University of Southern California Dissertations and Theses

Essays in corporate finance

(USC Thesis Other)

Essays in corporate finance

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Essays in Corporate Finance
by
Jianqiu (John) Bai
Submitted to the University of Southern California
in partial fulllment of the requirements for the degree of
Doctor of Philosophy in Finance and Business Economics
at the
UNIVERSITY OF SOUTHERN CALIFORNIA
May 2015
Copyrights c
Jianqiu (John) Bai, 2015
Essays in Corporate Finance
by
Jianqiu (John) Bai
Submitted to the University of Southern California
on May 15, 2015, in partial fulllment of the
requirements for the degree of
Doctor of Philosophy in Finance and Business Economics
Abstract
My thesis consists of two essays in corporate nance. The rst chapter studies how
dierences in organizational form aect rms' ability to respond to competitive pressure.
Using establishment level data from the U.S. Census and a dierence-in-dierences speci-
cation, I nd that relative to focused rms, conglomerate rms more actively restructure
during episodes of large import tari reductions. They restructure to focus on their core
competency and to improve productivity. Contrary to conventional wisdom, the internal
capital market primarily functions through the extensive margin, with plant opening and
closure decisions accounting for the majority (70%) of the productivity growth dierential
between conglomerates and standalones in the two years post tari shocks. Finally, the
coinsurance benets likely explain conglomerates' activeness in restructuring activities. The
second chapter studies the impact of state-level banking deregulation of local U.S. credit
markets on the reallocation of labor within local industries. In particular, reallocation of
labor towards rms with higher marginal products of labor signicantly increases after these
passages. Using rm production functions estimated with plant-level data, we propose and
examine an approach that quanties the industry productivity gains from labor reallocation
and nd that these gains are economically important.Our analysis suggests that labor real-
location is a signicant channel through which credit market conditions aect the aggregate
productivity and performance of local industries.
2
Dedication
To my beloved parents: my mother Yuan Cheng ( and my father Yue Bai, for teaching me
the value of knowledge, perseverance, and hard work.
To my wife Mengchao Ai, for her unwavering support.
To my children: Maggie (Fangwei) Bai and John Jr (Junhao) Bai for making Daddy so
fortunate! You are my jem!
3
Acknowledgments
I am grateful to my advisor, Gordon Phillips, for his kindness, guidance, and unwavering
support, and to the rest of my dissertation committee { Kenneth Ahern, John Matsusaka,
Yongxiang Wang, Nan Jia, and especially Daniel Carvalho { for their deep insights and en-
couragement. I thank Rene Stulz and participants at the FMA 2014 Doctoral Consortium
for helpful comments. I also thank seminar participants at the University of Colorado, Boul-
der, Cornell University (Hospitality), Syracuse University, University of Illinois at Chicago,
Texas University AM, Northeastern University, University of Hong Kong, Chinese University
of Hong Kong, Nanyang Technological University, and Australian National University for all
their helpful comments.
4
Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1 Firm Boundaries, Restructuring, and Productivity 11
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Related Literature & Hypothesis Development . . . . . . . . . . . . . . . . . 16
1.2.1 Overall Eciency of the Firm . . . . . . . . . . . . . . . . . . . . . . 17
1.2.2 Restructuring Activities of Conglomerates . . . . . . . . . . . . . . . 18
1.2.3 Causes of Observed Dierences between Conglomerates and Standalones 19
1.3 Data, Sample, and Empirical Methodology . . . . . . . . . . . . . . . . . . . 20
1.3.1 Data Sources & Sample Construction . . . . . . . . . . . . . . . . . . 20
1.3.2 Key Variable Denition . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.3.3 Empirical Specication . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4.1 Overall Eect of the Tari Shock on Productivity . . . . . . . . . . . 26
1.4.2 Restructuring Activities . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.4.3 Quantifying the Eect . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4.4 Causes of Dierences . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.4.5 Robustness Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Appendix 1.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Appendix 1.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5
2 The Impact of Bank Credit on Labor Reallocation and Aggregate Industry
Productivity 43
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.2 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.3 Methodological Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.3.1 Measuring Marginal Reallocation Gains . . . . . . . . . . . . . . . . . 51
2.3.2 Examining the Impact of Credit Market Reforms . . . . . . . . . . . 55
2.3.3 Alternative Measures of Industry Output and Productivity . . . . . . 57
2.3.4 Estimation of Production Functions . . . . . . . . . . . . . . . . . . . 58
2.4 Data and Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.5.1 Labor Reallocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.5.2 Potential Gains from Reallocation . . . . . . . . . . . . . . . . . . . . 64
2.5.3 Quantifying Cumulative Productivity Gains from Changes in Labor
Reallocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.5.4 Incorporating the Reallocation of Other Production Factors . . . . . 68
2.6 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.6.1 Identication Concerns . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.6.2 Measurement of Marginal Products . . . . . . . . . . . . . . . . . . . 75
2.6.3 Alternative Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
A Tables 81
B Figures 115
6
List of Figures
B-1 Figure 1-1: Distribution of Tari Cuts Over Time . . . . . . . . . . . . . . . 116
B-2 Figure 1-2: Restructuring Around Tari Shocks . . . . . . . . . . . . . . . . 116
B-3 Figure 1-3: Firm Level Productivity Around Tari Shocks . . . . . . . . . . 117
B-4 Figure 1-4: Average Productivity of Restructured Plants . . . . . . . . . . . 117
B-5 Figure 2-1: Dierences in Labor Allocation Prior to Interstate Deregulation . 118
7
THIS PAGE INTENTIONALLY LEFT BLANK
8
List of Tables
A.1 Table 1-1: Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 82
A.2 Table 1-2: Firm-level Productivity Around Tari Shocks . . . . . . . . . . . 83
A.3 Table 1-3: Productivity of the Treated Plants . . . . . . . . . . . . . . . . . 84
A.4 Table 1-4: Plant Exits and Divestitures . . . . . . . . . . . . . . . . . . . . . 85
A.5 Table 1-5: Plant Openings and Acquisitions . . . . . . . . . . . . . . . . . . 88
A.6 Table 1-6: Productivity of Conglomerates and Matched Focused Plants . . . 91
A.7 Table 1-7: Eect of Net Cash on Probability of Restructuring . . . . . . . . . 93
A.8 Table 1-8: Instrumental Variables Regression for Restructuring Probability . 94
A.9 Table 1-9: Treatment Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . 95
A.10 Table 1-10: First Stage for IV Regression . . . . . . . . . . . . . . . . . . . . 96
A.11 Table 2-1: State Banking Deregulation Dates . . . . . . . . . . . . . . . . . . 97
A.12 Table 2-2: Summary Statistics & Key Parameters . . . . . . . . . . . . . . . 99
A.13 Table 2-3: Credit Market Deregulation and the Sensitivity of Labor Realloca-
tion to Marginal Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
A.14 Table 2-4: Credit Market Deregulation and Potential Labor Reallocation Gains103
A.15 Table 2-5: Magnitude of Industry Productivity Gains from Increased Labor
Reallocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
A.16 Table 2-6: Credit Market Deregulation and Reallocation Gains from All Fac-
tors Industries with Estimated Returns to Scale Close to One . . . . . . . . 105
A.17 Table 2-7: Magnitude of Industry Productivity Gains from Increased Overall
Factor Reallocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
A.18 Table 2-8: Identication of Deregulation Eects . . . . . . . . . . . . . . . . 108
A.19 Table 2-9: Alternative Approaches to Estimate Production Functions . . . . 110
9
A.20 Table 2-10: Results Using Value-Added Production Functions . . . . . . . . 112
A.21 Table 2-11: Results Controlling for Dierences in Worker Skill . . . . . . . . 113
A.22 Table 2-12: Reallocation and Firm-Level Productivity Gains . . . . . . . . . 114
10
Chapter 1
Firm Boundaries, Restructuring, and
Productivity
1.1 Introduction
While previous research has clearly found that rm boundaries matter for their real
actions
1
and identied the internal capital market (ICM) as perhaps the key advantage of
the conglomerate form, much less is known about how the ICM works in reality. This paper
attempts to shed light on this issue.
The paucity of direct empirical evidence on ICM can perhaps be attributed to two factors:
data limitation and endogeneity. First, the use of Compustat data masks many micro-level
actions undertaken by the rm. For example, researchers would not observe the closure
of one of many plants within a business segment. A more serious issue is that the business
segment data from Compstat suers from strategic reporting as well as managerial incentives
problems (Bens, Berger, & Monahan (2011)). Secondly, the formation of a rm's ICM is an
evolving process that might be endogenously determined with other factors. Therefore, to
test the detailed actions of the ICM, one would, as J. Stein (1997) puts it, need a scenario
where: \[for] a company [that] owns two unrelated divisions A and B, the appeal of investing
1 See, for example, Maksimovic & Phillips (2001) for conglomerates' reaction to positive demand shocks,
Seru (2015) for the eect of the conglomerate form on rms' innovative activities, and Khanna & Tice
(2001) for response to competitive pressure.
11
in B suddenly increases...". (p.112)
This paper simultaneously addresses both issues by exploiting detailed plant-level data
from the U.S. Census Bureau, which provides comprehensive coverage on both public and
private rms at the establishment level. To capture episodes which suddenly alter the appeal
of investing in an given industry, I exploit a series of signicant declines in import tari
rates between 1976 and 2004 as a quasi-natural experiment. These import tari reductions
signicantly lower the cost of entry by foreign competitors and increase the expected level of
competition faced by domestic import-competing rms.
2
In this paper, I show that conglomerate rms, relative to their standalone counterparts,
experience a signicantly smaller drop in productive eciency during episodes of increased
competitive pressure. I further show that this is largely due to the active restructuring (e.g.,
plant opening and plant closure), as opposed to the reallocation of resources across existing
plants of conglomerates. I conduct the analysis on a longitudinal panel of approximately
871,000 manufacturing plants between 1976 and 2004. This data allows me to accurately
identify plant status and obtain a precise measure of total factor productivity (TFP here-
after), estimated at the establishment level.
Using a dierence-in-dierences regression specication, I start by documenting the
changes in TFP for conglomerate rms and focused rms around these competitive shocks.
Specically, while TFP of the conglomerate rms stays virtually unchanged, the focused rms
on average experience an economically large and statistically signicant drop in rm-level
TFP (4.2 log points).
3
To investigate the role of restructuring activities in explaining this observed dierence in
change in productivity, I examine both the probability of restructuring and the productivity
of the restructured plants. First, I nd that these import tari shocks signicantly alter
rms' propensity to restructure such as closing and opening of a plant, but much more so for
conglomerates. For example, the likelihood for conglomerates to clos down a plant aected
by the shock increases by approximately 120%, whereas that for focused rms goes up by
2 Importantly, as is shown in previous research, they are largely unanticipated (Melitz and Tre
er (2012))
and often accompanied by sizable increases in ex post imports (Xu (2012) and Fresard (2010)). Hence,
it is unlikely that rms alter their organizational form in anticipation of signicant drops in import tari
rates.
3 A similar pattern emerges if I only examine the segments in the aected industries
12
54%. Secondly, conglomerate rms' restructuring activities enhances rm level TFP. For
instance, plants closed down by the conglomerate rms during the tari shock are of much
lower productivity compared to surviving plants in the same local industry, whereas TFP of
the newly opened or closed plants by focused rms does not display any systematic pattern.
My next set of results concerns the exact manner in which conglomerate rms restructure.
The corporate diversication literature proposes that conglomerate rms possess comparative
advantage in their core industries (Maksimovic & Phillips (2002); J. G. Matsusaka (2001)).
Therefore, how conglomerate rms respond to the tari shocks likely depends on whether or
not they have comparative advantage in the aected industry. Indeed, conglomerate rms
restructure in a way so as to focus on their core competency. Specically, I nd that when
the core industries of the conglomerate rms are aected, they undertake comprehensive re-
structuring activities by actively shutting down peripheral operations and expanding further
into the core industries. In direct contrast, when noncore businesses are aected, conglomer-
ate rms simply shed the \fat" by ridding themselves of the fringe operations. Importantly,
all these actions seem to be eciency enhancing.
To further quantify how various adjustment margins (i.e. intensive margin, plant opening
and closure, acquisitions and divestitures) contribute to the overall dierence in TFP change
around tari shocks, I match conglomerate plants to focused plants in the same industry-year
that are similar in size and age. By tracking the status and productivity of the two groups
of plants over time, I nd that plant opening and closing decisions account for as much
as 70% of the total dierential in TFP change between the conglomerate rms and their
matched focused pairs around these tari shocks, which echoes the previous ndings that
restructuring activities play a critical role in conglomerate rms' response to competitive
pressure.
Finally, I investigate the underlying causes for conglomerate rms' activeness in restruc-
turing activities. Previous research suggests that conglomerate rms can rely on their inter-
nally generated cash across dierent divisions (Lewellen (1971)), the so-called \more money"
eect. A natural implication of this is that conglomerate rms' propensity to engage in re-
source intensive restructuring activities will be less sensitive to their internally held net cash
prior to the shock, compared to their focused cohorts. I nd evidence consistent with this
13
conjecture. Specically, conglomerate rms' likelihood to engage in acquisitions or opening
of new plants is completely unaected by their net cash holdings prior to the shock. In
contrast, focused rms in the High Net Cash group are signicantly more likely to acquire
or open new plants, and are less likely to shut down productive plants. Overall, the evi-
dence suggests that conglomerate rms' superior ability to cope with increased competitive
pressure partially stems from their nancial advantage.
This paper makes two important contributions to the existing literature on corporate di-
versication and ICM: First and foremost, this is the rst study to date that systematically
investigates the role and quanties the importance of restructuring activities in the func-
tioning of ICM. Existing literature has largely assumed that ICM operates purely through
a reallocation of resources between the existing units. This paper, on the other hand, sug-
gests that the extensive margin (i.e., plant opening, closure, acquisition, and divestiture)
can be as economically important as the intensive margin. Second, this paper enhances our
understanding of the ICM and how it can help companies respond to competitive shocks, ex-
tending earlier ndings by Khanna & Tice (2001) to a multi-industry setting. The evidence
sheds new light on how conglomerate rms exploit their ICM. Consistent with models such
as Maksimovic & Phillips (2001), J. Matsusaka & Nanda (2002), and Cestone & Fumagalli
(2005), conglomerate rms' \toughness" in their response to the treat of entry largely de-
pends on whether their core industry is aected. This further highlights the dynamic nature
of the ICM and underscores the importance of jointly considering cross subsidization and
winner picking in future theoretical models.
This paper builds on a large body of work on corporate diversication and ICM. The
literature to date has produced mixed evidence on the impact of organizational form on
rms' market value (Lang & Stulz (1994), Berger & Ofek (1995))
4
, capital investment (Shin
& Stulz (1998)), acquisition behavior (Maksimovic & Phillips (2008)), and nancial policies
(Duchin (2010), Hann, Ogneva, & Ozbas (2013)). This paper is related most closely to
Giroud & Mueller (2014), in which the authors examine the reallocation of resources such
as labor and capital through ICM when new airline routes are introduced between the head-
4 These earlier studies' ndings have been challenged by subsequent papers such as Villalonga (2004) and
Campa & Kedia (2002) which explicitly control for the endogeneity of the conglomerate status. In these
later studies, the diversication discount either shrinks or turns into a premium
14
quarter and plants. However, this paper diers from it in several important ways: First, this
paper explicitly investigates rms' restructuring activities, whereas Giroud & Mueller (2014)
only study reallocation across existing units. Second, Giroud & Mueller (2014) exploits a
\positive" shock, where the introduction of new airline routes presumably lowers the cost of
monitoring for the headquarter. This paper, on the other hand, examines rms' response to a
\negative" episode where the expected competitive pressure signicantly increases. Thirdly,
Giroud & Mueller (2014) only focus on public rms, whereas this paper exploits a compre-
hensive sample of both public and private rms, thus producing a more complete picture of
ICM.
5
This paper also adds to a small number of studies that use plant-level data to investi-
gate the interplay between organizational form, asset sales, and productivity. Maksimovic
& Phillips (2001) use plant level data from the Longitudinal Research Database
6
to study
asset sales and nd that most transactions result in productivity gains. The gains in pro-
ductivity depend signicantly on rms internal organization. Schoar (2002) also employs
census plant-level data and nds that conglomerate plants are approximately seven percent
more productive than their focused counterparts. She further examines productivity changes
around ownership changes to throw light on the dynamic eect of diversication decision on
rm productivity. Specically, she nds that while the productivity of the target plants
increases after the acquisition, incumbent plants experience a drop in eciency, resulting in
an overall net decline in rm productivity. This study departs from the extant literature
by focusing on episodes of import tari shocks and analyzing restructuring activities in all
margins. From a methodological standpoint, this is the rst paper, to my knowledge, that
employs advancements (i.e. Olley & Pakes (1996) methodology) in Industrial Organization
to evaluate the eect of conglomeration on rms productivity. The major advantage of the
OP methodology is that it addresses both the simultaneity and selection problems inherent
in the production function estimation. Controlling for both types of biases seems critical,
given that periods of major economic shocks are often accompanied by industry consolidation
5 In fact, the fact that the results are most pronounced in the private subsample highlights that ICM is most
valuable when rms face signicant costs in obtaining external nancing.
6 Longitudinal Research Database, or LRD, is now referred to as the Census of Manufacturers and the
Annual Survey of Manufacturers
15
and exits of rms.
Finally, this study relates to a growing body of research that documents the superior
performance of conglomerate rms relative to focused rms during dicult economic times.
In particular, conglomerate rms have been shown to exhibit higher sales growth, cash

