseas_SEOs -0.103 -0.092 0.119

(0.063) (0.078) (0.082)

Leverage 0.005 -0.009 0.062

(0.046) (0.041) (0.039)

PPE/TA -0.111** -0.078 -0.153***

(0.045) (0.048) (0.047)

Constant -1.283*** -1.633*** -1.464*** -0.002

(0.263) (0.296) (0.315) (0.269)

Firm & Year FE Y Y Y Y

Observations 17,437 17,003 16,071 16,159

58

Table 12: SEO Impact on Total Wages. This table reports the second-stage estimation of the impact of SEOs on total wages. The dependent variable is the log of total wages to all employees in Columns (1) and (2); the log of total wages to all non-executive employees in Column (3); the log of total wages to all executives in Column (4). Appendix 3 provides variable definitions and data sources. The sample period covers 2000 – 2012 for Columns (1) – (2) and 2001 – 2012 for Columns (3) – (4). All regressions include firm- and year fixed effects. Bootstrapped standard errors are reported in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

Ln(Payroll) Ln(Payroll_NonExe) Ln(Payroll_Exe)

VARIABLES (1) (2) (3) (4)

SEO� 0.006 0.013 0.018 -0.021

(0.049) (0.033) (0.037) (0.035)

Ln(NYEAR_LISTED) 0.247*** 0.084*** 0.082*** -0.035*

(0.023) (0.016) (0.019) (0.018)

Ln(MIN_WAGE) 0.195*** 0.014 -0.011 0.113**

(0.052) (0.040) (0.041) (0.050)

LAWSCORE -0.031*** -0.022*** -0.025*** -0.023***

(0.005) (0.003) (0.005) (0.005)

Labor_Law_Effect 0.003 0.000 0.003 0.014***

(0.003) (0.003) (0.003) (0.004)

Ln(SALES) 0.545*** 0.558*** 0.192***

(0.012) (0.014) (0.011)

RE/TotalEq 0.071*** 0.063*** 0.133***

(0.013) (0.016) (0.014)

ROA -0.398*** -0.413*** 0.414***

(0.108) (0.140) (0.129)

Inst_OWN 0.007 -0.002 0.184***

(0.020) (0.022) (0.025)

Ln(Tot_Volatility) 0.001 -0.001 -0.040**

(0.015) (0.017) (0.020)

%_LARGEST_SH 0.206*** 0.227*** -0.087

(0.068) (0.073) (0.059)

SALES_GR -0.100*** -0.099*** -0.040***

(0.010) (0.012) (0.009)

DIV_PR 0.005 0.005 0.006

(0.004) (0.005) (0.005)

%_STATE_OWN 0.195*** 0.198*** -0.004

(0.022) (0.023) (0.027)

%_IND_DIR 0.010 0.002 0.061

(0.035) (0.038) (0.056)

%_NONTRD_SH -0.026 -0.011 -0.088**

(0.029) (0.038) (0.043)

D_PRIVATE_PLACE 0.076*** 0.077*** -0.012

(0.015) (0.016) (0.018)

Overseas_SEOs 0.050 0.025 0.276**

(0.085) (0.086) (0.125)

Leverage 0.223*** 0.236*** 0.151***

(0.041) (0.049) (0.045)

PPE/TA 0.405*** 0.427*** -0.114**

(0.050) (0.049) (0.045)

Constant 2.189*** -0.277 -0.199 -2.271***

(0.313) (0.258) (0.261) (0.320)

Firm & Year FE Y Y Y Y

Observations 17,586 17,147 16,170 16,171

59

Table 13: Pre-Trends. This table reports the results of placebo tests for pre-trends with the full set of control variables. Appendix 3 provides variable definitions and data sources. Robust standard errors clustered at the firm level are reported in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

%_

Production

%_

Staff

%_

Tech_R&D

%_

S&M

%_

Finance

%_

Others

%_

Grad

%_

BA

Ln

(EMP)

Ln

(Inv_Tech_Assets)

