Nber Working Paper Series

Download as pdf or txt
Download as pdf or txt
You are on page 1of 67

NBER WORKING PAPER SERIES

MISALLOCATION, ESTABLISHMENT SIZE, AND PRODUCTIVITY

Pedro Bento
Diego Restuccia

Working Paper 22809


http://www.nber.org/papers/w22809

NATIONAL BUREAU OF ECONOMIC RESEARCH


1050 Massachusetts Avenue
Cambridge, MA 02138
November 2016

We thank David Atkin, Nick Bloom, Margarida Duarte, Peter Klenow, Virgiliu Midrigan,
Michael Peters, Xiaodong Zhu, two anonymous referees, and several seminar and conference
participants for helpful comments. All remaining errors are our own. Restuccia gratefully
acknowledges the financial support from the Canadian Research Chairs program and the Social
Sciences and Humanities Research Council of Canada. The views expressed herein are those of
the authors and do not necessarily reflect the views of the National Bureau of Economic
Research.

NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.

© 2016 by Pedro Bento and Diego Restuccia. All rights reserved. Short sections of text, not to
exceed two paragraphs, may be quoted without explicit permission provided that full credit,
including © notice, is given to the source.
Misallocation, Establishment Size, and Productivity
Pedro Bento and Diego Restuccia
NBER Working Paper No. 22809
November 2016
JEL No. O1,O4

ABSTRACT

We consider a tractable model of heterogeneous production units that features endogenous entry
and productivity investment to assess the quantitative impact of policy distortions on aggregate
output and establishment size. Relative to the standard factor misallocation framework, policy
distortions featuring a positive productivity elasticity of distortions imply larger reductions in
output through smaller investments in establishment productivity. A calibrated version of the
model implies that when the productivity elasticity of distortions increases from 0.09 in the U.S.
to 0.5 in India, aggregate output and average establishment size fall by 53 and 86 percent,
compared to 37 and 0 percent in the standard factor misallocation model. Entry productivity
investment and factor misallocation contribute equally to the reduction in output, whereas the
effect of lower life-cycle productivity growth is fully offset by increased entry and reduced
productivity dispersion. Establishment size differences in the model are consistent with evidence
from a comprehensive dataset we construct on average establishment size in manufacturing using
census data for 134 countries.

Pedro Bento
Texas A&M University
1101 Hollow Creek Dr. Unit 2208
Austin TX
78704
[email protected]

Diego Restuccia
Department of Economics
University of Toronto
150 St. George Street
Toronto, ON M5S 3G7
CANADA
and NBER
[email protected]
1 Introduction

A consensus view in the literature has emerged where the large variations in income per capita
across countries are mostly accounted for by differences in total factor productivity (TFP).1
What accounts for these TFP differences across countries? A prominent channel emphasized
in the literature generating differences in productivity is the misallocation of resources among
heterogeneous production units or establishments that differs across countries, e.g. Restuccia
and Rogerson (2008) and Hsieh and Klenow (2009). An important finding in the empirical
literature on misallocation is that not only is there evidence of large aggregate effects from mis-
allocation, but also that the implied distortions in developing countries feature stronger implicit
or explicit taxes on the more productive establishments, what Restuccia and Rogerson (2008)
called correlated idiosyncratic distortions.2 Whereas the misallocation literature has mostly
focused on analyzing the effect of such distortions on aggregate productivity for a fixed distri-
bution of establishment-level productivity, in this paper, we broaden this scope by emphasizing
the effect that correlated distortions have on productivity investment by establishments and
hence on the implied distribution of productivity in the economy. Our emphasis is motivated
by the empirical literature that finds substantial cross-country differences in establishment-level
productivity and investment in intangible capital.3 In our framework, policy distortions not
only misallocate resources across heterogeneous establishments but also affect the productivity
distribution, generating larger effects on aggregate productivity. We show through a calibrated
version of the model that correlated distortions generate substantial effects on the productiv-
ity distribution such that they roughly double the impact on aggregate productivity of factor
misallocation alone. We also show that, unlike in environments with fixed productivity dis-
tributions, the effect of correlated distortions on establishment productivity works to reduce
1
See, for instance, Klenow and Rodriguez-Clare (1997), Prescott (1998), and Jones (2015).
2
Larger productivity elasticities of distortions in poor countries is also highlighted in Hsieh and Klenow
(2014).
3
For instance, Hsieh and Klenow (2009), Bloom and Van Reenen (2010), Bloom et al. (2010), Pagés (2010),
Gal (2013), and Bloom et al. (2013) present evidence of establishment-level productivity differences across
countries and Corrado et al. (2012) document cross-country differences in intangible capital.

2
establishment size, an implication that is consistent with the evidence of smaller establishment
sizes in developing countries.

We consider a standard model of heterogeneous production units that builds from Hopenhayn
(1992). For comparability, the setup follows closely the monopolistic competition framework
used in the empirical analysis of Hsieh and Klenow (2009). The basic framework is extended
along three important dimensions in order to address differences in entry and establishment-level
productivity. We incorporate an endogenous entry decision of establishments, an initial invest-
ment decision determining establishment-level productivity upon entry, and investment over
time determining the growth of establishment productivity over the life cycle. There is a large
number of potential entrants that draw their idiosyncratic productivity from a known distribu-
tion at a cost. Establishments can improve their initial productivity through investment, but
only before the realization of their idiosyncratic productivity. In the theory, ex-ante identical
entering establishments make the same productivity investment decision but are ex-post hetero-
geneous in their idiosyncratic productivity. The theory connects policy distortions, institutions,
and frictions that discourage establishment-level investment. The key emphasis in the model is
the extent to which distortions that effectively penalize more productive relative to less produc-
tive establishments—correlated idiosyncratic distortions—discourage productivity investment
by all establishments.4 In the model, we show there is a strong connection between the extent
of correlated distortions, entrant productivity, establishment-level productivity growth, and
the mass of entrants in the economy. These effects work to lower establishment size, entrant
productivity, establishment productivity growth over the life cycle, and aggregate productivity.
We emphasize that with no entrant investment and constant establishment productivity, the
model would imply no differences in establishment productivity and establishment size from
4
The set of policies and institutions that effectively create correlated idiosyncratic distortions is very large
and has been extensively discussed in the literature, see for instance Restuccia and Rogerson (2013), Restuccia
(2013a), and Hopenhayn (2014) for discussions of these policies and institutions. Some examples are small
business subsidies, financial constraints, trade restrictions, and the ability of establishments to remain informal
to avoid taxes. Hsieh and Klenow (2009, 2014) discuss a number of possible extensions to a standard model
through which correlated distortions can emerge without adversely affecting productivity investment, such as
markups that increase in the size of an establishment as in Melitz and Ottaviano (2008) or Peters (2013), and
argue that these alternative mechanisms lead to counterfactual predictions.

3
factor misallocation. As a result, allowing for entry and investment not only amplifies the losses
in output and productivity from misallocation, but also rationalizes the impact of distortions
on establishment size as observed in the cross-country data. To the extent that misallocation
is reduced within a country over time, the model also contributes to understanding trends in
establishment size. In the United States, for example, Poschke (2014) reports a doubling of
average firm size since the early twentieth century, while the results in Ziebarth (2013) and
Hsieh and Klenow (2009) suggest a significant reduction in misallocation in the U.S. over the
same time period.

We calibrate a benchmark economy to U.S. data and show that reasonable variations in the
extent of correlated distortions across countries have substantial negative effects on establish-
ment size, entrant productivity, establishment growth over the life cycle, and aggregate output
per capita. In particular, compared to the calibrated U.S. benchmark economy, increasing
correlated distortions to 0.5—the elasticity between wedges and establishment productivity in
India—generates a reduction in establishment size from 22 workers in the U.S. benchmark to
3 workers, which represents an 86 percent reduction in average establishment size and a fac-
tor difference in average establishment size between the U.S. and India of 7. The increase
in correlated distortions generates a reduction in entrant productivity of 58 percent and in
establishment-level productivity growth from 5 to 2 percent, which together with the effect
of factor misallocation implies a drop in aggregate TFP of 53 percent. To put it differently,
in this experiment, a 1.6-fold difference in aggregate productivity between the U.S. and India
generated by static factor misallocation is amplified by 34 percent due to the added effects on
productivity investment.5

Our model is very tractable and nests the standard model of misallocation without life-cycle
growth. As a result, we are able to explicitly decompose the effects of correlated distortions into
5
Interestingly, our parsimonious measure of correlated distortions generates a reduction in aggregate TFP
from factor misallocation that compares in magnitude to the estimates in Hsieh and Klenow (2009) for India
using detailed establishment-by-establishment wedges, suggesting that our summary measure of distortions
captures the bulk of their effects on factor misallocation.

4
those working through the entry-investment channel emphasized in our paper, through the life-
cycle growth channel analyzed in Hsieh and Klenow (2014), and through the factor misallocation
channel analyzed in Restuccia and Rogerson (2008) and Hsieh and Klenow (2009). We show that
accounting for the life-cycle investments of establishments does not by itself amplify the impact
of correlated distortions on aggregate TFP, relative to the effects of static factor misallocation in
Restuccia and Rogerson (2008).6 Accounting for exogenous life-cycle growth reduces the impact
of correlated distortions because of their offsetting effect on entry.7 Correlated distortions
reduce the productivity growth of establishments which lowers aggregate TFP, but the net
impact on TFP is negligible (if not positive) because the lower productivity growth encourages
entry and compresses the productivity distribution reducing the impact of factor misallocation.
Correlated distortions also reduce investment in establishment-level productivity upon entry
but the effect on aggregate TFP is not mitigated by reduced misallocation. We show that the
reduction in entrant productivity brought about by correlated distortions substantially reduces
aggregate TFP, approximately doubling the impact of misallocation relative to the measured
impact in an economy without entrant investment, such as that in Hsieh and Klenow (2014).
Moreover, relative to the existing literature, our framework generates large establishment size
and productivity effects that are more in line with the cross-country data.

To assess the ability of correlated distortions to quantitatively explain productivity differences


across a large set of countries, we document evidence from cross-country micro data for the
elasticity between distortions (wedges) and establishment productivity, using establishment-
level data from the World Bank’s Enterprise Surveys. We show that the elasticity of distortions
with respect to productivity in the micro data is strongly negatively related to both average
establishment size and GDP per capita across 63 countries. We also provide evidence that the
establishment size effects in the model are consistent with data on establishment sizes. Evi-
6
Restuccia and Rogerson (2008) emphasize the impact of misallocation on aggregate TFP in an environment
where establishment-level productivity is exogenous and constant and entry is not affected by distortions. In
Hsieh and Klenow (2009) entry is constant and establishment-level productivity is also exogenous but given by
the micro data in each country.
7
A similar result in the context of a model with exogenous establishment growth is emphasized in Fattal-Jaef
(2015).

5
dence of the relationship between development and establishment size has been both sparse and
inconclusive due to the lack of standardized size data for a large group of countries.8 We address
this by constructing a standardized database on establishment size based on individual-country
data from manufacturing censuses and representative surveys and registries. Using hundreds
of separate sources, we have assembled data for 134 countries with comparable employment-
size data. In contrast to Alfaro et al. (2009) and Bollard et al. (2014), who use international
data plagued by cross-country differences in the size of sampled firms, we show that average
establishment size is strongly positively correlated with GDP per capita. For instance, whereas
average establishment size in U.S. manufacturing is 22 workers, in Benin and Sierra Leone it
is about 2 workers, an 11-fold difference. As a summary measure of the relationship between
development and size, we compute the income elasticity of establishment size to be 0.29. Sim-
ilarly, Poschke (2014) reports a positive income elasticity of firm size for the entire economy
across a smaller set of countries of 0.45.9 By constructing a harmonized dataset with compa-
rable numbers for the average establishment size in the manufacturing sector across a large set
of countries, we also hope to contribute to the literature by providing data that can be used
for calibration exercises and serve as an additional source of discipline to quantitative models.

Our paper is closely related to the broad literature on misallocation and productivity discussed
earlier. Specifically, our paper relates to the recent literature emphasizing the impact of misal-
location on establishment-level productivity.10 We highlight two contributions of our work to
this literature. First, we emphasize the effect of misallocation on entrant investment, which we
find is quantitatively substantial, roughly doubling the impact of factor misallocation. Second,
8
For instance, in recent studies, Poschke (2014) finds a strong positive relationship between average size and
development using two separate economy-wide samples of small/medium firms and large firms, while Alfaro
et al. (2009) and Bollard et al. (2014) find a negative relationship between size and development. We discuss
further the literature on establishment size in Section 2.
9
We note that large differences in operational scales are also found in other sectors such as agriculture from
Census data where the operational scale of farms in rich countries is 34 times that of poor countries (see for
instance, Adamopoulos and Restuccia, 2014).
10
Early examples of this literature include Restuccia (2013b) and Bello et al. (2011) with more elaborate
analysis in Ranasinghe (2014), Bhattacharya et al. (2013), Gabler and Poschke (2013), Hsieh and Klenow
(2014), and Da-Rocha et al. (2015). Closely related to our work is Hsieh and Klenow (2014) who consider the
impact of correlated distortions on life-cycle productivity growth of establishments and aggregate TFP.

6
we develop a parsimonious model that allows us to explicitly and analytically disentangle the
effects of correlated distortions working through the entry-investment channel, the life-cycle
growth channel, and factor misallocation. Investment by entrants has been an under-explored
mechanism through which policies and institutions can affect aggregate productivity, but a
recent paper by Moreira (2015) suggests it is important. Analyzing the size and investment
decisions of entrants over the business cycle in the U.S. data, Moreira finds that the average
size of establishments entering during expansions is larger (both at entry and over their life cy-
cle) than those entering during recessions. She concludes that firm investment decisions upon
entry depend on the state of the economy and that the productivity effects that result are very
persistent over time. In generating differences in establishment size, our paper is related to the
seminal work of Lucas Jr. (1978) who showed that an elasticity of substitution less than one
may be needed between capital and labor in the production function in order to rationalize the
larger operational scales in rich countries. In our framework, even with Cobb-Douglas tech-
nology, establishment size can vary with correlated distortions. The view that differences in
size across countries can arise from distortions shares with the work of Guner et al. (2008) who
emphasize size-dependent distortions, i.e., distortions such as taxes and regulations that apply
to establishments above a threshold size in terms of the number of workers. We differ from
Guner et al. (2008) in that in our framework any correlated distortion causes productivity at the
establishment level to drop for all establishments, adding to the potential factor misallocation
effects typically emphasized in the literature. For this reason, the size and productivity impact
of correlated distortions in our framework are orders of magnitude larger than those emphasized
in Guner et al. (2008). More generally, a number of papers have developed quantitative models
that generate differences in establishment size across countries.11 A common finding in these
papers, however, is a relatively small impact of distortions on size. Our model with entrant
investment generates large quantitative differences in average establishment across countries,
closer to the differences we document in the data. The literature has also explored many specific
policies thought to explain income differences across countries such as firing costs, entry costs,
11
For example, Bhattacharya et al. (2013), Poschke (2014), and Hsieh and Klenow (2014).

7
or average tax rates. But in a standard framework all of these policies lead to larger establish-
ment sizes in poor countries.12 To the extent that poor countries have both harmful policies
and correlated distortions, our paper helps to rationalize why establishments are smaller in
countries even when facing higher average costs of doing business.

The paper is organized as follows. In the next section, we present the facts from our constructed
dataset of 134 countries to establish that establishment size increases substantially with the level
of development across countries. Section 3 presents the model and characterizes the qualitative
implications. In Section 4, we calibrate the model to data for the United States and show the
quantitative implications of the model for hypothetical variations in the extent of correlated
distortions. We then construct and document measures of correlated distortions across countries
and assess their potential to generate differences in size and productivity. We also discuss our
results for reasonable extensions in the model and reasonable variations in key parameter values.
We conclude in Section 5.