ow, and expenditure on research and development during industry downturns (Gopalan
& Xie (2011)) and to invest more aggressively relative to their focused counterparts during
the recent nancial crisis (Kuppuswamy & Villalonga (2012)) as well as during periods of
heightened competitive pressure brought about by Walmart's entry into the local retail mar-
ket (Khanna & Tice (2001)). A recent working paper by Swanburg (2014) uses Compustat
segment data to study the investment policies and product market outcomes of conglomerate
and focused rms also in the context of import tari shocks, and nds that conglomerate
rms tend to shift resources away from the aected industries. This paper dierentiates
from the above studies by focusing on the restructuring activities of conglomerate rms and
quantifying their importance, overcoming the limitations with the Compustat data which
has been shown by papers such as Villalonga (2004) and Bens et al. (2011) to be subject to
strategic reporting and managerial agency problems. In addition, the asymmetric pattern
in restructuring activities within conglomerate rms, which depends on the importance of
the segment, is new to the literature. Overall, my paper proposes a potential mechanism
through which conglomerate rms can outperform their focused peers during poor economic
conditions.
The remainder of this paper is organized as follows. Section 1.2 provides a selective
review of the literature and develops testable hypotheses. Section 1.3 discusses data sources,
variable construction, and empirical specications. The main results are presented in Section
1.4, and Section 1.5 concludes.
1.2 Related Literature & Hypothesis Development
In this paper, I attempt to address the following questions:
1. How do preexisting dierences in organizational form aect rms' overall ability to
respond during episodes of increased competitive pressure?
16
2. How does the ICM work in reality? In particular, does it work through a reallocation
of resources across existing divisions or does it involve active restructuring through
decisions such as plant opening, closure, divestiture, and acquisitions?
In this section, I discuss the extant literature and develop the specic hypotheses that I
examine in the data.
1.2.1 Overall Eciency of the Firm
The sudden large reduction in import taris increases the expected competitive pressure
that domestic producers face. Two facts about these episodes are noteworthy: First, Fresard
(2010), Fresard & Valta (2014), and Xu (2012) document that in the aftermath of these
tari shocks, imports increase signicantly and aected rms adopt a more conservative
nancial policy. Such large increases in ex post imports provide validation for the use of
such a setting. Secondly, Fresard & Valta (2014) nd that the occurrence of these events is
unrelated to the overall investment prospects of the aected industries.
7
Moreover, to the
extent that rms often lobby to protect their respective industries, it is unlikely that the
corporate sector actively seeks to reduce the existing import tari which shields them from
foreign competition.
Existing theories on corporate diversication generate competing predictions on rms'
response to these competitive shocks. On the one hand, conglomerate rms may respond
aggressively by exploiting their ICM to pull resources together and investing heavily in the
aected industry (Faure-Grimaud & Inderst (2005)). On the other hand, conglomerate rms
might act \softly" by redeploying resources away from the aected industry (J. Matsusaka
& Nanda (2002), Cestone & Fumagalli (2005)). However, since the aggressiveness of con-
glomerates to \defend" a given industry does not generate any value implications, one would
intead have to draw on the two in
uential views of ICM.
7 Fresard & Valta (2014) examine the mean and median growth rates of investment over the three years
that precede the tari reductions between the treated and matched rms and nd no apparent dierences;
They also nd similar that the earnings estimates for the next scal year, investment recommendations,
and ve-year earnings growth rates to be similar one year prior to the tari shock. They further nd that
rms recognize these events to be important so as to change their competitive environment, as re
ected in
their Management's Discussion and Analysis.
17
On the one hand, the winner-picking camp, as rst advanced by Williamson (1975) and
formalized by J. Stein (1997), contends that by maintaining an active internal capital market,
conglomerate rms can channel resources to the most productive use. In the context of the
tari shocks, conglomerate rms can reshue resources across their asset portfolio so as to
minimize the productivity loss. I thus expect the following to hold:
H1: Conglomerate rms suer from a smaller loss in productivity than fo-
cused rms during the tari shocks.
Standing directly in contrast to the winner picking hypothesis is the in
uential cross
subsidization view, which maintains that conglomerate rms are more likely to be plagued
by agency problems. In particular, operating across dierent business lines creates a battle
ground by divisional managers for corporate resources (see Scharfstein & Stein (2002) and
Rajan, Servaes, & Zingales (2000)). In the context of an import tari shock, conglomerate
rms might ineciently retain low productivity plants and expand into new business lines
in which they have no comparative advantage. Regardless of the exact mechanism, the cross
subsidization hypothesis leads to the following prediction:
H1a: Conglomerate rms suer from a larger loss in productivity than focused
rms during the tari shocks.
1.2.2 Restructuring Activities of Conglomerates
Related to the winner picking hypothesis is the theory of comparative advantage devel-
oped in Maksimovic & Phillips (2002). Specically, in a neoclassical framework, they model
optimal rm allocation of resources in response to demand shocks and exploit the dierences
in productivity within the rms to test the model. The central prediction of their model is
that conglomerate rms, when faced with an unanticipated negative shock such as the tari
shock, will remain in businesses where they have an comparative advantage to operate.
Specic to the context of tari shocks, I expect conglomerate rms to behave dierently
when their core business, as opposed to peripheral operations, are aected. In particular, if
one thinks of the formation of a conglomerate rm as a matching exercise between existing
managerial skills and the underlying assets as in J. G. Matsusaka (2001), the business that
a conglomerate has comparative advantage to operate in is its largest segment. Therefore,
18
conglomerate rms might channel additional resources into the aected core segment but
retract from the aected peripheral industries. Summarizing the above argument leads to
the following hypothesis:
H2: Conglomerate rms' restructuring activities depend on whether their
core business is aected. In particular, they may expand more into aected core
segments, but retract from aected peripheral businesses.
On the other hand, the cross subsidization view conjectures that conglomerate rms'
restructuring activities are agency driven (e.g. empire building incentives). In the context of
a tari cut, a conglomerate rm may channel additional resources into the aected industry
irrespective of its comparative advantage. In this case, a conglomerate rm's restructuring
activities may or may not improve the rm's overall eciency. This leads to the following
hypothesis:
H2a: Conglomerate rms' restructuring activities do not depend on whether
their core business is aected. The conglomerate rms always commit more
resources to the aected industries regardless of where their core competency
lies.
1.2.3 Causes of Observed Dierences between Conglomerates and
Standalones
One of the key advantages of conglomerate rms is that by operating in industries with
imperfectly correlated cash
ows, conglomerate rms can pool resources across dierent
divisions and channel them into one centralized usage, the idea of which dates back to at
least Lewellen (1971). A natural implication of this idea is that relative to focused rms,
conglomerate rms' propensity to engage in restructuring activities is less in
uenced by their
liquidity on hand. This leads to the following prediction:
H3: Conglomerate rms' propensity to engage in resource intensive restruc-
turing activities (e.g., plant opening and acquisition) is less correlated with their
net cash holdings prior to the shock than focused rms.
19
1.3 Data, Sample, and Empirical Methodology
1.3.1 Data Sources & Sample Construction
Import Tari Data
Data on import tari is available from 1972 to 2001 at the 10-digit Harmonized System
(HS) Code level. The data is compiled by Feenstra (1996), Feenstra, Romalis, & Schott
(2002), and Schott (2010) and is publicly available from both the National Bureau of Eco-
nomic Research and the Center or International Data at University of California, Davis. To
construct import tari rates at the four-digit SIC industry, I rst use the concordance table
developed in Feenstra et al. (2002) to map the product level data on imports and duties
into 1987 four digit Standard Industry Classication (SIC) categories. I then calculate, for
each industry year, the ad valorem tari as the duties collected divided by the Free-on-
Board
8
value of imports. This procedure yields a longitudinal panel of import tari rates
(in percentage terms) at the industry-year level.
[Figure 1-1 About Here]
To qualify for a \tari cut", the following conditions need to be satised: First, the
decline in the annual tari rate in an industry must exceed two times its historical mean
annual rate of change over the sample period. Second, in order to capture non-transitory
changes in import taris, I also exclude cases when the \cut" is followed by large increases
of the same magnitude. Finally, due to a change of coding in imports in 1989, no tari cut is
dened between 1988 and 1989. Evident from Figure 1-1, these tari cuts are not clustered
over a specic time period or concentrated within a small set of industries. Importantly, these
import tari shocks are often accompanied by increases in ex post imports and increase the
competitive pressure faced by import competing rms.
[Figure 1-2 About Here]
8 The ad valorem tari implicitly assumes that the transportation costs associated with the imports count
towards the total value of the imports.
20
Figure 1-2 graphs the number of exiting and entering plants, as well as acquisitions and
divestitures around these tari cuts. It is clear that these are periods of intensive industry
restructuring. For example, the number of exiting plants in the aected industry surges
from approximately 43 to around 50 after the tari shock, a 16% increase. This provides
additional validation for the empirical setting. In particular, if rms can anticipate the tari
cuts and the increased competition associated with them, one should see a large jump in the
number of exiting plants before the shock occurs.
Main Analytical Sample
The primary data source comes from the Annual Survey of Manufacturers (ASM) and the
Census of Manufacturers (CMF)
9
, both maintained by the U.S. Census Bureau. The ASM
provides comprehensive coverage for the universe of manufacturing plants (SIC 2000-3999)
in the U.S. with more than 250 employees, but maintains a randomly selected rotating panel
for smaller plants. In Census years, the ASM is supplanted by the CMF, which covers the
universe of all manufacturing plants in the U.S. regardless of their size. Merging these two
databases allows me to construct a longitudinal panel containing key plant-level variables
such as capital expenditure, material inputs, employment, total assets, value of shipments,
and industry.
To construct the analytical sample, I start with the universe of the merged ASM and CMF
databases from 1976 to 2004. From this initial sample, I exclude observations if the plant
operates in an industry-year for which no import tari data can be obtained, if the plant has
a four-digit SIC code ending in 9, or if the data for the plant is imputed from administrative
records. I also drop observations for which no productivity (TFP) can be estimated. Finally,
plants with less than two years of continuous data are dropped to minimize sampling error.
In addition, I also note that the sampling methodology of ASM make it dicult to accurately
identify entries, exits, and changes of ownership, particularly those of plants that are small
in size. Thus I use the Longitudinal Business Database (LBD), which provides employment
and payroll information on the universe of nonfarm private sector from 1976 to 2005, to
identify plant status. I also restrict the sample to rms for which plants in the ASM/CMF
9 CMF and ASM together were known as the Longitudinal Research Database (LRD) prior to 2002
21
comprise at least 70% of the rm's total employment.
10
To correct for the oversampling
(undersampling) of large (small) plants associated with the ASM/CMF, all results in this
paper are weighted regressions, where the weights are computed as the reciprocal of the
probability weights.
11
Applying the above selection criteria yields a nal sample of 871,000 establishment-year
observations. Table 1-1 presents means and standard deviations of the key variables in the
analyses. Column 1 reports summary statistics for the whole sample. Columns 2 and 3
break down the data into subsamples of plant-year observations belonging to conglomerate
rms and standalone rms, respectively. All dollar gures are expressed in 1997 dollars (in
thousands). One thing worth noting is that plants that are part of a conglomerate rm are
on average older and larger in size. Therefore, all regression analyses include age and size as
controls.
Matched Sample
To quantify the contribution of various adjustment margins to the TFP change around
tari shocks, I also construct a matched sample from this main analytical sample. For each
year with a major import tari shock, I match plants that belong to a conglomerate rm
to standalone plants that are in the same industry, size group, and age cohort. Specically,
I control for the interaction between 143 four-digit industries, 10 rm size groups, and 6
age categories, which amounts to over 8,000 control cells for each year. Hence, the control
group for each conglomerate plant consists of all single-segment plants that are in the same
industry-size-age control cell.
1.3.2 Key Variable Denition
In this subsection, I dene the key variables used in the analyses. Appendix 1.1 contains
the detailed denition for all other variables.
10Giroud & Mueller (2014) use a similar restriction criterion but limit their sample to public rms.
11For example, an observation that has probability 1/3 of being included in a sample has probability weight
equal to 3.
22
Conglomerate Dummy
I classify a rm as a conglomerate if it operates in more than one unique three-digit
SIC industry. All main results of the paper remain qualitatively unchanged if I dene the
conglomerate dummy based on the two-digit SIC as in Schoar (2002).
Measures of Total Factor Productivity
The primary measure of eciency used in the analyses is total factor productivity (TFP)
estimated with a tranlog specication. The advantage of the translog form is that it can be
thought of as a second degree approximation to any arbitrary production function and relaxes
the assumption of constant factor elasticities as in the case of a Cobb-Douglas specication.
12
The rst approach exploits a xed eect specication using a rolling window. Specically,
I estimate the following regression:
lnQ
it
=A +f
i
+
N
X
j=1
c
j
lnL
jit
+
N
X
j=1
N
X
k=j
c
j
lnL
jit
lnL
kit
(1.1)
where i and t index plant and year, respectively. Q
it
is the output of plant i in year t, and
L
jit
is the quantity of input j employed in production at plant i in year t. A is an industry
specic technology shift parameter. f
i
is a plant-rm specic xed eect, and c
j
=
P
N
j=1
c
ji
represents returns to scale. In estimating the above equation
13
, I use ve years of data for
each plant. The measure of TFP employed in the main analyses is computed as the sum
of the residual from Eq. (1) and the plant-rm xed eect f
i
. In order to control for the
varying degree of precision in estimating TFP across dierent industries, I also standardize
the measure by subtracting the average TFP in each industry-year and dividing by the
corresponding standard deviation. The inclusion of the xed eect termf
i
has the advantage
of capturing persistent productivity eects due to managerial ability (e.g. Griliches (1957)
and Mundlak (1978)).
The second approach follows Olley & Pakes (1996) (OP henceforth). The advantage of the
12In unreported results, I reestimate all results using the Cobb-Douglas case and nd similar results.
13The advantage of assuming a log-linear production function is that it can be thought of as a second
degree approximation to any arbitrary production function and explicitly models rms choice of inputs.
In contrast, a Cobb-Douglas specication assumes constant factor elasticities. See Maksimovic & Phillips
(2002, 2008) for more details.
23
OP methodology is that it controls for the simultaneity and selection biases. To the extent
that periods of major economic shocks are often accompanied by industry consolidation and
exits of rms, it is crucial, in the productivity estimation, to control for the likelihood of exit
during these episodes.
14
Appendix 1.2 provides more details on the OP estimation procedure.
Treated Dummy
A plant is dened as treated if it operates in an industry that experiences a tari shock
in any of the past three years (t, t-1, or t-2). Accordingly, a rm is a treated rm if one or
more of its plants experiences an import tari shock within the past three years.
Core vs. Noncore
I classify a plant as a \core" asset if it operates in a four-digit SIC industry whose total
output accounts for more than 25% of the rm's total output. In untabulated results, I
experiment with other thresholds (i.e., 20%, 30%, up to 50%) and nd qualitatively similar
results.
Plant Opening, Closure, Acquisition, and Divestiture
All plant statuses are dened using data from the LBD. Specically, plant opening is de-
ned as a dummy variable that equals one in a plant's rst year of existence. Symmetrically,
I dene plant closure as occurring if a plant is in its last year of existence. Importantly,
I separate cases where a plant is rst erected from those when an existing plant changes
ownership due to an acquisition. Similarly, I distinguish between the permanent closure of a
plant and the sale of a plant to an acquiring rm. Since one of the main advantages of con-
glomerate rms over standalones is their ability to engage in acquisitions (e.g., Maksimovic
& Phillips (2008)), dierentiating new openings (closing down) of plants from a change in
ownership through the market for corporate control is important.
14See Pavcnik (2002) for an example that applies the OP estimation method to Chilean manufacturing plants
during periods of trade liberalization.
24
1.3.3 Empirical Specication
Firm-level Regressions
To investigate the eect of the tari shocks on rms' overall productivity, I employ the
following dierence-in-dierences specication:
y
jt
=
t
+
j
+
1
Treated
jt
+
0
X
jt
+"
jt
(1.2)
wherey
jt
is the rm-level productivity, computed as the capital-stock weighted sum of plant-
level productivity. j indexes rms and t indexes time.
t
and
j
are time and rm xed
eects, respectively. treated
jt
is an indicator variable that equals one if rm j has one or
more plants that have experienced a tari cut in the past three years. X is a vector of control
variables. The coecient of interest is
1
, which measures the eect of the tari shock on
rm level productivity (TFP).
Plant-level Regressions
A. Productivity Changes
In examining the dierential eects of the shock on the TFP of plants belonging to con-
glomerate versus focused rms, I estimate the following dierence-in-dierence-in-dierence
regression:
y
ijkt
=
ks
+
i
+
t
+
1
(Conglomerate
ijt
Treated
ikt
) +
2
Treated
ikt
+
3
Conglomerate
ijt
+
0
X
ijkt
+"
ijkt
(1.3)
where i, j, k, and s index plant, rm, industry, and state respectively.
i
are plant xed
eects to control for time-invariant plant level characteristics;
ks
are industry-state xed
eects to ensure that the comparison is among plants within the same local industry. X
ijkt
are control variables (i.e. plant age and plant size). Treated
ikt
is a dummy variable that
equals one if a plant operates in an industry that has a tari shock within the past three
years (t, t-1, and t-2). The parameter of interest is
1
which estimates the dierential TFP
change between conglomerate and focused plants during the tari shock.
25
B. Likelihood of Restructuring
To investigate the probability that plants are closed, divested, acquired, or opened, and
whether this varies with the importance of the segment, I estimate the following regression:
y
ijkt
=
ks
+
j
+
t
+
1
(Treated
ikt
Core
ijt
) +
2
(Treated
ikt
Noncore
ijt
)
+
3
(Other
ikt
Core
ijt
) +
3
(Other
ikt
Noncore
ijt
) +
0
X
ijkt
+"
ijkt
(1.4)
where i, j, k, and s index plant, rm, industry, and state respectively.
j
are rm xed
eects to control for time-invariant plant level characteristics;
t
are year xed eects;
ks
are industry-state xed eects to ensure that the comparison is among plants within the
same local industry. X
ijkt
are control variables (i.e. plant age and plant size). Treated
ikt
is
a dummy variable that equals one if a plant operates in an industry that has a tari shock
within the past three years (t, t-1, and t-2). Other
ikt
is a dummy variable that equals one if
a plant is unaected by the tari shock itself but belongs to a rm that has one or more of
its plants treated.
1.4 Empirical Results
1.4.1 Overall Eect of the Tari Shock on Productivity
I start with an examination of how rm level productivity is aected by the import
tari shocks and how this varies across rms with distinct organizational forms. Figure 1-3
graphs the evolution of (capital weighted) rm level productivity surrounding the tari cut.
While conglomerate and focused rms follow a similar path prior to the tari reduction,
their TFP gap becomes larger as the shock unfolds. Importantly, the TFP of conglomerate
rms drops at the onset of the tari shock, but quickly reverts back to the level prior to
the cut. Table 1-2 examines this more formally using a regression approach. The key result
is that while conglomerate rms impacted by the tari cut experience a slightly positive
change in TFP (col. 1 and 2), their focused counterparts suer from an economically and
statistically signicant drop in TFP (col. 3 and 4). Taken together, the results suggest that
conglomerate rms engage in actions that somehow mitigate the productivity loss. The rest
26
of this paper attempts to determine why this is the case and how restructuring activities
contribute to this.
I now turn to the impact of the tari shock on TFP of the aected plants. Panel A
of Table 1-3 presents the results. I control for plant xed eects, year xed eects, and
industry-state xed eects in these regressions.
15
Plant xed eects control for the time-
invariant plant specic characteristics and year xed eects control for any macroeconomic
conditions that aect all plants at the same time. Finally, industry-state xed eects ensures
that the comparison is between plants within the same local industry. This is important for
two reasons: First, the same drop in tari rates likely has dierent impacts on plants located
in dierent regions. For example, between 1989 and 1995 when most tari shocks correspond
to the phasing in of the North American Free Trade Agreement (NAFTA), plants in states
bordering Canada or Mexico can be expected to be more in
uenced by the tari cuts. Second,
the urban economics literature
16
has long documented the geographical concentration of
industries in the U.S.. Therefore, comparing plants within the same local industry allows for
a tighter identication. As Column [1] shows, treated plants aliated with a conglomerate
rm experience a much smaller TFP drop relative to the focused plants. In fact, when
estimated in subsamples, as in Columns [2] and [3], conglomerate plants experience a drop
in TFP that is both economically small (0.6 log points) and statistically insignicant. In
contrast, the drop in TFP at treated focused plants is large in magnitude (5.3 log points)
and statistically signicant at the 1% level. Columns [4] to [6] repeat the exercise with the
rolling window TFP and nd similar results.
1.4.2 Restructuring Activities
While the results so far suggest that conglomerate rms can somehow mitigate the impact
of the increased competitive pressure, they mask the underlying channel through which this
is achieved. In particular, it is unclear whether this is accomplished through a reallocation
of resources (intensive margin) across the existing units or via a systematic restructuring
15In unreported results, I also include year and state xed eects interacted with the conglomerate dummy
and the results remain qualitatively the same.
16See Duranton & Puga (2004) for a review of the literature
27
eort (extensive margin) by actively opening up (shutting down) productive (unproductive)
plants.
Table 1-4 directly investigates this possibility by examining how the likelihood of plant
closure and divestiture is aected by the competitive pressure and how this varies across rms
of distinct organizational forms. All regressions control for plant size and age, as well as rm,
year and industry-state xed eects. Several results stand out from Columns [1] to [3] of
Panel A: First, the probability of plant closure signicantly rises for establishments aected
by the tari cuts (col. 2 and 3). Second, although plants that are part of a conglomerate rm
have a lower probability of being shut down (Conglomerate
jt
), conglomerate plants aected
by the tari shock indeed have a higher likelihood of exiting, as is evident from the negative
and signicant coecient estimate on the interaction (Conglomerate
jt
Treated
kt
) term
(col. 1). Third, these eects are economically large. For example, the treated conglomerate
plants experience a 7.8% increase in probability of being shut down, which corresponds to a
120 percent increase of the mean exit rate (6.4%). Columns [4] to [6] concentrate on plant
divestiture and nd that treated conglomerate plants are more likely to be divested than
their focused cohorts.
17
Panel B focuses on the conglomerate subsample to examine the impact of the shock on
not only the treated plants but also other unaected plants that belong to the same rm.
This is evident in Column [1], where the tari cuts result in a 6.8% increase in the likelihood
of the aected plants being shut down, but also have a spillover eect on the \other" plants
that are part of the same rm but do not experience the shocks themselves. Column [2] inves-
tigates whether the response depends on the importance of the segment in the conglomerate
rm. Previous research suggests that the core segments are more productive than periph-
eral operations (Maksimovic & Phillips (2002)) and might represent a good match between
managerial skills and the underlying assets (J. G. Matsusaka (2001)).Hence, as discussed in
Section 1.2.2, one might expect conglomerate rms to react to shocks aecting their core
operations in dierent ways compared to events impacting their peripheral assets. This is
indeed what I nd: First, plants in core industries aected by the shock (Treated Core)
17While dierentiating cases of plant closure from divestiture is inherently interesting, it is outside the scope
of this paper. However, it is worth noting is that divesture represents a more complex process than closure,
as it depends on outside factors such as the availability of potential buyers in the market.
28
experience only an increase of plant closure indistinguishable from zero (0.002). In contrast,
noncore assets aected by the shock (Treated Noncore) are signicantly more likely to be
shut down (0.092). Second, tari shocks result in spillover eects in plant closure, but only
so for noncore assets. For example, a peripheral plant that is unaected itself but belongs to
a rm that has one or more plants aected (Other Noncore) has a much higher likelihood
(0.078) of being shut down, but if such a plant is in one of the core industry (Other Core),
it actually experiences a drop (-0.007) in the probability of being shut down. Columns [3]
and [4] focus on plant divestiture and nd results similar to plant closure but with smaller
magnitudes.
Finally, Panel C investigates the productivity (TFP) of the plants that are shut down
by rms aected by the tari cuts. Columns [1] and [2] examine the conglomerate rms
and focused rms, respectively. Plants shut down by conglomerate rms aected by the
shock are approximately 8 log points less productive than surviving plants in the same local
industry (i.e. within the same industry-state). In contrast, the TFP of plants shut down
by treated focused rms is not dierent from that of the surviving plants. Again, I observe
similar patterns in Columns [3] and [4] in the case of divestiture, but with weaker statistical
signicance.
Table 1-5 closely resembles Table 1-4 but concentrates on episodes of plant opening and
plant acquisitions. Overall, the results in Panel A suggest that treated conglomerate rms are
little aected in their propensity to open up new plants, and in the case of plant purchase,
treated conglomerate rms are indeed more likely to add additional plants to their asset
portfolio. The coecient of 0.008 (col. 2) represents a 20% increase in probability of being
acquired relative to the mean of 0.04. To the contrary, standalone rms are much less likely
to open up new plants (-0.084 in col. 3), but are unaected in the probability to acquire
new ones.
In available but uncleared (from Census) results in Panel B of Table 1-5, I nd that
conglomerate rms are indeed more likely to open up new plants in the core industries
that are treated (Treated Core), but are signicantly less likely to erect new plants in
the treated noncore industries (Treated Noncore). No clear pattern is found for plant
acquisitions. One possibility is that adjustment through plant acquisitions is a more time
29
consuming process and depends on other economic factors. Hence, the short event window
in this paper might not be enough to discern such acquisitive actions.
Finally, the results in Panel C show that plants that are newly opened by treated con-
glomerate rms tend to be of relatively higher TFP than incumbent plants. In contrast, no
signicant change is observed for plants opened by focused rms. Again, the estimates are
noisy in the case of acquisitions.
The results from Tables 1-2 to 1-5 convey the central message of the paper and paint
a comprehensive picture of how conglomerate rms restructure. First, despite their lower
unconditional propensity of engaging in certain restructuring activities (e.g. plant opening
and closure), treated conglomerate rms are more active on all adjustment margins than
their focused cohorts. In addition to this dierential propensity to engage in restructuring
activities, conglomerate rms manage to dispose of (build) relatively less (more) produc-
tive plants relative to surviving (incumbent) plants in the same local industry. Figure 1-4
conducts a straightforward univariate analysis and plots the average TFP of restructured
plants by conglomerate rms and focused rms on and o the tari shock. The same pattern
emerges, that conglomerate rms can somehow make wiser restructuring decisions when af-
fected by the competitive shock when compared to their focused peers. Second, the exact
mechanism through which treated conglomerate rms restructure depend on the importance
of the segment that is aected. Consistent with the idea that conglomerate rms restruc-
ture to focus on their core competency, when peripheral plants experience an increase in
competitive pressure, conglomerate rms simply cut the extra \fat" by divesting or shutting
down these aected assets. On the other hand, when the core segment is aected by the
shock, conglomerate rms engage in a systematic restructuring by shedding assets in pe-
ripheral industries and opening up more plants in the core industry, further reinforcing their
core competency. All these actions together improve conglomerate rms' overall productive
eciency relative to the focused rms.
1.4.3 Quantifying the Eect
To quantify the eect of conglomeration on rms' productivity changes around tari
cuts, I exploit the matched sample detailed in Section 1.3.1.3. Conceptually, by constructing
30
a matched sample which consists of focused plants that are similar in terms of plant-level
characteristics to the conglomerate plants, I can compare the productivity changes around
tari shocks between the actual conglomerate rms and a hypothetical focused rm formed
by a portfolio of focused plants and assess the role of various adjustment margins. While
there are many decomposition methods that have been used in the literature, I exploit one
used in a recent paper by Davis et al. (2014). Specically, let P denote the change in
productivity express average the two-year productivity change as:
P
t
= [S
C
t+2
P
C
t+2
S
C
t
P
C
t
] + [S
N
t+2
P
N
t+2
S
X
t
P
X
t
] + [S
A
t+2
P
A
t+2
S
D
t
P
D
t
] (1.5)
where S denotes an employment share, P denotes a TFP value, and C, N, X, A, and D
denote continuers, entrants, exits, acquisitions, and divestitures, respectively. For instance,
P
N
t+2
is the average TFP of the newly opened plants two years after the tari shock. The
average TFP change for controls (
~
P
t
), is dened analogously. Combine the two expressions
and express all TFP terms as deviations from TFP values for the control continuers, cancel
redundant terms, and rearrange to obtain the following expression:
P
t

