Outsourcing Decision and Intra-rm Wage Bargaining∗ Yuseob Lee† Department of Economics University of Wisconsin, Madison

This version: November 27, 2017 [Link to the latest version]

Abstract Outsourcing can change the match surplus to be split as well as the rule for splitting the surplus with employees. This study proposes a simple wage bargaining model that tracks down the time variation of revenue, cost, and input variables while taking the outsourcing patterns as given. The model is examined using a rm-level panel data linked to administrative information on income statement and balance sheet provided by the National Tax Service of South Korea. Evidence suggests that outsourcing rms tend to face i) lower bargaining power of workers, ii) a higher xed cost of bargaining failure, and iii) match surplus more responsive to the cost of purchases. These association

is strong for larger rms with 300 or more employees.

Keywords: Outsourcing, Collective bargaining, Wages, Labor costs JEL Classication: J31, L24, M55

∗

I am very grateful to my adviser, Chris Taber for his support and guidance. I also thank all seminar participants at

the University of Wisconsin-Madison for helpful comments and suggestions. This work is preliminary and incomplete. Please do not circulate. All remaining errors are my own. † Department of Economics, University of Wisconsin-Madison, 7230 Social Science Building, 1180 Observatory Drive, Madison, WI 53706. Electronic correspondence: [email protected]

1

1

Introduction

Does outsourcing tilt the bargaining table in favor of an employer? If so, is it more pronounced with production task outsourcing as opposed to service task outsourcing? Outsourcing is essentially an alternative to internalization or integration of associated tasks as has been extensively discussed in the rm boundary literature.1 Technological progress has boosted the outsourcing of various type of tasks over the last decade. We use outsourcing to refer to arm's length transaction of products and services, which is not restricted to the denition of Bhagwati et al. (2004) as arm's length trade of services.2 While varying welfare consequences and labor market outcomes across skill spectrum have been discussed,3 little attention has been paid to how task outsourcing decision shapes intrarm wage bargaining. The empirical diculty lies in that a rm's outsourcing decision determines not only the rule for splitting the surplus with employees but also the match surplus. In this paper, we present evidence on the association between the outsourcing decision and wage bargaining outcomes using an intra-rm wage bargaining model. It has been common to model the total match surplus as the value added net of static replacement cost (Abowd and Lemieux, 1993; Blanchower et al., 1996) or as the value added net of the capital stock that can be liquidated upon the rm's exit (Card et al., 2014; Grout, 1984). These modeling choices depend on how we dene the bargaining failure and its impact on the participating entities. Abowd and Lemieux (1993); Blanchower et al. (1996) assume frictionless worker replacement of the laid-o workers, while Card et al. (2014); Grout (1984) assume instant rm exit and capital liquidation upon the bargaining failure. Instead, our modeling choice is that the production process of a rm temporarily stops when a collective agreement is not reached. Then the match surplus is dened as the value added 1

Vertical integration incentives and outsourcing incentives have been discussed from various point of views: trans-

action cost (Williamson, 1985, 2002), regulation (Autor, 2003), specialization (Grossman and Helpman, 2002; Levine, 2012), incomplete contract and property rights (Grossman and Hart, 1986; Blanchard and Kremer, 1997), organization structure (Garicano, 2000), productivity and coordination cost (Alfaro et al., 2016; Legros and Newman, 2013), nancial constraints (Acemoglu et al., 2009; McMillan and Woodru, 1999). Abraham et al. (1996); Holmstrom and Tirole (1989); Lafontaine and Slade (2007); Acemoglu et al. (2010) provide overview and Dey et al. (2012); Atalay et al. (2014); Katz (1989); Perry (1989) discuss empirical outsourcing pattern.

2

Our denition includes both oshore outsourcing and domestic outsourcing. Oshoring is reserved to indicate

both oshore insourcing and oshore outsourcing. For example, in case of pure oshoring insourcing, global production chain operates across borders but inside a multinational company. Prevalent oshoring practices, boosted by technological progress, have drawn academic attention. (Antras and Helpman, 2004; Acemoglu et al., 2015; Grossman and Rossi-Hansberg, 2008; Antràs, 2003; Antràs and Chor, 2013; Antras, 2015; Antràs et al., 2012; Acemoglu et al., 2007; Helpman, 2006; Nunn, 2007; Yi, 2003; Costinot, 2009)

3

Amiti and Davis (2011); Baumgarten et al. (2013); Becker et al. (2013); Bhagwati et al. (2004); Eckel and

Irlacher (2017); Egger et al. (2015); Feenstra and Hanson (1996, 1999); Groizard et al. (2014); Mitra and Ranjan (2010); Ranjan (2013); Ritter (2014); Sethupathy (2013); Song et al. (2015); Wright (2014) to name a few.

2

net of the cost of shutting down and restarting the production process of a rm. An outsourcing decision is equivalent to a decision on the tasks to be performed internally, which determines the skill mix of employees and the associated cost of temporal shutdown as well as the bargaining power of workers. Using rm-level panel data, we track down realized sequence of input cost, revenue, number of employees, capital stock, and bargaining outcomes to understand rent-sharing itself rather than solving a full-edged model of stochastic dynamic adjustment. Our empirical strategy is to identify parameters specic to each outsourcing pattern, instead of studying the outsourcing incentives.4 Using time variation of revenue and cost through the lens of a simple collective wage bargaining model5 , we study how the match surplus is dened and how it is split between the employer and employees separately for each outsourcing pattern. Our wage bargaining model reduces to a xedeect model that relates rm-level average wage of workers to per-worker revenue, per-worker cost of purchases, and per-worker capital stock. We use the Survey of Business Activities that links administrative nancial information from National Tax Service of South Korea, such as income statement and balance sheet, with other surveyed items including the outsourcing decision. Our empirical analysis boils down to three main conclusion. First, workers' bargaining power is negatively correlated with a rm's outsourcing decision. In our baseline specication, the workers' bargaining power parameter is around 0.10 when the rm does not use outsourcing at all, but it reduces to around 0.03 when the rm use both production and non-production task outsourcing. Second, the cost of shutdown is closely related with outsourcing decision. Firms using outsourcing tend to face large shutdown cost and have match surplus more responsive to the cost of purchases. Third, the association of intra-rm wage bargaining and outsourcing decision is strong for large-sized rms with 300 or more employees. 4

There are a few studies addressing intra-rm bargaining with outsourcing option that primarily aim to understand

outsourcing incentives. Sly and Soderbery (2014) discuss the production allocation decision of multinational rms facing varying worker bargaining power across locations and varying markup across products. Stenbacka and Tombak (2012) study the optimal outsourcing decision when the rm has to bargain with both workers and subcontractors. Lommerud et al. (2009) study deunionization as weakening worker bargaining power and its eect on outsourcing decision.

5

Alternatively, Christodes and Oswald (1992) directly assume specic form of wage equation to gauge the cor-

relation of current wage and past protability.

Hildreth and Oswald (1997) loosely specify underlying bargaining

framework and estimate autoregressive equations to relate current wage to past protability and wage sequence. With emphasis on labor adjustment cost, Saint-Paul (1995); Ljungqvist (2002) show that bargaining outcome depend on the regulatory cost.

3

The rest of this study is structured as follows: Section 2 describes our collective wage bargaining model and estimation strategy. Section 3 provides descriptive statistics for the data set and describes the institutional settings. The last section concludes. The Appendix contains additional details and tables.

2

A Simple Model of Wage Bargaining

2.1

Model Setup

This section describes a simple wage bargaining model between a rm and a collection of workers. In reality, facing idiosyncratic and macroeconomic uctuations, a rm will solve the dynamic labor and capital adjustment problem in addition to the factor allocation problem governing the scope and intensity of task outsourcing, while acknowledging the upcoming wage bargaining with its employees. The lumpy capital adjustment due to xed adjustment cost and the lop-sided labor adjustment cost due to institutional regulations make it costly to write down and solve a fulledged model.6 Assuming that all bargaining participants take the short-term input and output level as given, we can use our stripped down version of wage bargaining model to study rent-sharing inside a rm. Time is discrete. Firms have access to the same production technology but the skill level of employees depends on the outsourcing and other production choice. Fix a rm, say j ∈ J , and denote its workers with index i ∈ Ij, t where Ij, t represents the set of workers employed by rm

j at period t. A worker i earns wage wi, t with reservation wage bi, t representing one's skill level. P Firm j 's total wage bill i∈Ij, t wi, t is determined through contemporaneous7 collective bargaining given the rm's short-term input and output level. We assume that bargaining failure shutdown 6

Factor adjustment model has extensive studies from the labor demand and investment literature (Nickell, 1986;

Hamermesh, 1986, 1996; Cooper and Willis, 2009; Asker et al., 2014; Asphjell et al., 2014; Caballero et al., 1997; Cooper and Haltiwanger, 2006; Bond and Van Reenen, 2007; Hall, 2004; Blatter et al., 2012). Eberly and Van Mieghem (1997) show that factor adjustment does not happens unless the shock is suciently large in the presence of adjustment cost. As was discussed in Bloom et al. (2007); Bloom (2009), investment becomes less responsive to demand shocks facing higher uncertainty.

7

Since the body of workers changes over time, period-wise bargaining is our modeling choice rather than bargaining

on long-term contract. Also yearly wage negotiation is common in Korea in case that establishment is covered by collective agreement. On the other hand, Hildreth and Oswald (1997); Christodes and Oswald (1992) show that past protability is realized as long-run changes in wages. This could be accrued to existing long-term employment contracts before renegotiation.