2 Average Establishment Size across Countries

We describe the construction of a newly-assembled dataset for the average employment size of
manufacturing establishments across a large sample of countries using census or representative
survey data to show that average establishment size is strongly positively related to the level of
development. While this finding is not entirely new or surprising, we discuss how our dataset
is able to circumvent some of the limitations from earlier studies on establishment size and
development.
12
See for instance Hopenhayn and Rogerson (1993), Barseghyan and DiCecio (2011), and Moscoso Boedo and
Mukoyama (2012), among others.

8
2.1 Data

We construct a dataset for the average employment size of manufacturing establishments across
countries using hundreds of reports from economic censuses and nationally-representative sur-
veys.13 Our goal in the construction of this dataset is to obtain an internationally-comparable
measure of average establishment size for a large sample of countries that is representative of
the world income distribution. The challenges of course are data availability—which typically
biases country samples towards rich countries— and international comparability—which often
involves having data reported using different definitions of employment and production units
or having data that disproportionally include larger firms.

Of crucial importance for the assessment of the relationship between establishment size and
development is the inclusion in the data of all establishments regardless of whether the estab-
lishments are registered or not, and whether the establishments have paid employees or not,
as a substantial portion of establishments in poor countries are unregistered and own account
businesses and may include unpaid family workers. In Sierra Leone, for example, 83 percent of
establishments have no paid employees, and in Ghana, unpaid workers account for almost half
of the manufacturing workforce. As a result, excluding non-employer establishments would gen-
erate a highly distorted picture of establishment size differences across countries. Throughout
our data collection process, we have made an effort to search for evidence from methodology
documents and other published reports that small establishments are not included. Any coun-
try for which such evidence exists is not included in our sample. We include all countries with
publicly-available data representative of all manufacturing establishments or firms.14 Estab-
lishments in the manufacturing sector include businesses with a fixed location. It also includes
businesses operating out of households when a sign is posted on the premises.15
13
In Appendix A we provide greater detail about how we construct the dataset. We also provide a list of
countries included and a list of the sources used for each country.
14
We also include in the dataset all territories such as French Guiana, Hong Kong, and Puerto Rico. We use
the word “country” solely for ease of exposition.
15
One exception to this rule is the United States. Although U.S. employer data uses a standard definition of
“establishment”, the data for non-employers (i.e., self-employed) includes businesses with no fixed location like

9
We collected data for as many years as possible for each country from 2000 to 2012 to construct
our dataset. Our standardized definition of size is the average number of persons engaged per
establishment. For the vast majority of countries in our sample, the data are reported as total
number of persons engaged and total number of establishments. But for some countries, about
a quarter of our sample, the data are reported differently as the total number of employees,
the total number of full-time equivalents, the total number of firms, or a combination of these
instead of persons engaged and establishments. In these cases, we impute persons engaged per
establishment using the reported data as follows.16 To impute the number of persons engaged
in countries that only report paid employees, we regress persons engaged on employees using
a large set of country-years for which both measures are reported. We then use the resulting
coefficient to calculate the number of employees for each year in countries that report only
employees. We do a similar imputation of persons engaged for countries that report full-time
equivalents or both employees and full-time equivalents. Using our measures of persons engaged
(both reported and imputed), the number of establishments, and the number of firms, we then
calculate the number of persons engaged per establishment and per firm for each country-year.
To impute the number of persons engaged per establishment for countries that only report the
number of firms, we first regress persons engaged per establishment on persons engaged per
firm using all country-years that report both firm and establishment counts, and then use the
resulting coefficient to impute the number of persons engaged per establishment for each year
in countries that report only firm counts. We emphasize that not only do these imputations
involve a small subset of our sample, but also that they are robust to whether the imputations
are based on cross-country relationships for poor vs. rich countries.

In our final dataset we report the average of persons engaged per establishment across all
available years for each country, resulting in final sample of 134 countries.
food trucks or sub-contractors in construction. Our focus on manufacturing should prevent this from being an
issue, but our reported employment size for the U.S. may as a result be slightly biased downwards.
16
See Appendix A for greater detail of the imputation procedure.

10
2.2 Findings

Table 1 reports some descriptive statistics concerning average establishment size from our
dataset, GDP per capita, and population for all the countries in our sample.17

Table 1: Descriptive Statistics

Poorest Richest
Mean Median Decile Decile
Establishment Size 12 9 6 19
GDP per capita (thousands) 18 13 1.2 55
Population (millions) 32 6 28 25
Notes: “Poorest” and “Richest” deciles refer to the ten percent of countries
with the lowest and highest GDP per capita. Data from multiple sources, see
text for details.

Figure 1 shows average establishment size for 134 countries in relation to GDP per capita.
The data clearly show a positive correlation between average establishment size and GDP per
capita. In particular, the elasticity of establishment size with respect to GDP per capita is
0.29. Figure 2 shows that the correlation between size and income is even stronger if we omit
small countries with populations less than half of one million. In this case, the elasticity rises
to 0.35. Each of these elasticities is remarkably robust to controlling for openness to trade and
quality of institutions.18 Recent models linking market size and markups predict that both
GDP per capita and establishment size should increase with population, suggesting that the
relationship illustrated in Figures 1 and 2 could be explained by differences in population size
across countries.19 But Figure 3 shows that establishment size is not systematically related to
17
GDP per capita (adjusted for purchasing power parity, PPP) is from Penn World Table v. 8.0 for 105
countries, the IMF’s World Economic Outlook 2013 for 7 countries, and the CIA World Factbook for 17 countries.
For four countries (actually overseas departments of France), GDP per capita is from France’s National Institute
of Statistics and Economic Studies and is made relative to the U.S. GDP per capita using market exchange
rates. GDP per capita for Âland Islands is from Statistics and Research Âland, and adjusted for purchasing
power parity using Finland’s PPP exchange rate from Penn World Table v. 8.0. Population data is from Penn
World Table v. 8.0 (105 countries), the World Bank’s World Development Indicators (21 countries), the CIA
World Factbook (7 countries), and Statistics and Research Âland (for Âland Islands).
18
Our measure of openness to trade is from Penn World Table v. 8.0. Our measure of institutional quality is
the Heritage Foundation’s Index of Economic Freedom (2014).
19
Melitz and Ottaviano (2008) and Desmet and Parente (2012), for example, each develop models in which

11
ASM

MYS

50
SGP
ARE LUX LIE
TTO DEU

Establishment Size (log scale)


PRI MCO

25
UKR MAC QAT
AUT
CANUSA
BGRMNP DNK
GBR KWT
NLD
JPN
BRAVEN LVARUS TWN FRO CHE
VIR GUM BEL IRL
FRA
PHL GEO MKD ROULTU PYF ABW
ISR SWE
POL
KAZ NZL HKG
KGZMDA TON ARG MEX HUN
HRV SVN

10
MUS ESTBHRKOR
ESP
MDG GHA
NPL STP LKA BTN
PERZAF
THA
SLVCOL URY PRT
SAU BRNSMR
AUS
CPVJOR TUR MLT
CZE ITAFIN
BGDSDN KHMVNM TUNBIH SRB PAN
DZA LBY SVK NOR
RWA CMR HNDBOL
UGA PRYMNGALB
ECU CYP AND ALA BMU
MAR IRN GRL
LAO MNE
PSENICIDN
4
GRC
REU NCL
SYR UVK
IND
YEM
PLWMTQ
GUF
SLE MDV
2

ETH
MWI GLP
BEN
1

500 2500 10000 50000


GDP per Capita (log scale)

Figure 1: Establishment Size and GDP per Capita

population.20

To confirm that the observed relationship between establishment size and GDP per capita
(elasticity of 0.29) is not being driven by our construction of establishment size data using
imputations, we separately test the relationship between size and GDP per capita for persons
engaged per establishment, persons engaged per firm, employees per establishment, and employ-
ees per firm, using only the raw source data for each country. The corresponding elasticities are
all positive and of comparable magnitude: 0.38 for persons engaged per establishment (data for
64 countries), 0.34 for persons engaged per firm (data for 48 countries), 0.32 for employees per
establishment (data for 45 countries), and 0.28 for employees per firm (data for 52 countries).

2.3 Related Literature

We now compare the implications of our data relative to the existing work in the literature with
emphasis on the connection between average establishment size and development. There are
larger populations can lead to both higher output per capita and larger establishments.
20
The regression slope coefficient (standard error) in Figure 1 is 0.29 (0.04) and in Figure 2 is 0.35 (0.04). In
Figure 3 the slope coefficient is an insignificant -0.003 (0.03).

12
MYS

50
SGP

ARE
TTO DEU
PRI

25
UKR QAT

Establishment Size (log scale)


AUT
CANUSA
BGR DNK
GBR
NLD KWT
JPN
TWN
BRAVEN LVA CHE
RUS BEL
PHL GEO MKD ROULTU FRA IRL
POL
KAZ ISR SWE
KGZMDA ARGMEX HUN
HRV NZL
SVN HKG
EST
10
GHA PERZAF
THAMUS BHRKOR
ESP
MDG
NPL LKA BTNSLVCOL URY PRT
SAU AUS
JOR TUR CZE ITA
FIN
BGDSDN KHMVNM TUN PAN
BIH SRB LBY
CMR BOLPRY DZA
MNG ALB SVK NOR
RWA
UGA HND ECU CYP
MAR IRN
LAO
PSE MNE
NICIDN
4

GRC
REU
SYR UVK
IND
YEM

SLE
2

ETH
MWI
BEN
1

500 2500 10000 50000


GDP per Capita (log scale)

Figure 2: Establishment Size and GDP per Capita (small countries removed)

ASM

MYS
50

SGP
LIE LUX ARE
TTO DEU
Establishment Size (log scale)

MCO PRI
25

MAC QAT UKR


AUT CAN USA
MNP KWT DNK BGR NLD GBR JPN
FRO LVA CHE TWN VEN BRA
RUS
VIR IRL BEL
ABW GUM
PYF MKDLTU
GEO ISR
SWE ROU FRA PHL
NZL
SVNMDA
HRV HKGHUN POL
KAZ ARG
KGZ MEX
10

TON EST
BHR
MUS KOR
ESP
SMR BRN PER
GHA ZAF
THA
STP MLTBTN URY SLV PRTMDG
LKA
SAU
AUS COL
NPL ITA
CPV JOR
FIN CZE TUR
PAN
BIH SRB KHMSDN
TUN BGD
ALB
MNG LBY
NOR
SVK
PRY BOL CMRDZA VNM
ALA BMU CYP HND RWA
ECU UGA
AND
GRL MAR IRN
MNE LAO
PSENIC GRC IDN
4

NCL REU UVK SYR


YEM IND
PLW GUF MTQ
MDV SLE
2

GLP MWI ETH


BEN
1

.01 .1 1 10 100 1000


Population (millions, log scale)

Figure 3: Establishment Size and Population

13
numerous studies emphasizing firm size across countries, for instance Tybout (2000) surveys
a broad literature addressing the size of manufacturing firms in developing countries, but as
recognized in this literature, for the vast majority of work the evidence on firm size comes from
a relatively small sample of countries and from data sources that may not deal systematically
with comparability issues across countries.21 Similarly, the large empirical literature addressing
the misallocation of resources across establishments in developing countries such as that in
Hsieh and Klenow (2009, 2014) and Pagés (2010) focus on a handful of countries with varying
degrees of data comparability across the countries.22

A widely cited reference for the relationship between firm size and income is Alfaro et al.
(2009). They use Dun & Bradstreet’s WorldBase data (DB) to document a negative relationship
between firm size and income per capita across 79 countries. More recently, Bollard et al.
(2014) report the same negative relationship using data from the United Nations Industrial
Development Organization’s (UNIDO) Industrial Statistics Database for 72 countries. These
observations are in direct contrast to those just documented from our data.23 To understand
Alfaro et al. (2009), it is useful to first emphasize that DB is comprised of business data
aggregated from multiple sources that is typically used to provide credit and market-assessment
services. A key issue is that DB has sparse coverage of small firms in poor countries relative
to rich countries, with no attempt to make the data representative of all establishments. As a
result, when calculating average firm size in poor countries, the under-representation of small
firms biases the average upwards. In a sense, Alfaro et al. are comparing average size across most
firms in rich countries with the average size of only large firms in poor countries. The UNIDO
data used by Bollard et al. (2014) similarly include countries with unbalanced populations
of firms, with some countries reporting data for all firms and other countries reporting data
21
For example, Hsieh and Olken (2014) study comprehensive firm-level data to study firm size distributions
across countries but the focus involves three countries: India, Indonesia, and Mexico.
22
In addition, in the studies of misallocation the data required to measure TFP at the establishment level
implies that many small establishments with missing or unreliable observations are excluded.
23
We note that the negative relationship between average firm size and development found using DB and
UNIDO data is also at odds with the relationship found for specific sectors such as agriculture where census
data indicates much smaller farm size operations in poor countries relative to rich, see for instance Adamopoulos
and Restuccia (2014).

14
only for larger firms.24 More importantly, our data contains information for 59 countries from
Alfaro et al.’s sample and 59 countries from Bollard et al.’s sample, and the result of a positive
relationship between establishment size and development is even stronger in these subsamples
than for all 134 countries.

The closest empirical application to our paper is Poschke (2014) who uses two separate datasets,
one from the Global Entrepreneurship Monitor (GEM) for small and medium firms in 47 coun-
tries, and one from Amadeus for large firms in 34 countries. It is the closest empirical application
to us because in choosing these two datasets, Poschke (2014) attempts to harmonize the cover-
age of firms across the entire cross-country income distribution. Not surprisingly then, Poschke
(2014) finds that firm size is strongly positively related with development, consistent with our
evidence. Unlike Alfaro et al. (2009) and Bollard et al. (2014), the survey data used in Poschke
is constructed to be representative of firms within each size class for each country. Although
with a smaller sample of countries, Poschke shows that average firm size in each of these datasets
is increasing in development across multiple sectors of the economy. For instance, the implied
income elasticities of firm size are 0.45 in both datasets in Poschke (2014), higher than the 0.29
elasticity in our dataset.25 While the cross-country patterns that arise between size and income
are similar, there are important differences between our dataset and the two datasets used in
Poschke (2014). First, our dataset provides a number on the average establishment size for each
country in a large sample of 134 countries, whereas the data in Poschke (2014) involves two
separate numbers for the average establishment size of small and medium firms from GEM and
of large firms from Amadeus for a relatively small sample of countries (47 and 34 countries in
each case). This distinction is important, because characterizing the relationship between size
and development is far from the only use for our data. In many applications a specific mea-
sure of establishment size is relevant for calibration and quantitative assessment, as is the case
in our paper for evaluating our model’s quantitative predictions for establishment size across
countries. Second, our dataset provides average establishment size in the manufacturing sector
24
For this reason, some of the countries used in Bollard et al. (2014) have been excluded from our dataset.
25
We thank Markus Poschke for generously providing the implied size-income elasticities in his two samples.

15
across countries, therefore implicitly controlling for changes in the structure of the economy
which vary systematically across countries, whereas the average size in Poschke (2014) includes
firms in all sectors of the economy.26

Comparing the results of our analysis with those of the previous literature makes it clear that
analyzing standardized, representative size data for a specific sector, especially with respect to
the smallest establishments in poor countries, is crucial to obtaining an accurate measure of
the average employment size of establishments across countries and how it varies with the level
of development.