~
P
t
=S
C
t+2
(P
C
t+2

~
P
C
t+2
)S
C
t
(P
C
t

~
P
C
t
)
+S
N
t+2
(P
N
t+2

~
P
C
t+2
)
~
S
N
t+2
(
~
P
N
t+2

~
P
C
t+2
)S
X
t
(P
X
t

~
P
C
t
) +
~
S
X
t
(
~
P
X
t

~
P
C
t
)
+S
A
t+2
(P
A
t+2

~
P
C
t+2
)
~
S
A
t+2
(
~
P
A
t+2

~
P
C
t+2
)S
D
t
(P
D
t

~
P
C
t
) +
~
S
D
t
(
~
P
D
t

~
P
C
t
)
(1.6)
The above expression decomposes the dierence in change in TFP of the conglomerate
rms relative to their matched focused peers into three components. The top line of the
above expression captures the contribution of treated-control dierences among continuing
plants, the second line estimates the change attributed to plant opening and closure, and the
third line isolates the adjustment from plant acquisitions and divestitures.
18
The employment
shares variables can be computed by taking the ratio of the employment of plants by their
status to the total employment of the rm. For example, to obtain
~
S
N
t+2
, one can compute
18Note that the employment shares variables depend on the plant status. For example, for plants that are
opened up during the two-year horizon, I use their employment share at t+2, the end of the horizon. To
the contrary, for plants that are closed down, I use their employment share at t, the beginning of the
horizon. (see Foster, Haltiwanger, & Krizan (2001) for a detailed discussion)
31
the employment share of the newly opened plants at time t+2 for each matched control rm
and take the average across all rms. As discussed in Davis et al. (2014), the main advantage
of this decomposition is it combines di-in-di estimates with a composition that is often
employed in the literature on rm-level productivity dynamics.
Table 1-6 estimates the di-in-di estimates where the dependent variable is the TFP
adjusted by the same-year control-continuer productivity and the dependent variables are
plant statuses. Remarkably, the estimates in Table 1-6 echo the results from the full sample
estimation. In particular, as is shown in Panel A, relative to their matched focused counter-
parts, conglomerate rms shut down and divest plants of lower TFP. For example, the plants
closed down by conglomerate rms are on average 3.8 log points less productive than the
control continuers, while those shut down by focused rms are 1.1 log points more produc-
tive. The dierence between the two is large and signicant at the 1%. As can be easily seen
from the previous decomposition, shutting down plants that are less productive than the
surviving plants will enhance rm productivity. Panel B concentrate on plant opening and
acquisitions. Again, conglomerate rms tend to open up more productive plants than their
matched focused cohorts, but no clear dierence is detected for acquisitions. Panel C of Ta-
ble 1-6 exploits the above decomposition by applying values from the computed employment
shares and TFP changes from Panels A and B. Overall, conglomerate rms out perform the
focused controls with respect to post-tari shock TFP growth by 2.36 log points. It is worth
noting that summing Summing over terms in the second line of Equation (6) yields a value
of 1.645 log points, implying that plant opening and closure eects account for 69 percent
of the superior TFP growth at conglomerate rms.
1.4.4 Causes of Dierences
In this section, I investigate the underlying causes that lead to the more activeness
of conglomerate rms relative to their focused counterparts in engaging in restructuring
activities as well as their ability to close down (open up) less (more) productive assets during
tari shocks. One possibility, as discussed in Section 1.2.3, is that conglomerate rms can
pool their internally generated cash
ow across divisions to fund operations in a particular
segment. This has been termed the \more money" eect by J. C. Stein (2003). Focused
32
rms, on the other hand, have to rely on cash holdings on hand. Therefore, relative to their
focused counterparts, conglomerate rms' propensity to engage in restructuring activities
might be less sensitive to their net cash positions prior to these tari shocks.
To test this hypothesis, I focus on a subsample of rms which have nancial information
from the Quarterly Financial Reports (QFR). QFR contains rm-level balance sheet items
such as assets, debt, both short and long term, and income statement variables such as
operating cash
ow and net income. One major advantage of QFR over the Compustat is
that it provides comprehensive coverage on both public and private rms, whereas Compustat
contains data on only publicly listed rms. However, Two caveat about QFR need to be
noted: First, QFR only provides continuous coverage for manufacturing companies with
assets greater than or equal to $250,000. Therefore, the QFR sample is more skewed towards
larger rms. Second, the Internal Revenue Agency (IRS) permits the merge between QFR
and other micro-level economic databases (e.g. ASM and CMF) only in census years
19
. I thus
create, for each industry in census years, an indicator variable that equals one if a rm's net
cash ratio ((CashHoldings -ShortTermDebt)/TotalAsset) is above the industry median,
and examine whether this aects rms' likelihood to engage in restructuring activities. Table
1-7 contains the results.
Overall, estimates from Table 1-7 are consistent with the \more money" hypothesis.
Conglomerate rms' likelihood to engage in plant opening, closure, and acquisitions is com-
pletely insensitive to their pre-shock nancial position. To the contrary, focused rms with
\High Net Cash" are signicantly more likely to engage in resource intensive activities such
as plant opening and acquisitions, but much less likely to close down plants. In unreported
results, I nd that the lower likelihood of \High Net cash" focused rms to shut down is
mostly concentrated within \High TFP" plants. However, no clear evidence is found with
regard to plant openings and acquisitions. Hence, nancial
exibility cannot solely explain
the observed discrepancy between conglomerate and focused rms during the competitive
shock, especially conglomerate rms' ability to open up high-TFP plants.
19Years ending in \2" and \7"
33
1.4.5 Robustness Tests
In this section, I consider several potential factors that might be driving the main results.
Using an instrumental variables approach, I rst address the concern that rms are not
randomly assigned to be conglomerate or focused. I next examine whether rms could be
adjusting their corporate behavior in anticipation of the tari shocks. Furthermore, I consider
whether the individual demand curves faced by rms of distinct organizational forms could
be driving the results.
Endogenizing the Conglomerate Status
One important issue that can cast doubt on the results so far is that organizational form
is an endogenous choice of rms. In particular, if there exist some unobserved time-varying
plant-level characteristics (e.g. plant manager's ability) that aect both the likelihood of
a plant to be part of a conglomerate rm and the outcome variables examined (e.g. in-
vestment), my conclusions will be biased. To mitigate this issue, I instrument for the or-
ganizational form by exploiting variables that have been identied in the prior literature to
directly aect the decision to conglomerate but not the outcome variables such as invest-
ment. Specically, I identifyCapitalintensity,Longtermchangeinindustryshipments,
and Laggedfirmsize from Maksimovic & Phillips (2008) and Number of mergers from
Gopalan & Xie (2011) as variables to instrument for the conglomerate status.
Since the conglomerate dummy is also interacted with the treated dummy on the right
hand side of the regressions, I follow the advice in Wooldridge (2002) (p. 236) in imple-
menting the IV. Specically, in the rst stage, I use the above variables and the interactions
between them and the treated dummy to instrument for the Conglomerate Dummy and
Conglomerate Treated Dummy, respectively. Columns [1] and [2] of Table 1-10 show
that the identied instruments are strong, as evident in the large values of the F-statistic of
the excluded instruments
20
. Table X contains the second stage IV regressions on some of the
main specications. Overall, all main results are robust to controlling for the endogeneity of
the conglomerate organizational form.
20The generally accepted threshold is 10.
34
Endogeneity of the Import Tari Reduction
Although there are compelling reasons to believe that the tari cuts as dened in this
paper are largely exogenous, one could still worry that there exist some political economy
reasons that might lead to these changes. In particular, if rms alter their corporate be-
havior in the period leading up to the shock, it would be indicative of such anticipation.
To formally rule out this possibility, I estimate a DD regression to examine the treatment
dynamics on a variety of rm level variables, including productivity (TFP), return on cap-
ital (ROC), employment, and sales growth. Table 1-9 presents the results. Note that the
parameters of interest are the coecients on Treated (2) and Treated (1), as well as
ConglomerateTreated (2) and ConglomerateTreated (1). In particular, if any of
these coecients are signicant, it would be indicative of anticipatory behavior and reverse
causality. Importantly, none of these coecients are signicant at the traditional level of
condence. I also conduct a placebo test where I randomly \assign" industries into experi-
encing the import tari reduction and nd no detectable changes in rms' TFP or propensity
to restructure.
Homogeneous Goods
One of the most widely documented eects of trade liberalization is that it often leads
to lower prices of goods in the aected industries. Since all productivity measures used so
far are revenue-based productivity measures, there is a possibility that rms with individ-
ual demand curves might be heterogeneously aected by the tari shock. In particular, if
conglomerate plants face a dierent demand compared to focused plants, the results so far
could simply capture the possibility that prices for goods produced by conglomerate rms
drop less than their focused counterparts. To alleviate this concern, I follow the approach in
Foster, Haltiwanger, & Syverson (2008) and reestimate all the main results on the subsample
of plants producing homogeneous products for which quality and thus prices are unlikely to
vary much across rms. In available but currently uncleared results (from Census), I nd
that the main conclusions remain unchanged.
35
Heterogeneity in the Treatment Eect
To provide additional support for the validity of the empirical setting as well as to shed
light on the underlying causes for rms' dierential reactions, I conduct a battery of cross
sectional tests: First, I separate the sample into conglomerates that are have divisions that
are truly unrelated and those that operate across industries that have potential vertical re-
lationships. Consistent with the coinsurance hypothesis, the results are mostly concentrated
in conglomerates which have a relatively
at structure and operate in unrelated industries.
Second, since the trade literature has documented substantial growth by exporting rms fol-
lowing trade liberalization, I reestimate the baseline specication using a subsample where I
exclude the exporting plants and nd similar results. Third, when I decompose the sample
into public and private rms, I nd that the results are stronger among the private subsam-
ple, which is consistent with the idea that being publicly listed reduces the wedge in nancial

exibility between conglomerates and standalones.
Other Robustness Tests
In unreported results, I also examine whether some of the other empirical methodological
choices could be driving my results. First, I reestimate TFP with a Cobb-Douglas speci-
cation using the methodology as in Schoar (2002) and Li (2013) and nd similar results.
Second, I dene the conglomerate dummy at the two-digit SIC instead of three-digit SIC
and the results remain unchanged. Third, I reestimate all results using only the eventually
treated sample, i.e. plants that eventually experience the tari shock during the sample
period, and the conclusions are unaected. Finally, I consider dierent thresholds of import
tari reduction to dene the treatment and nd similar results.
1.5 Conclusion
By concentrating on episodes of import tari reductions, this paper attempts to provide
some broad based evidence on how conglomerate rms and standalone rms react to compet-
itive shocks. By exploiting the unique feature of the plant-level data at Census, I focus on the
36
restructuring activities of conglomerates and standalones and their eects on productivity,
a largely unexplored question by the existing literature because of data limitations.
The rst main nding is that conglomerate rms navigate through these competitive
shocks better than their focused counterparts, as they experience a signicantly smaller
productivity loss around the tari cuts. Further tests show that conglomerate rms aected
by the shocks are signicantly more active in restructuring activities such as plant opening,
closing, acquisitions, and divestitures. Moreover, treated conglomerate rms tend to dispose
of plants of lower productivity (relative to surviving plants) and open up new plants of higher
productivity (relative to incumbent plants). Interestingly, conglomerate rms' restructuring
activities are asymmetric. That is, their restructuring depends heavily on whether their
core operations are aected. Specically, when core assets are aected, conglomerate rms
shut down unaected peripheral business and further expand in the core industry, seemingly
exhibiting \cross subsidization". When peripheral operations are threatened, conglomerate
rms simply shed the "extra fat" by getting out of the fringe business, consistent with \winner
picking". Overall, the combination of conglomerate rms' restructuring improve the rms'
productive eciency. Decomposition of TFP change dierentials between conglomerates and
matched focused rms reveal that almost 70% of the dierential change can be attributed
to plant opening and/or closing decisions.
The ndings in this paper shed light on a number of issues in the corporate diversication
literature. Most importantly, the evidence underscores the importance to take into account
their active restructuring as opposed to a simple reallocation of resources across existing
units. Furthermore, as demonstrated by the tests, winner picking and cross subsidization
need not be mutually exclusive as they are in the traditional models. Exactly which one
prevails is state contingent and largely re
ects conglomerate rms' desire to focus on their
core comparative advantage. Finally, although conglomerate rms' ability to pool resources
together can partially explain their activeness in restructuring relative to their focused peers,
their identication of assets of high productivity during these shocks remains an interesting
puzzle for future research.
37
Appendix 1.1
Total value of shipments: Total sales de
ated using the four-digit SIC industry ship-
ment de
ators from the NBER-CES Manufacturing Industry Database expressed in
$1997 dollars.
Investment/Capital Stock: Capital expenditure divided by the capital stock, which is
computed using the perpetual inventory method. Both capital expenditure and capital
stock are de
ated using the 4-digit SIC deator from the NBER-CES Manufacturing
Industry Database.
Logarithm of Employment: Logarithm of the number of the total number of people
employed.
Firm Size: Logarithm of the total value of shipments.
Plant Age: The number of years since a plant's rst appearance in the LBD. (all plants
appear in the sample only after 1976).
Sales Growth: The annual percentage change in the total value of shipments: Sales (t)
- Sales (t-1)/Sales(t-1)
TFP (Rolling Window): Marginal productivity estimated using Eq. (1)
TFP (OP): Marginal productivity estimated using the methodology detailed in Ap-
pendix 2.
Return on Capital (ROC): Total value of shipments less labor and material costs di-
vided by the capital stock.
Opening Dummy: A dummy variable that equals one if a plant is in its rst year of
existence (LBD).
Closure Dummy: A dummy variable that equals one if a plant is in its last year of
existence (LBD).
38
Sale Dummy: A dummy variable if a plant changes ownership between year t and year
t+1, and zero otherwise.
Acquisition Dummy: A dummy variable that equals one if a plant is in its rst year of
existence for Firm i in year t but belonged to a dierent rm in year t-1.
Conglomerate Dummy: A dummy variable that equals one if a rm operates in more
than one three-SIC industry in a given year.
Treated Dummy: A dummy variable that equals one if a plant operates in an industry
that has experienced a signicant import tari reduction in the past three years.
Treated (-2): A dummy variable that equals one if a plant experiences a signicant
tari cut in two years from now.
Treated (-1): A dummy variable that equals one if a plant experiences a signicant
tari cut in one year from now.
Treated (0): A dummy variable that equals one if a plant experiences a signicant tari
cut in the current year.
Treated (+1): A dummy variable that equals one if a plant experiences a signicant
tari cut one year ago.
Treated (+2): A dummy variable that equals one if a plant experiences a signicant
tari cut two years ago.
Lagged High Debt Dummy: A dummy variable that equals one if a rm's lagged short
term bank debt is above the industry-year median. The lagged information is from the
latest available census year.
Number of Mergers: Number of plant acquisitions in a given year.
Industry Long-run Change in Shipments: Change in total industry value of shipments
between 1976 and 2004.
Industry Capital Intensity: Capital expenditures divided by industry sales at the three-
digit SIC level.
39
Appendix 1.2
I follow the setup and assumptions discussed in Ackerberg, Benkard, Berry, & Pakes
(2007) in estimating TFP using the OP methodology. Although the main results are based
on TFP estimated using a translog specication, I illustrate the steps and procedures in
estimation using a Cobb-Douglas specication for simplicity in exposition:
Let x
ijt
=logX
ijt
, I can then write the production function in log as follows:
y
ijt
=
0
+
a
age
ijt
+
k
k
ijt
+
l
l
ijt
+
m
m
ijt
+!
ijt
+
ijt
where
ijt
is a shock revealed to rms at time t after all decisions have been made, age
ijt
is the rms age and !
ijt
is a productivity component observed by the rm before making
decisions in year t. Note that rm TFP is given by a
ijt

0
+
a
age
ijt
+!
ijt
+
ijt
and can be
inferred as y
ijt

k
k
ijt

l
l
ijt

m
m
ijt
if we know the production function parameters.
Let i
ijt
denote rm investment. A rst important condition for this approach is that,
conditional on the sample of rms with positive investment i
ijt
> 0, I can write !
ijt
=
h
t
(age
ijt
;k
ijt
;i
ijt
). In other words, conditional on a rms age and capital stock, rms in-
vestment i
ijt
allow us to uniquely determine !
ijt
. Moreover, conditional on all information
available for rms at year t, !
ijt
is a sucient statistic for predicting !
ijt+1
. A second im-
portant condition for this approach is that rms decide to operate in year t if and only if
!
ijt
>=
t
(age
ijt
;k
ijt
). This means that the decision to operate is monotonic on !
ijt
and
l
ijt1
and m
ijt1
are not state variables.
Let ((age
ijt
;k
ijt
;i
ijt
)) = h
t
(age
ijt
;k
ijt
;i
ijt
) +
0
+
a
age
ijt
+
k
k
ijt
In the rst stage, I
estimate y
ijt
= ((age
ijt
;k
ijt
;i
ijt
)) +
l
l
ijt
+
m
m
ijt
+
ijt
. This allows me to estimate
l
and
m
, as well as obtain a tted value for
^

ijt
. I estimate this equation using a polynomial
and on the sample with i
ijt
> 0. Let X
ijt
be an indicator that equals one of the rm
decides to operate in year t and I
ijt
denote the rms entire information set at year t. Let
P
ijt
=P (X
ijt
= 1jI
ijt1
). In the second stage, I estimate a tted value for P
ijt
. Under the
OP assumptions, I can write P
ijt
= (age
ijt1
;k
ijt1
;i
ijt1
) and estimate a tted value
^

ijt
for this expression using a probit model with a polynomial. In the third stage, I estimate
the following equation:
y
ijt
b
l
l
ijt
b
m
m
ijt
=
0
+
a
age
ijt
+
k
k
ijt
+g(
ijt1

0

a
age
ijt1

k
k
ijt1
;P
ijt
) +
ijt
40
Under the OP assumptions, I have that E(
ijt
j I
ijt1
;X
ijt
= 1) = 0. I use the previous
tted values for
^