4

the production process at the rm for a period rather than inducing the rm to exit the market for good.8

Mj, t represents the cost of purchases for rm j at period t. During the elapsed period of shutdown, θ ∈ [0, 1] fraction of purchases becomes obsolete9 and existing supply chains break down to incur additional cost to resume the normal operation of a rm. Moreover, further maintenance and worker training schedule are indispensable to restart the plant. Γ(Nj, t , Kj, t , Mj, t ) indicates the sum of all the costs associated with restarting this mothballed production process as well as the expected value of foregone future revenue streams. Nj, t and Kj, t indicate the number of workers and the stock of capital for rm j at period t. Note that the more costly the shutdown to both parties, the larger is the match surplus to split. We can write the surplus of an operating rm j as opposed to period shutdown as (Rj, t − rt Kj, t −

X

wi, t − Mj, t ) − (−rt Kj, t − θMj, t − Γ(Nj, t , Kj, t , Mj, t ))

i∈Ij, t

=Rj, t −

X

(1)

wi, t − (1 − θ)Mj, t + Γ(Nj, t , Kj, t , Mj, t )

i∈Ij, t

where Rj, t is the period revenue,

P

i∈Ij, t

wi, t is the total labor expense, and rt Kj, t is the total

capital expense when rt is the period rental rate of capital. Physical capital is assumed to be irreversible in the short run and does not constitute the rm's surplus directly. The workers' surplus as opposed to the mass layos is X 

1 EVi, t+1 1+ρ



X

1 EUi, t+1 1+ρ i∈Ij, t i∈It  X  1 = (EVi, t+1 − EUi, t+1 ) wi, t − bi, t + 1+ρ wi, t +

−



bi, t +

(2)

i∈Ij, t

where ρ > 0 is the common time discount rate. EVi, t+1 and EUi, t+1 represent the expected option 8

Had we assumed that the rm exit the market instead, the extent that the rm can liquidate current capital stock

will alter the bargaining outcome, as was embedded in Card et al. (2014) to study hold-up problem in two-period investment model. If we assume that

τ

fraction of period will be shutdown instead, then our identication would

be dierent. In particular, we will identify

τβ

instead of worker bargaining power

β

by associating time variation of

revenue and wages.

9

Some supply contracts are long-term contract to enable rm-specic design and necessary technology transfer for

intermediate inputs.

θ

could be interpreted as the degree that the purchased intermediate inputs are relation-specic.

To sell them to a third party, additional costs are entailed because extra treatments and quality verication are required to make the intermediate inputs compatible with another product specications and to enable transaction itself.

5

value for i when employed and unemployed respectively. If the agreement is not reached, all current employees are laid o at the end of period. Lower reservation wages or lower values of unemployment contribute to larger workers' surplus. Then, the total surplus of current matches is dened by Ωj, t

=

Rj, t − (1 − θ)Mj, t + Γ(Nj, t , Kj, t , Mj, t )  X  1 − bi, t − (EVi, t+1 − EUi, t+1 ) 1+ρ

(3)

i∈Ij, t

and generalized Nash bargaining between the collection of workers and the rm yields10 X

wi, t

=

i∈Ij, t

X 

bi, t −

i∈Ij, t

=

(1 − β)

X i∈Ij, t

 1 (EVi, t+1 − EUi, t+1 ) + βΩj, t 1+ρ bi, t −

1−β X (EVi, t+1 − EUi, t+1 ) + βRj, t − β(1 − θ)Mj, t 1+ρ

(4)

i∈Ij, t

+βΓ(Nj, t , Kj, t , Mj, t )

where β ∈ (0, 1) represents the (collective) bargaining power of workers. Dividing both sides of (4) by the rm size Nj, t , we have 1−β ¯ ¯j, t+1 ) + β Rj, t − β(1 − θ) Mj, t + β Γ(Nj, t , Kj, t , Mj, t ) (Vj, t+1 − U w ¯j, t = (1 − β)¯bj, t − 1+ρ Nj, t Nj, t Nj, t

where w ¯j, t =

P

i wi, t /Nj, t

(5)

P is the rm-level average wage, ¯bj, t = i bi, t /Nj, t is the average reser-

P ¯j, t+1 = vation wage, V¯j, t+1 = i EVi, t+1 /Nj, t is the average option value of employment, and U P i EUi, t+1 /Nj, t is the average option value of unemployment. Next, we introduce notations for outsourcing-specic parameters. χj, t = 0, 1, 2, 3 indicate the rm j 's decision of no outsourcing, non-production task outsourcing only, production task outsourcing only, and both production and non-production task outsourcing, respectively. Then (5) becomes w ¯j, t = (1 − βχ )¯bj, t − 10

1 − βχ ¯ ¯j, t+1 ) + βχ Rj, t − βχ (1 − θχ ) Mj, t + βχ Γχ (Nj, t , Kj, t , Mj, t ) (6) (Vj, t+1 − U 1+ρ Nj, t Nj, t Nj, t

We can conceive of pairwise bargaining following Brügemann et al. (2015); Stole and Zwiebel (1996a,b) instead;

however, pairwise bargaining requires extra assumptions on production technology. In particular, we need complete knowledge of worker's output contribution in all feasible combination of workers when workers are not homogeneous. Similar to Manning (1987), sequential bargaining by task type is another possible bargaining model. Still we need to gauge the contribution of each task category to understand the bargaining outcome. We conceive our discussion based on collective bargaining model as a parallel to growth accounting idea.

6

where χ = χj, t for notational simplicity. For later use, we parametrize the shutdown cost with

Γχ (N, K, M ) = γ0, χ + γ1, χ N + γ2, χ K + γ3, χ N K + γ4, χ M

(7)

to get11 w ¯j, t

1 − βχ ¯ ¯j, t+1 ) + βχ Rj, t − βχ (1 − θχ ) Mj, t (Vj, t+1 − U (1 − βχ )¯bj, t − 1+ρ Nj, t Nj, t   1 Kj, t Mj, t +βχ γ0, χ + γ1, χ + γ2, χ + γ3, χ Kj, t + γ4, χ Nj, t Nj, t Nj, t 1 − βχ ¯ ¯j, t+1 ) + βχ Rj, t + βχ (γ4, χ + θχ − 1) Mj, t = (1 − βχ )¯bj, t + βχ γ1, χ − (Vj, t+1 − U 1+ρ Nj, t Nj, t Kj, t 1 +βχ γ2, χ + βχ γ0, χ + βχ γ3, χ Kj, t Nj, t Nj, t

=

(8)

where γ0, χ represents the xed cost of shutdown conditional on the outsourcing choice χ. The contribution of the rm size, capital stock, interaction of rm size and capital stock, and cost of purchases to the cost of shutdown are represented by γ1, χ , γ2, χ , γ3, χ , γ4, χ . Note that we cannot separately identify γ4, χ and θχ . This is because a temporal shutdown not only deteriorates some fraction of purchased components and materials but also breaks down the existing input supply chains.

2.2

Estimation Strategy

We intend to translate (8) into a xed-eect framework, 

   Rj, t Mj, t w ¯j, t = µj + δχ + αt + λ1, χ + λ2, χ Nj, t Nj, t   Kj, t 1 + λ3, χ + λ4, χ + λ5, χ Kj, t + j, t Nj, t Nj, t

(9)

where χ = χj, t . The key identifying assumption is strict exogeneity when unobserved heterogeneity µj and idiosyncratic disturbance j, t are dened by 1 − βχ ¯ ¯j, t+1 ) µj + δχ + αt + j, t = (1 − βχ )¯bj, t + βχ γ1, χ − (Vj, t+1 − U 1+r 11

Alternatively, we use

γ4, χ M + γ5, χ K 2

γ1, χ N + γ2, χ K + γ3, χ N K + γ4, χ M + γ5, χ N M

as our robustness check in the Online Appendix.

7

and

γ0, χ + γ1, χ N + γ2, χ K + γ3, χ N K +

That is12 E [j, t |µj, χ , δχ , {ατ , Rj, τ , Mj, τ , Nj, τ , Kj, τ }τ ≤T ] = 0, t = 1, · · · , T

With a sequence of {Rj, t , Nj, t , Kj, t , Mj, t , w ¯j, t , χj, t }j∈J, t≤T , our xed-eect estimator enables the

ˆ 1, χ . Other point estimates identication of the collective bargaining power of workers via βˆχ = λ of model parameters can be recovered through non-linear transformations: θˆχ + γˆ4, χ =

γˆ0, χ =

ˆ 4, χ λ ˆ 1, χ , λ

γˆ2, χ =

ˆ 3, χ λ , ˆ λ1, χ

and γˆ3, χ =

ˆ 5, χ λ ˆ 1, χ . λ

ˆ 2, χ λ ˆ 1, χ λ

+ 1,

In turn, the corresponding standard errors are calculated

using the delta method. Our estimation strategy and reduced form equation (8) share some key features of investment model of Card et al. (2014) which solves a two-period investment problem to derive testable implications. They relates individual wage with the rm characteristics and reservation wage. Using matched employer-employee data, Card et al. (2014) estimated bargaining parameters with a proxy for reservation wage.13 Instead, we use rst dierences to single out the term containing individual reservation wages. Our empirical strategy shares some features of Hildreth and Oswald (1997); Christodes and Oswald (1992) in that time variation is the source of identication given rm-level or establishment-level panel data. Their primary focus is on the translation of the productivity enhancement into long-run changes in wage. 12

A rm has certain mix of worker skill levels depending on its outsourcing choice, which determines the average

reservation wage. If the skill mix is stable over time conditional on the outsourcing choice, then we expect that this strict exogeneity assumption would not be too extreme.