3 The Model

We consider an economy where time is discrete and indexed by t. A representative final-good


firm uses a variety of imperfectly substitutable inputs from intermediate-good firms to produce
the final consumption good, which also serves as the numéraire.27 There is a stand-in household
endowed with a continuum of members (of measure one), each supplying one unit of labor each
period. There are a large number of potential intermediate firms who are free to enter, but must
pay a fixed entry cost and make a costly productivity-investment decision before producing.
Each period after entry firms invest to further increase their productivity. Firms face output
distortions which may be correlated with firm-level productivity, and take these distortions into
account when investing in productivity. We assume an exogenous probability of exit and, as
a result, there is ongoing entry and exit in steady state. We study the steady state of the
economy in which firms take the wage, the interest rate, and the size of the economy as given,
and free entry ensures the value of entry is driven to zero. We then consider how the extent
of correlated distortions affects the number of firms, investment, firm growth, and aggregate
26
See for instance Duarte and Restuccia (2010) and Herrendorf et al. (2014) for a characterization of differences
and changes in the sectoral structure of the economy across countries and over time.
27
Throughout we use “firm” and “establishment” interchangeably in reference to intermediate-good producers.

16
output.28 We begin by describing the environment in more detail.

3.1 Environment

The representative final-good firm produces output using a variety of inputs from intermediate-
good firms according to the following production function;

σ
Z N σ−1
 σ−1
Y = yiσ
di ,
0

where N is the number of intermediate-good firms, yi the demand for input i, and σ > 1 the
constant elasticity of substitution between varieties.

Each intermediate-good firm has access to the following production function;

y = sz`,

where sz is productivity and ` is labor. An entrant’s realized z is drawn from a known exogenous
distribution, while s is determined in the following way. After paying an entry cost ce Y , but
before realizing z, an entrant chooses its initial s0 by incurring a cost equal to cS Y sθ0 , where
the subscript on s refers to the age of the firm, and both cS > 0 and θ > 1 are exogenous
and common to all firms.29 At the beginning of each period after entry, firms increase their
productivity by a factor of 1 + g by incurring a cost equal to cg (1 + g)φ Ω(s−1 , z), where s−1 z is
a firm’s productivity from the previous period, cg > 0, φ > 1, and;30

(s−1 z)σ(1−γ)−1
 
Y−1
Ω(s−1 , z) = · R N−1 .
N−1 1
(s−1,i zi )σ(1−γ)−1 di
N−1 0

28
We refer to the mass of firms as the number of firms for ease of exposition.
29
Our specification of the entry cost as a multiple of aggregate output is consistent with Bollard et al. (2014)
who argue, using time-series data, that entry costs should scale up with secular development. Note that if
population were not normalized to one, we would need to make the entry and investment costs scale up with
output per capita.
30
The productivity of an n-year-old firm is therefore equal to z · s0 · (1 + g1 ) · (1 + g2 ) · · · (1 + gn ).

17
That the cost of per-period productivity investment is increasing in the relative profitability
of a firm ensures that Gibrat’s Law is satisfied, that is that firm growth is independent of the
initial size of a firm.31 At the end of each period, each intermediate producer faces an exogenous
probability of exit equal to λ.

Output distortions are such that each firm retains a fraction (1 − τ ) of its output, and we
assume τ depends on firm-level productivity as follows;

(1 − τ ) = (sz)−γ ,

where the parameter γ is the elasticity of a firm’s distortion with respect to its productivity.

Given our assumptions, producers engage in monopolistic competition, each entrant chooses the
same s0 , each incumbent chooses the same g each period, all entrants choose to continue oper-
ating, and the cross-sectional distribution of firm productivities remain invariant. We abstract
from the household’s inter-temporal consumption decision and simply assume an exogenous
interest rate R.

3.2 Equilibrium

We focus on the steady-state decentralized equilibrium of the economy in which the distributions
of prices and allocations are invariant.32 A steady-state decentralized equilibrium is defined as a
wage rate w, distributions of firm-level productivities sz, intermediate-good prices P , output y,
labor demand `, operating profits π, productivity growth g, number of firms N , and aggregate
output Y , such that;
31
We follow Atkeson and Burstein (2010) in ensuring Gibrat’s Law is satisfied.
32
The equilibrium of this economy differs from its optimal allocation. As noted in Atkeson and Burstein
(2010), specifying entry and investment costs in terms of goods rather than labor results in a wedge between the
equilibrium allocation and the allocation chosen by a social planner. We solve for the social planner’s allocation
in Appendix B.

18
(i) given each P , the final-good firm demands intermediate-good inputs to maximize profits
in each period,

(ii) given w, R, and Y , intermediate-good producers choose labor to maximize per-period


profits,

(iii) given w, R, and Y , incumbents choose a factor increase in productivity (1+g) to maximize
the expected present value of lifetime profits,

(iv) given w, R, and Y , entrants choose initial productivity s0 to maximize the expected
present value of lifetime profits,

(v) free entry ensures the expected present value of lifetime profits for an entrant is equal to
the expected present value of all productivity investments plus the entry cost,

(vi) markets clear, i.e., the supply of labor (equal to one) is equal to the quantity of labor
demanded by firms.

The final-good firm takes input prices as given and maximizes profits in each period, generating
the following inverted demand function for each input i;

1 −1
P i = Y σ yi σ .

Operating profits in each period for an incumbent firm-i are therefore;

1 σ−1
πi = (1 − τi )Y σ yi σ − w`i , where yi = si zi `i .

Firms choose labor to maximize operating profits each period, generating the following demand
for labor and optimal output;


(1 − τi )σ Y (si zi )σ−1

σ−1
`i = ,
wσ σ

19

(1 − τi )σ Y (si zi )σ

σ−1
yi = .
wσ σ

Per-period operating profits for firm-i, given si zi , are therefore;

(1 − τi )σ Y (si zi )σ−1 (σ − 1)σ−1


πi = . (1)
wσ−1 σ σ

Combining yi above with the final-good production function results in the following expression;

1
  Z N  σ−1
σ−1
w= (si zi )σ−1 (1 − τi )σ−1 di . (2)
σ 0

Labor-market clearing results in;

 σ Z N 
σ σ−1 σ−1 σ
w =Y (si zi ) (1 − τi ) di . (3)
σ 0

Combining equations (2) and (3) and rearranging results in expressions for aggregate output
and the wage rate;
"Z 1
# σ−1
N  σ−1
1 − τi
Y = (si zi )σ−1 di , (4)
0 1−τ
 
σ−1
w = (1 − τ ) Y, (5)
σ

where (1 − τ ) is the weighted average of (1 − τi ) across all firms, weighted by each firm’s share
of aggregate output;
Z N
Pi y i
(1 − τ ) = (1 − τi )di.
0 Y

We digress to more precisely explain the counterfactual experiment we are interested in. In
Figure 4 the curve labeled ‘low γ’ represents the relationship between a firm’s distortion τi
and productivity si zi when γ is relatively low. In log scale, the slope of this curve is equal
to γ, the elasticity of distortions with respect to productivity. To investigate the impact of
more correlated distortions (a higher γ), we increase γ and then compare the two steady-state

20
equilibriums. While an increase in γ implies a pivoting of the curve in Figure 4, we must choose
a point to pivot around. Following Hsieh and Klenow (2009), we choose (1 − τ )−1 as the pivot.
This means that as γ is increased in our counterfactual, τ is kept constant. This allows us to
focus on the effects of an increase in γ while abstracting from the already well-studied effects
of a change in the average distortion.33

Figure 4: Firm-Level Distortions and Productivity

Notice that aggregate output in equation (4) can be rewritten as;

1
"Z
N  σ−1 # σ−1
M RP L
Y = si zi di , (6)
0 M RP Li

where a firm’s revenue marginal product of labor and the average revenue marginal product of
labor are defined as in Hsieh and Klenow (2009);

Pi yi 1
M RP Li = ∝ ,
`i (1 − τi )
33
There are additional reasons to keep τ constant in our counterfactual experiment. First, several of the
situations discussed in the literature that contribute to the existence of correlated distortions (for example,
subsidies to small or unproductive firms) need not raise the average distortion faced by firms. Second, to the
extent that correlated distortions are the result of explicit tax schedules, there does not appear to be a systematic
relationship between development and tax revenue as a fraction of GDP (Easterly and Rebelo, 1993). Third,
the method developed by Hsieh and Klenow (2009) to infer the distribution of distortions across establishments
from micro data (which we use in Section 4.2) can not identify the average distortion faced by establishments.

21
 Z N −1
1 Pi y i 1
M RP L = M RP L−1
i · di ∝ .
N 0 Y (1 − τ )

Equation (6) makes clear that if firm-level productivity were exogenous and constant over
the life cycle of the firm, removing misallocation by setting each firm’s M RP Li to M RP L
(bringing γ to zero while maintaining τ ) would have the same effect on aggregate output as in
Hsieh and Klenow (2009), as long as the number of firms N is not affected. To see that N is
indeed unaffected if productivity is fixed, we use equations (1) and (2) to derive the expected
per-period operating profits of an entrant;34

N
Y (σ − 1)σ−1
 
Y (1 − τ )
Z
1 σ−1 σ
E[π̂] = (si zi ) (1 − τi ) di = . (7)
wσ−1 σ σ N 0 σN

As long as the cost of entry scales up with aggregate output as is the case in our framework
and (1 − τ ) is not affected by the removal of misallocation as we assume in our counterfactual
experiments, then equation (7) shows that the number of firms is independent of the extent of
misallocation when productivity is exogenous and constant.

To determine an incumbent’s optimal increase in productivity in any given period, we note that
an increase in productivity in the current period affects an incumbent’s operating profits and
cost of investment in all future periods. Denoting all future increases in productivity by (1 + g 0 )
and taking into account that (1 − τ ) is equal to (sz)−γ , the value of an incumbent firm is;

V (s−1 , z) = π−1 (1 + g)σ(1−γ)−1 · Ψ − cg (1 + g)φ Ω(s−1 , z) (8)


σ(1−γ)−1
(1 − λ)(1 + g)
−cg (1 + g 0 )φ Ω(s−1 , z) · · Ψ,
1+R

where
∞  t
X (1 − λ)(1 + g 0 )σ(1−γ)−1 1+R
Ψ≡ = .
t=0
1+R [1 + R − (1 − λ)(1 + g 0 )σ(1−γ)−1 ]

The first term in equation (8) represents the expected present value of current and future oper-
34
If productivity was assumed to grow at an exogenous rate, then equation (15) shows the number of firms
would increase with γ.

22
ating profits, the second term represents the current cost of investment in productivity growth
g, and the third term represents the expected present value of all future costs of investment in
productivity growth g 0 . The subscripts on π and s refer to values from the previous period.

Maximizing equation (8) with respect to (1 + g) and taking into account that an incumbent
makes the same choice each period (so g = g 0 ) results in the following condition after some
rearranging;
[σ(1 − γ) − 1](1 + g)σ(1−γ)−1 (1 − τ )
cg (1 + g)φ = · Θ, (9)
σ

where
1+R
Θ≡ .
φ(1 + R) − [φ + 1 − σ(1 − γ)](1 − λ)(1 + g)σ(1−γ)−1

The value of entry for a potential firm comprises the cost of entry, expected profits upon
entry given the choice of entry productivity s0 and net of the cost of entry investment, and
the present discounted value of all future net profits (operational profits net of productivity
investment costs). The value of entry can be written as:

Ve = −ce Y + E[π0 |s0 ] − cS Y sθ0 (10)

  
σ(1−γ)−1 φ σ 1−λ
+ E[π0 |s0 ](1 + g) − cg (1 + g) E[π0 |s0 ] Ψ
1−τ 1+R

or
Ve = −ce Y − cS Y sθ0 + φE[π0 |s0 ] · Θ. (11)

An entrant chooses its initial productivity s0 to maximize Ve , resulting in the following condi-
tion;
[σ(1 − γ) − 1]φE[π0 ]
cS Y sθ0 = · Θ. (12)
θ

Free entry guarantees that the value of entry Ve is zero in equilibrium, resulting in the following

23
free-entry condition;
[θ + 1 − σ(1 − γ)]φE[π0 ]
ce Y = · Θ. (13)
θ

From equation (1), the expected operating profits of an entrant are equal to;

σ(1−γ)−1
Y (σ − 1)σ−1 s0 E[z σ(1−γ)−1 ]
E[π0 ] = . (14)
wσ−1 σ σ

Using equation (2) for w, this becomes;

 
Y (1 − τ̄ ) 1 − (1 − λ)(1 + g)σ(1−γ)−1
E[π0 ] = . (15)
σλN

Using equations (9), (12), (13), and (15), we can now characterize entrant productivity, per-
period firm growth, the number of firms, and aggregate investment in productivity (as a share
of output) in a stationary equilibrium;

  θ1
ce [σ(1 − γ) − 1]
s0 = , (16)
cS [θ + 1 − σ(1 − γ)]

[σ(1 − γ) − 1](1 − τ )
(1 + g)φ+1−σ(1−γ) = · Θ, (17)
cg σ

[θ + 1 − σ(1 − γ)](1 − τ )φ[1 − (1 − λ)(1 + g)σ(1−γ)−1 ]


N= · Θ, (18)
λce σθ

λN cS sθ0 + (1 − λ)cg (1 + g)φ (19)


 
[σ(1 − γ) − 1](1 − τ̄ ) φξ σ(1−γ)−1
= ·Θ· + (1 − λ)(1 + g) ,
σ θ

where ξ = [1 − (1 − λ)(1 + g)σ(1−γ)−1 ].35


35
We note that since population is constant and normalized to one, the average employment size of firms is
simply the inverse of the number of firms.

24
3.3 Comparative Statics

We are interested in the equilibrium response to changes in the productivity elasticity of dis-
tortions γ. Equations (16) through (18) imply that entrant productivity s0 , life-cycle growth g,
and average firm size 1/N are all decreasing in the productivity elasticity of distortions (γ).36
To understand the impact on productivity, note that if the productivity elasticity of distortions
γ increases, the return to investing in productivity decreases as a given investment results in
a larger distortion. As a result, firms enter with lower productivity s0 and their productivity
grows less over their lifecycle g.

To understand the impact of γ on the number of firms N and hence on firm size 1/N , we
note that operating profits of incumbents in equation (7) indicate that, for a given number of
firms, lower average productivity induced by lower s0 and g is exactly offset by a lower wage,
that is average operating profits across incumbents does not depend on γ holding N constant.
However, lower investment in productivity (both entry and life-cycle investment) induced by a
higher γ increases the value of entry which encourages more entry. In addition to the effect on
entry through lower investment, a higher productivity elasticity of distortions also has a direct
effect on entry. Under the counterfactual we consider, in which we increase γ while keeping
the average distortion constant, an increase in γ does not change the average distortion in
the economy (by construction) but it reduces the distortions faced by low productivity firms
relative to those faced by high productivity firms. In this context, and since entrants are less
productive than incumbents in expectation, a higher γ increases the relative operating profits
of entrants and thus encourages entry.37 Both the direct and indirect effects of an increased γ
on entry imply a lower average firm size 1/N .
36
Appendix C provides proofs of these and other comparative statics.
37
The positive impact of higher operating profits for entrants on the value of entry generated by increasing γ
is somewhat offset by the higher tax that an entrant can expect later in its life cycle. But with positive interest
rates, the positive effect of increased operating profits early in the life cycle dominate and the value of entry
increases. This higher value of entry encourages entry in equilibrium. Fattal-Jaef (2015) makes this point in
the context of a model with exogenous life-cycle growth.

25
The share of aggregate investment on output in equation (19) may exhibit a non-monotonic
relationship with respect to γ because of the differential impact of γ on investment and output.
We note however that even if the investment share increases with γ, productivity s0 and g
always fall with increases in γ.

Equations (18) and (19) also indicate that both entry and investment are decreasing in τ , and
that the number of firms is decreasing in the cost of entry ce . These are common features of
models with free entry, and reinforce the point that many of the policies often emphasized to
rationalize low productivity in poor countries would tend to increase the average size of firms
not reduce it as documented in our newly constructed data set in Section 2.