ijt
and
^
P
ijt
, and estimate
0
,
a
and
k
using non-linear least squares.
41
THIS PAGE INTENTIONALLY LEFT BLANK
42
Chapter 2
The Impact of Bank Credit on Labor
Reallocation and Aggregate Industry
Productivity
2.1 Introduction
An important question in economics and nance is to understand how nancial markets
aect real economic activity. Given the role of nancial markets in moving resources towards
the best economic opportunities, previous research has focused on how nancing frictions
may impact the allocation of resources and, as a consequence, aggregate productivity.Two
main channels have been posed and debated.
1
Namely, nancing frictions can lower aggregate
productivity by leading to a misallocation of capital across existing rms or by distorting
rms entry and exit decisions. Despite the central importance of labor as a production factor,
limited attention has been paid to the role of nancing markets in facilitating the reallocation
of labor towards the most productive rms. Indeed, existing research typically assumes that
nancing frictions do not directly aect rms ability to adjust their labor decisions, and that
these frictions in
uence the allocation of labor only indirectly through their impact on the
allocation of capital. According to this view, nancial markets will not have a rst-order
1 Recent examples include Hsieh & Klenow (2009), Bartelsman, Haltiwanger, & Scarpetta (2013), Buera,
Kaboski, & Shin (2011), Collard-Wexler & Loecker (2015) and Midrigan & Xu (2014).
43
eect on aggregate productivity by facilitating the reallocation of labor towards the most
productive rms.
2
In this paper, we study the role of nancial markets in in
uencing aggregate productivity
by shaping the reallocation of labor across rms. Using a dierence-in-dierence analysis, we
examine how reforms in U.S. local credit markets through major state-level banking dereg-
ulations aect the aggregate productivity of local industries by shaping the reallocation of
labor across rms. We nd that these state-level banking deregulation events are associated
with signicant increases in the within industry reallocation of labor towards higher marginal
product of labor rms and that labor reallocation is associated with large gains in aggregate
industry productivity.
Intuitively, labor reallocation will only aect the aggregate productivity of an industry to
the extent that these reallocations are correlated with dierences in rms marginal products
of labor. We propose and estimate an approach to formalize this intuition and measure
the overall impact of within-industry labor reallocations on industry productivity growth,
which we label labor reallocation gains. We build on previous research suggesting how to
use plant-level data to decompose aggregate industry productivity growth into its dierent
determinants and isolate the contribution of labor reallocation to this growth.
We argue that nancing frictions can potentially have signicant eects on aggregate
productivity by directly aecting labor reallocations.First, there are dierent reasons to
expect nancing frictions to directly aect rms employment decisions. To begin, rms
will need nancing to employ more labor if there is a timing delay between payments to
workers and the additional cash
ows generated by the use of more labor.Firms also often
face training and hiring costs, and rm-specic investments by workers can be important,
so expanding labor often requires upfront costs.
3
Unlike physical capital which can serve as
collateral, it can be harder for labor intensive rms to provide as much collateral to banks as
capital intensive rms can provide.Capital also has an additional nancing advantage over
2 If nancing frictions are not preventing labor to move across rms with diverging returns in using labor,
there is no reason to expect nancing frictions to have rst-order eects on aggregate productivity through
labor misallocation.
3 Even if some of these returns are generated over short-term horizons, Paravisini, Rappoport, Schnabl, &
Wolfenzon (2015) suggests that rms can face signicant nancing frictions in raising short-term working
capital.
44
labor as physical capital is frequently leased directly from capital providers.
Financially constrained rms can also expose workers to greater labor income risks and
workers might factor this issue into account when choosing among potential employers.
4
Since rms with higher returns in expanding their labor are likely to be the ones with
greater employment growth in the absence of nancing frictions, these frictions can limit
the extent to which labor is reallocated towards rms with the highest returns in using
labor and, as a consequence, lower aggregate productivity.Even if nancing constraints in
expanding labor were smaller than the ones involved in the nancing of long-term capital,
their impact on aggregate productivity could still be important when compared to the impact
of nancing frictions on aggregate productivity through the misallocation of capital, as labor
is a signicantly larger share of production relative to capital.
General equilibrium eects are also potentially important for labor. As more productive
rms expand and drive up factor prices, they trigger greater reallocation by crowding out
less productive rms (e.g., Melitz (2003)). To the extent that the aggregate supply of labor
is more inelastic than the one of capital, these eects will be more important in labor
markets. Similarly, frictions in redeploying factors across rms could be less important for
labor. Therefore, whether nancing frictions can have an economically signicant impact on
aggregate productivity by constraining the reallocation of labor is ultimately an empirical
question.
We focus on the within-industry resource allocation.
5
Reallocation of labor is dened in
broad terms to include any change in the shares of labor allocated to dierent rms in an
industry. These changes in labor shares will incorporate both direct reallocations of labor
across rms, where workers switch rms, but also the dierential employment growth rates
of rms within an industry. Labor reallocation gains are then the component of industry
productivity growth that can be explained by changes in the labor shares of rms over time.
Our approach allows us to quantify the impact that these major state-level credit market
reforms have on the aggregate output of local industries through the labor reallocation chan-
4 Cestone & Fumagalli (2013) and Brown & Matsa (2013) provide evidence supporting this idea.
5 Our approach follows Olley & Pakes (1996) and Hsieh & Klenow (2009). This focus on the within industry
allocation of resources is often motivated by the existence of signicant and persistent gaps in productivity
within industries Bartelsman et al. (2013).
45
nel. By considering dierent decompositions of industry productivity growth, our approach
also allows us to compare the economic importance of this eect to alternative channels
through which credit markets can aect the aggregate productivity and performance of local
industries. Credit markets can aect the aggregate productivity growth of local industries
through changes in the reallocation of capital, changes in rm-level productivity growth, or
changes in the entry and exit decisions of rms. Finally, in addition to changes in the real-
location of labor, credit markets can aect the performance of local industries by allowing
them to expand their aggregate labor.
We implement this analysis with plant-level data from the U.S. Census Bureau on a
broad sample of small U.S. manufacturing rms.The essential requirements for the imple-
mentation of our approach are measuring within industry gaps in rms marginal products
and empirically isolate the impact of credit market reforms. In order to measure dierences
in rms marginal products, we build on previous research in empirical industrial organization
which explicitly addresses the simultaneity and selection biases involved in the estimation of
production functions (Olley & Pakes (1996), Levinsohn & Petrin (2003), and Ackerberg et
al. (2007)).
We examine the within industry reallocation of labor and the magnitude of industry
productivity changes after state-level deregulation in credit markets,when compared to in-
dustries in states that did not deregulate credit markets around the same time.The state-level
deregulations that we study allowed banks to operate across state borders, as well as reduced
local bank monopolies. During our sample period, small U.S. rms heavily relied on loans
from local banks as a source of external nancing (e.g., Petersen & Rajan (1994)). Previous
research has suggested that these reforms aected local credit markets, leading to higher lo-
cal economic growth and mattered especially for small local rms (e.g., Jayaratne & Strahan
(1996) and Cetorelli & Strahan (2006)). These state-level deregulations have the advantage
that they are staggered across states over time. Kroszner & Strahan (1999) (hereafter KS)
provide evidence suggesting that these dierences in timing across states were not related to
contemporaneous changes in state-level economic or banking conditions.
6
6 Kroszner & Strahan (1999) argue that these reforms were triggered by national-level technological changes,
which weakened local banking monopolies and reduced their incentives to ght against deregulation, and
that dierences in the timing of deregulation across states largely capture long-term state characteristics
46
We estimate that this state banking deregulation is associated with economically impor-
tant increases in labor reallocation gains.Across dierent deregulation episodes and speci-
cations, these increases represent between 20%-45% additional increases in productivity over
time relative to pre-deregulation changes in productivity.We show that our results are robust
to examining geographically close markets that span multiple geographically close states that
experience dierent timing of state banking deregulation. By examining how credit market
reforms aects a specic component of aggregate industry productivity growth, labor re-
allocation gains, we isolate how important shifts in credit conditions matter for aggregate
industry productivity through the labor reallocation channel.
We then quantify how these additional reallocation gains associated with credit market
deregulation aect the level of industry output and productivity. The scope for such eects
is arguably more limited in the U.S. relative to many other countries in which resource
misallocation has been studied. We therefore evaluate these previous magnitudes not only
on the average local industry in our sample, but also in subsamples where the scope for
such gains is predicted to be larger. We predict such gains using data prior to deregulation
and measures of potential reallocation gains using our framework. Intuitively, industries
with high potential gains are industries with higher dispersion of marginal products prior
to deregulation. We nd that these changes in labor reallocation lead to economically large
increases in industry productivity especially in industries with high dispersion of marginal
products.
We nd that these results are robust to several checks on the two essential requirements of
our analysis. First, we address a potential concern regarding the accuracy of our measured
marginal product of labor dierences across rms. We nd that our results are robust
across a wide range of specications and approaches to estimating production functions,
including evidence that our results are not driven by omitted dierences in worker skill
across rms. Second, we address a concern with the identication of the eect of local
banking deregulation. In our basic ndings, identication comes from the staggered nature
of deregulation episodes across states.
Our identication hinges on the assumption that state-level banking deregulation is not
predicting the response of interest groups to these national-level changes.
47
related to other changes dierentially aecting the growth of higher marginal product rms
within local industries. We provide direct evidence that deregulation is not correlated with
prior changes in this dierential growth. We then examine these ndings in depth by con-
structing a sample of geographically and economically closely matched industries. For each
local industry in a state that deregulated credit markets during our sample (treated in-
dustry), we construct a group of control industries which include only geographically close
industries located in states that did not deregulate credit markets around the same period.
We nd that, relative to matched control industries, treated industries signicantly increase
their resource reallocation towards higher marginal product rms in the years immediately
after their deregulation episodes. Moreover, we nd that the magnitudes of these eects
match the ones from our basic results.
We also consider the impact of credit market deregulation on industry productivity
through the alternative channels previously discussed. We nd that labor reallocation gains
are important when compared to the productivity gains associated with capital reallocations
and these alternative channels. Consistent with prior research, we nd that credit mar-
ket deregulation is associated with increases in rm-level productivity (Krishnan, Nandy,
& Puri (in press)). For the average industry, we nd that the magnitude of the previous
rm-level productivity eect is comparable to the ones of the reallocation of production fac-
tors, but economically smaller. Moreover, in industries more likely to have misallocation,
the magnitude of the labor reallocation channel is signicantly larger than the rm-level
channel documented by Krishnan, Nandi and Puri. These results suggest the importance of
studying the implications of nancing frictions for productivity at the industry level and the
importance of labor reallocations in driving this gap.
While we nd evidence that rms entry and exit decisions change with deregulation,
our analysis suggests that the implications of these eects for industry productivity are
limited when compared to the intensive margin eects we document. In the context of the
U.S. banking deregulation experience, these ndings support the view that changes in credit
markets aect industry productivity more by improving the resource allocation of rms at
later stages of their life instead of improving selection of more productive rms at birth.
This is consistent with Kerr & Nanda (2009) who show that it is hard to predict the hard
48
to predict the quality of new rms before they start operating and producing results.
Overall, our paper makes two main contributions to a growing literature on the impact
of nance on resource allocation and aggregate productivity, which we discuss in greater
detail in the next section. First, we provide evidence that the labor reallocation channel
can be an economically important channel through which nancial markets aect aggregate
productivity. Second, we provide direct evidence that changes in nancial markets can
have economically important eects on aggregate productivity through their impact on the
intensive margin allocation of resources. Finally, our results suggest that such eects can be
signicant even in the context of the U.S..
These ndings have dierent broad implications. They suggest that nancing frictions
directly aect rms labor decisions and that incorporating such eects can be important for
understanding their real eects.
7
They also provide new evidence on the specic mechanisms
through which important reforms in credit markets can matter for the real economy and
relate to previous research on nancial development and growth. While this literature has
emphasized that nancial markets matter for economic growth, the specic mechanisms
driving this eect is still a topic of open debate.
8
Finally, they provide new evidence on the
determinants of dierences in aggregate productivity across economies. A growing body of
research has emphasized that dierences in the within-industry allocation of resources play
a signicant role in explaining aggregate productivity gaps at the industry or country level,
but has not converged on the underlying mechanisms driving these dierences in resource
allocation nor whether these productivity gaps may be mitigated by improvements in credit
markets.
9
7 Previous research has examined the impact of nancing frictions on rm employment and aggregate un-
employment (Benmelech, Bergman, & Seru (2012) and Chodorow-Reich (2014), as well as dispersions in
employment growth rates across industries (Pagano & Pica (2012), but has not on examined the eect on
rm and aggregate industry productivity.
8 For example, see Levine (1997), Rajan & Zingales (1998), Levine, Loayza, & Beck (2000) and the references
therein for the eect of nancial development on growth.
9 For example, see Olley & Pakes (1996), Hsieh & Klenow (2009), Bartelsman et al. (2013), and Asker,
Collard-Wexler, & Loecker (2014)
49
2.2 Related Literature
In this section we discuss in greater detail the connection between our paper and previous
research on how nancial markets aect the allocation of resources and aggregate produc-
tivity. Previous research has estimated calibrated models with nancing frictions and used
them to quantify the channels through these frictions aect aggregate productivity (Buera et
al. (2011), Midrigan & Xu (2014), and the references therein). A rst way that the analysis
in this paper complements these papers is by considering the role of the labor reallocation
channel. We provide evidence on how dierences in nancial markets aect the reallocation
of resources and aggregate productivity conditional on the importance of other factors. In
practice, there is a range of frictions potentially distorting the allocation of resources within
an industry, such as labor and product market regulations, and political institutions. For
tractability, calibrated exercises typically assume these frictions are not present and attribute
all deviations from benchmarks in resource allocation to nancing frictions.
10
A nal way
that our analysis complements these exercises is providing direct evidence on how signicant
changes in credit markets aect the dierent determinants of industry productivity growth.
Other papers have also connected credit markets reforms or measures of nancial devel-
opment to dierences in resource allocation within and across industries. Wurgler (2000)
relates cross-country dierences in nancial development to a measure of how eciently
countries allocate capital across their industries. Bertrand, Schoar, & Thesmar (2007) an-
alyze how French banking deregulation reforms aect the entry and exit decisions of rms
and the link between their product market shares and operating performance. Cetorelli &
Strahan (2006) and Kerr & Nanda (2009) study how U.S. state-level banking deregulations
aect the size distribution of rms and their entry and exit decisions, respectively. While the
eects documented in this previous research are likely to have implications for aggregate pro-
ductivity, these implications are not explicitly analyzed. In the absence of such analysis, the
quantitative implications of these results for the dierent channels through which nancial
markets aect aggregate productivity are unclear. More specically, it is unclear from this
10While we do not have a calibrated model, Moll (2014) emphasizes that tractability issues limit researchers
ability to evaluate the robustness of such quantitative exercises to dierent specications of the environment
and illustrates how changes in some commonly used assumptions, such as a focus on steady-state outcomes,
can have rst-order eects on the results.
50
evidence whether nancial markets can have a rst-order eect on aggregate productivity
by aecting the reallocation of labor.
Larrain & Stumpner (2013) explicitly analyze how cross-country dierences in nancial
development across Eastern European countries aect dierent components of aggregate
industry productivity. They do not consider the role of nancial markets in aecting aggre-
gate industry productivity through the reallocation of labor and assume that rms marginal
products of labor are equalized to wages, what implies that such gains are equal to zero.
Their analysis also does not separate the eect of nancial markets on industry productivity
through intensive margin reallocations from their eects through changes in the entry and
exit decisions of rms in the data due both to market selection and data coverage.
2.3 Methodological Framework
In this section, we describe our methodology to quantify the signicance of the labor real-
location channel in greater detail and then present the results implementing our methodology.
2.3.1 Measuring Marginal Reallocation Gains
We start by illustrating how to isolate the contribution of resource reallocation to marginal
changes in industry productivity using rst-order approximations for changes in industry out-
put over time. A rm i in industry j and time t can produce output Y
ijt
with a production
function given by:
Y
ijt
=A
ijt
F (K
ijt
;L
ijt
;M
ijt
)
(2.1)
where A
ijt
is a time-variant and rm-specic productivity component, K
ijt
is the rm's
capital stock, L
ijt
denotes the labor used in production, and M
ijt
denotes materials. As is
common in the productivity literature, productivityA
ijt
is modelled as a Hicks-neutral term.
As is also common in this literature, we dene rms' output as their total revenues de
ated
with an industry-specic price de
ator. Firm total factor productivity (TFP) is dened as
A
ijt
.
We dene dierences in industry productivity as dierences in industries value added
51
given the same aggregate industry factors. Similarly, we dene industry productivity growth
as the industry value added growth in excess of what can be predicted by the aggregate
growth of industry production factors. We focus on value added because it avoids double
counting output across industries. In our main results, percentage dierences in industry
value added are measured at a xed price for industries output.
11
We are interested in an-
alyzing how the reallocation of resources across an industrys existing rms contributes to
industry productivity growth. In our initial analysis of marginal changes in industry pro-
ductivity, we focus on industry productivity gains conditional on a given sample of industry
rms. When we extend our current analysis to examine the contribution of resource real-
location to cumulative changes in industry productivity, we explicitly take into account the
fact that this sample of rms changes over time due to entry and exit. LetI
jt
denote a xed
set of rms that exist in industry j around time t. Our rst denition of industry output is
Y
jt
=
P
i2I
jt
Y
ijt
.
12
For any production factorX
ijt
, letX
jt
=
P
i2I
jt
X
ijt
denote the industry
aggregate factor. Notice that, in general, the aggregation of rms production functions will
not necessarily lead to an industry production function with a separable TFP term as in (1).
In general, the simple aggregation of rms individual outputs gives us:
Y
jt
=G(N
jt
;A
ijt
;SK
ijt
;SL
ijt
;SM
ijt
;K
jt
;L
jt
;M
jt
)
(2.2)
whereSF
ijt
=
F
ijt
F
jt
is a rms industry share of production factorF ,N
jt
is the number of rms
inI
jt
, andA
ijt
;SK
ijt
;SL
ijt
;SM
ijt
denotes the joint distribution of these variables acrossN
jt
observations.
The allocation of resources in this framework is dened in broad terms and captures any
dierences in the shares of factors allocated to dierent rms within an industry.
13
Changes
11We will also consider measuring dierences in industry productivity using simple dierences in industry
total sales minus material costs. Evaluating dierences in output at xed prices is common in measures of
aggregate productivity (e.g., Basu & Fernald (2002), and Petrin & Levinsohn (2012)
12If rms face dierent relative prices within an industry, this measure will not necessarily capture real
industry output. We argue that if dispersion in rm-specic prices is important in a typical industry, it
is likely to capture dierentiated goods to a great extent. Therefore, one cannot simply sum real output
across rms to construct a measure of real industry output. We will consider an alternative measure of
industry output below.
13This broad denition of resource allocation is commonly used in studies of industry productivity growth
(e.g.,Olley & Pakes (1996)) and the literature linking within-industry resource allocation to aggregate
productivity (e.g., Hsieh & Klenow (2009))
52
in these shares, which we label resource reallocation, will incorporate both direct reallocations
of resources across rms, such as asset sales, but also the dierential growth rates of rms
within an industry. By using a rst-order approximation, we can isolate the importance of
changes in the allocation of resources in explaining marginal changes in industry productivity
over time. More formally, industry productivity growth is dened as:
IPG
jt
=

1
1sm
jt

dln(Y
jt
)
dt

jt
dln(K
jt
)
dt

jt
dln(L
jt
)
dt

jt
dln(M
jt
)
dt

(2.3)
where sm
jt
is the ratio of industry material costs to industry revenue and
jt
,
jt
, and

jt
denote industries capital, labor and materials' elasticity, respectively. The elasticity of
each of these factors is computed using the marginal product of the aggregate factor in
(2). For example, industry capital elasticity can be dened as
jt
=
K
jt
Y
jt
dY
jt
dK
jt
. This will tell
us the increase in aggregate output predicted by an increase in aggregate factors, holding
constant these other determinants of aggregate output. The term

1
1sm
jt

converts these
industry output gains into value added gains measured at current industry prices. Note that

1
1sm
jt

=
Y
jt
VA
jt
, whereVA
jt
is industry value added. In the simple case where the industry
production function has a separable TFP term as in (1), then (3) will estimate industry
productivity growth as TFP growth scaled by

1
1sm
jt

.
In Appendix A we show that one can write (3) as:
IPG
jt
=

1
1sm
jt

0
@
X
i2I
jt
Y
ijt
Y
jt
dln(A
ijt
)
dt
+LRG
jt
+KRG
jt
+MRG
jt
1
A
(2.4)
where LRG
jt
=
L
jt
Y
jt
P
i2I
jt
dY
ijt
dL
dSL
ijt
dt
denotes labor reallocation gains and the other two
terms are dened analogously based on capital and materials. The rst term in (2.4) captures
the contribution of rm-level productivity growth to industry growth. The other three terms
capture the contribution of resource allocation to industry productivity growth, which we
label as reallocation gains. These gains capture the additional growth in industry output
due to shifts in rms factor shares. More precisely, they capture the dierence between
the realized marginal growth of industry output and the growth we would observe in the
absence of any changes in factor shares. To illustrate the intuition for these gains, consider
53
the case of labor reallocation gains. Since
dSL
ijt
dt
has to add up to zero in the industry, these
gains capture an industry covariance between rms marginal products and
dSL
ijt
dt
. Intuitively,
reallocation gains are positive (negative) only to the extent that higher marginal product
rms grow faster (slower) within an industry.
We emphasize the dierent potential determinants of reallocation gains. In Appendix A
we show that one can approximate LRG
jt
as:
LRG
jt

Var

dY
ijt
dL

E
dY
ijt
dL
L
jt
Y
jt
LRSens
jt
(2.5)
whereVar(:) andE(:) capture variance and expected values measured using the industry
distribution and LRSens
jt
is the sensitivity of labor reallocation to the marginal product
of labor in the industry. LRSens
jt
is the additional increase in
dlog(SL
jt
)
dt
predicted by a
given percentage increase in
dY
ijt
dL
. More formally, it is the coecient on the log of
dY
ijt
dL
in
a linear regression of
dlog(SL
jt
)
dt
on the previous variable and a constant.
14
This sensitivity
measures the extent to which industries reallocate resources in response to a given gap in
the marginal product of its rms and, intuitively,captures dierences in the way industries
allocate resources across given opportunities.
As equation (5) illustrates, the impact of changes inLRSens
jt
onLRG
jt
depends on the
degree of dispersion in marginal products within the industry and the labor-to-output ratio in
the industry. The same sensitivity of reallocation to gaps in marginal products translates into
higher productivity gains when there are larger gaps in marginal products in the rst place.
The output gains from changing these shares are also more important when the industry
relies more on the factor per unit of output. These eects are measured by
LRG
jt
LRSens
jt
, which
captures dierences in the potential industry productivity gains from reallocating resources
across opportunities in a given way. We label this ratio as the potential reallocation gains.
14The approximation comes from the fact that we replace a regression coecient in levels by one measured
in logs adjusted based on the average value of the variables.
54
2.3.2 Examining the Impact of Credit Market Reforms
We examine the impact of a signicant credit market reform on our previous reallocation
gains.By doing this, we can evaluate how these changes in credit markets aect industry pro-
ductivity through their impact on previous components of industry productivity changes.
15
Moreover, we analyze this eect on the dierent components of marginal reallocation gains -
either by reallocation gains of existing rms or other determinants of industry productivity,
such as rm entry and exit decisions. This allows us to better understand how credit markets
impact reallocation gains. To the extent that credit markets matter by in
uencing the allo-
cation of resources across given opportunities, we should expectthem to aect reallocation
gains through the sensitivity of resource reallocation to marginal products. Notice that all
the terms in this analysis can be measured if we have estimated the production function
specied in (1).
The credit market reforms we examine are state-level banking deregulations. Prior to the
1970s most U.S. states had restrictions on banks ability to operate within and across state
borders that had remained historically stable. Given that small U.S. rms mostly relied on
geographically close banks as a source of external nancing until the early 1990s (Kroszner
& Strahan (1999), hereafter KS). Between the early 1970s and early 1990s states relaxed
these restrictions in a staggered way. Following previous research on U.S. state banking
deregulation, we focus on two main types of restrictions imposed by states. First, states
imposed restrictions on intrastate branching. For example, these included restrictions on
the ability of multibank holding companies to convert branches of acquired subsidiary banks
into branches of a single bank, as well as restrictions on banks ability to open new branches.
As Jayaratne & Strahan (1996) and others, we choose the date of intrastate deregulation as
the date in which a state permits branching through mergers and acquisitions. Second, the
Douglas amendment to the Bank Holding Act of 1956 prevented a bank holding company
from acquiring banks in another state unless that state explicitly permitted such acquisitions
by statute. No state allowed such acquisitions until the late 1970s. States then entered
15After presenting our main analysis, we also provide some evidence on the relative importance of this channel
versus other channels through which productivity can be impacted by credit markets, such as changes in
rm-level productivity and rms entry and exit decisions.
55
reciprocal regional or national arrangements which allowed their banks to be acquired by
banks in any other state in the arrangement. Except for Hawaii, all states had entered such
agreements in 1993. These episodes of interstate deregulation culminated with the passage of
the 1994 Riegle-Neal Interstate Banking and Branching Eciency Act, which codied these
state-level changes at the national level. As emphasized by Cetorelli & Strahan (2006),
because of national-level deregulation and changes in lending technology (Petersen & Rajan
(2002), it becomes increasingly less plausible to view banking markets as local after this
period. Our data is available from 1976 and, motivated by the above timeline, we end our
sample in 1993.
We follow Amel (1993) and Kroszner & Strahan (1999) in determining the dates of
interstate and intrastate deregulation. Table 2-1 shows these dates and illustrates the large
number of interstate deregulation episodes during our sample period. Given that previous
research has provided direct evidence that these deregulation episodes are associated with
changes in the borrowing terms of small local rms, e.g. reductions on interest rates, we will
focus our analysis on the industry productivity consequences of these deregulation episodes.
16
We are interested in linking changes in aggregate industry productivity to overall credit
market conditions faced by an industry. Therefore, the unit of analysis in our results will be
an industry-state, which we label as a local industry. In our analysis, we only include small
rms with a strong geographic exposure to a given state. More specically, when dening
each local industry, we include only single-plant rms. As we discuss below, these rms
represent a signicant portion of the aggregate sales and factors in their industry-state. Our
results then analyze changes in the aggregate productivity of these local industries using our
previous framework.
17
16See Kroszner & Strahan (1999), Kerr & Nanda (2009), and the references therein for a more detailed
discussion of state banking deregulation and previous research documenting its eects.
17One issue with this approach is that, in addition to entry and exit in an industry, rms might transition
between being single-plant and a multi-plant rm. We found that these transitions are empirically limited
and their implication for the aggregate productivity of local industries is limited.
56
2.3.3 Alternative Measures of Industry Output and Productivity
We also consider alternative measures of industry output. One important issue is that
dierences in rm outputY
ijt
=P
ijt
Q
ijt
in equation (1) can re
ect dierences in the physical
quantity of output Q
ijt
but also capture dierences in rm-specic relative prices P
ijt
((as
emphasized by Foster et al. (2008)). If this dispersion in rm-specic prices is important
then our previous industry output measure will not capture real industry output. We argue
that if dierences in relative prices are important in a typical industry, they are likely to
capture dierentiated goods to a great extent. Therefore, one cannot simply sum real output
across rms to construct a measure of real industry output. One alternative approach to
aggregate rm output into a measure of industry output is to dene industry output growth
as the weighted average growth of rm real output in the industry. More precisely, we can
dene industry output growth as
dln(Y
jt
)
dt
=
P
i2I
jt
Y
ijt
Y
jt
dln(Q
ijt
)
dt
.
18
Given an initial value forY
jt
,
then this measure of output will be uniquely determined by the values of Y
ijt
and Q
ijt
over
time. We can then follow all of our previous steps with this alternative measure of industry
output. Our analysis will not depend on the initial value we set for industries output in an
initial year.
There are dierent advantages from using this approach to measure industry output.
First, as in previous studies measuring aggregate productivity growth with plant-level data
(e.g., Petrin & Levinsohn (2012), hereafter LP), the contribution of changes in a rms
real output to aggregate value added is evaluated using current rm prices. Moreover, our
measure of (marginal) industry productivity growth in this context can be derived from the
framework proposed by LP to measure economy-wide productivity growth.
19
Second, this
measure of industry output can be derived as a measure of industry real output in previous
models used to study resource misallocation, where heterogeneous rms sell dierentiated
intermediate inputs. This will be the case if industry output is produced in a competitive
18A motivation for weighting rm output growth in this way is that prices might capture users marginal
valuation of dierentiated goods.
19The framework proposed by PL allows one to measure the contribution of an industry to aggregate pro-
ductivity growth, which might come from expanding industry aggregate factors. We are only interested
in productivity gains conditional on the aggregate factors of an industry and show in Appendix A that
our measure of industry productivity growth can be derived as a component of the PL measure that only
captures this eect.
57
market and is a CES aggregator of these intermediate inputs, as in the frameworks proposed
by Hsieh & Klenow (2009) and Bartelsman et al. (2013).
20
In principle, one challenge for implementing our analysis in this context is the absence of
extensive data on rm-specic pricesP
ijt
and real quantitiesQ
ijt
across industries. However,
under plausible assumptions, we can make inferences about our results with this industry
output measure using our previous analysis, which do not require measuring P
ijt
and Q
ijt
.
Our analysis of of geographically close markets using a dierence-in-dierence methodology
also allows us to control for time and industry xed eects. In addition, in Appendix A
we show that if rms face the same elasticity of demand for their dierentiated products
j
within each industry then we have the following results. Reallocation gains in this context
are given by our previous gains multiplied by

j

j
1

. Moreover, we can decompose these
gains in an analogous way to our previous case. In this decomposition, the sensitivity of
resource reallocation to marginal products remains the same as before, and potential gains
from reallocation can be obtained by multiplying our previous value by

j

j
1

.
21
Intuitively,
these results all come from the fact that, for any given factor F , reallocation gains are now
evaluated by replacing
dY
ijt
dF
with P
ijt
dQ
ijt
dF
and P
ijt
dQ
ijt
dF
=
dY
ijt
dF