Ln

(Total_Patent)t+3

Ln

(AWAGE)

Ln

(Payroll)

VARIABLES (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)

Affected*Year01 0.016 -0.008 0.008 -0.005 0.002 -0.022 0.003 0.002 0.015 0.006 -0.046 -0.025

(0.018) (0.007) (0.008) (0.008) (0.002) (0.021) (0.005) (0.012) (0.032) (0.033) (0.039) (0.032)

Affected*Year02 0.004 -0.003 0.015 -0.010 0.004 -0.026 0.003 0.009 -0.021 -0.048 -0.021 -0.037

(0.020) (0.008) (0.010) (0.010) (0.002) (0.024) (0.005) (0.016) (0.040) (0.039) (0.043) (0.034)

Affected*Year03 0.005 -0.001 0.009 -0.008 0.003 -0.031 -0.001 -0.005 -0.016 -0.050 -0.046 -0.059

(0.021) (0.008) (0.011) (0.011) (0.002) (0.026) (0.006) (0.016) (0.045) (0.044) (0.046) (0.037)

Affected*Year04 -0.004 0.006 0.015 -0.013 0.004 -0.020 -0.002 -0.010 -0.029 0.018 -0.056 -0.038 -0.066

(0.023) (0.009) (0.011) (0.012) (0.003) (0.027) (0.006) (0.018) (0.051) (0.133) (0.049) (0.051) (0.042)

Affected*Year05 0.005 -0.004 0.010 -0.010 0.004 -0.010 -0.000 -0.010 -0.026 -0.103 -0.082 -0.037 -0.064

(0.025) (0.009) (0.012) (0.014) (0.003) (0.029) (0.005) (0.019) (0.053) (0.141) (0.054) (0.052) (0.046)

Firm FE/YearFE Y Y Y Y Y Y Y Y Y Y Y Y Y

Full Set of

Controls Y Y Y Y Y Y Y Y Y Y Y Y Y

Observations 5,765 5,765 4,852 4,700 4,867 5,765 2,083 2,989 5,765 3,160 5,710 5,753 5,839

Adjusted R2 0.752 0.626 0.744 0.878 0.716 0.508 0.881 0.908 0.924 0.741 0.760 0.845 0.945

60

Table 14: Can Past Payout Ratios Explain Our Results? This table reports reestimation results while controlling for the most-recent past three years’ payout ratio with the full set of control variables. Appendix 3 provides variable definitions and data sources. All regressions include firm- and year fixed effects. Bootstrapped standard errors are reported in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

%_

Production

%_

Staff

%_

Tech_R&D

%_

S&M

%_

Finance

%_

Others

%_

Grad

%_

BA

Ln

(EMP)

Ln

(Inv_Tech_Assets)

Ln

(Total_Patent)t+3

Ln

(AWAGE)

Ln

(Payroll)

VARIABLES (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)

SEO� -0.038*** -0.014** 0.026*** 0.025*** 0.002 0.014 0.006** 0.025** -0.064* 0.188* 0.145** 0.070** 0.013

(0.014) (0.006) (0.008) (0.009) (0.002) (0.016) (0.003) (0.011) (0.037) (0.098) (0.069) (0.032) (0.041)

P3_PR 0.006*** 0.002 -0.001 0.000 -0.001*** -0.008*** -0.000 -0.000 0.004 0.003 0.001 0.017*** 0.022***

(0.002) (0.001) (0.001) (0.002) (0.000) (0.002) (0.000) (0.001) (0.006) (0.022) (0.008) (0.005) (0.005)

P3_PR_D 0.005 0.002 -0.009** 0.004 -0.001 -0.005 0.000 -0.008 -0.011 -0.111 0.038 0.035* 0.030

(0.008) (0.004) (0.004) (0.003) (0.001) (0.011) (0.001) (0.005) (0.029) (0.073) (0.028) (0.020) (0.019)