3.4 Decomposition of Aggregate Output and Firm Size

The impact of correlated distortions γ on aggregate output can be decomposed in our tractable
framework into effects working through the entry-investment channel, which is the focus of our
paper; the firm life-cycle investment channel, analyzed in Hsieh and Klenow (2014); and the
factor misallocation channel, which is the focus of much of the earlier literature on misallocation
such as Restuccia and Rogerson (2008) and Hsieh and Klenow (2009).38 To disentangle the
impact of correlated distortions γ on output and firm size working through the various channels,
we characterize aggregate output and average firm size separately for economies with fixed
productivity and no investment so as to isolate the impact of factor misallocation; exogenous life-
cycle growth and no entry investment to assess the additional impact of exogenous firm growth;
endogenous life-cycle productivity growth and no entry investment to assess the additional
impact of endogenous firm growth; and both endogenous life-cycle productivity growth and
entry productivity investment (our baseline model). We then compare aggregate output and
firm size in each case. Below we use gU S to denote the growth rate of firm productivity in the
benchmark economy of our baseline model, while g depends endogenously on γ.
38
We continue to consider the counterfactual exercise wherein γ is increased while τ is held constant.

26
We start with the model with just the factor misallocation channel. In this case, firm produc-
tivities are fixed and all investment is shut down. As discussed earlier for this case, the number
of firms is unaffected by changes in γ (see equation 7). Let ẑi denote the productivity of a
firm i in the stationary distribution of the benchmark economy in the baseline model. Then
aggregate output in this case is;

 R σ
 σ−1
1 N (σ−1)(1−γ)
N 0
ẑi di
YF M = ∆ · RN σ(1−γ)−1
, (20)
1
N 0
ẑi di

where F M refers to ‘factor misallocation’, and ∆ is a constant independent of γ. When firm


productivities are fixed and exogenous, then equation (20) represents the impact on aggregate
output through the factor misallocation channel analyzed by Restuccia and Rogerson (2008) and
Hsieh and Klenow (2009).39 To make a more straightforward mapping with the following three
versions of the model that feature firm growth, we can write firm productivity as the product of
a fixed component and a life-cycle component, that is if the productivity of each firm is growing
at an exogenous rate gU S , then the productivity of an a-year old firm is ẑi = zi · (1 + gU S )a .
Then we can re-write aggregate output in equation (20) as follows;40

RN σ
! σ−1
1 (σ−1)(1−γ)
N 0
zi di 1 − (1 − λ)(1 + gU S )σ(1−γ)−1
YF M = ∆ · (σ−1)(1−γ) R N σ(1−γ)−1 . (21)
1 − (1 − λ)(1 + gU S ) 1
zi di
N 0

We now extend the the above ‘factor misallocation’ economy to allow for exogenous productivity
growth over the life cycle of firms and assess how firm growth affects aggregate output and firm
39
Note however that whereas Restuccia and Rogerson (2008) assume the same productivity distribution
across economies, the gains from reallocation in Hsieh and Klenow (2009) are computed for a given productivity
distribution in each country.
40
We note that because the exit rate is independent of age, the expression for output does not depend on the
age distribution of firms. Note also that ∆ is not generally constant with firm growth but in this expression
for output we keep it constant to the number of firms in the benchmark economy for comparability to equation
(20).

27
size in conjunction with factor misallocation. In this case, aggregate output is given by;41

YF M 1
σ−1
YXG = · NXG , (22)

and average firm size is equal to;

−1 1 + R − (1 − λ)(1 + gU S )σ(1−γ)−1
NXG ∝ (ξU S · ΨU S )−1 ∝ . (23)
1 − (1 − λ)(1 + gU S )σ(1−γ)−1

When life-cycle growth is exogenous and firms cannot invest in productivity, then the factor
misallocation caused by correlated distortions has the same impact on aggregate output as in
equation (21). But this impact is at least somewhat offset by an increase in the number of firms
NXG (as discussed in Section 3.3 above).

We now extend the above ‘exogenous firm growth’ economy to allow for endogenous investment
in productivity over a firm’s life cycle, as in Hsieh and Klenow (2014). We refer to this economy
as ‘endogenous firm growth’ and in this case aggregate output is;

RN σ
! σ−1
1 (σ−1)(1−γ)
1
σ−1 N 0
zi di 1 − (1 − λ)(1 + g)σ(1−γ)−1
YN G = ·NN G · R N σ(1−γ)−1 , (24)
1 − (1 − λ)(1 + g)(σ−1)(1−γ) 1
zi di
N 0

and average firm size is;

φ(1 + R) − [φ + 1 − σ(1 − γ)](1 − λ)(1 + g)σ(1−γ)−1


NN−1G ∝ (ξ · Θ)−1 ∝ . (25)
1 − (1 − λ)(1 + g)σ(1−γ)−1

Note that the expression for output is identical to equation (22), except that life-cycle pro-
ductivity g is now endogenous and decreasing in γ. The numerator in the expression for firm
size also differs from that in equation (23). Relative to an economy with exogenous life-cycle
growth, the lower average productivity induced by higher γ magnifies the impact of γ on ag-
gregate output. But at the same time, the impact of γ on aggregate output is also dampened
41
We solve explicitly for the expressions of output and firm size in Appendix D.

28
in three ways. First, by compressing the productivity distribution, a lower g reduces the im-
pact of factor misallocation.42 Second, lower growth implies that the average productivity of
entrants is closer to the average productivity of incumbents. This matters because firm profits
(which depend positively on productivity) increase over the life of a firm. For a given level of
average profits over the life cycle, a positive discount rate implies that the discounted value of
expected lifetime profits is decreasing in g. By lowering g, a higher γ therefore induces more
entry and increases the number of firms NN G , which increases aggregate output. Third, the
numerator in equation (25) accounts for the fact that a higher γ reduces firm-level investment
in productivity without increasing the average tax burden of firms. For a given number of firms
this lower investment increases the value of entry, thus encouraging entry further increasing
the number of firms.43

Our ‘baseline model’ allows for productivity investment upon entry in addition to life cycle
productivity investment. With entry productivity investment, aggregate output can now be
expressed as;
 1
 σ−1
NBM
YBM = YN G · · s0 , (26)
NN G

and average firm size is;


−1
NBM = NN−1G · [θ + 1 − σ(1 − γ)]−1 , (27)

where s0 is decreasing in γ, as discussed in Section 3.3. The drop in entrant productivity


from a higher γ reduces aggregate output, magnifying the impact of correlated distortions. As
with life-cycle investment, this magnification in aggregate output is somewhat mitigated by
an increase in the number of firms implied by equation (27). But in this case, the additional
increase in the number of firms is driven solely by lower investment in initial productivity. Lower
entrant productivity does not compress the productivity distribution or change the distortion
42
See Appendix D for a proof of this proposition.
43
A more standard tax on investment would be an increase in cS in the cost function for entrant productivity.
Equations (16) and (18) show that such a tax would lower average productivity while leaving the number of
firms unchanged. This is because the lower investment by firms would be exactly offset by the higher tax burden
they face, leaving the value of entry unchanged.

29
faced by entrants relative to incumbents which are additional mitigating forces in the life-cycle
investment channel.44 This suggests that the entry-investment channel has the potential to
account for a more substantial portion of the impact of correlated distortions than the life-cycle
investment channel.

In summary, the decomposition above offers three main points. First, accounting for exogenous
productivity growth over the life cycle of firms reduces the impact of correlated distortions
relative to the static impact of factor misallocation emphasized in the early misallocation liter-
ature. Second, allowing for endogenous firm-level productivity growth can increase the impact
of correlated distortions, but this increase is offset through various mechanisms involving a com-
pression of the productivity distribution and increases in the number of firms. Third, extending
the model to allow for endogenous entrant productivity has the potential to greatly amplify the
impact of correlated distortions. In Section 4 we use the above expressions to decompose our
quantitative results and quantify the relative importance of each of these channels.

Even though we did not include a separate term for aggregate productivity in our model, it is
worth discussing how variation in aggregate productivity affects outcomes as poor countries may
be characterized by having policies and institutions other than correlated distortions that may
impact aggregate productivity. Imagine that an entrant-i’s productivity is equal to s0 ẑi A, where
zi = ẑi A and A is common to all firms. The absence of z in equations (16) through (19) implies
that cross-country differences in A do not generate differences in investment, firm productivity
growth, or the number of firms in our framework. The only variables of interest affected by A
are aggregate output and the real wage rate. This is the result of our assumptions that the
costs of entry and investment scale up with development. An increase in A increases these
costs and the operating profits of firms proportionately, leaving all firm decisions unchanged.
If entry and investment costs were constant and independent of development, as in for example
Bhattacharya et al. (2013), then differences in A would affect operating profits but not entry
44
Lower entrant productivity simply shifts the entire productivity distribution, leaving the relative produc-
tivities and relative distortions of entrants to incumbents unchanged.

30
or investment costs, leading to differences in life-cycle growth and entry. In particular, higher
aggregate productivity would lead to the counterfactual prediction of more entry and smaller
firms, which is inconsistent with our findings in Section 2. Hence, we conclude that the effects
of correlated distortions on entrant productivity, life-cycle growth, and factor misallocation
emphasized in our model are separate from other policies and institutions that may contribute
to lower capital accumulation and aggregate TFP in poor countries.

Note that if greater misallocation was generated simply by more dispersion in random idiosyn-
cratic distortions (uncorrelated with productivity), then equation (17) implies that life-cycle
investment and productivity growth would remain unaffected. Similarly, equations (16) and
(18) imply that entrant investment and the number of firms would not be affected. This is the
case because random distortions affect neither the average distortion τ nor the marginal return
to investment. This echoes the finding of Restuccia and Rogerson (2008) that simple random
dispersion in idiosyncratic distortions cannot explain much variation in aggregate TFP, and
that the strength of correlated distortions (γ in this paper) is what generates the large poten-
tial impact from factor misallocation. While the subsequent literature has shown that Restuccia
and Rogerson’s finding may not hold for all distributions of productivity and distortions, the
importance of correlated distortions for the investment channels is unambiguous, in the context
of our model, only correlated distortions (not random distortions) reduce the marginal benefit
of investing in productivity, thereby reducing productivity and decreasing average employment
across all firms.

4 Quantitative Analysis

In this section, we calibrate the model to U.S. data and show the quantitative implications for
establishment size, productivity, and aggregate output of hypothetical variations in the degree
of correlated distortions across countries. We decompose the total effect on aggregate output

31
that arises through the different channels in the model: entry investment, life-cycle growth, and
factor misallocation. We then use establishment-level data to estimate empirically the extent
of correlated distortions across countries and their implications for cross-country variations in
establishment size, initial productivity, life-cycle growth, and output. We end the section with
a discussion of these results for variations in the model setup as well as some robustness checks
on parameter values.

4.1 Calibration

We calibrate the model to manufacturing data for the United States in order to quantify the
cross-country effects of correlated distortions on average establishment size, productivity, and
aggregate output. The effects of distortions working through the investment channel depend
on five key parameters in our model:

• θ: the elasticity of entrant investment with respect to initial productivity,

• φ: the elasticity of life-cycle investment with respect to life-cycle productivity growth,

• cg : a level parameter in the cost of life-cycle investment,

• γ: the elasticity of distortions with respect to productivity,

• σ: the elasticity of substitution between varieties.

In order to keep a close tie with the literature for comparison, we follow Hsieh and Klenow (2009,
2014) in setting σ = 3. For U.S. manufacturing, Hsieh and Klenow (2014) report γU S = 0.09.
To obtain values for φ and cg , the convexity and level parameters from the life-cycle growth
investment function, we target an average rate of productivity growth for U.S. firms of 5 percent
from Hsieh and Klenow (2014) and an elasticity of output with respect to R&D equal to 0.028

32
from Hall et al. (2010).45 In our model, the elasticity of productivity with respect to life-cycle
investment is [σ(1 − γ) − 1]/φ. Using the values for σ and γU S discussed above, we obtain a
value for φ equal to 61.8. We then obtain a value for cg equal to 0.005 from equation (9).

The productivity elasticity of initial investment, θ, plays a prominent role in determining the
aggregate share of output invested in productivity, along with φ, cg , the exit rate λ, the real
interest rate R, and the average level of distortions τ .46 Given values for λ, R, and τ , we
choose a value for θ to match the share of value added invested in intangible capital estimated
by McGrattan and Prescott (2010), equal to 0.135. We set λ and R equal to 0.1 and 0.05,
standard values in the literature. Our value for τ is taken from the World Bank’s Doing
Business Surveys, which reports an average tax rate of 9 percent.47 Given each of these values,
we use equation (17) to back out a value for θ of 2.53. We note that this value for θ is relatively
close to the value of 2.01 estimated using trade data in Rubini (2014).

The effect of distortions working through the factor misallocation channel depends on the
parameters above, as well as on the cross-sectional distribution of productivity in our benchmark
economy. In the model, given our parsimonious representation of correlated distortions, there
is a simple mapping between productivity sz and employment, such that the demand for labor
of establishment i relative to that of establishment j is;

 σ(1−γ)−1
`i si zi
= .
`j sj zj

Given values for σ and γU S , we therefore use the above mapping to back out a distribution for sz
45
Hall et al. (2010, Table 2b) survey several studies estimating within-firm R&D elasticities, and report a
median elasticity of 0.028 across studies using recent data (post-1990). We do not include estimates from
cross-sectional studies as they generally neglect to control for industry.
46
Note that while entrant productivity s0 depends on cS (the scale parameter in the initial-investment cost
function), equation (19) shows that the aggregate share of output invested in productivity is independent of
cS . Since all entry and investment costs scale up with aggregate output, any change in s0 driven by a change
in cS affects these costs and the operating profits of establishments proportionately. As a result, the entire
adjustment to a change in cS will be through s0 .
47
The World Bank calculates an average tax rate on revenue, makes an assumption about markups, then
reports the implied tax rate on profits. We back out our average tax on revenue by reversing these steps.

33
using data from the U.S. Census Bureau for the employment distribution of U.S. manufacturing
establishments. Once we adjust the data to be representative of both paid and unpaid workers,
we obtain a distribution of persons engaged per establishment ranging from 1 to 3,000 persons,
and back out a distribution for productivity using σ = 3 and γU S = 0.09.48

Using the above distribution and our calibrated parameter values, we quantify how average
establishment size, productivity, investment, and aggregate output per worker (TFP) change
with the extent of correlated distortions, i.e., when γ is increased above the U.S. level (keeping
other parameters constant). We follow Hsieh and Klenow (2009) in maintaining a constant
average distortion τ as γ is increased. We report the results of this exercise in Table 2. The main
finding is that the model implies large variations in average establishment size, productivity,
and output per worker across economies with different correlated distortions.49

Table 2: Model Results across Correlated Distortions γ

γ Establishment Entrant Life-Cycle Investment Relative


Size Productivity Growth (%) Share (%) Output
0.09 (γU S ) 22 1 5.0 13.5 1
0.15 12 0.88 4.6 13.3 0.97
0.2 8.4 0.80 4.3 12.7 0.93
0.3 5.3 0.66 3.7 10.8 0.83
0.4 3.8 0.54 3.0 8.3 0.66
0.5 (γIndia ) 3.0 0.42 2.1 5.4 0.47
0.6 2.4 0.28 0.5 2.3 0.26
Notes: Columns report equilibrium values of average establishment size (1/N ), entrant productivity (s0 ),
life-cycle productivity growth (g), share of aggregate output invested in productivity, and aggregate output
(Y ). Results in columns 2 and 5 are reported relative to the benchmark U.S. economy.