j

j
1

.
22
2.3.4 Estimation of Production Functions
In order to implement the previous analysis, we need to rst estimate the production
function specied in equation (1) for each industry. We build on previous research in empir-
ical industrial organization which explicitly addresses the simultaneity and selection biases
involved in the estimation of production functions (Olley & Pakes (1996), Levinsohn & Petrin
(2003), and Ackerberg et al. (2007)). Our analysis uses these estimates as inputs and does
not imply anything about how these production functions should be estimated. We therefore
20Under these assumptions, real industry output growth will be given by the previous weighted average of
rms real output growth. Intuitively, rm-specic prices will measure the marginal rate of transformation
between rm intermediate goods and industry output.
21This assumption can be interpreted as an approximation and is common in recent models linking within-
industry resource allocation to aggregate productivity (e.g., Hsieh & Klenow (2009) and Bartelsman et al.
(2013)
22After presenting our initial evidence, we also measure dierences in industry productivity using total
dierences in industry value added conditional on aggregate factors or dierences in the weighted average
of rm productivity in the industry (e.g., Olley & Pakes (1996)
58
consider a range of approaches to estimate production functions. We consider both translog
and Cobb-Douglas specications.The key advantage of the translog specication is that it
can be thought as a second-order approximation to any production function specied in (1).
It does not impose the assumption that the factor elasticity of labor, capital and inputs
are constant as the Cobb-Douglas does impose. Instead, it allows this elasticity to depend
on rms choices of all inputs. This is important as a factors elasticity plays an important
role in determining its marginal product and the central aspect of our analysis is modeling
heterogeneity across rms in marginal products.
In our main results, we estimate (1) using the two approaches. First, we consider the ap-
proach proposed by Olley & Pakes (1996) (hereafter OP). We then consider extensions of this
approach building on the insights of Levinsohn & Petrin (2003) and Ackerberg et al. (2007)
(hereafter ABBP). The approach in OP controls for the simultaneity and selection problems
involved in the estimation of (1) by using a proxy method where one uses rms investment
decisions to construct proxies for their unobserved productivity parameters. Levinsohn &
Petrin (2003) suggest using rms choices of other inputs as proxy variables. ABBP discuss
some of the issues with this approach and suggest directions to accommodate them. We
build on these insights and extend the OP approach to use both investment and materials
as proxy variables. We term this approach as LP.
We note that the explicit assumptions on primitives that we make when using these dif-
ferent approaches are consistent with the importance of nancing frictions analyzed in this
paper. A key assumption across these approaches is that one can uniquely pin down rm
productivity, A
ijt
after conditioning on rms choices and characteristics at time t. In the
context of OP, this assumption means that there must be a unique mapping between rms
investment and its productivity at period tfor rms with positive investment, after condi-
tioning on its initial capital stock and age. This condition is consistent with the existence of
nancing frictions. While it does not allow rms exposure to these frictions to be arbitrarily
heterogeneous across rms, it allows this exposure to be a function of rm age, size and
productivity.
23
23A simpler and alternative approach to estimate (1) is to assume that labor and materials factor shares are
equal to their respective elasticity and recover the capital elasticity assuming constant returns to scale.
However, this approach relies on the assumption that rms equate their marginal products of labor and
59
We estimate (1) separately for each 3-digit SIC code using plant-level panel data, which
we describe in greater detail in Section 2.2. In our robustness analysis, we consider additional
variations of OP, also discussed by ABBP, that rely on dierent assumptions. We also
consider simple alternative approaches such as OLS regressions. Appendix B describes these
approaches and their implementation in greater detail.
2.4 Data and Summary Statistics
Our main data sources are the Longitudinal Business Database (LBD),the Census of
Manufacturers (CM) and the Annual Survey of Manufacturers (ASM) from the U.S. Census
Bureau.The CM provides information on the sales and inputs used by all manufacturing rms
every ve years.Our analysis tracks over time the allocation of resources within industries
across small rms, what requires data over time on a comprehensive number of small rms
in these industries. Higher frequency data on small rms is useful in our analysis as it allows
one to more precisely link changes in credit markets to changes in resource allocation. The
ASM allows one to track this same information for a subsample of manufacturing rms in
non-census years through rotating ve-year panels.However, while large plants are sampled
with probability one, small plants are sampled randomly with probabilities that decline
with their size.When compared to samples of local industries in the CM, samples of local
industries constructed in this way capture less than 10% of the rms of interest for our
purposes. This issue is particularly relevant in the context of this paper because we need
to measure within industry correlations over time. We address this challenge by combining
the CM with the LBD. The LBD provides annual employment and payroll information for
every private establishment from 1976 onward. The underlying data are sourced from U.S.
tax records and Census Bureau surveys. We use the LBD to annually track over time the
within-industry reallocation of labor and link to the CM to relate this reallocation to rm
marginal products and rm productivities. We can only directly measure the reallocation of
labor at an annual frequency, an issue that we explicitly address in our analysis. We measure
materials to their respective factor costs. This assumption is inconsistent with the analysis in this paper,
which is motivated by the existence of wedges between rms marginal products of labor and labor costs
(wages).
60
rms marginal products and productivities in a given year using data from the last available
Census and address the potential measurement issues associated with this approach. We
also use the LBD to track the entry and exit of rms.
We construct our initial data by matching single-plant rms in the LBD and CM. As
previously discussed in Section 2.1.6, we focus on smaller single-establishment rms. Most
establishments in manufacturing belong to a single-establishment rm. While these rms are
small, in aggregate they represent closeto 50% of the overall sales and employment of their
industry-state on average across all years. Therefore, this sample of U.S. local industries
captures a large portion of the U.S. economy.
Table 2-2 provides summary statistics on our main sample. It also provides information
on the estimated average factor elasticity across factors and our dierent production function
specications. Additionally, it shows the within-industry dispersion of estimated marginal
products and rm TFP across these dierent approaches. Since the methods outlined in
Section 2.1.3 require panel data, we estimate the industry-level parameters of the production
functions specied in (1) using the ASM. We construct our measures of marginal products
and rm productivity by combining data from the CM with these estimated industry-level
parameters. Variable denitions are in Appendix C.
2.5 Results
2.5.1 Labor Reallocation
Following our methodological framework, we examine how local credit market deregula-
tion relates to changes in within-industry labor reallocation gains. We start by examining
how the sensitivity of labor reallocation to the marginal product of labor in local industries
relates to local credit market deregulation. A rst approach to examine this relationship is
to estimate:
EmpShare
isjt
=
sjt
+
0
MPL
isjt
+
1
Dereg
st
MPL
isjt
+X
isjt
+
isjt
(2.6)
61
where EmpShare
isjt
is the change in the employment share of rm i in industry j, state
s and time t,
sjt
is a state-industry-year xed eect, MPL is the log of rm marginal
product of labor, Dereg is an indicator that equals one if credit market deregulation has
been passed in the state andX denotes age controls. These controls age variables as well as
their interaction with X. Employment share is the ratio of rm employment to the overall
employment of a rms industry-state. EmpShare
isjt
is measured as the log dierence of
this share between year t and t 1. Only rms present in the industry-state in both year t
and t 1 are included in the sample and the computation of the employment share.
Notice that
0
tells us the sensitivity of employment reallocation to the marginal product
of labor for industries located in states that have not deregulated credit markets, i.e. it mea-
sures an average value of LRSens
jt
across these industries (See Section 2.1.2). Also notice
that the state-industry-year xed eects ensure that this relationship captures a correlation
within an industry-state-year.
The coecient of interest is
1
and tells us the dierential value of this sensitivity for
industries located in states with deregulated credit markets. The age controls X include
the one-year lag of age, its squared value, as well as the interactions of both these variables
withDereg. There are important life-cycle patterns in productivity, and we want to capture
dierences between the marginal products of rms at the same stage of their life cycle.
One potential issue with this approach is that
1
might be capturing cross-state dierences
and times-series trends in the employment reallocation of industries.We address these issues
by controlling for both xed dierences across states and time-series changes in the employ-
ment reallocation of local industries. This is done by adding state and year xed eects
interacted with MPL as controls in the estimation of (7). After we add these controls, the
estimation of
1
can be thought as a dierence-in-dierences estimation of how state credit
market deregulation aects the labor reallocation sensitivity of local industries. Intuitively,
one can think about this estimation as involving two steps. First, we estimate the sensitiv-
ity of labor reallocation to the marginal product of labor within each industry-state-year.
We then estimate how deregulation aects this relationship using a dierence-in-dierences
specication. We are implementing these two steps together in a single regression.
24
If
24Notice that the sample of rms used to estimate this relationship is changing over time and can be aected
62
dierences in the timing of deregulation across states capture long-term dierences across
them, as argued by Kroszner & Strahan (1999), this approach will isolate the impact of
deregulation on LRSens.
In addition to these controls, we also include rm xed eects to control for xed dier-
ences across rms in their employment growth. This leads us to estimate:
EmpShare
isjt
=
sjt
+
i
+
s
MPL
isjt
+
t
Dereg
st
+
1
Dereg
st
MPL
isjt
+X
isjt
+
isjt
(2.7)
where
i
denotes rm xed eects,
t
denotes year xed eects,
s
denotes state xed eects,
and the other variables are dened as in equation (6).
Table 2-3 reports results of the estimation of equations (7) and (8). We consider both
intrastate and interstate credit market deregulation episodes, and use both translog and
Cobb-Douglas production function specications. Furthermore, we estimate these produc-
tion functions based on both the OP and LP approaches. Panel A of Table 2-3 reports
the estimated coecients for
1
, which capture changes in LRSens. Panel B of Table 2-3
quanties the magnitude of the percentage changes in LRSens implied by these eects. We
compare our estimates for
1
to the average sensitivity of employment reallocation to the
marginal product of labor prior to deregulation. We nd that local credit market dereg-
ulation is associated with both economically and statistically signicant dierences in the
sensitivity of labor reallocation to the marginal product of labor. We nd that credit market
deregulation leads to percentage increases in LRSens between 27%-32% and 46%-49% in
the context of intrastate and interstate deregulation episodes, respectively. This evidence
suggests that credit market deregulation is associated with signicant changes in the extent
to which industries reallocate resources in response to a given gap in marginal products.
by deregulation. Motivated by our analysis in Section 2.1, we are interested in analyzing how an industry
measure (LRSens) changes with credit market deregulation. At any given year, this measure has to be
computed using all existing rms in an industry.
63
2.5.2 Potential Gains from Reallocation
We now examine whether credit market deregulation is associated with changes in the
potential reallocation gains in equation (5). As equation (5) illustrates, reallocation gains are
the product of potential gains and LRSens. We examine the extent to which credit market
deregulation is associated with percentage changes in potential labor reallocation gains. By
combing these results with our previous estimates for the percentage changes inLRSens, we
can analyze the extent to which credit market deregulation is associated with overall changes
in labor reallocation gains. One reason to expect changes in potential reallocation gains is
that, as resources move towards higher marginal product rms, marginal products might
become more equalized across rms. However, in practice, the signicance of this eect
is unclear for at least two reasons. First, the extent to which labor reallocation feedbacks
into lower dispersion in rm marginal products will depend on the curvature of production
functions and how rms adjust other factors.
25
Second, credit market deregulation might also
aect the distribution of rm productivity in an industry, for example, because of changes
in individual rm-level productivity.
We address this issue by estimating:
Log(PotentialLRG)
sjt
=
sj
+
t
+Dereg
st
+
sjt
(2.8)
where PotentialLRG are potential labor reallocation gains in industry j, state s, and
time t,
sj
is a state-industry xed eect,
t
are year xed eects, and Dereg
st
is dened
as before. This approach is similar to the one in our previous results where our analysis is
equivalent to estimating a dierence-in-dierences specication with LRSens.
Table 2-4 reports the results. We nd that percentage changes in potential reallocation
gains are signicantly smaller in magnitude than our previously estimated the percentage
increases in LRSens. For example,in the case of interstate deregulation we estimate drops
in PotentialLRG between 5-8% and increases in LRSens between 45%-49%. We conclude
that the percentage changes in LRSens associated with credit market deregulation mostly
translate into percentage increases in labor reallocation gains. Therefore, credit market
25For example, if rms adjust all factors together and returns to scale are close to one, then there will be a
limited drop in the dispersion of rms marginal products.
64
deregulation is associated with signicant percentage changes in labor reallocation gains and
this eect is driven by changes in the sensitivity of reallocation to marginal products.
2.5.3 Quantifying Cumulative Productivity Gains from Changes
in Labor Reallocation
We quantify the cumulative industry productivity gains implied by these changes in
marginal labor reallocation gains. We quantify these gains by considering the contribution
of resource reallocation to cumulative changes in industry productivity over a given period
of time. More precisely, we want to analyze the dierence between the realized growth
rate of industry value added between years t and t + and the growth rate we would
have observed in the absence of any changes in factor shares over these years. As in our
previous analysis, we evaluate this dierence conditional on other determinants of changes
in industry output and at constant industry output prices. More specically, as before,
we hold constant changes over time in industry total factors and rm-level productivity.
We now explicitly consider entry and exit decisions, which will change the composition of
rms over time. This additional growth in industry value added between years t and t +
will capture the contribution of resource reallocation in the intensive margin to changes in
industry productivity over this period. LetVA
j
(t;t+) denote this value-added contribution,
measured att+ industry output prices. Let alsoRG
js
=LRG
js
+KRG
js
+MRG
js
denote
a discrete time approximation for our previous reallocation gains, computed in the sample
of rms that exist in the industry in both years sand s 1. In Appendix A we show that,
under plausible conditions, a rst-order approximation for VA
j
(t;t +) is given by:
VA
j
(t +)

1
1sm
jt+

(1s)(RG
jt+
+RG
jt+1
+ +
1
RG
jt
(2.9)
where s is the share of new entrants in the industrys output, =
(1s)
1+
, captures the per-
sistence of rm productivity and is the growth rate of rm-level productivity.
26
In general,
all these parameters can change by year. We have set them as constant for expositional
26The approximation comes from the fact that we use rst-order approximations for the output growth of
the industry in each year.
65
simplicity.
Intuitively, dierences in resource reallocation over time only matter by aecting the
nal allocation of resources in year t +, and their impact on industry output in year t +
needs to be evaluated using the values of rm productivity in that same year. Equation (9)
illustrates the dierent reasons to expect a reallocation gain today to have an impact on
the future level of industry productivity that goes away over time. First, the distribution
of productivity across existing rms in an industry will change over time. For example, if
dierences in rm productivity within an industry today are uncorrelated with dierences in
rm productivity in ten years, then reallocation gains taking place today should not aect
the level of industry productivity in ten years. Second, gains from reallocating resources
across existing rms in an initial year might diminish in importance in future years, as
entry and exit reduce the importance of the initial rms in the industry. Additionally, the
existence of rm-level productivity growth will lead to intensive margin changes in output
even in the absence of reallocation. Percentage output gains due to reallocation in the past
will represent a smaller percentage of future output.
These eects are captured by the discount rates , 1s, and
1
1+
, respectively. These
discount rates capture simple moments in the data that we directly measure.
We use Equation (9) to compute the cumulative gains over the sample due to the greater
labor reallocation gains associated with credit market deregulation. More specically, we
determine the discount rates in Equation (9) using moments from our overall sample. We
estimate additional reallocation gains using our estimated percentage changes in these gains
combined with the average value of these gains prior to deregulation. For each type of dereg-
ulation episode, we accumulate these eects over the average number of deregulated years
in the sample. On average, states have credit markets that already passed interstate and
intrastate deregulation during 7.3 and 14.8 years in our sample, respectively. As discussed
in Section 2.1, we consider two measures of industry output and equation (9) captures these
gains using our rst measure. We translate reallocation gains using this rst measure into
gains with our second measure by scaling them by

1
where is the average demand elasticity
for rms products across industries. The literature tends to use values for the average de-
66
mand elasticity between three and ve and we set this value equal to four.
27
Table 2-5 reports
the magnitudes of cumulative productivity gains across dierent specications and industry
productivity measures. We denote these gains computed with our rst and second measures
of industry output as Industry Productivity Gain
1
and Industry Productivity Gain
2
, re-
spectively. Panel A of Table 2-5 reports estimates of these magnitudes using all industries
in our sample. The magnitudes of productivity gains are signicantly more important for
interstate deregulation episodes.
28
Notice that our goal is examine if some signicant re-
form in credit markets leads to substantial reallocation gains. Our null hypothesis is that
the reallocation channel is not quantitatively important and, therefore, reforms in credit
markets cannot lead to sizeable productivity gains through this channel. We estimate that
the increase in labor reallocation gains associated with interstate deregulation leads to an
increase in the value-added of local industries between 1.5% and 2.0% over a horizon of
approximately seven years. To place these estimates in perspective, we note that Hsieh
& Klenow (2009) estimate that fully equalizing rms marginal products across all factors
in the U.S. manufacturing sector during the late 1980s would lead to increases in industry
productivity of approximately 31%. These estimated gains are directly comparable to the
ones based on our measure Industry Productivity Gain
2
. Relative to this benchmark, our
results suggest that, in seven years, changes in labor reallocation associated with interstate
deregulation generate approximately 6.5% of possible long-term gains from reallocating all
production factors.
We note that this gain is an average across all U.S. local industries and that the scope
for such reallocation gains is likely to be smaller in the U.S. than in other settings. Indeed,
the U.S. is used as a low frictions benchmark to calibrate the model in many studies of
misallocation (e.g., Hsieh & Klenow (2009)). We then evaluate our previous magnitudes on
a subset of local industries where the scope for such gains is predicted to be larger. We argue
that the eect in such industries is more likely to be representative of eects in environments
outside the U.S. where resource misallocation issues are likely to be more pronounced. We
27For example, see Hsieh & Klenow (2009), Bartelsman et al. (2013) and Asker et al. (2014). Using the
previous expression, it is simple to see that our estimates will be not very sensitive to dierent choices
within this range of values.
28This is consistent with Cetorelli & Strahan (2006) that provide evidence that interstate deregulation
episodes matter more for small manufacturing rms.
67
predict the scope for such gains using two criteria. First, we restrict our sample to industries
in the top 50% and 33% of potential gains from reallocation prior to deregulation. Intuitively,
these are industries with greater dispersion in marginal products. Second, we implement this
analysis only among industries in the top tercile of estimated returns to scale. As previously
discussed, the impact of changes in within industry resource reallocation might be mitigated
by drops in the potential gains from reallocation, and this eect is likely to be less relevant
in such industries. In this subsample, average estimated returns to scale are approximately
0.95. The nal subsamples in these results represent between 17% and 11% of our initial
sample.
In each of these exercises, we follow our previous steps, examining both percentage
changes in LRSens and potential gains from labor reallocation and then compute the mag-
nitude of cumulative industry value added gains. As in our previous results, we found
signicant increases in LRSens and no evidence of signicant drops in the potential gains
from reallocation in these subsamples. Panels B to D of Table 2-5 report the magnitude of
the cumulative productivity gains implied by these eects. The magnitude of these gains
in the overall sample of higher returns to scale is similar to the one in our previous results.
Panel C shows that the economic magnitudes of these gains in industries with high potential
reallocation gains is associated with productivity gains between 3.3% and 4.4% of industry
value signicantly larger than the gains in the overall sample.
Together, this evidence suggests that the productivity gains due tolabor reallocation after
credit market deregulation are signicant for the average U.S. industry and are economically
large for an important subset of U.S.industries with high dispersion in marginal products
and close to constant returns to scale.
2.5.4 Incorporating the Reallocation of Other Production Factors
We now incorporate potential productivity gains from the reallocation of other fac-
tors.Under additional assumptions, we can infer reallocation gains across all factors from
the reallocation of labor.
29
These additional assumptions are the following. First, rms
29We have also directly examined capital reallocation gains and found evidence that incorporating these
eects amplies our previously estimated labor reallocation gains. One major challenge in following this
68
industry shares of factors are constant across factors. In the Internet Appendix, we provide
direct evidence that, within our sample of small rms, industry shares across factors are
close to constant for a given rm. Second, plants returns to scale must be close to one. We
address this issue by restricting our analysis to industries with estimated returns to scale
close to one. Our previous results suggest that labor reallocation gains are similar in this
subset of industries.In Appendix A we show that, under these conditions, we can write in-
dustry productivity growth using our rst output measure as IPG
jt
=

1
1sm
jt

dln(A
jt
dt

,
whereA
jt
is a weighted average of rm productivity across the industrys existing rms. This
average is a commonly used measure of industry productivity in the literature (e.g, Olley &
Pakes (1996)) and the weights used to compute A
jt
are the industry shares of any produc-
tion factor. In our analysis we use labor. Moreover, let RG
jt
=LRG
jt
+KRG
jt
+MRG
jt
denote reallocation gains across all factors. In Appendix A, we also show that, under these
conditions,we can write these gains as RG
jt
=
1
A
jt
P
i2I
jt
A
ijt
dSL
ijt
dt
. Intuitively, we can now
evaluate reallocation gains using rm productivity, as opposed to marginal products.We can
also further decompose these gains in a similar way to equation (5):
RG
jt