13

With the current set of notations, the testable implication of Card et al. (2014) is

wi, t = (1 − β)bi, t + β where

π

  Kj, t+1 Kj, t Rj, t 1 (1 − τ )(1 − π)rt+1 − β πrt + Nj, t Nj, t 1+r Nj, t+1

is the fraction of capital stock that can be liquidated and

τ

is the fraction of scheduled investment that

can be costlessly annulled. With matched employer-employee data, the authors estimate relevant parameters using log-linearized equation

Kj, t φ4 + ξi, t Nj, t characteristics xi, t are used

log wi, t = log(bi, t )φ1 + x0i, t φ2 + V Aj, t φ3 +

V A in place of revenue. Various individual to capture bi, t . Estimated ˆ φ represents the elasticity of wage with respect to rents and ˆ4 represents the degree of capital liquidation. φ3

with value added

φˆ3 V Aj, t

8

3

Institutional Setting and Data

3.1

Institutional Setting

The union density14 of South Korea is approximately 10% which is even lower than that of the U.S. (Figure 1) Coverage rate of collective agreement is approximately 12% in Korea which is comparable to the U.S. rate. Such low level of coverage rate beguiles the indirect collective bargaining coverage due to establishment-level

rules of employment

that is uniformly applied to all workers of the

same kind of job inside an establishment(Park et al., 2010) It is illegal for a rm to provide dierent compensation packages to two comparable workers doing the same kind of job.15 The dominant form of wage bargaining has been rm-level negotiation, except hospital, transportation, and banking sectors, due to employer's reluctance to sector-wide negotiations (Grubb et al., 2007; Yoon et al., 2009)

10

10

Coverage Rate(%) 15 20 25

Union Density(%) 15 20 25

30

30

Figure 1: Union Density and Collective Agreement Coverage Rate

2000

2005

2010

2015

2000

2005

Year Germany Korea United States

2010 Time

Japan United Kingdom

Germany Korea United States

Source: OECD(2016) "Trade Unions: Trade union density", KLI(2017) "Labor Statistics Archive"

Japan United Kingdom

Source: OECD(2016) "Trade Unions: Trade union density"

Prevalent use of xed-term contracts and temporary help agency workers has segmented Korean labor market and initiated major labor reform of 2006. This reform was two-folded: i) to enforce equal treatment of permanent and temporary/xed-term workers performing the same task ii) to restrict the length of continuous use of xed-term contracts to be less than 2 years. Its implementation has been gradual over the following years based on the number of employees. Firms 14 15

Union density indicates the fraction of trade union members among wage and salary earners. Rules of employment stipulates various aspects of employment contract ranging from working conditions including

pecuniary and non-pecuniary compensations to layos. Korean Labor Standard Act requires that rules of employment should not conict with collective agreement(Article 96) and Trade Union And Labor Relations Adjustment Act requires that a collective agreement applies to all workers of the same kind of job if it applied to a majority of workers of the same kind of job(Article 35)

9

with 300 or more employees had been regulated from mid 2007, 100 or more from mid 2008, and 5 or more from mid 2009 while any newly signed, renewed, or revised contracts were immediately under the restrictions. The regulation package and its variable enforcement can be embedded in our framework by introducing new bargaining power parameters βχ, τ where τ = 0, 1 indicate without and with enforced regulations.

3.2

Data

The main data set is the Survey of Business Activities(SBA) of South Korea from 2006 to 2013 that interviews all rms with at least 50 full-time employees and with equity value larger than $270,000 USD approximately.16 The SBA population is dened based on the Census on Establishment of Korea that surveys all existing establishments.17 From 2010, survey data were linked to administrative information on income statement and balance sheet provided by the National Tax Service of Korea. 2008 and 2009 surveys were revised using linked administrative information. The SBA includes breakdown of the asset, capital, and debt along with breakdown of the revenue, cost of goods sold, and selling and administrative expenses. Moreover, the indicator variables for various outsourced tasks as well as the number of employees are available. Only rms that can be identied across all years from 2006 to 2013, without missing information on key variables, constitute our balanced panel data set. Firms that can be identied across multiple years constitutes our unbalanced panel data set. We restrict our sample to manufacturing rms in the balanced panel data set. Our detailed sample selection steps and the resulting loss of observations are reported in the Appendix A. It is a nontrivial task to dene a rm's industry because a rm is typically composed of multiple establishments. The SBA dene a rm's industry to be the highest revenue generating industry of establishments, based on the establishment-level revenue aggregation. Depending on the performance of establishments, the highest revenue generating industry might change over time. We use the initially observed industry category of a rm to represent the rm's industry. By the same token, initially observed number of employees is used to dene the rm size category of a rm. 16

If the equity value exceeds $900,000 USD approximately, a rm with at least 5 full-time employees is included

for wholesale and retail trade and other service industry.

17

Responses to the SBA are enforced by law and survey response rates are relatively high. For example, only 106

out of 13,096 rms in 2012 and 90 out of 13,053 rms in 2013 refused interview.

10

All cost and revenue variables are converted to 2010 KRW (in millions) using the Consumer Price Index. The SBA denes outsourcing as the use of long-term contracts with outside rms to perform certain tasks. Responding rms indicate all currently outsourced tasks.18 Panel (a) of Figure 2 shows how detailed task outsourcing pattern has evolved across years for manufacturing and nonmanufacturing rms. The growth of production, distribution, and maintenance task outsourcing has been pronounced for manufacturing rms while that of maintenance, human resource management, and distribution task outsourcing has been strong for non-manufacturing rms. Panel (b) of Figure 2 shows the trends among rms outsourcing at least one task. Since rms outsource multiple tasks at the same time, Figure 3 collapses detailed task outsourcing patterns into joint outsourcing indicators for production task and non-production task. We denote no outsourcing as χ = 0, outsourcing only non-production task as χ = 1, outsourcing only production task as χ = 2, and outsourcing both tasks as χ = 3. Panel (a) shows the trends for manufacturing rms across rm size categories. The qualitative feature is that the combination of production task outsourcing and non-production task outsourcing becomes prevalent regardless of rm size. Panel (b) shows the corresponding trends for non-manufacturing rms. The majority of outsourced tasks are non-production tasks and the share of rms not relying on outsourcing at all is on the decline. 18

Originally, rms indicate their outsourcing practice using the twelve task categories.

We collapse some task

categories for the ease of presentation. For example, the HRM category includes hiring, training, and benet administration tasks.

11

Figure 2: Task Outsourcing Trend: 2006-2013

Non−manufacturing .8

.8

Manufacturing distribution

financial

HRM

IT

distribution HRM

IT

maintenance

production

maintenance

production

.4 .2 0

0

.2

.4

.6

research

.6

research

financial

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

Source: Survey of Business Activities(2015), Balanced Panel of Firms (a) All Firms

Non−manufacturing .8

.8

Manufacturing distribution

financial

HRM

IT

distribution HRM

IT

maintenance

production

maintenance

production

.4 .2 0

0

.2

.4

.6

research

.6

research

financial

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

Source: Survey of Business Activities(2015), Balanced Panel of Firms Using Outsourcing (b) Firms Outsourcing at Least One Task

Table 1 reports the descriptive statistics for our main sample. The rst column shows the moments for all observations without missing information on the key variables. Next two columns 12

report the moments for unbalanced panel and the last two columns for the balanced panel data set. The rst four rows report the number of rms, rm by year observations, employees, and average number of employees. Then rm characteristics are reported as averages weighted by the number of employees. The bottom rows report the sample fraction of rms using specic task outsourcing. Outsourcing of production task and distribution task is relatively more common for manufacturing rms. The value added per worker is comparable between manufacturing and non-manufacturing rms but the capital intensity and cost of purchases per worker are much higher for manufacturing rms. The average wage is higher for manufacturing rms. The rst panel of gure 4 (a) shows that the average revenue per worker has been highest for rms outsourcing only non-production tasks, which converges to that of rms outsourcing both tasks. The average revenue per worker for rms outsourcing nothing and that for rms outsourcing only production task converge to a low level. Similar pattern persists in the second panel representing the per-worker value added and the fourth panel representing the average labor cost per worker. The third panel shows that rms outsourcing non-production task or both tasks are relatively large in size. The fth panel shows that the overall capital intensity is on the rise and rms outsourcing only non-productive tasks are operating under relatively more capital-intensive production technology. Capital intensity for rms that do not use outsourcing at all show the lowest level of capital stock per worker. Figure 5 and Figure 6 show corresponding average characteristics for manufacturing rms across size categories, where qualitative features persist. Figure 4 (b) shows the characteristics of non-manufacturing rms over time. Compared to nonoutsourcing rms, those outsourcing only non-production task or both tasks show higher average revenue per worker, value added per worker, labor cost per worker, and capital stock per worker as well as rm size itself. Firms outsourcing only production tasks show distinctive pattern mainly due to their small sample size.

13

Table 1: Descriptive Statistics for Firms Sample

all

Industry

unbalanced

unbalanced

balanced

balanced

manuf.

non-manuf.

manuf.

non-manuf.