For instance, an economy with γ = 0.4 features an average establishment size that is 17 percent
48
We transform the employment distribution into a distribution of persons engaged by assuming non-employer
establishments employ between 1 and 3 persons, while employer establishments with 1 to 4 employees are
assumed to employ 3 to 5 persons. The employment data contain 10 employment ranges, and we assume
establishment-level productivity is uniformly distributed within each range. The last range is open-ended (at
least 1,000 employees), so we choose an upper bound of 3,000 to match an average employment size of 2,000
employees.
49
Table 2 reports the impact of increasing γ as high as 0.6. Equation (1) makes it clear that γ must be less
than (σ − 1)/σ to ensure that the after-tax profitability of establishments remains an increasing function of
establishment productivity.

34
of that in the United States. This economy also features an entry-level productivity that is 54
percent of the benchmark due to lower initial investment in productivity. As a result, output
per capita is 66 percent of the benchmark economy. These are large differences in size and
productivity compared to the findings in the broad literature on misallocation. The γ’s in
Table 2 are hypothetical, but the range is plausible. As a point of reference, consider that
Hsieh and Klenow (2014) report γ = 0.5 for India using a large micro dataset of manufacturing
plants. Given this value for γ, the model predicts an average establishment size of 3 workers,
close to the value of 3.1 workers found in the data, and a growth rate of establishment-level
productivity of 2.1 percent, also very close to that reported for India by Hsieh and Klenow
(2014). The model also predicts India should have an aggregate output of about 47 percent of
the U.S. level, generating a factor-difference in output about 40 percent higher than that found
from static factor misallocation only between the U.S. and India in Hsieh and Klenow (2009).

It is important to note that the effects of correlated distortions on establishment size and pro-
ductivity work solely through the entry and life-cycle investment channels, while the impact on
aggregate output works also through the factor misallocation channel. In Table 3 we decompose
the impact on aggregate output into the effects working through each channel. The numbers in
each column are derived from the corresponding expressions for aggregate output in Section 3.4
(equations 21, 22, 24, and 26). The first column reports the impact of correlated distortions on
output solely through factor misallocation. The second column shows that the impact of factor
misallocation on aggregate output is offset by increased entry with exogenous life-cycle pro-
ductivity growth. Allowing for endogenous life-cycle productivity growth (column 3) increases
the implied impact of distortions (relative to the model with exogenous firm growth), but the
net impact on output is still weaker than that calculated for the factor misallocation economy
without life-cycle growth in column 1. The impact of γ in the baseline model is substantially
larger because of the entry productivity investment channel. In particular, consider the impact
on aggregate output of increasing γ from the U.S. level (0.09) to that of India (0.5). Factor
misallocation reduces output to 63% of the U.S. economy, while adding exogenous life-cycle

35
growth reduces it to 91% which implies that exogenous firm growth increases output by a
factor 1.44 (0.91/0.63), partially compensating the reduction through factor misallocation. Al-
lowing for endogenous growth, whereby firms in India would growth slower than in the United
States, reduces output to 70% of the U.S. economy, still implying a net increase in output
due to endogenous firm growth by a factor 1.11 (0.70/0.63). Allowing for entry productivity
investment (our baseline model) generates a strong reduction in aggregate output to 47% of the
U.S. economy. The entry productivity channel reduces output by a factor of 0.67 (0.47/0.70),
roughly doubling the contribution of factor misallocation to the reduction in aggregate output
in this economy.

Table 3: Decomposition: γ and Aggregate Output

γ Factor + Exogenous + Endogenous + Entry


Misallocation Life-Cycle Life-Cycle Productivity
Growth Investment Investment
0.09 (γU S ) 1 1 1 1
0.15 0.98 1.10 0.99 0.97
0.2 0.94 1.13 0.97 0.93
0.3 0.86 1.13 0.93 0.83
0.4 0.74 1.02 0.82 0.66
0.5 (γIndia ) 0.63 0.91 0.70 0.47
0.6 0.52 0.77 0.54 0.26
Notes: Columns 1 through 4 report the cumulative impact on aggregate output (Y ) of distortions (higher
γ’s) working through the factor misallocation channel, through entry when life-cycle growth is exogenous,
through the life-cycle investment channel, and through the entry productivity channel in the baseline
model. All results are relative to the benchmark U.S. economy.

Note that the impact of distortions working through the static factor misallocation channel for
India is the same as that estimated for India by Hsieh and Klenow (2009), who use compre-
hensive micro data to back out distributions of distortions and productivity. We interpret this
finding as suggestive that our parsimonious representation of idiosyncratic distortions through
γ (the elasticity of distortions with respect to productivity) works extremely well as a summary
measure of empirical distortions (actual wedges in the data).

36
Table 4: Decomposition: γ and Establishment Size

γ Factor + Exogenous + Endogenous + Entry


Misallocation Life-Cycle Life-Cycle Productivity
Growth Investment Investment
0.09 (γU S ) 22 22 22 22
0.15 22 17 14 12
0.2 22 15 12 8.4
0.3 22 13 9.4 5.3
0.4 22 11 8.2 3.8
0.5 (γIndia ) 22 11 7.5 3.0
0.6 22 9.9 7.1 2.4
Notes: Columns 1 through 4 report the cumulative impact on average establishment size (1/N ) of distor-
tions (higher γ’s) working through the factor misallocation channel, through entry when life-cycle growth
is exogenous, through the life-cycle investment channel, and through the entry productivity channel in the
baseline model. All results are relative to the benchmark U.S. economy.

Table 4 reports the contribution of each channel to the total effect of γ on average establishment
size. Again taking India as an example, increasing γ from 0.09 to 0.5 results in a substantial
drop in average establishment size in a model with exogenous growth in life-cycle productivity
but no investment. Allowing for life-cycle investment amplifies this effect moderately, while
extending the model to include investment at entry substantially increases the impact of γ on
average size. The combined effect of these channels is to decrease establishment size by a factor
of 7.

We note that the quantitative impact of correlated distortions on aggregate output via life-
cycle growth and factor misallocation is fairly stable across different model configurations. For
instance, in our model, Gibrat’s law holds in all economies and hence correlated distortions
reduce growth for all establishments in the same proportion, whereas in Hsieh and Klenow
(2014), Gibrat’s law only holds in an economy without distortions. So in their setup, correlated
distortions reduce growth for all establishments but more so for high productivity establish-
ments, compressing the productivity distribution. While this compression in the productivity
distribution reduces the impact of factor misallocation, it generates less of a positive impact on

37
the number of establishments, leading to similar quantitative effects. To put it differently, our
decomposition of the life-cycle growth effect contain two opposing effects on aggregate output
(one working through the number of establishments and the other one through reduced factor
misallocation) that roughly offset each other in different configurations of life-cycle growth. The
implications of these configurations for average establishment size are quite different. Whereas
the life-cycle growth in Hsieh and Klenow (2014) implies modest reductions in size via changes
in γ (a reduction of about 15 percent in India relative to the United States), in our model
establishment size is reduced through the life-cycle channel by a factor of three.

An important takeaway from Tables 3 and 4 is that accounting for the growth of establishments
over their life cycle enriches our understanding of establishment dynamics and its interaction
with misallocation, but does little to amplify the overall impact of misallocation on aggre-
gate output and TFP. In contrast, accounting for investment decisions at entry approximately
doubles the impact of correlated distortions on aggregate output.

4.2 Correlated Distortions

The calibrated model shows how correlated distortions encourage smaller establishments, lower
aggregate output, and lower investment in productivity. In this section, we provide systematic
evidence that the productivity elasticity of distortions is indeed higher in poor countries. We
then provide evidence consistent with the mechanism highlighted in Section 3, using cross-
country data to show that both average establishment size and aggregate R&D investment are
decreasing in the extent of correlated distortions.

Our measure of correlated distortions is constructed using establishment-level data from the
World Bank’s Enterprise Surveys. Enterprise Surveys is an ongoing project of the World Bank
to collect establishment-level data from mostly low and middle-income countries through face-
to-face surveys. The dataset contains standardized information about sales, intermediate pur-

38
chases, inputs, and a host of other variables for establishments in over 100 countries for at least
one year since 2002. In each country, between 150 and more than 1000 establishments have been
surveyed, and efforts have been made to make these samples representative of the population
of establishments with at least five employees.50 Importantly for our purposes, manufactur-
ing establishments are classified into fifteen industries. From this dataset, we use observations
containing values for industry classification, sales, number of employees, total wage bill, and
purchases of materials and intermediate goods, for all countries which are also in our dataset
for establishment size described in Section 2.

We back out our measure of establishment-level distortions and productivity for each estab-
lishment within a country-industry following Hsieh and Klenow (2009), except that we do not
use capital data. Abstracting from capital allows us to increase the number of usable countries
substantially, as a large number of establishments in the Surveys do not report capital (more
on this below). From Section 3, labor productivity for some establishment i is;

 
Pi yi w σ 1
= ∝ ,
`i (1 − τi ) σ−1 (1 − τi )

where Pi yi is an establishment’s value added (sales minus intermediate inputs) and `i is em-
ployment.51 As in Hsieh and Klenow, we remove the constant in the above expression by using
labor productivity relative to the weighted average of labor productivity across all establish-
ments within the same industry.52 We infer an establishment’s productivity si zi by exploiting
50
Given the absence of very small establishments in the Enterprise Surveys data, we need to assume (as we
do in Section 3) that the elasticity of distortions with respect to productivity is constant.
51
Following Hsieh and Klenow (2009), we use an establishment’s total wage bill (including benefits) in our
computations instead of employment in order to control for differences in human capital across establishments.
52
More precisely, we measure the distortion faced by establsihment i as;
N
" −1 #
Pi yi X Pi0 yi0 Pi0 yi0
PN .
`i 0 `i0 i0 =1 Pi0 yi0
i =1

P 1
 σ−1
N σ−1
Productivity si zi is similarly measured relative to i0 =1 (si0 zi0 ) .

39
the following relationship;
σ
yi (Pi yi ) σ−1
si zi = ∝ .
`i `i

Following Hsieh and Klenow (2014), we then do a simple OLS regression of logged distortions
on logged productivity to obtain each country’s productivity elasticity of distortions (γ).53
Some countries have data for two or even three years, so we average elasticities over all years,
weighting by the number of observations in each year. We obtain elasticities for 93 countries,
62 of which are included in the establishment-size data from Section 2.54 Among these 62
countries, elasticities range from 0.22 to 0.74, averaging 0.52. Among all 93 countries the
average elasticity remains a close 0.51. It is reassuring to note that our computed elasticity for
India is 0.56, close to the value Hsieh and Klenow (2014) obtain using much more comprehensive
micro data. To check the sensitivity of our measures to abstracting from capital, we also
calculate elasticities using Hsieh and Klenow’s (2009) TFPR and TFPQ as our measures of
distortions and productivity. Among the 50 countries which satisfy the criteria above, the
average elasticity is 0.56. If we recalculate these elasticities abstracting from capital data (but
only using observations that report capital) we find the same average, and the correlation
between the two measures is 0.89.55

Figures 5 and 6 show how GDP per capita and average establishment size are related to the
productivity elasticity of distortions in 63 countries. The elasticity for the U.S. of 0.09 is taken
from Hsieh and Klenow (2014).56 The data show a clear link between the elasticity and both
GDP per capita and average size, consistent with the model. Our measure of each country’s
distortion elasticity is estimated from a regression, so we also have information about the
standard error associated with each country’s estimate. For robustness we perform a weighted
53
Hsieh and Klenow (2014) perform this procedure for the U.S., India, and Mexico. Before doing the regres-
sions, we first trim the 1 percent tails of both distortions and productivity for each country to remove outliers.
We then recalculate the averages as above.
54
We do not use countries with fewer than 100 observations. Over the 62 countries with size data, we use a
total of 37,410 establishment-level observations in our regressions.
55
This is consistent with Gal (2013, Table 9), who calculates both labor productivity and TFPR for firms in
a handful of OECD countries and reports correlations between the two statistics ranging from 0.8 to 0.9.
56
The regression coefficients (standard errors) in Figures 5 and 6 are -3.02 (0.80) and -1.95 (0.46).

40
50000
USA IRL

ESP

GDP per Capita (log scale)


SVN CZE
SVK
EST
HRVRUS
POL HUN LTU PAN
TTO
LVA

10000
MEX KAZ ROU TUR
URYBGR SRBMUS
ARG MKD
ZAF THA BRAUKR
COLPER ALB ECU BIH
SLV UVK
DZA MNG
GEO
PRY IDN
MDA MAR
NIC JOR
BOL
LKA
2500 PHL
KGZVNMHND
IND PSE YEM
LAO
GHA
BGD BEN
UGANPL
MDG
ETHMWI
500

0 .2 .4 .6 .8
Productivity Elasticity of Distortions

Figure 5: GDP per Capita and Correlated Distortions

least squares regression on the observations in both Figures 5 and 6, weighting each observation
by the inverse of its standard error. The resulting coefficients are very similar and remain
significant at the one percent level. The dashed lines in Figures 5 and 6 show the analogous
relationships predicted by the model. In Figure 5 the model matches the relationship between
the elasticity and GDP per capita well, although it does not capture the entire magnitude of
the differences in GDP per capita across countries. The dashed line in Figure 6 shows that
while the model comes close to predicting India’s average establishment size, it predicts average
sizes for most countries lower than those reported in the data. As discussed previously, to the
extent that output per capita and establishment size are affected by country features other than
correlated distortions (such as capital accumulation and entry costs differences) it is reassuring
that although correlated distortions are able to account for a large portion of the cross-country
patterns, it does not account for all the patterns leaving room for other plausible and relevant
explanations.

In the model developed in Section 3, the mechanism through which correlated distortions reduce
establishment size is the disincentive to invest in productivity. As a consequence, the model
also predicts the share of output invested in productivity should be lower in economies with

41
50
TTO

Establishment Size (log scale)


25
UKR
USA
BGR
BRA RUS LVA
LTUIRL ROU MKD
PHL
POL HUN KAZ GEO
MEX KGZMDA
ARG SVN HRV

10
ESP EST
MUS
ZAF THA
COLPER URY
GHA
MDGLKA
SLV NPLTUR
JORCZE
BGD ALB PAN
SRB BIH
VNM
SVK BOL DZA
PRY MNG
HNDUGA ECU
MAR
LAO
PSE IDN
4
NIC
UVK
IND YEM
2

ETHMWI
BEN
1

0 .2 .4 .6 .8
Productivity Elasticity of Distortions

Figure 6: Establishment Size and Correlated Distortions

high γ. Broad measures of investment in intangible capital have not yet been collected for a
large number of countries, but R&D intensity should provide a fair proxy. R&D is a significant
component of life-cycle investment, and Corrado et al. (2012) show that differences in life-cycle
investment (including R&D) across countries are highly correlated with proxies for entrant
investment (like early-stage venture capital investment, for example). Figure 7 shows how
establishment size across countries varies with R&D intensity, while Figure 8 shows how R&D
intensity is related to γ.57 The relationships illustrated in both figures are clearly consistent
with the predictions of the model. Again using India as a point of reference, the model predicts
an investment share in India about 40 percent of the U.S. level. In the data, India’s R&D
intensity is a relatively close 29 percent of the U.S. level. The dashed line in Figure 8 shows
that the investment share predicted by the model matches the rest of the data fairly well.
57
The regression coefficients (standard errors) in Figures 7 and 8 are 0.16 (0.04) and -4.23 (0.97). The
coefficient (standard error) from a weighted least squares regression corresponding to Figure 8 (see above) is
-3.30 (1.12). R&D data is taken from UNESCO, and is calculated as total investment in R&D as a share of
GDP, relative to the U.S.