Var(A
ijt
)
E(A
ijt
)
1
A
jt
TFPSens
jt
(2.10)
whereTFPSens
jt
is dened analogously toLRSens
jt
by replacing the marginal product
of labor with rm tfp. This measure captures the extent to which industries reallocate
resources towards higher tfp rms. The term
RG
jt
TFPSens
jt
captures the potential gains from
reallocating all factors.
We follow the same steps used in our previous results in this context. We rst estimate:
EmpShare
isjt
=
sjt
+
i
+
s
TFP
isjt
+
t
Dereg
st
+
1
Dereg
st
TFP
isjt
+X
isjt
+
isjt
(2.11)
approach is that we need to restrict our sample to the four years in which there was a Census of Manu-
facturers: 1977, 1982, 1987 and 1992. As discussed in Section 2.2, our analysis requires tracking a large
number of small rms. This reduces the precision of our estimates and makes it harder to identify the
impact of deregulation episodes.
69
where TFP
isjt
is rm TFP, and all other variables are dened as in equation (8). We
then estimate the implied percentage changes in TFPSens
jt
and combine them with direct
estimates of percentage changes in potential reallocation gains. Finally, we use equation (6)
to estimate the cumulative productivity gains associated with changes in the reallocation of
all factors.
Panel A of Table 2-6 reports results from the estimation of equation (11) and the implied
magnitudes for the percentage changes in TFPSens
jt
.We nd that deregulation is also
associated with signicant increases inTFPSens
jt
.The percentage increase in this sensitivity
is estimated to be 17.8% and 50.5% in the context of intrastate and interstate deregulations,
respectively. Panel B of Table 2-6 shows that these changes inTFPSens
jt
are associated with
much smaller and statistically insignicant changes in the potential gains from reallocating
factors. As before, changes in resource reallocation translate into signicant percentage
increases in marginal reallocation gains.
Finally, Table 2-7 computes the same magnitudes analyzed in Table 2-5 in the context of
this analysis.These magnitudes now capture productivity gains from reallocating all factors,
as opposed to only gains from reallocating labor. Overall, the magnitudes in this analysis
are larger than the magnitudes reported in Table 2-5. For example, we now estimate that
interstate deregulation leads to cumulative value added gains between 2.8% and 3.7%. These
results now imply that changes in factor reallocation associated with interstate deregulation
generate approximately 10.5% of all possible long-term gains from reallocation.We also es-
timate these gains in the industries with the largest productivity dierences and thus the
largest potential productivity gains. In the subsample of industries in with high productivity
dispersion, the results imply that interstate deregulation leads to productivity gains between
4.7% and 6.2% of industry value added.
Together, these ndings suggest that our previous labor reallocation gains play an im-
portant role in determining the overall eect of credit market deregulation on aggregate
productivity through the reallocation of production factors.
70
2.6 Robustness
Our analysis relies on two key ingredients. First, there is an identication concern. We
need to be able to empirically identify the eect of credit market deregulation on industry
outcomes and credit market deregulation cannot be correlated with other state-level changes
that aect the relative growth of higher marginal product rms.Second, we need to measure
gaps in the marginal products of rms. This raises a misspecication concern. We extensively
address each of these concerns in the context of our previous labor reallocation results. In
this analysis, we focus on interstate deregulation episodes, where we found our strongest
eects.
30
2.6.1 Identication Concerns
Our analysis requires isolating the impact of changes in credit markets on the dierential
growth of higher marginal product rms. An identication concern is that this relative
growth may be correlated with other state-level changes. We rst address this identication
concern in the context of our previous methodology.We start by examining trends inLRSens
prior to credit market deregulation. We analyze this issue by adding Dereg (1to 5) to
the estimation of (8). This variable is an indicator that equals one in the ve years prior to
deregulation and is included in an analogous way to Dereg. Columns (1) and (2) in Panel
A of Table 2-8 show that states do not experience dierential changes inLRSens in the ve
years prior to deregulation. Figure 2-1 breaks down this eect across the ve years prior to
deregulation, normalized by our previously estimated eects associated with Dereg. These
results further show that deregulation is not associated with a positive dierential trend in
LRSens in the years prior to deregulation. These results provide support to the view that
deregulation is not capturing previous positive trends dierentially aecting higher marginal
product rms within local industries.
Second, we rene our previous estimates and comparing only industries located in the
30To the extent that these eects do capture the impact of deregulation, they should be more easily detected
in rened results. An additional reason to focus on interstate deregulation when addressing identication
concerns is this analysis requires many deregulation episodes during our sample with data available for
many years prior to deregulation.
71
same Census region. One can think of these results as estimating the previous eects for
each of these ve regions and then averaging the eects across the ve cases. The previous
identication assumption now only needs to be applied to the timing of deregulation within
each region. Columns (3) and (4) in Panel A of Table 2-8 shows these results, which are
estimated by adding region-year xed eects and their interaction with MPL as additional
controls in the estimation of (8). These coecients are directly comparable and similar to the
ones in columns (5) and (6) in Panel B of Table 2-3. These results show that our ndings are
robust to applying our previous identication assumption only to the timing of deregulation
within each region.
Our third way to address the identication concern is to use a matching approach. We
identify local industries that experienced deregulation in their states and construct a matched
sample of geographically close industries in adjoining states that did not experience deregu-
lation over that same period. An example would be examining the Washington area SMSA
and comparing the same industry in the adjacent states of Maryland and Virginia. We then
examine if the sensitivity of labor reallocation to the marginal product of labor dierentially
changed in treated industries, when compared to matched industries, around the time of
their deregulation episode.
For each industry that experiences deregulation during our sample, we construct a group
of matched industries in the following way. We nd the ten closest industries in the same 2-
digit SIC code and Census region but in dierent states that did not experience a deregulation
episode around the treated industrys episode. More precisely, we only consider industries
that did not experience a deregulation episode in a seven-year period centered in the treated
industries deregulation year. We measure the distance between two local industries as the
average distance between their plants. We construct dierent samples of matched industries,
which impose dierent constraints on the maximum allowed distance between treated and
control industries.
This approach is motivated by the idea that, among the small manufacturing rms in
our sample period, credit markets are more local than product markets. Petersen & Rajan
(2002) estimate that the average distance between small rms and their bank lenders is
approximately 50 miles during our sample period. Moreover, their estimate for this distance
72
in early 1990s is 68 miles. Using plant level data from the commodity
ow survey, Holmes
& Stevens (2012) estimate average shipment distances for manufacturing plants in the size
range of our sample between 330 and 420 miles in 1997. Therefore, if control and treated
industries are geographically close within a certain distance range, they are arguably exposed
to dierent credit markets but face similar product market conditions.
Motivated by these previous numbers, we exclude industries closer than 50 miles from
treated industries while constructing control industries. We also impose dierent upper
bounds on their distance to treated industries. By imposing upper bounds of 1,000 and 500
miles, we construct two groups of treated and control industries with average distances equal
to 292 and 215 miles, respectively. In each of these samples, we have found that most treated
and control industries have a distance below these average values. We denote these samples
of treated and control industries as Sample
1
and Sample
2
, respectively.
After constructing these samples of matched treated and control industries for each in-
terstate deregulation episode, we estimate the following specication:
EmpShare
isjt
=
sjct
+
0
Treated
c
MPL
isjt
+
1
Post
ct
MPL
isjt
+Treated
c
Post
ct
MPL
isjt
+X
isjct
+
isjct
(2.12)
whereEmpShare
isjt
is the change in the employment share of rmi in industryj, state
s, timet, and episodec. The deregulation of the credit markets faced by each industry-state
is indexed as a separate episode c. For any given episode, both the treated industry and
the matched controls for that episode are included and the data covers a seven-year period
centered in the deregulation year of the treated industry. The data for all episodes is then
stacked. Notice that, by construction, control industries do not experience deregulation
during a given episode. Therefore, a given industry-state-year cannot be used as treated
local industry in one episode and a control local industry in another episode. However, it
might be used as a control for dierent episodes and appear multiple times in the data.
31
The remaining variables are dened as follows.
sjct
is a state-industry-episode-year xed
31We address the implications of this issue for statistical inference by clustering standard errors at the
industry level.
73
eect, Treated is an indicator that equals one for the treated industry in a given episode,
Post is an indicator that equals one during the years after the treated industrys deregulation,
MPL is the log of rm marginal product of labor, and X denotes age controls.
The coecient of interest is and tells us whether the sensitivity of labor reallocation to
the marginal product of labor dierentially changes in treated industries after their deregu-
lation, relative to geographically and economically close control industries. As in the context
of equation (8), one can think about the estimation of this eect as capturing a dierences-
in dierence estimator of changes in LRSens around deregulation. The central dierence
between these results and our previous results is the choice of the control groups. In the pre-
vious results, for each industry in a state that deregulated credit markets, we used all other
industries that did not pass deregulation around that time as controls. Another important
dierence is that we are focusing now on shorter window around deregulation dates.
32
Panel B of Table 2-8 reports the results. We nd a signicant increase in LRSens for
treated industries versus control industries in the years immediately following deregulation.
The magnitude of this increase is directly comparable and similar to the ones in columns (5)
and (6) in Panel B of Table 2-3. This magnitude is also stable across dierent specications
using alternative distances between treated and control industries.
As a nal check on this analysis, we formally test whether treated and control industries
have dierential trends in labor reallocation prior to deregulation. We extend our previous
sample to six years prior to deregulation years and keep only control industries that did
not experience deregulation over these additional years. We use the upper bound of 1,000
miles to maximize our sample size. Panel C of Table 2-8 reports these results, which show
no statistically signicant dierence in pre-trends between treated and control industries.
Prior to deregulation, treated industries experience lower increases in LRSens, and these
dierences are economically small when compared to the eects in the opposite direction
after deregulation.
Together, this evidence provides support to the view that our previous evidence on in-
creased reallocation gains after interstate deregulation captures the eect of banking dereg-
32Note that the group of treated industries is essentially the same as before, as almost all states passed
interstate deregulation in the middle of our sample.
74
ulation.
2.6.2 Measurement of Marginal Products
We implement several robustness checks to address the concern that we might not be
accurately measuring dierences or gaps in the marginal product of rms. We rst consider
alternative approaches to estimate the production function specied in (1).
33
Following the
discussion in Ackerberg et al. (2007), we modify the OP approach to allow labor as a dy-
namic input. We label this estimation approach as OP2.We also consider simple alternative
approaches to estimate (1). More specically, we consider OLS regressions with only time
xed eects and panel data estimates including plant and time xed eects. We label these
estimation approaches as OLS and FE, respectively. Panel A and B of Table 2-9 reports
results replicating the estimates of Table 2-3 with these dierent approaches. We nd that
credit market deregulation is associated with percentage increases in the sensitivity of labor
reallocation to marginal products that are similar to the ones in our previous results. In the
Internet Appendix we show that, as in our previous results, these increases in LRSens are
associated with much smaller and statistically insignicant changes in the potential gains
from reallocation. Panel C of Table 2-9 then quanties the magnitude of productivity gains
implied by these eects, following the same steps used in Table 2-5. For expositional simplic-
ity, we focus only on the magnitudes for the average industry in the sample and normalize the
estimated magnitudes by the average of respective values in Table 2-5. These results suggest
that our previous magnitudes are robust across a range of approaches for the estimation of
(1).
We then consider value-added production functions. In this approach, since we measure
value added directly, we do not need to adjust changes in industry output with the
1
1sm
term as we did in Section 2.1. In this approach, dierences in industry productivity simply
capture gaps in the total value added of industries given the same aggregate factors.
34
We
estimate value-added production functions also using translog and Cobb-Douglas specica-
33Note that the translog production function can be thought as a second-order approximation to any pro-
duction function specied in (1).
34In contrast to our main results, dierences in value added here are not measured at constant industry
output prices.
75
tions, as well as the OP, OLS and FE estimation approaches. Table 2-10 reports these results
in an analogous way to Table 2-9. The results show that credit market deregulation is as-
sociated with larger percentage increases in the sensitivity of labor reallocation to marginal
products when marginal products are estimated using value-added production functions. In
the Internet Appendix we show once more that these increases in LRSens are associated
with smaller changes in the potential gains from reallocation. Moreover, the magnitudes of
productivity gains estimated with this approach are similar to and approximately 30 percent
larger than the ones estimated in Table 2-5.
An additional concern with our measurement of marginal products is that a higher
marginal product of labor might be capturing a more skilled workforce. According to this
view, our main results capture a dierential increase in the growth of rms using higher
skilled labor after credit market deregulation. We note that, in contrast with this view,
previous research has provided evidence that these same deregulation episodes lead to an in-
crease in the demand for unskilled labor (Beck, Levine, & Levkov (2010)). We then directly
address this possibility using average worker wages in a rm as a control for average worker
skill. Previous research has suggested that wage dierentials across workers capture mostly
worker characteristics, as opposed to rm characteristics. More specically, we include rm
wages controls in the estimation of an analogous way to rm age controls in the estimation
of equation (8). Since previous research has suggested that wage dierentials are positively
correlated with rm productivity, this approach might lead us to underestimate the impor-
tance of labor reallocation gains. Table 2-11 report results following this approach. We nd
that both percentage increases in LRSens and the magnitude of productivity gains implied
by these changes remain similar to the ones in Table 2-3 and 2-5. These results suggest that
dierences in worker skill across rms are unlikely to be driving our previous ndings.
A nal concern with our measurement of marginal products comes from the fact that, as
previously discussed in Section 2.2, we measure rms marginal products using data from the
last available Census of Manufacturers. We note that the average distance between the last
census and the current year in our sample is two years. In the Internet Appendix, we provide
direct evidence that dierences in marginal products within an industry are highly persistent
at such horizon and also nd that our analysis is robust to including only years which are
76
closer in time to the years in which marginal products are measured. These ndings suggest
that this source of misspecication does not signicantly aect our analysis.
2.6.3 Alternative Channels
We close our analysis by considering alternative channels through which credit mar-
ket deregulation might aect the aggregate productivity of local industries. We consider
rm-level productivity changes and extensive margin changes through entry and exit deci-
sions.Because of space limitations, we only report our main ndings and discuss our basic
approach.We show our analysis in more detail in the Internet Appendix.
We rst consider changes in rm-level productivity. Previous research has provided evi-
dence that credit market deregulation is associated with increases in rm-level productivity
(Krishnan et al. (in press), hereafter KNP). One interpretation for such eect is that nanc-
ing constraints limit rms ability to adopt dierent technologies or management practices.
We use a dierences-in-dierence specication to examine how deregulation is associated
with changes in the productivity of a given rm in our sample. We nd that interstate
deregulation is associated with increases in rm-level productivity, with magnitudes similar
to the one reported in KNP. We then quantify the cumulative increases in industry value
added through this channel after interstate deregulation. Panel A of Table 2-12 reports these
results for the average industry in our sample. These results are directly comparable to our
previous estimates for IndustryProductivityGain
1
. When compared to our previous real-
location eects, these rm-level eects are the same sign but smaller in magnitude. These
estimates suggest that the intensive margin productivity increases associated with deregu-
lation mostly capture reallocation gains. Moreover, we have found no signicant dierences
in these rm-level eects for the subset industries where we found larger reallocation eects.
These ndings emphasize the importance of studying the implications of nancing frictions
for productivity at the industry level, as opposed to only at the rm level.
These two previous channels capture the intensive margins through which industry pro-
ductivity can change. We also extend our analysis to capture potential eects of credit mar-
ket deregulation on industry productivity through extensive margin eects due to changes
in rms entry and exit decisions.In the interest of space, we present these results in the Ap-
77
pendix A. Under the assumptions discussed in Section 2.3.4, we can write our rst measure
of industry output as Y
jt
= A
jt
H(K
jt
;L
jt
;M
jt
), where A
jt
=
P
i2I
jt
L
ijt
L
jt
A
ijt
is a measure
of industry productivity or total factor productivity. As discussed in Section 2.3.4, we can
address the assumption of constant returns to scale by restricting our analysis to indus-
tries with estimated returns to scale close to one. Once we have this simple measure of
industry productivity, we can build on previous dynamic decompositions which isolate the
contributions of resource reallocation in both the intensive and extensive margins to industry
productivity growth (e.g., Foster et al. (2001)). Summarizing these results, we nd changes
in entry and exit along the lines of Kerr & Nanda (2009). However, also consistent with their
ndings, our results suggest that these eects had a limited impact on industry productivity
growth. One simple explanation for these ndings is it can be hard to predict the quality
of new rms before they start operating and producing results. Therefore, changes in credit
markets have a limited impact in improving the selection of rms at birth and matter more
by shaping this selection at later stages.
2.7 Conclusion
We study how the deregulation of local credit markets in the U.S. aects the aggregate
productivity of local industries by shaping the allocation of labor among rms, a channel we
label as reallocation channel. We nd that the deregulation of these local U.S. credit markets
through the state banking deregulation leads to signicant increases in the reallocation of
labor within local industries towards rms with higher marginal products. We propose an
approach to quantify the industry productivity gains from such increased reallocation by
estimating rm marginal products and rm productivity using plant-level data.
We nd that these reallocation eects through labor lead to signicant increases on
the aggregate productivity of the average U.S. industry. Moreover, these eects can be
economically large for an important set of industries where such eects are predicted to be
larger. Across a range of tests, we show that our results are robust to extensive checks
addressing the two essential requirements for our analysis. Namely, measuring gaps in rms
marginal products and isolating the eect of credit market deregulation. Our results are
78
robust to conducting a dierence-in-dierence approach in geographically close markets that
span states that have deregulated at dierent times. Finally, we also compare these eects
to changes in industry productivity after credit market deregulation through other channels
including the entry of new rms. We nd evidence that the reallocation channel is signicant
when compared to these other channels.
Overall, our analysis suggests that the labor reallocation channel can be economically
important even in the United States which has relatively well-developed nancial markets
and where resource misallocation is often believed to be limited. The economic signicance
of these eects for industries more likely to face misallocation suggests that, more broadly,
changes in credit markets can have a rst-order impact on aggregate productivity through
changes in the intensive margin and the reallocation of resources towards more productive
rms.
Our results not only suggest the quantitative importance of the reallocation channel,
but also have additional implications. For example, they suggest that reallocation eects
through labor, not only capital, can be important. They also suggest that, at least during
the U.S. banking deregulation experience, changes in credit markets matter more by aecting
resource allocation at later stages of rms life cycle versus at the selection of rms at their
birth.
79
THIS PAGE INTENTIONALLY LEFT BLANK
80
Appendix A
Tables
81
Table 1-1: Summary Statistics
This table lists the sample mean and standard deviation (in parentheses) of the main
variables in the analyses. The sample period is from 1976 to 2004. Conglomerates are
plants that belong to a rm that operate in more than one distinct three-digit SIC
industries in a given year. Standalones are plants that belong to a rm that operate in one
distinct three-digit SIC industry in a given year. All variable denitions are in Appendix 1.
All Plants Conglomerates Standalones
Shipments 49,199 90,287 24,330
(543,624) (429,339) (600,992)
Capital Investment 0.126 0.216 0.149
(0.192) (0.137) (0.216)
Total Employment 215 377 117
(605) (897) (275)
Logged Employment 4.088 4.694 3.721
(1.547) (1.557) (1.420)
Size 8.906 10.029 8.226
(1.926) (1.637) (1.763)
Age 8.030 11.564 5.892
(7.519) (8.171) (6.182)
Capital Stock 7.679 9.098 6.820
(2.400) (1.982) (2.216)
Sales Growth 0.007 0.005 0.009
(0.355) (0.314) (0.381)
Return on Capital 0.233 0.162 0.276
(0.629) (0.380) (0.737)
Entry 0.089 0.051 0.132
(0.337) (0.221) (0.383)
Exit 0.092 0.064 0.115
(0.361) (0.277) (0.397)
Sale 0.029 0.040 0.022
(0.168) (0.197) (0.147)
Purchase 0.028 0.041 0.020
(0.166) (0.199) (0.142)
TFP (Rolling Window) -0.010 -0.076 0.098
(1.011) (1.016) (0.993)
TFP (Olley and Pakes (1996)) 3.254 3.192 3.292
(0.782) (0.857) (0.731)
Number of Observations 871,000 329,000 543,000
82
Table 1-2: Firm-level Productivity Around Tari Shocks
This table presents rm-level dierence-in-dierences regressions where the dependent
variable is the weighted TFP of the rm, where the weight is the capital stock of
each plant. All regressions are based on a sample from 1976 to 2004 and include
rm size and rm age as control variables. Firm Treated is an indicator variable
that equals one if a rm has one or more of its plants experiencing a tari shock in
the past three years (t, t-1 or t-2). Capital stock is computed using the perpetual
inventory method. All other variables are dened in Appendix 1. Heteroscadasticity-
robust standard errors are in parentheses and correct for clustering at the industry
level. *, **, and *** denotes signicance at the 10%, 5%, and 1% level, respectively.
Conglomerate Subsample Focused Subsample
Dependent Variable Rolling Window TFP OP TFP Rolling Window TFP OP TFP
[1] [2] [3] [4]
Firm Treated 0.0003 0.0003 -0.046*** -0.037***
(0.003) (0.002) (0.007) (0.006)
Firm Size -0.012*** -0.011*** -0.081*** -0.068***
(0.001) (0.001) (0.001) (0.001)
Firm Age -0.005*** -0.005*** -0.011*** -0.010***
(0.000) (0.000) (0.001) (0.001)
Firm FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
Observations 40,000 40,000 480,000 480,000
Adj. R-squared 0.07 0.07 0.18 0.18
83
Table 1-3: Productivity of the Treated Plants
This table presents plant-level dierence-in-dierences regressions where the dependent variable is TFP of the plant.
All regressions are based on a sample from 1976 to 2004 and include plant size and plant age as control variables. Firm
Treated is an indicator variable that equals one if a rm has one or more of its plants experiencing a tari shock in the past
three years (t, t-1 or t-2). Treated is an indicator variable that equals one if a plant operates in an industry that experiences
a tari shock in the past three years (t, t-1 or t-2). Conglomerate is a dummy variable that equals one if a plant belongs
to a rm that operates in more than one unique three-digit SIC industry. All other variables are dened in Appendix 1.
Heteroscadasticity robust standard errors are in parentheses and are corrected for clustering at the industry level. *, **, and
*** denotes signicance at the 10%, 5%, and 1% level, respectively.
Dependent Variable: OP TFP Rolling Window TFP
Full Sample Conglomerates Focused Full Sample Conglomerates Focused
[1] [2] [3] [4] [5] [6]
Treated -0.033*** -0.006 -0.053*** -0.041*** -0.005 -0.063***
(0.015) (0.005) (0.013) (0.006) (0.007) (0.008)
Treated x Conglomerate 0.025*** 0.038**
(0.0090) (0.0071)
Conglomerate -0.016 -0.007
(0.006) (0.005)
Size and Age Control Yes Yes Yes Yes Yes Yes
Plant F.E. Yes Yes Yes Yes Yes Yes
Year F.E. Yes Yes Yes Yes Yes Yes
Observations 871,000 329,000 543,000 871,000 329,000 543,000
Adj. R-squared 0.27 0.25 0.31 0.32 0.31
84
Table 1-4: Plant Exits and Divestitures
This table presents dierence-in-dierences regressions based on a sample from 1976 to 2004 and include plant size and
plant age as control variables. Firm Treated is an indicator variable that equals one if a rm has one or more of its plants
experiencing a tari shock in the past three years (t, t-1 or t-2). Treated is an indicator variable that equals one if a plant
operates in an industry that experiences a tari shock in the past three years (t, t-1 or t-2). Conglomerate is a dummy variable
that equals one if a plant belongs to a rm that operates in more than one unique three-digit SIC industry. All other variables
are dened in Appendix 1. Heteroscadasticity robust standard errors are in parentheses and are corrected for clustering at the
industry level. *, **, and *** denotes signicance at the 10%, 5%, and 1% level, respectively.
Panel A: Probability of Exits & Divestiture
Dependent Variable: Exit Dummy Divest Dummy
Full Sample Conglomerates Focused Full Sample Conglomerates Focused
[1] [2] [3] [4] [5] [6]
Treated Dummy 0.068*** 0.078*** 0.062*** 0.004 0.011*** 0.004
(0.016) (0.018) (0.016) (0.003) (0.001) (0.003)
Conglomerate Dummy -0.040*** 0.011***
(0.004) (0.001)
Treated x Conglomerate 0.024** 0.014**
(0.006) (0.007)
Plant Size and Age Control Yes Yes Yes Yes Yes Yes
Firm Fixed Eects Yes Yes Yes Yes Yes Yes
Year Fixed Eects Yes Yes Yes Yes Yes Yes
Industry-state Fixed Eects Yes Yes Yes Yes Yes Yes
Observations 871,000 329,000 543,000 871,000 329,000 543,000
Adj. R-squared 0.36 0.29 0.38 0.06 0.08 0.05
85
Panel B: Core vs. Noncore Segments
Dependent Variable: Exit Dummy Exit Dummy Divest Dummy Divest Dummy
[1] [2] [3] [4]
Treated 0.068*** 0.008***
(0.017) (0.001)
Other 0.059*** 0.012***
(0.019) (0.001)
Treated x Core 0.002 0.000
(0.013) (0.002)
Treated x Noncore 0.092*** 0.011***
(0.017) (0.001)
Other x Core -0.007 -0.002
(0.020) (0.001)
Other x Noncore 0.078*** 0.015***
(0.017) (0.002)
Size and Age Control Yes Yes Yes Yes
Firm F.E. Yes Yes Yes Yes
Year F.E. Yes Yes Yes Yes
Industry-state F.E. Yes Yes Yes Yes
Observations 329,000 329,000 329,000 329,000
Adj. R-squared 0.28 0.26 0.08 0.08
86
Panel C: TFP of Closed & Divested Plants
Dependent Variable: OP TFP
Conglomerates Focused Conglomerates Focused
[1] [2] [3] [4]
Firm Treated Dummy -0.013 -0.027*** -0.006 -0.025**
(0.011) (0.006) (0.023) (0.012)
Exit Dummy -0.086*** -0.128***
(0.020) (0.012)
Firm Treated Dummy x Exit Dummy -0.081** 0.024
(0.032) (0.058)
Divest Dummy -0.090*** -0.112***
(0.017) (0.026)
Firm Treated Dummy x Divest Dummy -0.059* 0.090*
(0.034) (0.049)
Plant Size and Age Control Yes Yes Yes Yes
Firm Fixed Eects Yes Yes Yes Yes
Year Fixed Eects Yes Yes Yes Yes
Industry-state Fixed Eects Yes Yes Yes Yes
Observations 329,000 543,000 329,000 543,000
Adj. R-squared 0.12 0.14 0.18 0.19
87
Table 1-5: Plant Openings and Acquisitions
This table presents dierence-in-dierences regressions based on a sample from 1976 to 2004 and include plant size and
plant age as control variables. Firm Treated is an indicator variable that equals one if a rm has one or more of its plants
experiencing a tari shock in the past three years (t, t-1 or t-2). Treated is an indicator variable that equals one if a plant
operates in an industry that experiences a tari shock in the past three years (t, t-1 or t-2). Conglomerate is a dummy variable
that equals one if a plant belongs to a rm that operates in more than one unique three-digit SIC industry. All other variables
are dened in Appendix 1. Heteroscadasticity robust standard errors are in parentheses and are corrected for clustering at the
industry level. *, **, and *** denotes signicance at the 10%, 5%, and 1% level, respectively.
Panel A: Probability of Opening & Acquisition
Dependent Variable: Opening Dummy Acquisition Dummy
Full Sample Conglomerates Focused Full Sample Conglomerates Focused
[1] [2] [3] [4] [5] [6]
Firm Treated Dummy -0.086*** -0.004* -0.084*** 0.000 0.008*** 0.003
(0.028) (0.002) (0.026) (0.002) (0.002) (0.002)
Lagged Conglomerate Dummy -0.017*** 0.021***
(0.003) (0.002)
Firm Treated x (Lagged) Conglomerate 0.081*** 0.008**
(0.022) (0.003)
Plant Size and Age Control Yes Yes Yes Yes Yes Yes
Firm Fixed Eects Yes Yes Yes Yes Yes Yes
Year Fixed Eects Yes Yes Yes Yes Yes Yes
Industry-state Fixed Eects Yes Yes Yes Yes Yes Yes
Observations 871,000 329,000 543,000 871,000 329,000 543,000
Adj. R-squared 0.09 0.10 0.08 0.03 0.01 0.04
88
Panel B: Core vs. Noncore Segments
Dependent Variable: Opening Dummy Opening Dummy Acquisition Dummy Acquisition Dummy
[1] [2] [3] [4]
Treated -0.032 0.002
(0.023) (0.002)
Other -0.056*** 0.000
(0.021) (0.003)
Treated x (Lagged) Core 0.076*** 0.003*
(0.022) (0.002)
Treated x (Lagged) Noncore -0.122*** -0.000
(0.019) (0.003)
Other x (Lagged) Core 0.024 -0.001
(0.021) (0.003)
Other x (Lagged) Noncore -0.162*** -0.000
(0.028) (0.003)
Size and Age Control Yes Yes Yes Yes
Firm F.E. Yes Yes Yes Yes
Year F.E. Yes Yes Yes Yes
Industry-state F.E. Yes Yes Yes Yes
Observations 329,000 329,000 329,000 329,000
Adj. R-squared 0.10 0.10 0.03 0.02
89
Panel C: TFP of Opened & Acquired Plants
Dependent Variable: OP TFP
Conglomerates Focused Conglomerates Focused
[1] [2] [3] [4]
Firm Treated Dummy -0.006 -0.015*** -0.007 -0.013***
(0.008) (0.003) (0.010) (0.002)
Entry Dummy -0.082*** -0.051***
(0.014) (0.011)
Firm Treated Dummy x Entry Dummy 0.064*** 0.000
(0.018) (0.029)
Purchase Dummy -0.047 -0.013
(0.031) (0.019)
Firm Treated Dummy x Purchase Dummy 0.005 0.010
(0.014) (0.018)
Plant Size and Age Control Yes Yes Yes Yes
Yes Yes Yes Yes
Firm Fixed Eects Yes Yes Yes Yes
Year Fixed Eects Yes Yes Yes Yes
Industry-state Fixed Eects Yes Yes Yes Yes
Observations 329,000 543,000 329,000 543,000
Adj. R-squared 0.04 0.06 0.02 0.05
90
Table 1-6: Productivity of Conglomerates and Matched Focused Plants
This table presents dierence-in-dierences regressions based on a matched sample from 1976 to 2004. All specica-
tions include industry-year eects as well as size and age as controls. Continuers are plants that continue operations between
time t and t+2; Exiters (divestitures) are plants that exist at t but are shut down (divested) at t+1 or t+2; Entrants
(acquisitions) are plants that exist at time t+2 but are opened (acquired) at t or t+1. Heteroscadasticity robust standard
errors are in parentheses and are corrected for clustering at the industry level. *, **, and *** denotes signicance at the 10%,
5%, and 1% level, respectively.
Panel A. TFP In Tari Cut t by Plant Status in t+2
Dependent Variable: Plant-level Log TFP in Year t
Plant Status Conglomerates Focused Control T-value Dierence P Value
Continuers -0.008 Omitted Group -0.77 0.441
(0.011)
Exits -0.038 0.011 -3.55 0.000
(0.011) (0.011)
Divestitures -0.025 -0.011 -2.90 0.004
(0.007) (0.008)
Ind-Year Fixed Eects Yes
Number of Observations 205,000
91
Panel B. TFP In Tari Cut t+2 by Plant Status in t+2
Dependent Variable: Plant-level Log TFP in Year t+2
Plant Status Conglomerates Focused Control T-value Dierence P Value
Continuers -0.007 Omitted Group -0.85 0.396
(0.008)
Entrants 0.051 -0.012 2.90 0.004
(0.009) (0.015)
Acquisitions 0.018 0.020 -0.31 0.757
(0.011) (0.018)
Ind-Year Fixed Eects Yes
Number of Observations 190,000
Panel C: Estimated Average Two-Year Post Tari Change in TFP at
Conglomerate Firms Relative to Focused Controls, Log Points
TFP Log Change Dierential 2.36
Share of Total TFP Two-Year Change
Dierential By Margin of Adjustment
Continuing Establishments 0.13
Entry and Exit 0.69
Acquisitions and Divestitures 0.18
92
Table 1-7: Eect of Net Cash on Probability of Restructuring
This table presents dierence-in-dierences regressions based on a sample from 1976 to 2004, with information available
from both the CMF/ASM and QFR. All specications include size and age as controls. Continuers are plants that continue
operations between time t and t+2; Exiters (divestitures) are plants that exist at t but are shut down (divested) at t+1 or
t+2; Entrants (acquisitions) are plants that exist at time t+2 but are opened (acquired) at t or t+1. Heteroscadasticity robust
standard errors are in parentheses and are corrected for clustering at the industry level. *, **, and *** denotes signicance at
the 10%, 5%, and 1% level, respectively.
Dependent Variable Plant Opening Dummy Plant Acquisition Dummy Plant Closure Dummy
Focused Conglomeates Focused Conglomeates Focused Conglomeates
[1] [2] [3] [4] [5] [6]
Firm Treated Dummy -0.013* -0.002 0.001 0.020*** 0.014* 0.010**
(0.006) (0.004) (0.007) (0.005) (0.008) (0.005)
Firm Treated x High Net Cash Dummy 0.010** 0.000 0.009** -0.008 -0.025*** 0.004
(0.005) (0.000) (0.005) (0.005) (0.009) (0.005)
High Net Cash Dummy 0.016 -0.003 -0.001 -0.001 0.006 -0.014
(0.007) (0.004) (0.002) (0.005) (0.006) (0.006)
Size and Age Controls Yes Yes Yes Yes Yes Yes
Firm F.E. Yes Yes Yes Yes Yes Yes
Year F.E. Yes Yes Yes Yes Yes Yes
Observations 116,000 50,000 116,000 50,000 116,000 50,000
Adj. R-squared 0.09 0.12 0.01 0.00 0.06 0.15
93
Table 1-8: Instrumental Variables Regression for Restructuring Probability
This table presents dierence-in-dierences regressions based on a sample from 1976 to 2004. The ConglomerateDummy and
Firm TreatedConglomerate Dummy are instrumented using the variables described in Section 4.5.1. Heteroscadasticity
robust standard errors are in parentheses and are corrected for clustering at the industry level. *, **, and *** denotes