Firm & Year FE Y Y Y Y Y Y Y Y Y Y Y Y Y

Full Set of

Controls Y Y Y Y Y Y Y Y Y Y Y Y Y

Observations 17,006 17,006 13,953 10,609 13,374 17,006 8,121 11,671 17,006 14,468 10,804 16,998 17,142

61

Table 15: Other Robustness Checks. This table reports the second-stage estimation results using alternative instruments and definitions of SEO. Column (1) lists dependentvariables. Only coefficients on the predicted SEO, standard errors, and sample size are reported for each robustness test. Column (2) includes the short-lived 2004 regulation in constructing the instrument. Column (3) turns on the instrument only for firms affected by the 2006 regulation in 2006 and firms affected by the 2008 regulation in 2008. Column (4) uses one-year lag between the beginning of an SEO process and the availability of SEO proceeds. Column (5) relies only on the 2006 regulation to construct the instrument. Column (6) excludes small SEOs with proceeds in the bottom decile. Appendix 3 provides variable definitions and data sources. Appendix 5 reports first-stage estimation results. Bootstrapped standard errors are reported in parentheses. Coefficients marked with *, **, and *** are significant at 10%, 5%, and 1%, respectively.

DEPENDENT

VARIABLES

Including 2004

regulation in IV

construction

IV based on

teatments only

in 2006 and

2008

Using one-

year lag

IV based only

on the 2006

regulation

Excluding

small SEOs

(1) (2) (3) (4) (5) (6)

%_Production -0.037** -0.037** -0.035** -0.037** -0.037**

(0.018) (0.014) (0.015) (0.017) (0.015)

N 17,011 17,011 17,011 17,011 17,011

%_Staff -0.015** -0.014** -0.015*** -0.014** -0.014**

(0.006) (0.007) (0.006) (0.007) (0.007)

N 17,011 17,011 17,011 17,011 17,011

%_Tech_R&D 0.027*** 0.026*** 0.029*** 0.026*** 0.026***

(0.008) (0.007) (0.009) (0.009) (0.009)

N 13,957 13,957 13,957 13,957 13,957

%_S&M 0.026*** 0.026*** 0.025*** 0.026** 0.026***

(0.010) (0.008) (0.005) (0.010) (0.008)

N 10,612 10,612 10,612 10,612 10,612

%_Finance 0.002 0.002 0.002 0.002 0.002

(0.002) (0.002) (0.002) (0.002) (0.002)

N 13,378 13,378 13,378 13,378 13,378

%_Others 0.011 0.012 0.010 0.012 0.012

(0.017) (0.019) (0.019) (0.017) (0.019)

N 17,011 17,011 17,011 17,011 17,011

%_Grad 0.007** 0.006*** 0.008** 0.006** 0.007**

(0.003) (0.002) (0.003) (0.003) (0.003)

N 8,122 8,122 8,122 8,122 8,122

%_BA 0.027*** 0.026*** 0.025*** 0.026*** 0.027***

(0.010) (0.009) (0.009) (0.009) (0.009)

N 11,673 11,673 11,673 11,673 11,673

Ln(EMP) -0.061* -0.070* -0.076* -0.070** -0.077**

(0.032) (0.041) (0.042) (0.034) (0.038)

N 17,011 17,011 17,011 17,011 17,011

Ln(Inv_Tech_Assets) 0.185** 0.188* 0.162 0.188* 0.189** (0.087) (0.099) (0.112) (0.101) (0.093)

N 14,469 14,469 14,469 14,469 14,469

Ln(Total_Patent)t+3 0.147* 0.131** 0.119* 0.131** 0.148**

(0.080) (0.065) (0.071) (0.055) (0.072)

N 10,808 10,808 10,808 10,808 10,808

Ln(AWAGE) 0.077** 0.078** 0.083** 0.078** 0.080**

(0.031) (0.034) (0.037) (0.037) (0.034)

N 17,003 17,003 17,003 17,003 17,003

Ln(Payroll) 0.020 0.013 0.011 0.013 0.009 (0.036) (0.031) (0.043) (0.039) (0.032)

N 17,147 17,147 17,147 17,147 17,147