42
MYS

50
SGP
ARE LUX
TTO DEU

Establishment Size (log scale)


MCO PRI

25
MAC UKR
CANAUT
USA
KWT BGR NLD DNK
GBR JPN
LVA FROBRA CHE
RUS IRLBEL
PHL MKD ROU LTU FRA
GEO SWEISR
KAZ
KGZ MDAPOL
ARG
MEX HKG
HRVHUNNZL SVN
10

EST KOR
BRNNPL PER GHA MUS
THA ZAF ESP
SAU LKAMDG URY
COL MLT
PRT
ITACZE AUS
JOR TUR FIN
BIH KHM DZA PAN SRB SDN TUN
ALBBOLVNM MNG SVK NOR
HND PRY BMU
ECU UGA
GRL
MAR
IRN
LAO MNE
IDN
4

NIC GRC
IND
2

ETH
1

.05 .2 1 5
R&D Intensity (%, log scale)

Figure 7: Establishment Size and R&D Intensity


5

USA
EST
IRL SVN
CZE
R&D Intensity (%, log scale)

ESP BRA RUS


HUN
1

ZAF LTU HRV


IND
UKR
POL SVK MAR TUR
LVA
MDA BGRROU
MEX ARG UGA
URY MUS
JOR
SRB
ECU MNG
THA GHA
MDG MKD
.2

VNM GEO
LKAPAN
COL
KGZ
PERKAZ ETH BOL
ALBPHL
TTO
IDN
DZA
PRY
.05

NPL
HND LAO
NIC
BIH

0 .2 .4 .6 .8
Productivity Elasticity of Distortions

Figure 8: R&D Intensity and Correlated Distortions

43
4.3 Measurement Error

It is worth considering whether and to what extent measurement error may be driving the
relationship between correlated distortions and development reported in Section 4.2. If mea-
surement error causes us to overestimate the elasticity of distortions with respect to produc-
tivity, then we might simply be picking up a negative relationship between measurement error
and development. There are two reasons why measurement error may be more prevalent in
poor countries. First, statistical agencies may be under-funded or otherwise less efficient. This
source of error should not be as important in our context, as the Enterprise Surveys data is
collected by the World Bank using a common methodology for all included countries. Second,
weaker management practices in poor countries may result in less-accurate record keeping by
firms, resulting in larger reporting errors (Bloom et al., 2012). We focus on this second source of
error by considering how measurement error is related to our measure of γ across the countries
in our sample.

Our estimate of a country’s elasticity of distortions with respect to productivity (γ) is from a
regression of each establishment’s (logged) inferred distortion on each establishment’s (logged)
inferred productivity. We cannot separate measurement error associated with estimated dis-
tortions from the true dispersion in idiosyncratic distortions. But our inferred measure of
productivity should be a combination of only an establishment’s true productivity and mea-
surement error, so we can test whether the variance of (logged) inferred productivity is higher
in countries with higher estimated γ’s. In our model, the variance of true logged productiv-
ity across all establishments is equal to the variance of log(z), and so is independent of γ. If
the variance of inferred productivity is correlated with our estimates of γ, then this can be
interpreted as evidence that measurement error may be driving our results.

One complication with this test is that the Enterprise Surveys data we use to estimate γ
only includes establishments with at least five workers. This is important because a higher γ
in our model decreases the size of establishments. This means that for a given productivity

44
distribution, the threshold productivity above which an establishment will be included in the
data should be increasing in γ. For our calibrated distribution of productivity, this leads to a
higher variance in logged productivity for included establishments.58 Looking at the data, we
find that the variance of logged productivity does indeed increase with γ. But this increase is
roughly in line with what the model predicts when only establishments with at least 5 workers
are included.59 We interpret this result as suggesting that measurement error, which we admit
may affect our average measure of γ across all countries, does not seem to be decreasing in
development and so does not seem to be driving the relationship between γ and development.

4.4 Discussion

We discuss our main results for reasonable extensions of the model and different values of key
parameters.

Model Extensions Extending the model to include capital and capital accumulation does
not change our results, as long as we interpret our baseline impact on aggregate output as an
impact on TFP. The total impact on aggregate output would be magnified in the usual way
through a change in the steady-state capital stock. Extending the model to allow entrants to
learn the exogenous portion of their productivity (z) before investing would generate a richer
relationship between γ and the productivity distribution across establishments, as the incentive
for more productive establishments to invest more than less productive establishments would be
dampened. We leave this as an interesting topic for future theoretical and empirical research.
58
We calculate the implied variance of log(z) in the model by calculating the productivity at which establish-
ments employ five workers in our benchmark U.S. economy, then calculating how this threshold increases when
average establishment size decreases due to γ, and then calculating the variance of log(z) across establishments
above each threshold. Note that these calculations do not take into account how the distribution of productivity
changes with life-cycle growth, which may dampen the implied increases in variance.
59
In fact, the model predicts that the variance in logged productivity increases with γ even faster than what
we observe in the data.

45
Robustness on θ A critical parameter determining the impact of correlated distortions on
establishment size and aggregate output is the elasticity of investment in entrant productivity
θ. Our baseline calibrated value for θ is 2.53. We explore the cross-country implications of the
model for two alternative values for θ: a 20 percent higher value than in the baseline calibration
which implies θ = 3.03, and a 20 percent lower value than the baseline which implies θ = 2.02.
We report the results of increasing correlated distortions from γU S = 0.09 to γIndia = 0.5 on
establishment size, entrant productivity, life-cycle growth, investment, and aggregate output
in Table 5. Note that the calibrated value for φ is independent of θ, and so remains the same
as in the baseline calibration. As a result, the effect of higher γ’s on life-cycle growth does
not differ from the baseline case in Table 2. Note also that since we calibrated θ to match an
aggregate share of intangible investment to output, the variations in θ amount to changes in the
aggregate share of intangible investment in the benchmark economy ranging from 12.7 to 14.7
percent (relative to our calibration target of 13.5 percent). While the quantitative magnitude
of changes in γ on the variables of interest depends sensibly on θ, overall the amplification effect
of entrant productivity on aggregate output remains quantitatively important.

Table 5: Model Results for Alternative Values of θ

γ Establishment Entrant Life-Cycle Investment Relative


Size Productivity Growth (%) Share Output
θ = 3.03
0.09 (γU S ) 22 1 5.0 12.7 1
0.5 (γIndia ) 3.9 0.53 2.1 4.8 0.52
θ = 2.02
0.09 (γU S ) 22 1 5.0 14.7 1
0.5 (γIndia ) 1.4 0.24 2.1 6.4 0.38
Notes: Columns report equilibrium values of average establishment size (1/N ), entrant productivity (s0 ),
life-cycle productivity growth (g), share of aggregate output invested in productivity, and aggregate output
(Y ). Results in columns 2 and 5 are reported relative to the benchmark U.S. economy.

Robustness on φ The elasticity of life-cycle investment with respect to productivity growth


φ is an important determinant of both life-cycle growth and establishment size. In Table 6

46
we show how our variables of interest are affected by correlated distortions when we increase
and decrease the calibrated value for φ by 20 percent. Note that each alternative value for φ
implies a different value for θ from our calibration. When φ = 74.1, θ must be lower than our
benchmark value in order to generate our target investment share. When φ = 49.4, θ must be
higher. As a result, a higher φ results in less variation in life-cycle growth, but more variation
in entrant productivity. The opposite holds for a lower value of φ. The net impact on aggregate
output is therefore close to the benchmark case.

Table 6: Model Results for Alternative Values of φ

γ Establishment Entrant Life-Cycle Investment Relative


Size Productivity Growth (%) Share Output
φ = 74.1
0.09 (γU S ) 22 1 5.0 13.5 1
0.5 (γIndia ) 1.9 0.29 2.6 5.9 0.42
φ = 49.4
0.09 (γU S ) 22 1 5.0 13.5 1
0.5 (γIndia ) 4.2 0.60 1.5 4.7 0.53
Notes: Columns report equilibrium values of average establishment size (1/N ), entrant productivity (s0 ),
life-cycle productivity growth (g), share of aggregate output invested in productivity, and aggregate output
(Y ). Results in columns 2 and 5 are reported relative to the benchmark U.S. economy.

5 Conclusion

We have assembled from hundreds of sources a unique dataset of manufacturing establishments


to document a strong positive association between average establishment size and GDP per
capita. The cross-country income elasticity of establishment size is 0.29 in our sample of 134
countries. We considered an otherwise standard model of heterogenous establishments with en-
dogenous entry and investment in establishment productivity. We showed that, in a reasonably
calibrated version of the model, cross country variation in the degree of correlation between
establishment distortions and productivity generates substantial differences in establishment

47
size consistent with our reported data and aggregate productivity. We also documented cross-
country variation in the degree of correlated distortions from micro data for a large set of
countries. Overall, the analysis in this paper puts us closer to understanding the patterns in
operational scale and productivity observed across countries.

Our model captures several mechanisms highlighted in the literature through which correlated
distortions can affect productivity and establishment size. By keeping the model tractable, we
have been able to analytically and quantitatively decompose the effects of correlated distor-
tions into those working through entry, entrant investment, life-cycle investment, and factor
misallocation. We found that accounting for life-cycle investment allows us to rationalize the
relationship between correlated distortions and lower life-cycle investment in productivity, but
does not amplify the effects of misallocation relative to those calculated in a setting without
life-cycle growth. In contrast, accounting for entrant investment substantially increases the
estimated impact of correlated distortions.

Our analysis has abstracted from many factors which may be worth exploring further. For ex-
ample, we have abstracted from different forms of entry and operation costs that seem to hinder
the operation of establishments in many poor countries. We have also abstracted from policies
and institutions that may generate differences in the productivity distribution of entrants across
countries. We leave a detailed exploration of these factors for future research.

48
References
Adamopoulos, Tasso and Diego Restuccia, “The Size Distribution of Farms and Interna-
tional Productivity Differences,” American Economic Review, 2014, 104 (6), 1667–1697.

Alfaro, Laura, Andrew Charlton, and Fabio Kanczuk, “Plant-Size Distribution and
Cross-Country Income Differences,” in “NBER International Seminar on Macroeconomics
2008” University of Chicago Press 2009, pp. 243–272.

Atkeson, Andrew and Ariel Tomás Burstein, “Innovation, Firm Dynamics, and Interna-
tional Trade,” Journal of Political Economy, 2010, 118 (3), 433–484.

Barseghyan, Levon and Riccardo DiCecio, “Entry costs, industry structure, and cross-
country income and TFP differences,” Journal of Economic Theory, 2011, 146 (5), 1828–1851.

Bello, Omar D., Juan S. Blyde, and Diego Restuccia, “Venezuela’s growth experience,”
Latin American Journal of Economics, 2011, 48 (2), 199–226.

Bhattacharya, Dhritiman, Nezih Guner, and Gustavo Ventura, “Distortions, Endoge-


nous Managerial Skills and Productivity Differences,” Review of Economic Dynamics, 2013,
16 (1), 11–25.

Bloom, Nicholas and John Van Reenen, “Why Do Management Practices Differ Across
Firms and Countries?,” Journal of Economic Perspectives, 2010, 24 (1), 203–224.

, Aprajit Mahajan, David McKenzie, and John Roberts, “Why Do firms in De-
veloping Countries Have Low Productivity?,” American Economic Review, 2010, 100 (2),
619–623.

, Benn Eifert, Aprajit Mahajan, David McKenzie, and John Roberts, “Does Man-
agement Matter? Evidence from India,” Quarterly Journal of Economics, 2013, 128 (1),
1–51.

, Raffaella Sadun, and John Van Reenen, “Management as a Technology?,” 2012.


Unpublished manuscript, London School of Economics.

Boedo, Hernan J. Moscoso and Toshihiko Mukoyama, “Evaluating the Effects of Entry
Regulations and Firing Costs on International Income Differences,” Journal of Economic
Growth, 2012, 17 (2), 143–170.

Bollard, Albert, Peter J Klenow, and Huiyu Li, “Entry Costs Rise with Development,”
2014. Unpublished manuscript, Stanford University.

Corrado, Carol, Jonathan Haskel, Cecilia Jona-Lasinio, and Massimiliano Iommi,


Intangible Capital and Growth in Advanced Economies: Measurement Methods and Compar-
ative Results, IZA, 2012.

Da-Rocha, José-Marıa, Marina Mendes Tavares, and Diego Restuccia, “Inaction,


Size Dependent Policies and Aggregate Productivity with Endogenous Establishment-Level
Productivity,” 2015. Working Paper, University of Toronto.

49
Desmet, Klaus and Stephen L Parente, “The Evolution of Markets and the Revolution of
Industry: A Unified Theory of Growth,” Journal of Economic Growth, 2012, 17 (3), 205–234.

Duarte, Margarida and Diego Restuccia, “The Role of the Structural Transformation in
Aggregate Productivity,” Quarterly Journal of Economics, 2010, 125 (1), 129–173.

Easterly, William and Sergio Rebelo, “Fiscal Policy and Economic Growth,” Journal of
Monetary Economics, 1993, 32, 417–458.

Fattal-Jaef, Roberto N., “Entry, Exit and Misallocation Frictions,” 2015. Unpublished
manuscript, World Bank.

Gabler, A. and M. Poschke, “Experimentation by Firms, Distortions, and Aggregate Pro-


ductivity,” Review of Economic Dynamics, 2013, 16 (1), 26–38.

Gal, Peter N, “Measuring Total Factor Productivity at the Firm Level Using OECD-ORBIS,”
2013. Working Paper, World Bank.

Guner, N., G. Ventura, and Y. Xu, “Macroeconomic Implications of Size-Dependent


Policies,” Review of Economic Dynamics, 2008, 11 (4), 721–744.

Hall, Bronwyn H., Jacques Mairesse, and Pierre Mohnen, “Measuring the Returns to
R&D,” Handbook of the Economics of Innovation, 2010, 2, 1033–1082.

Herrendorf, Berthold M., Richard Rogerson, and Ákos Valentinyi, “Growth and
Structural Transformation,” Handbook of Economic Growth, 2014, 2, 855–941.

Hopenhayn, Hugo A., “Entry, Exit, and Firm Dynamics in Long Run Equilibrium,” Econo-
metrica, 1992, 60, 1127–50.

Hopenhayn, Hugo A, “Firms, Misallocation, and Aggregate Productivity: A Review,” An-


nual Review of Economics, 2014, 6 (1), 735–770.

Hopenhayn, Hugo A. and Richard Rogerson, “Job Turnover and Policy Evaluation: A
General Equilibrium Analysis,” Journal of Political Economy, 1993, pp. 915–938.

Hsieh, Chang-Tai and Benjamin Olken, “The Missing” Missing Middle”,” Journal of
Economic Perspectives, 2014, 28 (3), 89–108.

and Peter J. Klenow, “Misallocation and Manufacturing TFP in China and India,”
Quarterly Journal of Economics, November 2009, 124 (4), 1403–1448.

and , “The Life Cycle of Plants in India and Mexico,” Quarterly Journal of Economics,
August 2014, 129 (3), 1035–1084.

Jones, Charles I, “The Facts of Economic Growth,” Technical Report, National Bureau of
Economic Research 2015.

Jr., Robert E. Lucas, “On the Size Distribution of Business Firms,” Bell Journal of Eco-
nomics, 1978, pp. 508–523.

50
Klenow, Peter J. and Andreas Rodriguez-Clare, “The Neoclassical Revival in Growth
Economics: Has it Gone Too Far?,” in B. Bernanke and J. Rotemberg, eds., NBER Macroe-
conomics Annual, Cambridge: MIT Press, 1997.

McGrattan, Ellen R. and Edward C. Prescott, “Technology Capital and the US Current
Account,” American Economic Review, 2010, 100, 1493–1522.

Melitz, Marc J. and Giancarlo I.P. Ottaviano, “Market Size, Trade, and Productivity,”
Review of Economic Studies, 2008, 75 (1), 295–316.

Moreira, Sara, “Firm Dynamics, Persistent Effects of Entry Conditions, and Business Cycles,”
2015. Unpublished manuscript, University of Chicago.