signicance at the 10%, 5%, and 1% level, respectively.
Dependent Variable Opening Dummy Closure Dummy Divestiture Dummy Acuisition Dummy
[1] [2] [3] [4]
Firm Treated Dummy -0.132*** 0.099*** 0.017 0.026
(0.036) (0.032) (0.027) (0.018)
Conglomerate Dummy (Predicted) -0.103*** -0.141*** 0.154*** 0.115***
(0.025) (0.028) (0.028) (0.002)
Conglomerate (Predicted) x Firm Treated 0.141*** 0.097** 0.095** 0.088**
(0.029) (0.026) (0.031) (0.027)
Size and Age Controls Yes Yes Yes Yes
Firm Fixed Eects Yes Yes Yes Yes
Year Fixed Eects Yes Yes Yes Yes
Industry-state FE Yes Yes Yes Yes
Observations 871,000 871,000 871,000 871,000
94
Table 1-9: Treatment Dynamics
This table presents dierence-in-dierences regressions based on a sample from 1976
to 2004. The Conglomerate Dummy and Firm TreatedConglomerate Dummy are
instrumented using the variables described in Section 4.5.1. Heteroscadasticity robust
standard errors are in parentheses and are corrected for clustering at the industry level. *,
**, and *** denotes signicance at the 10%, 5%, and 1% level, respectively.
Dependent Variable TFP ROC Employment Sales Growth
[2] [3] [4] [5]
Treated (-2) -0.0007 -0.0012 0.0022 -0.0025
(0.0070) (0.0080) (0.0086) (0.0021)
Treated (-1) -0.0027 -0.0088 0.0095 0.0013
(0.0069) (0.0060) (0.0087) (0.0032)
Treated (0) -0.0134* -0.0188*** 0.0126 0.0021
(0.0075) (0.0048) (0.0082) (0.0026)
Treated (+1) -0.0347*** -0.0483*** 0.0223** 0.0023
(0.0086) (0.0076) (0.0093) (0.0026)
Treated (+2) -0.0318*** -0.0272*** 0.0166* -0.0066**
(0.0093) (0.0102) (0.0091) (0.0028)
Conglomerate x Treated (-2) 0.0012 -0.0008 -0.0084 0.0016
(0.0133) (0.0151) (0.0155) (0.0033)
Conglomerate x Treated (-1) -0.0037 0.0043 0.0081 -0.0017
(0.0092) (0.0086) (0.0085) (0.0042)
Conglomerate x Treated (0) 0.0130 0.0170*** -0.0039 -0.0075**
(0.0095) (0.0063) (0.0084) (0.0035)
Conglomerate x Treated (+1) 0.0326*** 0.0300*** -0.0124 -0.0030
(0.0094) (0.0068) (0.0088) (0.0034)
Conglomerate x Treated (+2) 0.0244** 0.0085 -0.0022 0.0018
(0.0105) (0.0092) (0.0077) (0.0040)
Conglomerate Dummy -0.0615*** -0.0285*** -0.0003 -0.0255***
(0.0091) (0.0031) (0.0130) (0.0022)
Industry-Year Yes Yes Yes Yes
Control Variables Yes Yes Yes Yes
Plant Fixed Eects Yes Yes Yes Yes
Year Fixed Eects Yes Yes Yes Yes
Observations 871,000 871,000 871,000 871,000
Adj. R-squared 0.41 0.38 0.86 0.27
95
Table 1-10: First Stage for IV Regression
This table presents the rst stage results in the Instrumental Variables (IV) regres-
sion in Table 8. All variable denitions are in Appendix 1. *, **, and *** denotes
signicance at the 10%, 5%, and 1% level, respectively.
First Stage
Dependent Variable Conglomerate Conglomerate x Firm Treated
[1] [2]
Firm Treated -0.0152*** 0.3820***
(0.0029) (0.0014)
Conglomerate Dummy
Conglomerate x Firm Treated
Number of Mergers 0.0251*** 0.0531***
(0.0026) (0.0013)
Total Firm Sales (lagged) 0.0086*** -0.0014***
(0.0001) (0.0000)
Long-run Change in Industry Shipments 0.5083 0.4238***
(0.3286) (0.1600)
Industry Capital Intensity 0.3865*** -0.3620***
(0.0361) (0.0176)
Number of Mergers x Treated -0.0220*** -0.2225***
(0.0031) (0.0015)
Total Firm Sales (lagged) x Treated 0.0030*** 0.0204***
(0.0002) (0.0001)
Long-run Change in Industry Shipments x Treated 0.0128 -0.2677***
(0.0200) (0.0097)
Industry Capital Intensity x Treated 0.4763*** 3.0167***
(0.0612) (0.0298)
Plant Size 0.0722*** 0.0153***
(0.0003) (0.0002)
Plant Age 0.0107*** 0.0010***
(0.0001) (0.0000)
F-statistics of excluded instruments 150.60 51.24
Observations 871,000 871,000
96
Table 2-1: State Banking Deregulation Dates
This table presents the dates of interstate and intrastate deregulation events used in
our analysis. We follow Amel (1993) and Kroszner and Strahan (1999) in determining these
dates. See Section 1.3 for more details.
State Intrastate Deregulation Year Interstate Deregulation Year
Alabama 1981 1987
Alaska <1970 1982
Arizona <1970 1986
Arkansas 1994 1989
California <1970 1987
Colorado 1991 1988
Connecticut 1980 1983
Delaware <1970 1988
DC <1970 1985
Florida 1988 1985
Georgia 1983 1985
Hawaii 1986 >1993
Idaho <1970 1985
Illinois 1988 1986
Indiana 1989 1986
Iowa 1997 1991
Kansas 1987 1992
Kentucky 1990 1984
Louisiana 1988 1987
Maine 1975 1978
Maryland <1970 1985
Massachusetts 1984 1983
Michigan 1987 1986
Minnesota 1993 1986
Mississippi 1986 1988
Missouri 1990 1986
Montana 1990 1993
97
Table 2-1 Continued
State Intrastate Deregulation Year Interstate Deregulation Year
Nebraska 1985 1990
Nevada <1970 1985
New Hampshire 1987 1987
New Jersey 1977 1986
New Mexico 1991 1989
New York 1976 1982
North Carolina <1970 1985
North Dakota 1987 1991
Ohio 1979 1985
Oklahoma 1988 1987
Oregon 1985 1986
Pennsylvania 1982 1986
Rhode Island <1970 1984
South Carolina <1970 1986
South Dakota <1970 1988
Tennessee 1985 1985
Texas 1988 1987
Utah 1981 1984
Vermont 1970 1988
Virginia 1978 1985
Washington 1985 1987
West Virginia 1987 1988
Wisconsin 1990 1987
Wyoming 1988 1987
98
Table 2-2: Summary Statistics & Key Parameters
This table presents summary statistics on dierent variables and estimates used in
the paper. Panel A shows summary statistics for the main sample used in the paper. Sales
is the only variable using information from the Census of Manufacturers and available
only for a subset of sample years. Variable denitions are in Appendix C. Panel B reports
the average values of the factor elasticities estimated using dierent production function
specications and methods. Panel C, D, and E report the within industry dispersion in
the estimated marginal product of labor, marginal product of capital, and rm total factor
productivity across these approaches, respectively.
Panel A: Summary Statistics
Variable Mean Std Nobs
Employment Growth 0.0089 0.4621 2,287,100
Employment Share 0.0272 0.0834 2,287,100
Employment Share Growth -0.0131 0.4570 2,287,100
Employment 22.28 46.23 2,287,100
Sales ($1K 1987) 1,648 4,533 397,700
Age 5.20 4.50 2,795,000
Exit 0.0685 0.2526 2,287,100
Entry 0.0819 0.2743 2,795,000
Intra Deregulation 0.6215 0.4850 2,795,000
Inter Deregulation 0.4139 0.4925 2,795,000
Panel B: Estimated Factor Elasticities
Translog Cobb-Douglas
Factor OP LP OP LP
Capital 0.0848 0.1052 0.0491 0.0562
Labor 0.3717 0.3793 0.3264 0.3002
Materials 0.4023 0.4455 0.5021 0.6222
99
Panel C: Dispersion in MPL (within industry-state-year)
OP LP
Translog Specication 0.3722 0.3788
Cobb-Douglas Specication 0.5198 0.5198
Panel D: Dispersion in MPK (within industry-state-year)
OP LP
Translog Specication 0.4804 0.5472
Cobb-Douglas Specication 0.4581 0.4581
Panel E: Dispersion in TFP (within industry-state-year)
OP LP
Translog Specication 0.3253 0.2975
Cobb-Douglas Specication 0.3287 0.3088
100
Table 2-3: Credit Market Deregulation and the Sensitivity of Labor Reallocation to Marginal Products
This table presents results linking the sensitivity of labor reallocation to the marginal product of labor within an industry-state
(LRSens) to credit market deregulation. Panels A and B report results from the estimation of equations (7) and (8),
respectively. The dependent variable is the annual change in the log of the rm's industry-state employment share. For a given
year t, this change in share is computed including only rms present in both year t and t-1. MPL is the log of the marginal
product of labor, which can be based on a Translog or Cobb-Douglas production function, with parameters estimated using
the OP or LP approaches (see text for more details). Dereg is an indicator that equals one if the state has passed banking
deregulation (intrastate or interstate). The control variables in all regressions include the one-year lag of age, its squared value,
as well as the interactions of both these variables with Dereg. Standard errors are heteroskedasticity robust and clustered at
the industry level. *, **, and *** indicate statistical signicance at the 10%, 5%, and 1%, respectively. Panel C reports the
percentage changes in LRSens implied by the eects in Panel B. These percentage changes are computed as the ratio of the
eects in Panel B to the estimated value of LRSens in the subsample that has not passed deregulation.
Panel A: Initial Evidence
Outcome: Change in Log of Employment Share
Intrastate Deregulation Interstate Deregulation
Translog Cobb-Douglas Translog Cobb-Douglas
OP LP OP LP OP LP OP LP
(1) (2) (3) (4) (5) (6) (7) (8)
MPL 0.0283*** 0.0273*** 0.0262*** 0.0263*** 0.0309*** 0.0282*** 0.0277*** 0.0277***
(0.0023) (0.0027) (0.0014) (0.0014) (0.0021) (0.0026) (0.0013) (0.0013)
MPL Dereg 0.0186*** 0.0153*** 0.0116*** 0.0116*** 0.0208*** 0.0196*** 0.0129*** 0.0129***
(0.0019) (0.0022) (0.0016) (0.0016) (0.0020) (0.0021) (0.0016) (0.0016)
Nobs 2,287,100 2,287,100 2,287,100 2,287,100 2,287,100 2,287,100 2,287,100 2,287,100
R-square 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01
State-Industry-Year FE Yes Yes Yes Yes Yes Yes Yes Yes
101
Panel B: Main Specication
Outcome: Change in Log of Employment Share
Intrastate Deregulation Interstate Deregulation
Translog Cobb-Douglas Translog Cobb-Douglas
OP LP OP LP OP LP OP LP
(1) (2) (3) (4) (5) (6) (7) (8)
MPL Dereg 0.0081*** 0.0079*** 0.0068*** 0.0065*** 0.0151*** 0.0160*** 0.0117*** 0.0117***
(0.0012) (0.0012) (0.0014) (0.0015) (0.0020) (0.0022) (0.0019) (0.0020)
Nobs 2,287,100 2,287,100 2,287,100 2,287,100 2,287,100 2,287,100 2,287,100 2,287,100
R-square 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01
State-Industry-Year FE Yes Yes Yes Yes Yes Yes Yes Yes
State FE x MP Yes Yes Yes Yes Yes Yes Yes Yes
Year FE x MP Yes Yes Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes Yes Yes
Panel C: Magnitude of Changes in Labor Reallocation - Main Specication
Intrastate Dereg Interstate Dereg
OP LP OP LP
Percentage Change in LRSens 27.3% 26.8% 45.5% 49.4%
102
Table 2-4: Credit Market Deregulation and Potential Labor Reallocation Gains
This table presents results linking the potential gains from labor reallocation within
an industry-state to credit market deregulation. The results are the output from the
estimation of equation (9). Potential gains from labor reallocation are computed using
equation (5) (see the text for more details). Dereg is an indicator that equals one if
the state has passed banking deregulation (intrastate or interstate). Standard errors are
heteroskedasticity robust and clustered at the industry level. *, **, and *** indicate
statistical signicance at the 10%, 5%, and 1%, respectively.
Outcome: Log of Potential Labor Reallocation Gains
Intrastate Deregulation Interstate Deregulation
Translog Cobb-Douglas Translog Cobb-Douglas
OP LP OP LP OP LP OP LP
(1) (2) (3) (4) (5) (6) (7) (8)
Dereg -0.030 -0.021 0.025 0.025 -0.083 -0.047 -0.085 -0.085
(0.034) (0.034) (0.033) (0.033) (0.058) (0.062) (0.063) (0.063)
Nobs 11,000 11,000 11,000 11,000 11,000 11,000 11,000 11,000
R-square 0.10 0.08 0.12 0.12 0.10 0.08 0.12 0.12
State-Industry FE Yes Yes Yes Yes Yes Yes Yes Yes
Year FE Yes Yes Yes Yes Yes Yes Yes Yes
103
Table 2-5: Magnitude of Industry Productivity Gains from Increased Labor
Reallocation
This table presents results quantifying the cumulative industry productivity gains im-
plied by the changes in labor reallocation gains. These gains are additional percentage
increases in value added due to the additional intensive margin reallocation of labor, and
are estimated using equation (6) (see text for more details). Panel A reports the gains
implied by the results in Tables 3 and 4. Panels B, C, and D estimate these same gains
in dierent subsamples of industries. Panel B restricts the analysis to industries in the
top tercile of estimated returns to scale with the OP approach. Panels C and D further
restrict the sample from Panel B to industries in the top 50% and top 33% of potential
labor reallocation gains prior to deregulation (percentiles computed within the sample from
Panel B).
Panel A: All Industries
Intrastate Deregulation Interstate Deregulation
OP LP OP LP
Industry Productivity Gain 1 (%VA) 0.85% 0.76% 1.56% 1.52%
Industry Productivity Gain 2 (%VA) 1.13% 1.02% 2.08% 2.02%
Panel B: Industries with Estimated Returns to Scale Close to One
Intrastate Deregulation Interstate Deregulation
OP OP
Industry Productivity Gain 1 (%VA) 0.63% 1.66%
Industry Productivity Gain 2 (%VA) 0.85% 2.21%
Panel C: Industries with High Potential Labor Reallocation Gains (Top 50%)
Intrastate Deregulation Interstate Deregulation
OP OP
Industry Productivity Gain 1 (%VA) 1.72% 3.27%
Industry Productivity Gain 2 (%VA) 2.30% 4.36%
Panel D: Industries with High Potential Labor Reallocation Gains (Top 33%)
Intrastate Deregulation Interstate Deregulation
OP OP
Industry Productivity Gain 1 (%VA) 3.55% 4.53%
Industry Productivity Gain 2 (%VA) 4.74% 6.04%
104
Table 2-6: Credit Market Deregulation and Reallocation Gains from All Factors
Industries with Estimated Returns to Scale Close to One
This table presents two sets of results using the sample of industries in the top ter-
cile of estimated returns to scale with the OP approach. Panel A reports results linking
the sensitivity of labor reallocation to rm tfp within an industry-state (TFPSens) to
credit market deregulation. These results are the output from the estimation of equation
(11). The dependent variable is the annual change in the log of the rm's industry-state
employment share. For a given year t, this change in share is computed including only rms
present in both year t and t-1. TFP is the log of rm total factor productivity, which can
be based on a Translog or Cobb-Douglas production function, with parameters estimated
using the OP approach (see text for more details). Dereg is an indicator that equals one if
the state has passed banking deregulation (intrastate or interstate). The control variables in
all regressions include the one-year lag of age, its squared value, as well as the interactions
of both these variables with Dereg. Panel B reports the percentage changes in TFPSens
implied by the eects in Panel A in an analogous way to Panel C of Table 3. Panel C
presents results linking the potential gains from the reallocation of all factors within an
industry-state to credit market deregulation. The results are the output from the estimation
of equation (9), with potential gains from reallocation computed using equation (10) (see
the text for more details). Standard errors are heteroskedasticity robust and clustered at
the industry level. *, **, and *** indicate statistical signicance at the 10%, 5%, and 1%,
respectively.
Panel A: Changes in the Sensitivity of Reallocation to Firm Productivity
Outcome: Change in Log of Employment Share
Intrastate Deregulation Interstate Deregulation
Translog Cobb-Douglas Translog Cobb-Douglas
OP OP OP OP
(1) (3) (5) (7)
TFP Dereg 0.0067*** 0.0108*** 0.0185*** 0.0263***
(0.0014) (0.0024) (0.0023) (0.0033)
Nobs 755,000 755,000 755,000 755,000
R-square 0.01 0.01 0.01 0.01
State-Industry-Year FE Yes Yes Yes Yes
State FE x TFP Yes Yes Yes Yes
Year FE x TFP Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
105
Panel B: Magnitude of Changes in the Sensitivity of
Reallocation to Firm Productivity
Intrastate Interstate
OP OP
Percentage Change in TFPSens 17.8% 50.5%
Panel C: Changes in Potential Reallocation Gains
Outcome: Log of Potential Reallocation Gains
Intrastate Deregulation Interstate Deregulation
Translog Cobb-Douglas Translog Cobb-Douglas
OP OP OP OP
(1) (3) (5) (7)
Dereg -0.084 -0.026 -0.061 -0.043
(0.054) (0.044) (0.107) (0.086)
Nobs 7,700 7,700 7,700 7,700
R-square 0.04 0.03 0.04 0.03
State-Industry FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
106
Table 2-7: Magnitude of Industry Productivity Gains from Increased Overall
Factor Reallocation
This table presents results quantifying the cumulative industry productivity gains im-
plied by the reallocation gains of all factors. These gains are additional percentage increases
in value added due to the additional intensive margin reallocation of all factors, and are
estimated using equation (6) (see text for more details). The analysis is restricted to
industries in the top tercile of estimated returns to scale with the OP approach. Panels B
and C further restrict this sample to industries in the top 50% and top 33% of potential
reallocation gains prior to deregulation (percentiles computed within the sample of industries
with estimated returns to scale close to one).
Panel A: Industries with Estimated Returns to Scale Close to One
Intrastate Deregulation Interstate Deregulation
OP OP
Industry Productivity Gain 1 (%VA) 1.41% 2.82%
Industry Productivity Gain 2 (%VA) 1.89% 3.76%
Panel B: Industries with High Potential Overall Reallocation Gains (Top 50%)
Intrastate Deregulation Interstate Deregulation
OP OP
Industry Productivity Gain 1 (%VA) 3.46% 4.68%
Industry Productivity Gain 2 (%VA) 4.61% 6.24%
Panel C: Industries with High Potential Overall Reallocation Gains (Top 33%)
Intrastate Deregulation Interstate Deregulation
OP OP
Industry Productivity Gain 1 (%VA) 4.43% 6.62%
Industry Productivity Gain 2 (%VA) 5.91% 8.83%
107
Table 2-8: Identication of Deregulation Eects
This table presents results addressing the identication of the eect of credit market
deregulation on the sensitivity of labor reallocation to the marginal product of labor
(LRSens). Panel A reports results addressing the robustness of the eects in Panel B of
Table 3 (columns (5) and (6)). The results in columns (1) and (2) add the variable Dereg
(-1 to -5), as well as its interactions with MPL and age controls (see Table 3). Dereg (-1 to
-5) is an indicator that equals one in the ve years prior to state credit market deregulation.
The results in columns (3) and (4) add region-year xed eects as well as their interaction
with MPL. Panel B reports results using a matching approach. We examine if the sensitivity
of labor reallocation to the marginal product of labor dierentially changed in treated
industries, when compared to matched industries, around the time of their deregulation
episode. See the text for more details. These results are the output from the estimation of
equation (12). Treated is an indicator that equals one for industries in states that deregulate
credit markets. Post is an indicator that equals one after credit market deregulation
dates. MPL is the marginal product of labor. We also include interactions of age controls
(see Table 3) with Treated, Post, and Treated Post. Panel C reports results examining
the trends in LRSens prior to deregulation across the treated and control groups in our
matching analysis. These results are based on linear regressions linking Change in Log of
Employment Share to MPL Control, MPL Treated, MPL Time Control,and MPL Time
Treated. This analysis also includes analogous variables replacing MPL with age variables
(see Table 3) as controls and is based on the six years prior to the deregulation events
examined in Panel B. See the text for more details. Standard errors are heteroskedasticity
robust and clustered at the industry level. *, **, and *** indicate statistical signicance at
the 10%, 5%, and 1%, respectively..
Panel A: Robustness of Previous Results
Outcome: Change in Log of Employment Share
Interstate Deregulation - Translog Specication
OP LP OP LP
(1) (2) (3) (4)
MPL Dereg 0.0095** 0.0087** 0.0122*** 0.0134***
(0.0039) (0.0038) (0.0021) (0.0022)
MPL Dereg (-1 to -5) -0.0025 -0.0031
(0.0021) (0.0020)
Nobs 2,287,100 2,287,100 2,287,100 2,287,100
R-square 0.01 0.01 0.01 0.01
State-Industry-Year FE Yes Yes Yes Yes
State FE x MP Yes Yes Yes Yes
Year FE x MP Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
Region-Year FE x MP No No Yes Yes
108
Panel B: Results Using Matching Approach
Outcome: Change in Log of Employment Share
Interstate Deregulation -Translog Specication
Sample 1 Sample 2
OP LP OP LP
(1) (2) (3) (4)
MPL Treated Post 0.0227*** 0.0160*** 0.0217*** 0.0159***
(0.0064) (0.0059) (0.0065) (0.0063)
Nobs 914,500 914,500 704,000 704,000
R-squared 0.01 0.01 0.01 0.01
State-Industry-Year-Episode FE Yes Yes Yes Yes
Panel C: Are There Diential Trends in Treated Industries Prior to Deregulation?
Outcome: Change in Log of Employment Share
Interstate Deregulation -Translog Specication
6-Year Window Prior to Deregulation
OP LP
(1) (2)
MPL Time Control 0.0112*** 0.0131***
(0.0068) (0.0067)
MPL Time Treated 0.0107*** 0.0095***
(0.0027) (0.0027)
Dierence (Treated - Control) -0.0013 -0.0035
(0.0055) (0.0056)
Nobs 191,900 191,900
R-squared 0.01 0.01
State-Industry-Year-Episode FE Yes Yes
109
Table 2-9: Alternative Approaches to Estimate Production Functions
This table presents the results in Panels B and C of Table 3 and Panel A of Table 5
across additional approaches to estimate production functions. See the text for more details
on dierent estimation approaches. Panel C reports the magnitudes of cumulative output
gains divided by the same gains in Panel A of Table 5 (also for interstate deregulation).
Standard errors are heteroskedasticity robust and clustered at the industry level. *, **, and
*** indicate statistical signicance at the 10%, 5%, and 1%, respectively.
Panel A: Changes in Labor Reallocation Sensitivity
Outcome: Change in Log of Employment Share
Interstate Deregulation
Translog Specication Cobb-Douglas Specication
OP2 OLS FE OP2 OLS FE
(1) (2) (3) (4) (5) (6)
MPL x Dereg 0.0154*** 0.0288*** 0.0242*** 0.0125*** 0.0277*** 0.0277***
(0.0024) (0.0022) (0.0017) (0.0023) (0.0013) (0.0013)
Nobs 1,929,900 2,287,100 2,287,100 1,929,900 2,287,100 2,287,100
R-square 0.01 0.01 0.01 0.01 0.01 0.01
State-Industry-Year FE Yes Yes Yes Yes Yes Yes
State FE x MP Yes Yes Yes Yes Yes Yes
Year FE x MP Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes
110
Panel B: Magnitude of Changes in Labor Reallocation
OP2 OLS FE
Percentage Change in LRSens 53.2% 48.4% 48.7%
Panel C: Magnitude of Industry Productivity Gains
OP2 OLS FE
Industry Productivity Gain 1.10 1.08 1.23
111
Table 2-10: Results Using Value-Added Production Functions
This table presents the results in Panels B and C of Table 3 and Panel A of Table 5
using value-added production functions. Panel C reports the magnitudes of cumulative out-
put gains divided by the same gains in Panel A of Table 5 (also for interstate deregulation).
Standard errors are heteroskedasticity robust and clustered at the industry level. *, **, and
*** indicate statistical signicance at the 10%, 5%, and 1%, respectively.
Panel A: Changes in Labor Reallocation Sensitivity
Outcome: Change in Log of Employment Share
Interstate Deregulation
Translog Specication Cobb-Douglas Specication
OP OLS FE OP OLS FE
(1) (2) (3) (4) (5) (6)
MPL x Dereg 0.0156*** 0.0163*** 0.0150*** 0.0141*** 0.0140*** 0.0151***
(0.0021) (0.0019) (0.0022) (0.0018) (0.0017) (0.0019)
Nobs 2,287,100 2,287,100 2,287,100 2,287,100 2,287,100 2,287,100
R-square 0.01 0.01 0.01 0.01 0.01 0.01
State-Industry-Year FE Yes Yes Yes Yes Yes Yes
State FE x MP Yes Yes Yes Yes Yes Yes
Year FE x MP Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes
Panel B: Magnitude of Changes in Labor Reallocation
OP OLS FE
Percentage Change in LRSens 61.4% 69.9% 74.6%
Panel C: Magnitude of Industry Productivity Gains
OP OLS FE
Industry Productivity Gain 1.39 1.34 1.22
112
Table 2-11: Results Controlling for Dierences in Worker Skill
This table presents the results in Panels B and C of Table 3 and Panel A of Table 5
with additional controls for dierences in worker skill across rms. In addition to age
controls, we now also include the average wage of rms (wage) as controls in the estimation
of (8). These additional control variables are the one-year lag of wage, its squared value, as
well as the interactions of both these variables with Dereg. Panel C reports the magnitudes
of cumulative output gains divided by the same gains in Panel A of Table 5 (also for
interstate deregulation). Standard errors are heteroskedasticity robust and clustered at the
industry level. *, **, and *** indicate statistical signicance at the 10%, 5%, and 1%,
respectively.
Panel A: Changes in Labor Reallocation Sensitivity
Outcome: Change in Log of Employment Share
Interstate Deregulation
Translog Specication Cobb-Douglas Specication
OP LP OP LP
(1) (2) (3) (4)
MPL x Dereg 0.0150*** 0.0159*** 0.0116*** 0.0116***
(0.0020) (0.0022) (0.0019) (0.0020)
Nobs 2,287,100 2,287,100 2,287,100 2,287,100
R-square 0.01 0.01 0.01 0.01
State-Industry-Year FE Yes Yes Yes Yes
State FE x MP Yes Yes Yes Yes
Year FE x MP Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
Panel B: Magnitude of Changes in Labor Reallocation
OP LP
Percentage Change in LRSens 45.3% 49.1%
Panel C: Magnitude of Industry Productivity Gains
OP LP
Industry Productivity Gain 1.02 0.97
113
Table 2-12: Reallocation and Firm-Level Productivity Gains
This table combines previous estimates for the additional reallocation gains after
credit market deregulation across all factors (Panel A of Table 7) with estimated eects of
credit market deregulation on rm level tfp. These results are restricted to the sample of
industries in the top tercile of estimated returns to scale with the OP approach. See the
text for more details.
Interstate Deregulation - Industries with Estimated Returns to Scale Close to One
Industry Productivity Gain 1 (%VA)
OP
Firm-Level Productivity Gains 1.05%
Reallocation Gains Across All Factors 2.82%
Total Gains (Intensive Margin) 3.87%
Percentage of Gains from Firm-Level Channel 27.1%
Percentage of Gains from Reallocation Channel 72.9%
114
Appendix B
Figures
115
0 5 10 15 20
# of Tariff Cuts
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
Frequency of Import Tariff Reductions
Distribution of Tariff Cuts over Time
Figure 1-1: Distribution of Tari Cuts Over Time
42 44 46 48 50
Exiters
−2 −1 0 1 2
Event Time
Industry Year Means
Exits around Tariff Shock
35 40 45 50 55 60
Entrants
−2 −1 0 1 2
Event Time
Industry Year Means
Entry around Tariff Shock
3 3.5 4 4.5
Divestitures
−2 −1 0 1 2
Event Time
Industry Year Means
Divestitures around Tariff Shock
2.5 3 3.5 4
Acquisitions
−2 −1 0 1 2
Event Time
Industry Year Means
Acquisitions around Tariff Shock
Figure 1-2: Restructuring Around Tari Shocks
116
2.8 2.9 3 3.1 3.2 3.3
TFP
−2 −1 0 1 2
Event Time
Conglomerate Focused
Productivity around Tariff Shocks
Figure 1-3: Firm Level Productivity Around Tari Shocks
0 .5 1 1.5 2 2.5
TFP
On the Shock Off the Shock
On and Off the Tariff Shock
Productivity of Exiting Plants
Focus Conglomerate
0 1 2 3
TFP
On the Shock Off the Shock
On and Off the Tariff Shock
Productivity of Entering Plants
Focus Conglomerate
0 1 2 3
TFP
On the Shock Off the Shock
On and Off the Tariff Shock
Productivity of Divested Plants
Focus Conglomerate
0 1 2 3
TFP
On the Shock Off the Shock
On and Off the Tariff Shock
Productivity of Acquired Plants
Focus Conglomerate
Figure 1-4: Average Productivity of Restructured Plants
117
Figure 2-1: Dierences in Labor Allocation Prior to Interstate Deregulation
118
References
Ackerberg, D., Benkard, L., Berry, S., & Pakes, A. (2007). Econometric tools for analyzing market outcomes.
In J. Heckman & E. Leamer (Eds.), Handbook of econometrics. Elsevier: North-Holland.
Amel, D. (1993). State laws aecting the geographic expansion of commercial banks. manuscript.
Asker, J., Collard-Wexler, A., & Loecker, J. D. (2014). Dynamic inputs and resource (mis)allocation. Journal
of Political Economy, 122, 1013-1063.
Bartelsman, E., Haltiwanger, J., & Scarpetta, S. (2013). Cross-country dierences in productivity: The role
of allocation and selection. American Economic Review, 103, 305-334.
Basu, S., & Fernald, J. (2002). Aggregate productivity and aggregate technology. European Economic
Review, 46, 963-991.
Beck, T., Levine, R., & Levkov, A. (2010). Big bad banks? the winners and losers from bank deregulation
in the united states. Journal of Finance, 65(5), 16371667.
Benmelech, E., Bergman, N. K., & Seru, A. (2012). Financing labor. manuscript.
Bens, D., Berger, P., & Monahan, S. (2011). Discretionary disclosure in nancial reporting: An examination
comparing internal rm data to externally reported segment data. The Accounting Review, 86(2), 417-
479.
Berger, P., & Ofek, E. (1995). Diversications eect on rm value. Review of Financial Studies, 37, 39-65.
Bertrand, M., Schoar, A., & Thesmar, D. (2007). Banking deregulation and industry structure: Evidence
from the french banking reforms of 1985. Journal of Finance, 62, 597-628.
Brown, J., & Matsa, D. (2013). Boarding a sinking ship? an investigation of job applications to distressed
rms. manuscript.
Buera, F. J., Kaboski, J. P., & Shin, Y. (2011). Finance and development: A tale of two sectors. American
Economic Review, 101, 19642002.
Campa, J. M., & Kedia, S. (2002). Explaining the diversication discount. The Journal of Finance, 57(4),
1731-1762.
Cestone, G., & Fumagalli, C. (2005). The strategic impact of resource
exibility in business groups. Rand
Journal of Economics, 36, 193-214.
Cestone, G., & Fumagalli, C. (2013). Labor unemployment risk and corporate nancing decisions. Journal
of Financial Economics, 108, 449-470.
Cetorelli, N., & Strahan, P. E. (2006). Finance as a barrier to entry: Bank competition and industry
structure in local u.s. markets. Journal of Finance, 61, 437-461.
119
Chodorow-Reich, G. (2014). The employment eects of credit market disruptions: Firm-level evidence from
the 2008-09 nancial crisis. Quarterly Journal of Economics, 129(1), 1-59.
Collard-Wexler, A., & Loecker, J. D. (2015). Reallocation and technology: Evidence from the u.s. steel
industry. American Economic Review, Forthcoming.
Davis, S., Haltiwanger, J., Handley, K., Jarmin, R., Lerner, J., & Miranda, J. (2014). Private equity, jobs,
and productivity. American Economic Review, Forthcoming.
Duchin, R. (2010). Cash holdings and corporate diversication. Journal of Finance, 65(3), 955-992.
Duranton, G., & Puga, D. (2004). Micro-foundations of urban agglomeration economies. In J. V. Henderson
& J.-F. Thisse (Eds.), Handbook of econometrics. Elsevier: North-Holland.
Faure-Grimaud, A., & Inderst, R. (2005). Conglomerate entrenchment under optimal nancial contracting.
The American Economic Review, 95(3), 850-861.
Feenstra, R. (1996). Us imports,1972-1994:data and concordances. unpublished working paper,NBER.
Feenstra, R., Romalis, J., & Schott, P. K. (2002). U.s. imports, exports, and tari data: 1989-2001.
unpublished working paper, NBER.
Foster, L., Haltiwanger, J., & Krizan, C. (2001). Aggregate productivity growth: Lessons from microe-
conomic evidence. In E. Dean, M. Harper, & C. Hulten (Eds.), New directions in productivity analysis.
University of Chicago Press.
Foster, L., Haltiwanger, J., & Syverson, C. (2008). Reallocation, rm turnover, and eciency: Selection on
productivity or protability. American Economic Review, 98(1), 394-425.
Fresard, L. (2010). Financial strength and product market behavior: the real eects of corporate cash
holdings. Journal of Finance, 65, 1097-1122.
Fresard, L., & Valta, P. (2014). Competitive pressures and corporate policies: Evidence from trade liberal-
ization. Unpublished Working Paper.
Giroud, X., & Mueller, H. (2014). Capital and labor reallocation inside rms. The Journal of Finance,
Forthcoming.
Gopalan, R., & Xie, K. (2011). Conglomerates and industry distress. Review of Financial Studies, 24(12),
3642-3687.
Griliches, Z. (1957). Hybrid corn: An exploration in the economics of technological change. Econometrica,
25(4), 501-522.
Hann, R., Ogneva, M., & Ozbas, O. (2013). Corporate diversication and the cost of capital. Journal of
Finance, 68(5), 19611999.
Holmes, T. J., & Stevens, J. J. (2012). Exports, borders, distance, and plant size. Journal of International
Economics, 88, 91-103.
Hsieh, C.-T., & Klenow, P. J. (2009). Misallocation and manufacturing tfp in china and india. Quarterly
Journal of Economics, 124(4), 1403-1448.
Jayaratne, J., & Strahan, P. (1996). The nance-growth nexus: Evidence from bank branch deregulation.
Quarterly Journal of Economics, 111(3), 639-670.
Kerr, W., & Nanda, R. (2009). Banking deregulations, nancing constraints, and rm entry size. Journal
of Financial Economics, 94(1), 124-149.
120
Khanna, N., & Tice, S. (2001). The bright side of internal capital markets. Journal of Finance, 56(4),
14891528.
Krishnan, K., Nandy, D., & Puri, M. (in press). Does nancing spur small business productivity? evidence
from a natural experiment. Review of Financial Studies.
Kroszner, R. S., & Strahan, P. E. (1999). What drives deregulation? economics and politics of the relaxation
of banking branching restrictions. Quarterly Journal of Economics, 114(4), 1437-1467.
Kuppuswamy, V., & Villalonga, B. (2012). Does diversication create value in the presence of external
nancing constraints? evidence from the 20072009 nancial crisis. Working Paper.
Lang, L., & Stulz, R. (1994). Tobins q, corporate diversication, and rm performance. Journal of Political
Economy, 102(6), 1248-1280.
Larrain, M., & Stumpner, S. (2013). Financial liberalization and aggregate productivity: The microeconomic
channels. manuscript.
Levine, R. (1997). Financial development and economic growth: Views and agenda. Journal of Economic
Literature, 35, 688726.
Levine, R., Loayza, N., & Beck, T. (2000). Financial intermediation and growth: Causality and causes.
Journal of Monetary Economics, 46, 31-77.
Levinsohn, J., & Petrin, A. (2003). Estimating production functions using inputs to control for unobservables.
The Review of Economic Studies, 70(2), 317{341.
Lewellen, W. (1971). A pure nancial rationale for the conglomerate merger. Journal of Finance, 26(2),
521537.
Li, X. (2013). Productivity, restructuring, and the gains from takeovers. Journal of Financial Economics,
109(1), 250-271.
Maksimovic, V., & Phillips, G. (2001). The market for corporate assets: Who engages in mergers and asset
sales and are there gains? Journal of Finance, 6, 2019-2065.
Maksimovic, V., & Phillips, G. (2002). Do conglomerate rms allocate resources ineciently across indus-
tries? theory and evidence. Journal of Finance, 57, 721-767.
Maksimovic, V., & Phillips, G. (2008). The industry life-cycle, acquisitions and investment: Does rm
organization matter? Journal of Finance, 63(2), 673-707.
Matsusaka, J., & Nanda, V. (2002). Internal capital markets and corporate refocusing. Journal of Financial
Intermediation, 11(2), 176-211.
Matsusaka, J. G. (2001). Corporate diversication, value maximization and organizational capabilities.
Journal of Business, 74, 409-431.
Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate industry productivity.
Econometrica, 71(6), 1695{1725.
Midrigan, V., & Xu, D. Y. (2014). Finance and misallocation: Evidence from plant-level data. American
Economic Review, 104(2), 422-458.
Moll, B. (2014). Productivity losses from nancial frictions: Can self-nancing undo capital misallocation?
The American Economic Review, 104(10), 3186{3221.
121
Mundlak, Y. (1978). On the pooling of time series and cross section data. Econometrica, 46, 69-85.
Olley, G. S., & Pakes, A. (1996). The dynamics of productivity in the telecommunications equipment
industry. Econometrica, 64(6), 1263-1297.
Pagano, M., & Pica, G. (2012). Finance and employment. Economic Policy, 27(69), 5{55.
Paravisini, D., Rappoport, V., Schnabl, P., & Wolfenzon, D. (2015). Dissecting the eect of credit supply
on trade: Evidence from matched credit-export data. The Review of Economic Studies, 82(1), 333{359.
Pavcnik, N. (2002). Trade liberalization, exit, and productivity improvements: Evidence from chilean plants.
Review of Economic Studies, 69(4), 245-276.
Petersen, M., & Rajan, R. (1994). The benets of lending relationships: Evidence from small business data.
Journal of Finance, 49(1), 337.
Petersen, M., & Rajan, R. (2002). Does distance still matter? the information revolution in small business
lending. The Journal of Finance, 57(6), 2533{2570.
Petrin, A., & Levinsohn, J. (2012). Measuring aggregate productivity growth using plant-level data. The
RAND Journal of Economics, 43(4), 705725.
Rajan, R., Servaes, H., & Zingales, L. (2000). The cost of diversity: The diversication discount and
inecient investment. Journal of Finance, 55(1), 3580.
Rajan, R., & Zingales, L. (1998). Financial dependence and growth. American Economic Review, 88(3),
559-586.
Scharfstein, D., & Stein, J. (2002). The dark side of internal capital markets: divisional rent-seeking and
inecient investment. Journal of Finance, 55(6), 25372564.
Schoar, A. (2002). Eects of corporate diversication on productivity. Journal of Finance, 57(6), 2379-2404.
Schott, P. K. (2010). U.s. manufacturing exports and imports by sic or naics category and partner country,
1972-2005. unpublished working paper, Yale University.
Seru, A. (2015). Firm boundaries matter: Evidence from conglomerates and rd activity. Journal of Financial
Economics, Forthcoming.
Shin, H.-H., & Stulz, R. M. (1998). Are internal capital markets ecient? Quarterly Journal of Economics,
113(2), 531-552.
Stein, J. (1997). Internal capital markets and the competition for corporate resources. Journal of Finance,
52(1), 111-133.
Stein, J. C. (2003). Agency, information and corporate investment. In G. M. Constantinides, M. Harris, &
R. M. Stulz (Eds.), Handbook of the economics of nance. Amsterdam: Elsevier NH.
Swanburg, G. (2014). Internal capital markets and competitive threats. Working Paper.
Villalonga, B. (2004). Diversication discount or premium? new evidence from the business information
tracking series. Journal of Finance, 59, 479-506.
Williamson, O. (1975). Markets and hierarchies: Analysis and antitrust implications. New York: The Free
Press.
Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data. Cambridge, Mass: MIT
Press.
122
Wurgler, J. (2000). Financial markets and the allocation of capital. Journal of Financial Economics, 58(1),
187{214.
Xu, J. (2012). Protability and capital structure: evidence from import penetration. Journal of Financial
Economics, 106(2), 427-446.
123