(1)

(2)

(3)

(4)

(5)

Number of rms

16,113

7,693

6,560

3,650

2,530

Number of rm by year

83,553

44,544

36,449

29,200

20,240

Number of employees(thousands)

20,181

8,183

8,952

6,436

6,370

Average number of employees

241.53

183.71

245.61

220.42

314.71

Characteristics of rms

a

Revenue/worker (million KRWs) Value added/worker

347.87

448.43

301.35

461.50

(579.38)

(463.01)

(716.97)

(458.51)

72.05

74.54

71.10

78.71

(million KRWs)

(124.79)

(89.31)

(152.85)

(87.07)

Capital stock/worker

161.04

202.07

133.98

215.99

(million KRWs)

(408.28)

(266.87)

(501.97)

(261.36)

Purchase/worker (million KRWs) Labor cost/worker (million KRWs)

I{outsourcing I{outsourcing

production task}

non-production task}

I{outsourcing

IT task}

I{outsourcing

research/design task}

I{outsourcing I{outsourcing

HRM task}

distribution task}

I{outsourcing

nancial task}

I{outsourcing

maintenance task}

165.97

252.59

113.13

262.98

(430.01)

(414.11)

(487.14)

(427.67)

41.53

44.79

40.34

46.45

(26.87)

(20.90)

(31.77)

(21.20)

0.25

0.47

0.07

0.49

(0.43)

(0.50)

(0.26)

(0.50)

0.73

0.77

0.69

0.80

(0.44)

(0.42)

(0.46)

(0.40)

0.16

0.15

0.13

0.18

(0.37)

(0.36)

(0.34)

(0.38)

0.14

0.13

0.13

0.15

(0.35)

(0.34)

(0.33)

(0.36)

0.24

0.20

0.26

0.22

(0.43)

(0.40)

(0.44)

(0.42)

0.31

0.45

0.19

0.48

(0.46)

(0.50)

(0.39)

(0.50)

0.12

0.10

0.15

0.09

(0.33)

(0.30)

(0.36)

(0.29)

0.45

0.50

0.39

0.54

(0.50)

(0.50)

(0.49)

(0.50)

309.63 (706.68) 72.61 (143.54) 138.24 (356.82) 116.68 (456.13) 41.73 (32.27)

0.08 (0.27) 0.71 (0.45) 0.15 (0.36) 0.14 (0.35) 0.26 (0.44) 0.22 (0.41) 0.13 (0.34) 0.44 (0.50)

Notes: Sample in column 1 includes all yearly observations of rms with at least 5 employees in all observed years between 2006 and 2013. Sample in column 2 and 3 includes manufacturing and non-manufacturing rms observed at least twice between 2006 and 2013 with 5 or more employees in each observed year. Sample in column 4 and 5 includes manufacturing and nonmanufacturing rms observed in all years between 2006 and 2013 with 5 or more employees in each year. a weighted by the number of employees. Parenthesis represent (weighted) sample standard deviation.

14

4 4.1

Estimation Results Baseline Model

In essence, the time variation of rm-level average wage and per-worker revenue identify the bargaining power parameter βˆ holding others variables constant. The estimates for the rest of model parameters are derived from the xed-eect estimates and the estimated bargaining power parameter in turn. Corresponding standard errors are calculated using the delta method. As a reference point, model parameters of (5) are estimated using OLS and xed-eect regression framework whose parameters of interest are {β, θ, γ0 , γ1 , γ2 , γ3 }. Table 3 reports the OLS estimates and derived model parameter estimates in the rst column. The xed-eect estimates with and without year dummies are reported in the adjacent columns. The OLS and xed eect estimates show noticeable dierences in magnitude. In particular, the OLS estimate for the worker bargaining power parameter βˆ is smaller than the xed-eect counterpart. The OLS estimate for xed cost of shutdown γˆ0 is negative and not statistically signicant while the corresponding estimate from xed-eect regression is positive and statistically signicant across alternative specications. The estimated marginal contribution of capital to the cost of shutdown γˆ2 is higher with the OLS estimates compared to the xed-eect estimates. The signs for γˆ3 are negative and statistically signicant across the specications. Using the balanced or unbalanced panel data set lead to noticeable dierences in estimates while including year dummies does not substantially alter our estimates. The xed-eect estimates from the balanced panel data set show slightly lower bargaining power. Using the unbalanced panel data set leads to relatively low level of estimated xed cost of shutdown γˆ0 and low level of estimated marginal contribution of capital to the cost of shutdown γˆ2 . This could be accrued to the fact that the rms appearing only in the unbalanced panel data set are typically downsizing or closing down their businesses. Estimated θˆ+ γˆ4 are similar in magnitude from the balanced and unbalanced panel data set. The magnitude of our estimated bargaining power βˆ shows that wage is not very responsive to the revenue. Every dollar increase of revenue per employee is associated with ve-cent increase of average wage inside a rm. In our estimating equation, we cannot separately identify θ and γ4 and 15

we should interpret the estimates with caution. For example, θˆ + γˆ4 ≈ 0.79 is from the deterioration of purchased inputs as well as break-down of existing input supply chains when the bargaining reaches an impasse. Still we can interpret the estimates as the contribution of the cost of purchases to the total surplus,

∂Ωj, t ∂Mj, t

= −1 + θ + γ4 which is estimated to be around −0.21. Heavy dependence

on purchased components and materials decrease the dened surplus of matches other things remain the same. The magnitude of γˆ0 indicates that the xed cost of shutdown is not only statistically signicant but also not negligible in size. Estimated γˆ2 show that rms with higher capital intensity face higher cost of shutdown. For example, the estimated γˆ2 ≈ 1.3 from the balanced panel data set implies that every dollar investment in physical capital increases the shutdown cost by 1.3 dollars. The negative sign of γˆ3 shows that the marginal contribution of capital to the cost of shutdown is decreasing in rm size. By the same token, the marginal contribution of rm size to the cost of shutdown is decreasing in capital stock. The baseline specication of (9) can be estimated using xed-eect regression model. The corresponding parameters of interest are

{βχ , θχ , γ0, χ , γ1, χ , γ2, χ , γ3, χ , γ4, χ : χ = 0, 1, 2, 3} Table 2 reports only the model parameter estimates derived from the actual regression estimates. The rst column shows the estimates derived from the OLS estimates and the second and third column show those derived from the xed eect estimates. The use of unbalanced and balanced panel data set lead to similar bargaining power parameter but other parameters in Γχ (·) show some non-negligible dierences. In particular, the estimated xed cost of shutdown γˆ0 is much smaller in unbalanced panel data set. Since the rms appearing only in the unbalanced panel data set are typically downsizing or closing their businesses, the cost of shutdown as well as match surplus are dened in dierent ways. The temporal shutdown would not work as an eective threat to downsizing rms. It is noteworthy that the estimated workers' bargaining power βˆ is negatively associated with a rm's outsourcing decision. For example, when not using outsourcing at all, an employer split around 10% of match surplus to workers; however, when using both types of outsourcing, an employer split around 3% of match surplus instead. The association of θˆχ + γˆ4, χ with outsourcing decision χ = χj, t 16

shows another interesting pattern. The estimated θˆχ + γˆ4, χ is the lowest for non-outsourcing rms (χ = 0) and gets higher for rms using outsourcing. In particular, rms using production task outsourcing (χ = 2, 3) show higher level of estimates compared to rms using only non-production task outsourcing (χ = 1). Interpreting the estimates as the contribution of the cost of purchases to the total surplus.

∂Ωj, t ∂Mj, t

= −1 + θ + γ4 , we see that every dollar purchase increases the total surplus

by 16 to 20 cents for rms using production task outsourcing (χ = 2, 3). The estimated xed cost of shutdown γˆ0, χ show that the temporal shutdown is more costly for rms using outsourcing compared to a seemingly similar rms not using outsourcing at all. In terms of marginal contribution of capital stock to temporal shutdown cost γˆ2 , we observe that rms using only non-production task outsourcing face shutdown cost more sensitive to capital stock.

4.2

Size

Now we separately consider rms with dierent size. We divide the rms into ve size categories denoted by ξ = 1, 2, 3, 4, 5 with a choice of cutos to make categories similar in the number of observations. Then the parameters of our interest are

{βχ, ξ , θχ, ξ , γ0, χ ξ , γ1, χ ξ , γ2, χ ξ , γ3, χ ξ , γ4, χ ξ : χ = 0, 1, 2, 3 and ξ = 1, 2, 3, 4, 5} Table 4 shows only estimates for model parameters derived from the xed eect estimates using the balanced panel data set. It is noteworthy that the rst four columns from small and mediumsized rms are close to each other but the fth column from large-sized rms show distinctive features. In particular, the association of bargaining power and outsourcing decision is small or negligible for small and medium-sized rms, whereas it is strong for large-sized rms. For example, a dollar increase of revenue per employee translates into one to two cents wage increase for small and medium-sized rms. For a large-sized rm not using outsourcing at all (χ = 0), it becomes 20 cents; however, the magnitude decreases to 4 cents when both production and non-production tasks are outsourced (χ = 3).

17

Table 2: The OLS and xed-eect estimates of intra-rm wage bargaining model (1) Sample Model

(2)

(3)

unbalanced

unbalanced

balanced

OLS

xed eect

xed eect

Outsourcing none

(χ = 0)

β θ + γ4 γ0 γ2 γ3

non-prod. only

(χ = 1)

β θ + γ4 γ0 γ2 γ3

prod. only

(χ = 2)

β θ + γ4 γ0 γ2 γ3

both

(χ = 3)

β θ + γ4 γ0 γ2 γ3 N. Obs.