Pagés, Carmen, The Age of Productivity: Transforming Economies from the Bottom Up,
Palgrave Macmillan, 2010.

Peters, Michael, “Heterogeneous mark-ups, growth and endogenous misallocation,” Technical


Report, London School of Economics and Political Science, LSE Library 2013.

Poschke, Markus, “The Firm Size Distribution Across Countries and Skill-Biased Change in
Entrepreneurial Technology,” 2014. Discussion Paper, IZA.

Prescott, Edward C., “Needed: A Theory of Total Factor Productivity,” International Eco-
nomic Review, 1998, 39, 525–52.

Ranasinghe, Ashantha, “Impact of Policy Distortions on Firm-Level Innovation, Productiv-


ity Dynamics and TFP,” Journal of Economic Dynamics and Control, 2014, 46, 114–129.

Restuccia, Diego, “Factor Misallocation and Development,” in Online Edition, ed., The New
Palgrave Dictionary of Economics, Palgrave Macmillan: Steven N. Durlauf and Lawrence E.
Blume, 2013.

, “The Latin American Development Problem: An Interpretation,” Economı́a, 2013, 13 (2),


69–100.

and Richard Rogerson, “Policy Distortions and Aggregate Productivity with Heteroge-
neous Establishments,” Review of Economic Dynamics, 2008, 11 (4), 707–720.

and , “Misallocation and Productivity,” Review of Economic Dynamics, 2013, 16 (1),


1–10.

Rubini, Loris, “Innovation and the Trade Elasticity,” Journal of Monetary Economics, 2014,
66, 32–46.

Tybout, James R, “Manufacturing Firms in Developing Countries: How Well Do They Do,
and Why?,” Journal of Economic Literature, 2000, 38 (1), 11–44.

Ziebarth, Nicolas L., “Are China and India Backward? Evidence from the 19th Century US
Census of Manufactures,” Review of Economic Dynamics, 2013, 16 (1), 86–99.

51
On-line Appendix (Not for Publication)

A Establishment Size Data

We describe in more detail how we construct the establishment size data for the manufacturing
sector. Our standardized definition of establishment size is the number of persons engaged per
establishment. Persons engaged is defined as the average number of persons working for an
establishment, both paid and unpaid. A manufacturing establishment is defined as a physi-
cal location where the primary activity is manufacturing. Establishments include households
who have signs posted on the property indicating commercial activity. Not all countries report
persons engaged or the number of establishments, so we also use data on the number of paid
employees, the number of full-time equivalent employees, and the number of firms (collections
of one or more establishments under common ownership) to impute persons engaged and estab-
lishments for these countries. We explain in detail the exact procedure for these imputations
but we note that imputations are only involved in about one quarter of our sample of countries.

The source data for each country is from economic censuses, as well as surveys which use
comprehensive business registries to create sampling frames and as a result are representative
of the population of establishments.60 We use all publicly available data for the years 2000
through 2012.61 In an effort to maintain consistency across countries, we do not use data unless
efforts were made by a statistical agency to make the data representative of an economy’s
entire population of manufacturing establishments. We exclude any data collected without
accounting for small establishments, except in cases where only establishments without paid
employees are excluded. In the later case, we use U.S. data to adjust measured establishment
size (this is the case for eight countries). Further, we include data for any country that excludes
60
For some countries data is from EUROSTAT or OECD’s Structural Business Statistics, but we check each
country’s methodology to confirm the consistency of definitions.
61
In some cases countries have published only press releases or bulletins describing the census data. We
include these countries when the data meets our criteria.

52
establishments with low revenue, as long as the revenue threshold is lower than the country’s
GDP per capita (this is the case for four countries). Two countries (Algeria and Honduras)
do not report employment, but do report the distribution of establishments across multiple
employment tranches. In these two cases we estimate total employment by using an average
employment within each tranche consistent with data in comparable countries.62 We are left
with 134 countries with useable data for at least one year, with an average of six years per
country.63 Table 7 reports the total number of countries reporting each variable for at least
one year, as well as the total number of poor countries and the total number of rich countries
(defined as having GDP per capita below and above the median) doing the same.

Table 7: Sample of Countries

Total Number Number of Number of


Variable of Countries Poor Countries Rich Countries
persons engaged 101 54 47
employees 86 34 52
engaged and employees 53 21 32
full-time equivalents 25 2 23
establishments 83 45 38
firms 67 26 41
establishments and firms 16 4 12

Note: ‘Poor’ and ‘Rich’ refer to countries with GDP per capita below and above the median.
Data from multiple sources, see text for details.

We construct our standardized measure of persons engaged per establishment as follows. First,
the total number of persons engaged is reported for 101 countries. For the remaining 33
countries, we impute persons engaged based on each country’s reported data for the number of
employees and/or the number of full-time equivalent employment. We estimate the relationship
between persons engaged and employment from a regression of persons engaged on employees
and/or full-time equivalent employment using country-year data for the more than 50 countries
62
We assume average employment within a tranche to be one third of the distance from the lower to the
upper threshold. For the last open-ended tranche (for example, 200 or more employees) we assume an average
employment equal to twice the lower threshold.
63
Although size data is also available for Norfolk Island, it has been dropped for lack of any reliable measure
of GDP per capita.

53
that report all these variables. We then multiply the estimated coefficients by the reported
country-year data to obtain persons engaged for those countries. Hence, this first step produces
a number for the total persons engaged for each country-year in our sample. Second, we
compute persons engaged per establishment (83 countries) and persons engaged per firm (67
countries). This allowed us to estimate the coefficient from a regression of persons engaged per
establishment on persons engaged per firm for country-years that report both and then use the
estimated coefficient to impute persons engaged per establishment using the data of countries
that only report persons engaged per firm.

Each of the above regressions use all country-years which report the relevant variables. The
results of the four regressions described above are;

• persons engaged = 1.44 · employees − 0.40 · full-time equivalents

• persons engaged = 1.07 · employees

• persons engaged = 1.12 · full-time equivalents

• persons engaged per establishment = 0.89 · persons engaged per firm

There is a small number of countries for which the data exclude non-employer establishments
or that report a combination of manufacturing, extraction, and energy instead of just manufac-
turing. For these countries we do the following. To adjust persons engaged per establishment
in countries which exclude non-employer establishments (this is the case for eight countries),
we multiply these values by a factor equal to the average ratio of persons engaged per establish-
ment to persons engaged per establishment with paid employees across all years in the U.S. data
(this ratio is 0.51 in U.S. data). We similarly standardize persons engaged per establishment
for manufacturing for five countries which report statistics for a combination of manufacturing,
extraction, and energy using U.S. data for the ratio of persons engaged per establishment in
manufacturing relative to manufacturing, extraction and energy (this ratio is 1.14 in U.S. data).

54
In our final dataset, the resulting measures of persons engaged per establishment are averaged
over all years for each of the 134 countries.

Table 8 lists each country in the final dataset, the number of years for which data is available,
and the sources from which data has been collected.

55
Table 8: List of Countries and Sources

Country Code Years Sources


Âland Islands ALA 9 Statistics and Research Âland: Statistical Yearbooks of Âland
2006-2010 and 2013, and www.asub.ax
Albania ALB 8 Instituti i Statistikave: www.instat.gov.al/en/figures/statistical-
databases.aspx
Algeria DZA 1 Office National des Statistiques, Alger: Premier recensement
économique -2011- Résultats définitifs
American Samoa ASM 2 U.S. Census Bureau: 2002, 2007 County Business Patterns, and
2002, 2007 Nonemployer Statistics
Andorra AND 12 Departament d’Estadı́stica: 2010 Statistical Yearbook, and
www.estadistica.ad
Argentina ARG 1 Instituto Nacional de Estadı́stica y Censos: 2005 Economic Census
Aruba ABW 1 Central Bureau of Statistics: Business Count 2003
Australia AUS 5 Australian Bureau of Statistics: Counts of Australian Businesses
2003-2007, Labour Force Surveys (Quarterly)
Austria AUT 12 Statistik Austria: statcube.at, and OECD’s SDBS Structural
Business Statistics
Bahrain BHR 2 Kingdom of Bahrain Central Informatics Organization: Popula-
tion, Housing, Buildings, Establishments and Agriculture Census
Bangladesh BGD 1 Bangladesh Bureau of Statistics: Economic Census 2001 & 2003
Belgium BEL 11 Eurostat, and OECD’s SDBS Structural Business Statistics
Benin BEN 1 Institut National de la Statistique et de l’Analyse Economique:
General Census of Companies, and Les Entreprises Artisanales au
Benin
Bermuda BMU 11 Department of Statistics: www.govsubportal.com
Bhutan BTN 4 National Statistics Bureau: Statistical Yearbooks 2010-2013
Bolivia BOL 1 Instituto Nacional de Estadı́stica: Structural Statistics of the
Manufacturing Industry, Trade and Services - 2010, and Results
of the Quarterly Survey of Micro and Small Business 2010
Bosnia and Herze- BIH 8 Institute for Statistics of FB&H: Statistical Yearbooks 2008-2013
govina
Brazil BRA 13 Brazilian Institute of Geography and Statistics: Cadastro Central
de Empresas
Brunei BRN 1 Department of Economic Planning and Development: Brunei
Darussalam Statistical Yearbook 2010
Bulgaria BGR 12 Eurostat
Cambodia KHM 2 National Institute of Statistics: Economic Census 2011, and Es-
tablishment Listing 2009
Cameroon CMR 1 Institut National de la Statistique du Cameroun: Recensement
Général des Entreprises 2009
Canada CAN 7 Statistics Canada: CANSIM
Cape Verde CPV 4 Instituto Nacional de Estatı́stica: Business Census 2007, and An-
nual Business Surveys 2008-2009
Columbia COL 1 Departamento Administrativo Nacional de Estadı́stica: Encuesta
Annual Manufacturera, and www.dane.gov.co
Croatia CRV 4 Eurostat

56
Table 8: List of Countries and Sources

Country Code Years Sources


Cyprus CYP 12 Eurostat
Czech Republic CZE 10 Eurostat, and OECD’s SDBS Structural Business Statistics
Denmark DNK 12 Eurostat, and OECD’s SDBS Structural Business Statistics
Ecuador ECU 1 Instituto Nacional Estadı́stica y Censos: National Economic Cen-
sus 2010
El Salvador SLV 1 Ministerio de Economica: Tomo I de los VII Censos Económicos
Nacionales 2005
Estonia EST 1 Statistics Estonia: Statistical Yearbooks 2011-2013, and
pub.stat.ee
Ethiopia ETH 1 Central Statistical Agency: Report on Small Scale Manufac-
turing Industries Survey 2005/6, Report on Large and Medium
Scale Manufacturing and Electricity Industries Survey 2005/6, and
Labour Force Survey 2005
Faroe Islands FRO 12 Statistics Faroe Islands: www.hagstova.fo
Finland FIN 1 Statistics Finland: Labour Force Survey 2013, and www.stat.fi
France FRA 9 Institut National de la Statistique et des Études Économiques:
Tableaux de l’Économie Française - Édition 2005-6, 2010-2014,
L’industrie en France - édition 2007, 2008, and www.insee.fr
French Guiana GUF 1 Institut national de la statistique et des études économiques: Car-
actéristiques des entreprises et établissements
French Polynesia PYF 13 Institut de la Statistique de la Polynésie Française: www.ispf.pf
FYR Macedonia FYR 5 State Statistical Office: www.stat.gov.mk
Georgia GEO 11 National Statistics Office of Georgia: Statistical Yearbooks 2009-
2013, and www.geostat.ge
Germany DEU 12 Eurostat, and OECD’s SDBS Structural Business Statistics
Ghana GHA 1 Ghana Statistical Service: National Industrial Census 2003
Greece GRC 6 Eurostat, and OECD’s SDBS Structural Business Statistics
Greenland GRL 5 Statistics Greenland: bank.stat.gl
Guadeloupe GLP 1 Institut national de la statistique et des études économiques: Car-
actéristiques des entreprises et établissements
Guam GUM 7 U.S. Census Bureau: 2008-2011 County Business Patterns, and
2002, 2007, 2012 Economic Census of Island Areas
Honduras HND 1 Instituto Nacional de Estadı́stica y Censos: Directorio de Establec-
imientos Económicos
Hong Kong HKG 13 Census and Statistics Department: Annual Survey of Industrial
Production, and www.statistics.gov.hk
Hungary HUN 11 Eurostat, and OECD’s SDBS Structural Business Statistics
India IND 1 Central Statistics Office: 2005 Economic Census
Indonesia IDN 3 Statistics Indonesia: Statistical Yearbook 2013
Iran IRN 1 Statistical Centre of Iran: Statistical Yearbook 1382
Israel ISR 9 Central Bureau of Statistics: www1.cbs.gov.il, Eurostat, and
OECD’s SDBS Structural Business Statistics
Italy ITA 12 Eurostat, and OECD’s SDBS Structural Business Statistics
Japan JPN 3 Statistics Japan: Establishment and Enterprise Censuses 2001,
2004, 2006

57
Table 8: List of Countries and Sources

Country Code Years Sources


Jordan JOR 8 Department of Statistics: www.dos.gov.jo
Kazakhstan KAZ 1 Committee on Statistics: www.stat.gov.kz
Korea KOR 9 Statistics Korea: Censuses on Establishments 2007, 2009, 2011,
2012
Kosovo UVK 6 Statistical Agency of Kosovo: Statistical Register of Business
Kuwait KWT 10 Central Statistical Bureau: Annual Surveys of Establishments
2002-2011
Kyrgyzstan KGZ 1 National Statistical Committee of Kyrgyz Republic: stat.kg
Laos LAO 1 Lao Statistics Bureau: Economic Census 2006
Latvia LVA 10 Central Statistical Bureau of Latvia: www.csb.gov.lv, and Euro-
stat
Libya LBY 2 Bureau of Statistics and Census Libya: bsc.ly
Liechtenstein LIE 6 Statistical Office: Statistical Yearbooks 2007/2008, 2009-2012
Lithuania LTU 7 Eurostat
Luxembourg LUX 12 Eurostat
Macau MAC 13 Statistics and Census Service: www.dsec.gov.mo
Madagascar MDG 1 Institut National de la Statistique: Rapport de l’enquete sur les
Entreprises a Madagascar
Malawi MWI 6 National Statistical Office: www.nsomalawi.mw
Malaysia MYS 6 Department of Statistics Malaysia: Statistics Yearbooks 2007-2012
Maldives MDV 1 Department of National Planning: Economic Survey 2007/2008
Malta MLT 7 Eurostat
Martinique MTQ 1 Institut national de la statistique et des études économiques: Car-
actéristiques des entreprises et établissements
Mauritius MUS 2 Statistics Mauritius: Censuses of Economic Activity 2002, 2007,
Phases I and II
Mexico MEX 2 Instituto Nacional de Estadstica y Geografı́a: Censos Economicos
2004, 2009
Moldova MDA 8 Statistica Moldovei: www.statistica.md
Monaco MCO 13 Monaco Statistics: Observatoire de l’Economie 2012, 2013
Mongolia MNG 2 National Statistical Office of Mongolia: Monthly Bulletins of
Statistics 2011, 2012
Montenegro MNE 3 Statistical Office of Montenegro: www.monstat.org
Morocco MAR 1 Haut-Commissariat au Plan du Maroc: 2001-2 Economic Census
Nepal NPL 1 Central Bureau of Statistics: Census of Manufacturing Establish-
ments 2006/7, Survey of Small Manufacturing 2008/9
Netherlands NLD 11 Eurostat, Statistics Netherlands: Statistical Yearbooks 2004-2013
New Caledonia NCL 13 Institut de la Statistique et des Etudes Economique: www.isee.nc
New Zealand NZL 13 Statistics New Zealand: www.stats.govt.nz
Nicaragua NIC 1 Instituto Nacional de Información de Desarrollo: Urban Economic
Census
Norfolk Island NFK 1 Australian Business Statistics: www.ausstats.abs.gov.au
Northern Mariana MNP 6 U.S. Census Bureau: 2008-2011 County Business Patterns, and
Islands 2007, 2012 Economic Census of Island Areas
Norway NOR 8 Eurostat, and OECD’s SDBS Structural Business Statistics