Abstract (if available)

Abstract My thesis consists of two essays in corporate finance. The first chapter studies how differences in organizational form affect firms' ability to respond to competitive pressure. Using establishment level data from the U.S. Census and a difference-in-differences specification, I find that relative to focused firms, conglomerate firms more actively restructure during episodes of large import tariff reductions. They restructure to focus on their core competency and to improve productivity. Contrary to conventional wisdom, the internal capital market primarily functions through the extensive margin, with plant opening and closure decisions accounting for the majority (70%) of the productivity growth differential between conglomerates and standalones in the two years post tariff shocks. Finally, the coinsurance benefits likely explain conglomerates' activeness in restructuring activities. The second chapter studies the impact of state-level banking deregulation of local U.S. credit markets on the reallocation of labor within local industries. In particular, reallocation of labor towards firms with higher marginal products of labor significantly increases after these passages. Using firm production functions estimated with plant-level data, we propose and examine an approach that quantifies the industry productivity gains from labor reallocation and find that these gains are economically important.Our analysis suggests that labor reallocation is a significant channel through which credit market conditions affect the aggregate productivity and performance of local industries.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Internal capital markets and competitive threats

PDF

Two essays on financial crisis and banking

PDF

Slashing liquidity through asset purchases: evidence from collective bargaining

PDF

Essays on information and financial economics

PDF

Essays in financial intermediation

PDF

Essays in non-market strategy and entrepreneurship

PDF

How large are the coinsurance benefits of mergers?

PDF

Essays on real options

PDF

Essays in asset pricing

PDF

Essays on the role of entry strategy and quality strategy in market and consumer response

PDF

Essays on competition for customer memberships

PDF

Essays in corporate finance

PDF

Three essays on young entrepreneurial firms

PDF

Three essays on macro and labor finance

PDF

Evolution of returns to scale and investor flows during the life cycle of active asset management

PDF

Essays in financial economics

PDF

Essays on the competition between new and used durable products

PDF

Real asset liquidity and asset impairments

PDF

How product designs and presentations influence consumers’ post-acquisition decisions

PDF

Investment behavior of mutual fund managers

Asset Metadata

Creator Bai, Jianqiu (John) (author)

Core Title Essays in corporate finance

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Finance

Publication Date 04/14/2015

Defense Date 03/12/2015

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

Tag banking deregulation,internal capital market,labor reallocation,OAI-PMH Harvest,organizational form,productivity,restructuring

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Phillips, Gordon (committee chair), Ahern, Kenneth (committee member), Carvalho, Daniel (committee member), Jia, Nan (committee member), Wang, Yongxiang (committee member)

Creator Email jianqiub@usc.edu,john.baijianqiu@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-546995

Unique identifier UC11297650

Identifier etd-BaiJianqiu-3294.pdf (filename),usctheses-c3-546995 (legacy record id)

Legacy Identifier etd-BaiJianqiu-3294.pdf

Dmrecord 546995

Document Type Dissertation

Format application/pdf (imt)

Rights Bai, Jianqiu (John)

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

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