0.0590

0.0913

0.0985

(0.0015)

(0.0015)

(0.0021)

0.2452

0.6081

0.6060

(0.0172)

(0.0146)

(0.0176)

4,577

5,528

5,217

(1,044)

(829)

(1,297)

0.2186

0.2748

1.1335

(0.0276)

(0.0182)

(0.0535)

-0.0005

-0.001

-0.0028

(0.0002)

(0.0001)

(0.0002)

0.0335

0.0564

0.0612

(0.0006)

(0.0008)

(0.0009)

0.2113

0.7265

0.7699

(0.0121)

(0.0079)

(0.0076)

-7,451

-1,004

17,195

(1,182)

(1,042)

(1,478)

2.4986

2.0991

2.3258

(0.0618)

(0.0429)

(0.0504)

-0.0025

-0.0035

-0.003

(0.0001)

(0.0001)

(0.0001)

0.0180

0.0385

0.0390

(0.0017)

(0.0017)

(0.0020)

0.6196

1.0823

1.1625

(0.0907)

(0.0511)

(0.0601)

-603

16,451

20,535

(3,787)

(2,347)

(3,405)

0.8116

0.9846

1.5597

(0.1853)

(0.1002)

(0.1444)

0.0007

-0.0023

-0.0028

(0.0004)

(0.0002)

(0.0003)

0.0234

0.0336

0.0303

(0.0008)

(0.0009)

(0.0011)

0.6082

1.0904

1.1944

(0.0274)

(0.0236)

(0.0302)

-529

19,218

30,037

(1,881)

(1,929)

(3,223)

0.6343

0.5638

0.8573

(0.0529)

(0.0422)

(0.0713)

0.0003

-0.0015

-0.001

(0.0001)

(0.0001)

(0.0002)

44,544

44,544

29,200

Notes: Dependent variable in all specications is labor cost per worker that includes benet and pension cost as well as wage cost. A rm's industry is dened to be the initial industry to which the rm belongs. Column 1 corresponds to OLS estimates and column 2 and 3 correspond to xed-eect estimates. The rst two columns include all manufacturing rms observed at least twice between 2006 and 2013 with 5 or more employees in each observed year. The third column includes all manufacturing rms observed in all years between 2006 and 2013 with 5 or more employees in each year.

18

4.3

Labor Reform

Next we introduce regulation into our model. The 2006 labor reform and its variable enforcement are introduced by regulation dummy τ ∈ {0, 1}. Firms with 300 or more employees are regulated from mid 2007, 100 or more from mid 2008, and 5 or more from mid 2009.19 Then the parameters of our interest are

{βχ, ξ, τ , θχ, ξ, τ , γ0, χ ξ, τ , γ1, χ ξ, τ , γ2, χ ξ, τ , γ3, χ ξ, τ , γ4, χ ξ, τ : χ = 0, 1, 2, 3 , ξ = 1, 2, 3, 4, 5, and τ = 0, 1}

Table 5 shows the estimated model parameters. Introduction of labor reforms does not alter the qualitative patterns of our bargaining model estimates. The enforced regulation is associated with higher workers' bargaining power. The magnitude of such associations is the largest for rms not using outsourcing (χ = 0) or rms using only non-production task outsourcing (χ = 1). Among rms using production task outsourcing (χ = 2, 3), the magnitude is small.

4.4

Discussion

Since our empirical framework does not model the endogenous outsourcing decision, estimated parameters do not allow causal inferences. We describe possible directions of causality using our observed correlations. The dened match surplus is dierent across outsourcing patterns since the cost of shutdown Γχ (·) depends on a rm's outsourcing decision χ = χj, t . With our estimated baseline model, the xed cost of shutdown γ0 is the lowest but the workers' bargaining power β is the highest for non-outsourcing rms (χ = 0), whereas the xed cost of shutdown γ0 is the highest but the workers' bargaining power is the lowest β for rms outsourcing both tasks (χ = 3). This might imply that rms with high shutdown cost are more prone to adopt outsourcing of various tasks as a way to suppress workers' bargaining power. Alternatively, rms facing lower workers' bargaining power might tend to adopt business practices that utilize various task outsourcing and entail relatively high xed cost of shutdown. The heavy dependence on purchased components and materials increase the dened match surplus of (3). Our estimated θ + γ4 show that such association is the weakest for non-outsourcing 19

Unlike the size denition for

ξ ∈ {1, 2, 3, 4, 5},

we used the concurrent number of employees to dene regulation

enforcement.

19

rms (χ = 0) but the strongest for rms outsourcing both tasks (χ = 3). The implication might be that rms having match surplus which is highly responsive to the cost of purchases are more likely to outsource tasks to lower workers' bargaining power. Alternatively, rms with lower bargaining power might be more prone to adopt outsourcing practices that make match surplus more responsive to cost of purchases.

5

Conclusion

Widespread outsourcing practice has drawn academic attention to the long-term eect of outsourcing on unemployment and wage inequality. Either from oshore sourcing that provides access to cheap foreign labor or domestic sourcing that facilitate the replacement of internally performed tasks with specialized outside contractors or rms, newly available alternatives enhance the protability of all adopting rms. Therefore, the majority of earlier studies has emphasized the varying welfare consequences across worker skill levels rather than intra-rm bargaining. With adopted outsourcing practice, on the other hand, rms would not only end up splitting the rents from enhanced profitability but also redening the rents to be split through wage bargaining since the position of rms and workers when wage bargaining reaches an impasse is determined by outsourcing practice. We use a rm-level panel data from South Korea to estimate a wage bargaining model that denes the bargaining failure as a temporal shutdown of a rm, which later incurs cost to restart the mothballed production process as well as a loss to future revenue stream. We nd evidence that rent-sharing is correlated with outsourcing pattern through i) dierences in bargaining power and ii) the dened match surplus. These correlations are distinctively observed for rms with 300 or more employees. Our empirical analysis provides interesting correlations but are limited in providing causal inferences. Further empirical studies on how a rm's investment decision is associated with outsourcing decision are required to better understand the various sources of prevalent outsourcing practices, which should contribute as a stepping stone to build an equilibrium model that endogenize the outsourcing choice.

20

References Abowd, J. A. and Lemieux, T. (1993). The eects of product market competition on collective bargaining agreements: The case of foreign competition in canada.

The Quarterly Journal of

Economics, 108(4):9831014. Abraham, K. G., Taylor, S. K., et al. (1996). Firms' use of outside contractors: Theory and evidence.

Journal of Labor Economics, 14(3):394424. Acemoglu, D., Antràs, P., and Helpman, E. (2007). Contracts and technology adoption.

The

American economic review, 97(3):916943. Acemoglu, D., Gancia, G., and Zilibotti, F. (2015). Oshoring and directed technical change.

American Economic Journal: Macroeconomics, 7(3):84122. Acemoglu, D., Grith, R., Aghion, P., and Zilibotti, F. (2010). Vertical integration and technology: theory and evidence.

Journal of the european economic Association, 8(5):9891033.

Acemoglu, D., Johnson, S., and Mitton, T. (2009). Determinants of vertical integration: nancial development and contracting costs.

The Journal of Finance, 64(3):12511290.

Alfaro, L., Conconi, P., Fadinger, H., and Newman, A. F. (2016). Do prices determine vertical integration?

The Review of Economic Studies, 83(3):855888.

Amiti, M. and Davis, D. R. (2011). Trade, rms, and wages: Theory and evidence.

The Review of

economic studies, 79(1):136. Antràs, P. (2003). Firms, contracts, and trade structure.

The Quarterly Journal of Economics,

118(4):13751418. Antras, P. (2015).

Global Production: Firms, Contracts, and Trade Structure. Princeton University

Press. Antràs, P. and Chor, D. (2013). Organizing the global value chain.

Econometrica, 81(6):21272204.

Antràs, P., Chor, D., Fally, T., and Hillberry, R. (2012). Measuring the upstreamness of production and trade ows.

The American Economic Review, 102(3):412416. 21

Antras, P. and Helpman, E. (2004). Global sourcing.

Journal of Political Economy, 112(3).

Asker, J., Collard-Wexler, A., and De Loecker, J. (2014). Dynamic inputs and resource (mis) allocation.

Journal of Political Economy, 122(5):10131063.

Asphjell, M. K., Letterie, W., Nilsen, Ø. A., and Pfann, G. A. (2014). Sequentiality versus simultaneity: Interrelated factor demand.

Review of Economics and Statistics, 96(5):986998.

Atalay, E., Hortaçsu, A., and Syverson, C. (2014). Vertical integration and input ows.

The

American Economic Review, 104(4):11201148. Autor, D. H. (2003). Outsourcing at will: The contribution of unjust dismissal doctrine to the growth of employment outsourcing.

Journal of Labor Economics, 21(1):142.

Baumgarten, D., Geishecker, I., and Görg, H. (2013). Oshoring, tasks, and the skill-wage pattern.

European Economic Review, 61:132152. Becker, S. O., Ekholm, K., and Muendler, M.-A. (2013). Oshoring and the onshore composition of tasks and skills.

Journal of International Economics, 90(1):91106.

Bhagwati, J., Panagariya, A., and Srinivasan, T. N. (2004). The muddles over outsourcing.

The

Journal of Economic Perspectives, 18(4):93114. Blanchard, O. and Kremer, M. (1997). Disorganization.

The Quarterly Journal of Economics,

112(4):10911126. Blanchower, D. G., Oswald, A. J., and Sanfey, P. (1996). Wages, prots, and rent-sharing.

The

Quarterly Journal of Economics, 111(1):227251. Blatter, M., Muehlemann, S., and Schenker, S. (2012). The costs of hiring skilled workers.

European

Economic Review, 56(1):2035. Bloom, N. (2009). The impact of uncertainty shocks.

econometrica, 77(3):623685.

Bloom, N., Bond, S., and Van Reenen, J. (2007). Uncertainty and investment dynamics.

of economic studies, 74(2):391415. 22

The review

Bond, S. and Van Reenen, J. (2007). Microeconometric models of investment and employment.

Handbook of econometrics, 6:44174498. Brügemann, B., Gautier, P., and Menzio, G. (2015). Intra rm bargaining and shapley values. Technical report, National Bureau of Economic Research. Caballero, R. J., Engel, E. M., and Haltiwanger, J. (1997). Aggregate employment dynamics: Building from microeconomic evidence.

The American Economic Review, pages 115137.

Card, D., Devicienti, F., and Maida, A. (2014). Rent-sharing, holdup, and wages: Evidence from matched panel data.

Review of Economic Studies, 81(1):84111.

Christodes, L. N. and Oswald, A. J. (1992). Real wage determination and rent-sharing in collective bargaining agreements.

The Quarterly Journal of Economics, 107(3):9851002.

Cooper, R. and Willis, J. L. (2009). The cost of labor adjustment: Inferences from the gap.

Review

of Economic dynamics, 12(4):632647. Cooper, R. W. and Haltiwanger, J. C. (2006). On the nature of capital adjustment costs.

The

Review of Economic Studies, 73(3):611633. Costinot, A. (2009). On the origins of comparative advantage.

Journal of International Economics,

77(2):255264. Dey, M., Houseman, S. N., and Polivka, A. E. (2012). Manufacturers' outsourcing to stang services.

ILR Review, 65(3):533559.

Eberly, J. C. and Van Mieghem, J. A. (1997). Multi-factor dynamic investment under uncertainty.

journal of economic theory, 75(2):345387. Eckel, C. and Irlacher, M. (2017). Multi-product oshoring.

European Economic Review, 94:7189.

Egger, H., Kreickemeier, U., and Wrona, J. (2015). Oshoring domestic jobs.

Journal of Interna-

tional Economics, 97(1):112125. Feenstra, R. C. and Hanson, G. H. (1996). Globalization, outsourcing, and wage inequality.

American Economic Review, 86(2):240245. 23

The

Feenstra, R. C. and Hanson, G. H. (1999). The impact of outsourcing and high-technology capital on wages: estimates for the united states, 19791990.

The Quarterly Journal of Economics,

114(3):907940. Garicano, L. (2000). Hierarchies and the organization of knowledge in production.

Journal of

political economy, 108(5):874904. Groizard, J. L., Ranjan, P., and Rodriguez-Lopez, A. (2014). Oshoring and jobs: The myriad channels of inuence.

European Economic Review, 72:221239.

Grossman, G. M. and Helpman, E. (2002). Integration versus outsourcing in industry equilibrium.

Quarterly Journal of Economics, pages 85120. Grossman, G. M. and Rossi-Hansberg, E. (2008). Trading tasks: A simple theory of oshoring.

The

American Economic Review, 98(5):19781997. Grossman, S. J. and Hart, O. D. (1986). The costs and benets of ownership: A theory of vertical and lateral integration.

The Journal of Political Economy, pages 691719.

Grout, P. A. (1984). Investment and wages in the absence of binding contracts: A nash bargaining approach.

Econometrica: Journal of the Econometric Society, pages 449460.

Grubb, D., Lee, J.-K., and Tergeist, P. (2007). Addressing labour market duality in korea.

OECD

Social, Employment, and Migration Working Papers, (61). Hall, R. E. (2004). Measuring factor adjustment costs.

The Quarterly Journal of Economics,

119(3):899927. Hamermesh, D. S. (1986). The demand for labor in the long run.

Handbook of labor economics,

1:429471. Hamermesh, D. S. (1996).

Labor demand. princeton University press.

Helpman, E. (2006). Trade, fdi, and the organization of rms.

Journal of Economic Literature,

44(3):589630. Hildreth, A. K. and Oswald, A. J. (1997). Rent-sharing and wages: evidence from company and establishment panels.

Journal of Labor Economics, 15(2):318337. 24

Holmstrom, B. R. and Tirole, J. (1989). The theory of the rm.

Handbook of industrial organization,

1:61133. Katz, M. L. (1989). Vertical contractual relations.

Handbook of industrial organization, 1:655721.

Lafontaine, F. and Slade, M. (2007). Vertical integration and rm boundaries: The evidence.

Journal of Economic Literature, 45(3):629685. Legros, P. and Newman, A. F. (2013). A price theory of vertical and lateral integration.

The

Quarterly Journal of Economics, 128(2):725770. Levine, D. K. (2012). Production chains. Ljungqvist, L. (2002).

Review of Economic Dynamics, 15(3):271282.

How do lay-o costs aect employment?

The Economic Journal,

112(482):829853. Lommerud, K. E., Meland, F., and Straume, O. R. (2009). Can deunionization lead to international outsourcing?

Journal of International Economics, 77(1):109119.

Manning, A. (1987). Collective bargaining institutions and eciency: An application of a sequential bargaining model.

European Economic Review, 31(1-2):168176.

McMillan, J. and Woodru, C. (1999). Interrm relationships and informal credit in vietnam.

The

Quarterly Journal of Economics, 114(4):12851320. Mitra, D. and Ranjan, P. (2010). Oshoring and unemployment: The role of search frictions labor mobility.

Journal of International Economics, 81(2):219229.

Nickell, S. J. (1986). Dynamic models of labour demand.

Handbook of labor economics, 1:473522.

Nunn, N. (2007). Relationship-specicity, incomplete contracts, and the pattern of trade.

The

Quarterly Journal of Economics, 122(2):569600. Park, J., Kim, S., Bang, J., and Kim, M. (2010). Coverage rates for collective bargaining in korea: A comparative study. Technical report, Report prepared for Korea Ministry of Employment and Labor. 25

Perry, M. K. (1989). Vertical integration: determinants and eects.

Handbook of industrial organi-

zation, 1:183255. Ranjan, P. (2013). Oshoring, unemployment, and wages: The role of labor market institutions.

Journal of International Economics, 89(1):172186. Ritter, M. (2014). Oshoring and occupational specicity of human capital.

Review of Economic

Dynamics, 17(4):780798. Saint-Paul, G. (1995). The high unemployment trap.

The Quarterly Journal of Economics, pages

527550.

European

Sethupathy, G. (2013). Oshoring, wages, and employment: Theory and evidence.

Economic Review, 62:7397. Sly, N. and Soderbery, A. (2014). Strategic sourcing and wage bargaining.

Journal of Development

Economics, 109:172187. Song, J., Price, D. J., Guvenen, F., Bloom, N., and Von Wachter, T. (2015). Firming up inequality. Technical report, National Bureau of Economic Research. Stenbacka, R. and Tombak, M. (2012). Make and buy: Balancing bargaining power.

Journal of

Economic Behavior & Organization, 81(2):391402. Stole, L. A. and Zwiebel, J. (1996a). Intra-rm bargaining under non-binding contracts.

The Review

of Economic Studies, 63(3):375410. Stole, L. A. and Zwiebel, J. (1996b). Organizational design and technology choice under intrarm bargaining.

The American Economic Review, pages 195222.

Williamson, O. E. (1985).

The economic intstitutions of capitalism. Simon and Schuster.

Williamson, O. E. (2002). The theory of the rm as governance structure: from choice to contract.

The Journal of Economic Perspectives, 16(3):171195. Wright, G. C. (2014). Revisiting the employment impact of oshoring. 66:6383. 26

European Economic Review,

Yi, K.-M. (2003). Can vertical specialization explain the growth of world trade?

Journal of political

Economy, 111(1):52102. Yoon, Y. et al. (2009).

A comparative study on industrial relations and collective bargaining in East

Asian countries. ILO Geneva.

27

Appendix A

Sample Selection

We start with 17,040 rms and 95,662 rm by year observations. Excluding the observations from year of 2014, we lose 11,866 rm by year observations. Then we lose 243 rm by year observations by removing rms reporting its number of employee to be less than 5 at least once between 2006 and 2013. After excluding the top 1% rms in size, we select rms observed all years across 2006 and 2013 to dene the balanced panel data set with 49440 rm by year observations and select rms observed at least twice to dene the unbalanced panel data set with 80,993 rm by year observations. Then the manufacturing sample has 29,200 rm by year observations and and 3,650 rms in balanced panel data set. The non-manufacturing sample has 44,544 rm by year observations and 7,693 rms in unbalanced panel data set. 1. 95,662 rm by year observations and 7,040 rms 2. Exclude year 2014 (a) lost 11,866 observations (b) 83,796 rm by year observations. (c) 16,180 rms 3. Firms should have at least 5 employees across all observed years (a) lost 243 observations (b) 83,553 rm by year observations. (c) 16,113 rms 4. Across-period observation (a) Balanced panel : should have been observed all years across 2006 - 2013 and exclude the top 1% rms with size. i. lost 34,113 observations ii. 49,440 rm by year observations: 29,200 manufacturing / 20,240 non-manufacturing iii. 6,180 rms: 3,650 manufacturing / 2,530 non-manufacturing (b) Unbalanced panel: Observed at least twice across periods i. lost 2,560 observations ii. 80,993 rm by year observations: 44,544 manufacturing / 36,449 non-manufacturing iii. 14,253 rms: 7693 manufacturing / 6560 non-manufacturing 28

Additional Figures

Figure 3: Task Outsourcing Trend by Firm Size: 2006-2013

[5,100) both

.8

prod. only

.2 0

0

.2

.4

.6

non−prod. only

.4

.6

.8

[5,inf) no sourcing

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

2012

2014

2012

2014

2012

2014

.6 .4 .2 0

0

.2

.4

.6

.8

[300,inf)

.8

[100,300)

2006

2008

2010 year

2012

2014

2006

2008

2010 year

Source: Survey of Business Activities(2015), Balanced Panel of Manufacturing Firms (a) Trend by rm size (Manufacturing only)

[5,100) both

1

prod. only

.4 .2 0

0

.2

.4

.6

.8

non−prod. only

.6

.8

1

[5,inf) no sourcing

2006

2008

2010 year

2012

2014

2006

2008

2010 year

.8 .6 .4 .2 0

.2

.4

.6

.8

1

[300,inf)

1

[100,300)

0

B

2006

2008

2010 year

2012

2014

2006

2008

2010 year

Source: Survey of Business Activities(2015), Balanced Panel of Non−Manufacturing Firms (b) Trend by rm size (Non-manufacturing only)

29

Figure 4: Firm characteristics by outsourcing pattern: 2006-2013

Revenue/worker

2006

2008

2010 year

2012

400 200 100

60 2014

0

40

0

no sourcing non−prod. only prod. only both

Number of workers 300

120

200

80

400

100

600

Value added/worker

2006

2010 year

2012

2014

2006

2010 year

2012

2014

Purchase/worker 300 200 100 0

35 40 45 50 55

2008

400

Capital/worker 100 150 200 250 300

Labor cost/worker

2008

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

Source: Survey of Business Activities(2015), Balanced Panel of Manufacturing Firms (a) Manufacturing rms

400 300

60 2010 year

2012

2014

0

40 2008

2006

2008

2010 year

2012

2014

2012

2014

2010 year

2012

2014

400 300

300

200 0

100

100 0 2010 year

2008

Purchase/worker

200

50 40 30

2008

2006

Capital/worker

60

Labor cost/worker

2006

no sourcing non−prod. only prod. only both

100

200

120 100 80

400 200 0 2006

Number of workers

Value added/worker

600

Revenue/worker

2006

2008

2010 year

2012

2014

2006

2008

2010 year

Source: Survey of Business Activities(2015), Balanced Panel of Non−Manufacturing Firms (b) Non-manufacturing rms

30

2012

2014

Figure 5: Firm characteristics by outsourcing pattern: 2006-2013

Revenue/worker

2006

2008

2010 year

2012

400 200 100

60 40

0

no sourcing non−prod. only prod. only both 2014

0

200

80

400

300

120

Number of workers

100

600

Value added/worker

2006

2010 year

2012

2014

2006

Capital/worker

2008

2010 year

2012

2014

Purchase/worker

0

100

200

100 120 140 160 180 200

300

35 40 45 50 55

400

Labor cost/worker

2008

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

Source: Survey of Business Activities(2015), Balanced Panel of Manufacturing Firms (a) Manufacturing rms with size

Revenue/worker

[5, 100)

2006

2008

2010 year

2012

400 200 100

60 40

0

no sourcing non−prod. only prod. only both 2014

0

200

80

400

300

120

Number of workers

100

600

Value added/worker

2006

2010 year

2012

2014

2006

Capital/worker

2008

2010 year

2012

2014

Purchase/worker 300 200 100 0

363840424446

100 150 200 250

400

Labor cost/worker

2008

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

2006

2008

Source: Survey of Business Activities(2015), Balanced Panel of Manufacturing Firms (b) Manufacturing rms with size

31

[100, 300)

2010 year

2012

2014

Figure 6: Firm characteristics by outsourcing pattern: 2006-2013

Revenue/worker

Number of workers

2006

2008

2010 year

2012

600 400 200

60 40

0

no sourcing non−prod. only prod. only both 2014

0

200

80

400

100

600

120

Value added/worker

2006

2010 year

2012

2014

2006

Capital/worker

2010 year

2012

2014

Purchase/worker

0

100

200

300

150 200 250 300

35 40 45 50 55

2008

400

Labor cost/worker

2008

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

Source: Survey of Business Activities(2015), Balanced Panel of Manufacturing Firms (a) Manufacturing rms with size

Revenue/worker

[300, +∞)

2006

2008

2010 year

2012

400 200 100

60 2014

0

40

0

no sourcing non−prod. only prod. only both

Number of workers 300

120

200

80

400

100

600

Value added/worker

2006

2010 year

2012

2014

2006

2010 year

2012

2014

Purchase/worker 300 200 100 0

35 40 45 50 55

2008

400

Capital/worker 100 150 200 250 300

Labor cost/worker

2008

2006

2008

2010 year

2012

2014

2006

2008

2010 year

2012

2014

2006

2008

Source: Survey of Business Activities(2015), Balanced Panel of Manufacturing Firms (b) Manufacturing rms with all size

32

2010 year

2012

2014

C

Additional Tables Table 3: The OLS estimates and xed eects estimates of intra-rm wage bargaining model (1) Sample Model Revenue/worker,

Purchase/worker,

R/N M/N

Capital Stock/worker,

K/N

1/N Capital stock,

K

Year Dummies

Number of Observations

(2)

(3)

(4)

(5)

unbalanced

unbalanced

unbalanced

balanced

balanced

OLS

xed eect

xed eect

xed eect

xed eect

0.0343

0.0567

0.0564

0.0534

0.0527

(0.0005)

(0.0007)

(0.0007)

(0.0010)

(0.0010)

-0.0225

-0.0127

-0.0127

-0.0109

-0.0109

(0.0005)

(0.0005)

(0.0005)

(0.0005)

(0.0005)

0.0418

0.0476

0.0478

0.0693

0.0694

(0.0006)

(0.0009)

(0.0009)

(0.0013)

(0.0013)

-45.5773

508.4942

495.9620

788.1109

841.8942

(25.8346)

(50.4848)

(50.1107)

(89.5677)

(88.0370)

-0.0000

-0.0001

-0.0001

-0.0001

-0.0001

(0.0000)

(0.0000)

(0.0000)

(0.0000)

(0.0000)

Yes

Yes

No

Yes

No

44,544

44,544

44,544

29,200

29,200

0.0343

0.0567

0.0564

0.0534

0.0527

(0.0005)

(0.0007)

(0.0007)

(0.0010)

(0.0010)

0.3448

0.7753

0.7742

0.7960

0.7934

(0.0093)

(0.0075)

(0.0076)

(0.0095)

(0.0096)

8,786

14,764

15,961 (1,734)

Derived Estimates

β

: worker bargaining power

θ + γ4

: fraction of relation-

specic purchase and contribution of purchase in restart cost

shutdown cost (and foregone future revenue)

Γ(N, K, M ) = γ0 + γ1 N + γ2 K + γ3 N K + γ4 M γ0

-1,328 (753)

(918)

(913)

(1,742)

γ2

1.2177

0.8402

0.8470

1.2979

1.3153

(0.0288)

(0.0231)

(0.0232)

(0.0448)

(0.0454)

γ3

8,971

-0.0009

-0.0017

-0.0017

-0.0016

-0.0017

(0.0001)

(0.0001)

(0.0001)

(0.0001)

(0.0001)

Notes: Dependent variable in all specications is labor cost per worker that includes benet and pension cost as well as wage cost. A rm's industry is dened to be the initial industry to which the rm belongs. Column 1 corresponds to OLS estimates and column 2 to 5 correspond to xed-eect estimates. Column 1, 2, 4 contain set of year dummies. The rst three columns include all manufacturing rms observed at least twice between 2006 and 2013 with 5 or more employees in each observed year. Sample in column 4 and 5 includes all manufacturing rms observed in all years between 2006 and 2013 with 5 or more employees in each year.

33

Table 4: Derived estimates of intra-rm wage bargaining model: Firm Size Firm Size

(1)

(2)

(3)

(4)

(5)

[5, 75)

[75, 100)

[100, 150)

[150, 300)

[300, ∞]

Outsourcing none

(χ = 0)

β θ + γ4 γ0 γ2 γ3

non-prod. only

β

(χ = 1) θ + γ4 γ0 γ2 γ3 prod. only

β

(χ = 2) θ + γ4 γ0 γ2 γ3 both

(χ = 3)

β θ + γ4 γ0 γ2 γ3 N. Obs.

0.0196

0.0255

0.0239

0.0205

0.2023

(0.0023)

(0.0019)

(0.0021)

(0.0015)

(0.0078)

0.8347

0.6427

0.6445

1.1768

0.6623

(0.1039)

(0.0597)

(0.0619)

(0.0805)

(0.0546)

59,110

50,578

59,718

91,874

12,393

(10,010)

(6,439)

(8,445)

(11,072)

(7,368)

0.7040

1.4648

0.6445

0.6136

0.7800

(0.3518)

(0.2328)

(0.2282)

(0.2532)

(0.1125)

-0.0033

-0.006

-0.0026

-0.0015

-0.0012

(0.0038)

(0.0018)

(0.0019)

(0.0009)

(0.0002)

0.0201

0.0164

0.0148

0.0263

0.1518

(0.0017)

(0.0009)

(0.0008)

(0.0010)

(0.0024)

0.7271

0.9834

0.9244

0.7361

0.2486

(0.0583)

(0.0115)

(0.0447)

(0.0286)

(0.0137)

66,036

80,885

106,095

98,522

45,091

(7,244)

(7,052)

(8,346)

(5,831)

(4,191)

0.0483

0.8277

1.4250

0.7297

0.5327

(0.1515)

(0.1749)

(0.2083)

(0.0876)

(0.0243) -0.0005

0.004

0.0019

-0.0040

-0.0018

(0.0018)

(0.0014)

(0.0012)

(0.0003)

0.0000

0.0117

0.0161

0.0210

0.0272

0.0564

(0.0016)

(0.0017)

(0.0018)

(0.0017)

(0.0091)

1.1717

1.047

0.9196

1.0024

0.8776

(0.1259)

(0.1133)

(0.0920)

(0.0637)

(0.1641)

111,242

89,960

79,918

57,996

56,683

(19,157)

(12,958)

(9,622)

(7,295)

(17,976)

-0.5729

1.0709

1.2099

0.5485

0.3378

(0.8603)

(0.3779)

(0.2306)

(0.1847)

(0.2157)

0.0139

-0.0005

-0.0057

-0.0027

-0.0001

(0.0123)

(0.0030)

(0.0013)

(0.0007)

(0.0003)

0.0170

0.0207

0.0164

0.0267

0.0412

(0.0017)

(0.0010)

(0.0010)

(0.0012)

(0.0029)

0.8983

0.9116

1.0283

0.8192

1.0236

(0.0904)

(0.0354)

(0.0302)

(0.0352)

(0.0391)

61,229

94,655

111,260

64,109

28,927

(9,074)

(6,468)

(9,648)

(5,337)

(15,765)

0.6267

-0.5796

0.3062

0.9265

0.0231

(0.3567)

(0.1620)

(0.1917)

(0.0790)

(0.0954)

-0.0026

0.0109

0.0042

-0.0033

0.0004

(0.0041)

(0.0017)

(0.0014)

(0.0004)

(0.0002)

5,008

6,008

5,832

7,544

4,808

Notes: Dependent variable in all specications is labor cost per worker that includes benet and pension cost as well as wage cost. A rm's industry is dened to be the initial industry to which the rm belongs. Column 1 corresponds to OLS estimates and column 2 and 3 correspond to xed-eect estimates. Sample in column 1 includes all yearly observations of rms with at least 5 employees in all observed years between 2006 and 2013. Sample in column 2 includes rms observed at least twice between 2006 and 2013 with 5 or more employees in each observed year. Sample in column 3 includes rms observed in all years between 2006 and 2013 with 5 or more employees in each year.

34

35

(χ = 1)

unregulated

non-prod. only

γ3

γ2

γ0

θ + γ4

β

γ3

γ2

γ0

θ + γ4

0.0038 (0.0021)

(0.0046)

(0.2069)

(0.3127) -0.0006

0.2945

(8,518)

(8,281) 0.2389

68,390

49,264

0.6887 (0.0642)

0.7516

(0.0018)

(0.1110)

(0.0032)

0.0190

(0.0037)

(0.0121) 0.0260

-0.0024

(0.3552)

(0.7788) -0.0100

0.5258

(10,578)

(13,115) 1.4139

52,633

55,933

0.8996 (0.1261)

0.8119 (0.1621)

0.0205 (0.0028)

0.0205 (0.0036)

β

(χ = 0)

none

(0.0084)

-0.0079

(0.8137)

2.1264

(18,593)

114,593

(0.1631)

1.1645

(0.0019)

0.0140

(0.0059)

0.0013

(0.5868)

0.7929

(16,383)

70,268

(0.2067)

0.9595

(0.0042)

0.0217

unregulated

[75, 100) regulated

[5, 75) unregulated

Firm Size

Outsourcing

(2)

(1)

(0.0014)

0.002

(0.1832)

0.7167

(6,914)

74,131

(0.0111)

0.9838

(0.0010)

0.0168

(0.0022)

-0.0092

(0.3368)

2.1437

(7,313)

48,605

(0.0710)

0.4887

(0.0021)

0.0236

regulated

(0.0018)

-0.0072

(0.2319)

1.3207

(7,814)

62,877

(0.0608)

0.5411

(0.0025)

0.0295

(0.0075)

0.0087

(0.8074)

-0.4939

(34,552)

110,261

(0.3404)

1.2159

(0.0048)

0.0177

unregulated

[100, 150)

(3)

(0.0012)

-0.0038

(0.2229)

1.5488

(8,196)

93,509

(0.0464)

0.9339

(0.0008)

0.0146

(0.0018)

-0.0032

(0.2277)

0.7663

(7,387)

49,271

(0.0581)

0.6087

(0.0023)

0.0261

regulated

(4)

(0.0010)

-0.0007

(0.2171)

0.5587

(10,111)

103,301

(0.0607)

0.9579

(0.0017)

0.0263

(0.0046)

-0.0043

(1.0633)

1.6845

(42,452)

127,403

(0.5096)

1.7628

(0.0040)

0.0141

unregulated

[150, 300)

Table 5: Derived estimates of intra-rm wage bargaining model: Firm Size

(0.0002)

0.0000

(0.0679)

0.1668

(4,897)

91,188

(0.0254)

0.6858

(0.0010)

0.0298

(0.0007)

-0.0005

(0.2175)

0.1850

(9,073)

82,294

(0.0716)

1.1292

(0.0014)

0.0233

regulated

(0.0002)

-0.0005

(0.2229)

0.9718

(18,844)

69,927

(0.0670)

0.7394

(0.0076)

0.0817

(0.0004)

-0.0003

(0.2900)

0.3446

(40,319)

99,627

(0.3245)

0.6142

(0.0337)

0.1145

unregulated

[300, ∞]

(5)

0.0000

-0.0004

(0.0220)

0.4659

(3,905)

40,536

(0.0131)

0.1925

(0.0024)

0.1603

(0.0002)

-0.0016

(0.1356)

0.9985

(7,507)

4,516

(0.0599)

0.7195

(0.0082)

0.1943

regulated

36

(χ = 3)

unregulated

both

unregulated

N. Obs.

γ3

γ2

γ0

θ + γ4

β

γ3

γ2

γ0

θ + γ4

(0.0035)

(0.0151) 5,008

0.0005

(0.3111)

(1.0376) 0.0188

0.0315

0.0595

49,275 (7,438)

79,962

(0.0821)

(0.1929)

(18,113)

0.8872

(0.0019)

0.9273

(0.0027)

0.0212

(0.0110)

(0.0240) 0.0144

0.0170

(0.8049)

(1.4945) -0.0446

-0.3814

2.7466

90,161 (16,107)

56,713

(0.1103)

(0.3052)

(21,473)

1.0837

(0.0017)

0.0134

regulated

0.9749

(0.0067)

0.0215

β

unregulated

prod. only

(χ = 2)

6,008

(0.0072)

0.0167

(0.6240)

-1.0037

(16,813)

112,456

(0.1430)

1.0474

(0.0025)

0.0201

(0.0187)

-0.0044

(1.9839)

2.3826

(68,376)

174,156

(0.6259)

1.6244

(0.0037)

0.0103

unregulated

(2)

[75, 100)

(1)

[5, 75)

Outsourcing

Firm Size

(0.0017)

0.0106

(0.1629)

-0.5677

(6,402)

91,694

(0.0353)

0.9033

(0.0011)

0.0211

(0.0029)

-0.0010

(0.3724)

1.0239

(11,608)

74,073

(0.1095)

0.9789

(0.0018)

0.0173

regulated

5,832

(0.0043)

0.0094

(0.4636)

-0.7360

(19,009)

98,732

(0.1869)

1.0339

(0.0037)

0.0228

(0.0045)

-0.0184

(0.5810)

2.4152

(11,268)

73,834

(0.1163)

0.7744

(0.0031)

0.0282

unregulated

[100, 150)

(3)

(0.0013)

0.0032

(0.1929)

0.5302

(9,032)

101,771

(0.0290)

1.0278

(0.0010)

0.0168

(0.0014)

-0.0065

(0.2810)

1.4803

(11,103)

77,026

(0.1191)

0.9456

(0.0020)

0.0195

regulated

7,544

(0.0021)

-0.0137

(0.5729)

3.5902

(12,481)

58,640

(0.1535)

0.9721

(0.0037)

0.0253

(0.0030)

-0.0007

(0.6095)

-0.3600

(20,698)

86,010

(0.1910)

1.3267

(0.0038)

0.0229

unregulated

[150, 300)

(4)

(0.0004)

0.0002

(0.0798)

0.1195

(6,518)

85,467

(0.0407)

0.9297

(0.0012)

0.0239

(0.0007)

-0.0025

(0.1803)

0.5736

(7,040)

58,430

(0.0667)

0.9739

(0.0018)

0.0276

regulated

Table 6: Derived estimates of intra-rm wage bargaining model: Firm Size (Continued) (5)

4,808

(0.0006)

0.0004

(0.5310)

0.7273

(47,526)

-59,305

(0.0595)

0.9609

(0.0069)

0.0348

(0.0016)

-0.0013

(1.1509)

0.9203

(63,901)

2,768

(1.1904)

1.8462

(0.0237)

0.0403

unregulated

[300, ∞]

(0.0002)

0.0004

(0.0979)

0.0267

(15,818)

30,465

(0.0777)

1.0528

(0.0033)

0.0404

(0.0003)

0.0000

(0.2091)

0.2855

(17,647)

56,318

(0.1605)

0.8434

(0.0092)

0.0568

regulated

37