58
Table 8: List of Countries and Sources

Country Code Years Sources


Palau PLW 1 Office of Planning and Statistics: 2012 - 2nd, 3rd Quarters Eco-
nomic Indicators
Palestinian Terri- PSE 7 Palestinian Central Bureau of Statistics: Establishment Censuses
tories 2004, 2007, 2012, and Comparison Study on Industrial Activities
1999-2004
Panama PAN 1 Instituto Nacional de Estadı́stica y Censo: Preliminary Results of
Economic Census 2012
Paraguay PRY 1 Direccin General de Estadı́stica, Encuestas y Censos: National
Economic Census 2011
Peru PER 1 Instituto Nacional de Estadı́stica e Informática: IV Censo Na-
tional Economico 2008
Philippines PHL 2 National Statistics Office: NSO’s 2012 List of Establishments, and
2003 Annual Survey of Philippine Business and Industry (ASPBI)
Poland POL 12 Central Statistical Office of Poland: Statistical Yearbook 2011,
2012, Eurostat, and OECD’s SDBS Structural Business Statistics
Portugal PRT 11 Eurostat, and OECD’s SDBS Structural Business Statistics
Puerto Rico PRI 7 U.S. Census Bureau: 2006-2011 County Business Patterns, and
2002 Economic Census of Island Areas
Qatar QAT 3 Ministry of Development Planning and Statistics: Establishment
Censuses 2004, 2008, 2010
Réunion REU 3 Institut national de la statistique et des études économiques: Car-
actéristiques des entreprises et établissements
Romania ROU 6 National Institute of Statistics: Statistical Yearbooks 2006-2012
Russia RUS 3 Federal State Statistics Service: Industry of Russia 2008, 2009,
2011, and Small and Medium Businesses in Russia 2008, 2009,
2011
Rwanda RWA 1 National Institute of Statistics of Rwanda: Establishment Census
- 2011
San Marino SMR 8 Ufficio Informatica, Tecnologia, Dati e Statistica:
www.statistica.sm
São Tomé and STP 2 Instituto Nacional de Estatı́sticas de São Tomé e Prı́ncipe: Busi-
Prı́ncipe ness Statistics 2006, 2007
Saudi Arabia SAU 1 Central Department of Statistics and Information: 2010 Economic
Census
Serbia SRB 3 Republika Srpska Institute of Statistics: Statistical Yearbook of
Republika Srpska 2011, 2012, 2013
Sierra Leone SLE 1 Statistics Sierra Leone: Report of the Census of Business Estab-
lishments 2005
Singapore SGP 10 Department of Statistics Singapore: Census of Manufacturing Ac-
tivities 2012
Slovak Republic SVK 2 Eurostat
Slovenia SVN 12 Eurostat, and OECD’s SDBS Structural Business Statistics
South Africa ZAF 12 Statistics South Africa: Annual Financial Statistics 2010, 2012,
and Survey of Employers and the Self-Employed 2013
Spain ESP 12 Eurostat, and OECD’s SDBS Structural Business Statistics

59
Table 8: List of Countries and Sources

Country Code Years Sources


Sri Lanka LKA 1 Department of Census and Statistics - Sri Lanka: Census of In-
dustry 2003/4
Sudan SDN 1 Central Bureau of Statistics: Statistical Year Book for the Year
2009
Sweden SWE 12 Eurostat, and OECD’s SDBS Structural Business Statistics
Switzerland CHE 3 Swiss Statistics: www.bfs.admin.ch/bfs/portal/en/index.html
Syria SYR 4 Central Bureau of Statistics: www.cbssyr.sy
Taiwan TWN 3 National Statistics: Industry, Commerce and Service Censuses
2001, 2006, 2011
Thailand THA 2 National Statistical Office: Industrial Censuses 2007, 2012
Tonga TON 7 Tonga Department of Statistics: Manufacturing Output, Employ-
ment and Wages/Salaries 2000-2003, 2001-2005, 2002-2006
Trinidad and To- TTO 7 Central Statistical Office: Business Establishments in T & T by
bago Industry Economic Activity 2005-2007
Tunisia TUN 12 Institut National de la Statistique: Statistiques Issues du
Répertoire des Entreprises
Turkey TUR 8 Eurostat, and OECD’s SDBS Structural Business Statistics
Uganda UGA 2 Uganda Bureau of Statistics: Report on the Census of Business
Establishments 2010/2011, and Business Register 2001/02
Ukraine UKR 3 State Statistics Service of Ukraine: www.ukrstat.gov.ua
United Arab Emi- ARE 1 National Bureau of Statistics: www.uaestatistics.gov.ae
rates
United Kingdom GBR 12 Eurostat, and OECD’s SDBS Structural Business Statistics
United States USA 11 U.S. Census Bureau: 2002-2011 County Business Patterns, and
2002-2011 Nonemployer Statistics
Uruguay URY 9 Instituto Nacional de Estadı́stica: Anuario Estadı́stico 2000-2012
U.S. Virgin Is- VIR 2 U.S. Census Bureau: County Business Patterns, and 2002, 2012
lands Economic Census of Island Areas
Venezuela VEN 1 Instituto Nacional de Estadı́stica: IV Censo Económico
Vietnam VNM 3 General Statistics Office: Establishment Censuses 2002, 2007, and
2012
Yemen YEM 2 Central Statistical Organization: Results of Economic Surveys
2005-2006

60
B Social Planner

We solve the social planner’s problem for our model economy. In each period, the social planner

chooses the number of entrants (which we denote by E), entrant productivity s0 , the growth rate of

productivity for incumbents g, and labor for each producer (`) to maximize the discounted present

value of an infinite stream of consumption (C). Given s0 , g, and the number of firms N , the planner

chooses `i for each producer i in each period to maximize;

σ
Z N σ−1
 σ−1
C = Y · (1 − I) = yiσ
di · (1 − I),
0

Z N
subject to yi = si zi `i and 1 = `i di ,
0

where I is the investment rate, or the fraction of aggregate output spent to finance entry, initial

investment, and life-cycle investment each period. The optimal quantity of labor for each firm i is;

(si zi )σ−1
`i = R N .
σ−1 d
0 (sj zj ) j

Let hatted variables denote variables chosen in previous or future periods. The planner chooses E, s0 ,

and g to maximize;
 
Y0 · 1 − E(ce + cS sθ0 ) − Y−1 (1 − λ)cg (1 + g)φ

∞  
X Yt 
θ
 Yt−1 φ
+ 1 − Ê(ce + cS sˆ0 ) − (1 − λ)cg (1 + ĝ) .
(1 + R)t (1 + R)t
t=1

A bit of rearranging results in;

− Y−1 (1 − λ)cg (1 + g)φ (28)

(1 − λ)cg (1 + ĝ)φ
 
+Y0 1 − E(ce + cS sθ0 ) −
(1 + R)

(1 − λ)cg (1 + ĝ)φ
 
X Yt
+ 1 − Ê(ce + cS sˆ0 θ ) − ,
(1 + R)t (1 + R)
t=1

! 1
λN−1 z σ−1 sˆ0 σ−1 σ−1

where Y−1 = ,
1 − (1 − λ)(1 + ĝ)σ−1

61
!
σ−1 λN−1 z σ−1 sˆ0 σ−1
Yt≥0 = (1 + g)σ−1 (1 − λ)t+1 (1 + ĝ)t(σ−1)
1 − (1 − λ)(1 + ĝ)σ−1

+ Ez σ−1 sσ−1
0 (1 − λ)t (1 + ĝ)t(σ−1)
t
X
+ Êz σ−1 sˆ0 σ−1 (1 − λ)T (1 + ĝ)T (σ−1) ,
T =1
Z N
1
and z σ−1 ≡ ziσ−1 di.
N 0

Considering the fact that the planner’s choices of E, s0 , and g are identical for each period, the first

order conditions for this problem are;


(1 − λ)cg (1 + g)φ
 
  X ∂Yt /∂E
(E) : Y ce + cS sθ0 = θ
1 − ce E − cS s0 E − (29)
(1 + R)t (1 + R)
t=0


(1 − λ)cg (1 + g)φ
 
X ∂Yt /∂s0
(s0 ) : Y θcS sθ−1
0 = 1 − ce E − c sθ
S 0 E − (30)
(1 + R)t (1 + R)
t=0


φ(1 − λ)cg (1 + g)φ−1 X ∂Yt /∂g (1 − λ)cg (1 + g)φ
 
(g) : Y = 1 − ce E − cS sθ0 E − (31)
σ (1 + R)t (1 + R)
t=0

Combined with the condition that E = λN in steady state, the following conditions characterize the

planner’s optimal allocation;

1 − (1 − λ)(1 + g)σ−1
   
(E) : λN ce + cS sθ0 = · Ψ · (1 − I) (32)
σ−1

1 − (1 − λ)(1 + g)σ−1
 
(s0 ) : λN cS sθ0 = · Ψ · (1 − I) (33)
θ
(1 − λ)cg (1 + g)φ (1 − λ)(1 + g)σ−1
(g) : = · Ψ · (1 − I) (34)
σ φ

(1 − λ)cg (1 + g)φ
 
θ
where (1 − I) ≡ 1 − ce E − cS s0 E −
(1 + R)

1+R
and Ψ ≡ .
1 + R − (1 − λ)(1 + g)σ−1

62
The investment rate I is;

(1 + R)[φ − (φ + 1 − σ)(1 − λ)(1 + g)σ−1 ]


I= , (35)
(1 + R)[σφ − (φ + 1 − σ)(1 − λ)(1 + g)σ−1 ] − φ(σ − 1)(1 − λ)(1 + g)σ−1

which is higher than the equilibrium investment rate in an undistorted economy. The social planner

chooses the same entrant productivity s0 but allocates more resources to establishment entry and

life-cycle productivity growth relative to the equilibrium allocation. This wedge between the optimal

and equilibrium allocations is common in models with endogenous life-cycle growth when costs are

specified in terms of goods rather than labor (e.g., Atkeson and Burstein, 2010).

C Comparative Statics

We show that entrant productivity s0 , life-cycle growth g, and average firm size 1/N are all decreasing

in γ.

From equation (16), initial productivity is clearly decreasing in γ;

∂(sθ0 ) σθ
= −∆ · < 0,
∂γ [θ + 1 − σ(1 − γ)]2

where ∆ contains all terms that are independent of γ.

The effect of γ on firm productivity growth g can be analyzed from equation (17). We fully differentiate

equation (17) and rearrange to obtain the following expression;

 
∂(1 + g) −(1 + g)σ 1
= · φln(1 + g)Ψ + .
∂γ φ + 1 − σ(1 − γ) σ(1 − γ) − 1

Given σ(1 − γ) > 1 and φ > σ(1 − γ) − 1, productivity growth is unambiguously decreasing in γ.

63
Average firm size from equation (18) is equal to;

!
φ(1 + R) − [φ + 1 − σ(1 − γ)](1 − λ)(1 + g)σ(1−γ)−1
1/N = ∆ · [θ + 1 − σ(1 − γ)]−1 · ,
1 − (1 − λ)(1 + g)σ(1−γ)−1

or !
−1 1 + R − (1 − λ)(1 + g)σ(1−γ)−1
1/N = ∆ · [θ + 1 − σ(1 − γ)] φ·
1 − (1 − λ)(1 + g)σ(1−γ)−1
!
(1 − λ)(1 + g)σ(1−γ)−1
 
σ(1 − γ) − 1
+∆ · · .
θ + 1 − σ(1 − γ) 1 − (1 − λ)(1 + g)σ(1−γ)−1

Given g is decreasing in γ, average firm size is also decreasing in γ.

D Decomposition

We describe and solve two simplified variants of our model. The first is a model with exogenous

productivity growth over a firm’s life cycle, as in Fattal-Jaef (2015), with no firm investments in

productivity. The second is a model with endogenous productivity growth over a firm’s life cycle but

without a productivity investment at entry, as in Hsieh and Klenow (2014).

D.1 Exogenous Life-Cycle Growth

From equation (15), the expected operating profits of an entrant are equal to;

Y (1 − τ )
E[π0 ] = · ξU S ,
σλN

ξU S ≡ 1 − (1 − λ)(1 + gU S )σ(1−γ)−1 ,

where gU S is the exogenous growth rate of firm productivity. With no investments in productivity,

free entry requires that the present value of expected life-time profits be equal to the cost of entry;

1−τ
ce = · ξU S · ΨU S ,
σλN

64

!t
X (1 − λ)(1 + gU S )σ(1−γ)−1 1+R
ΨU S ≡ = .
t=0
1+R 1 + R − (1 − λ)(1 + gU S )σ(1−γ)−1

The above free entry condition can be rearranged to express average firm size 1/N as;

σλce
N −1 = · (ξU S · ΨU S )−1 .
1−τ

As in Fattal-Jaef (2015) an increase in γ when life-cycle growth is exogenous leads to an increase in

N , and therefore a decrease in firm size. Given that aggregate output is increasing in N , this partially

offsets the effect of misallocation on output through factor misallocation.

D.2 Endogenous Life-Cycle Growth

We extend the model of exogenous life-cycle productivity growth to allow for investment in productivity

in each period after a firm enters (but not at entry). From equation (11), the expected discounted

value of life-time operating profits for a firm net of investments in life-cycle productivity is;

E[π0 ] · φ · Θ,

1+R
Θ≡ ,
φ(1 + R) − [φ + 1 − σ(1 − γ)](1 − λ)(1 + g)σ(1−γ)−1

where E[π0 ] is defined as above in Section D.1. With no initial investment in entrant productivity,

free entry requires that the above net profits be equal to the cost of entry;

φ(1 − τ )
ce = · ξ · Θ.
σλN

Average firm size can now be expressed as;

σλce
N −1 = · (ξ · Θ)−1 .
φ(1 − τ )

65
Equation (9) shows that g is decreasing in γ. To prove that average size 1/N is decreasing in γ, we

therefore show that average size is decreasing in γ given g, and increasing in g given γ.

!
∂(N −1 ) ∂ φ(1 + R) − [φ + 1 − σ(1 − γ) − 1](1 − λ)(1 + g)σ(1−γ)−1
= ∆
∂γ ∂γ 1 − (1 − λ)(1 + g)σ(1−γ)−1

σ(1 − λ)(1 + g)σ(1−γ)−1


= −∆  2 · [ξ + ln(1 + g) (φ(1 + R) − [φ + 1 − σ(1 − γ)])] .
1 − (1 − λ)(1 + g)σ(1−γ)−1

Given ξ > 0, ln(1 + g) > 0, and γ < (σ − 1)/σ, the above derivative is indeed negative. And the

following expression shows that average firm size is indeed increasing in g, given γ;

∂(N −1 ) [σ(1 − γ) − 1](1 − λ)(1 + g)σ(1−γ)−2


=∆ · (φ(1 + R) − [φ + 1 − σ(1 − γ)]) > 0.
∂g ξ2

We now prove that (as we discuss in Section 3.4) a decrease in life-cycle growth g from an increase

in γ dampens the effect of factor misallocation on aggregate output by compressing the productivity

distribution. Using equation (21), the percentage decrease in Y through factor misallocation from a

small increase in γ is;

− (1 + g)(σ−1)(1−γ) − (1 + g)σ(1−γ)−1
 
∂YF M −1
Y =∆·  · (1 − λ)σln(1 + g)
∂γ F M

1 − (1 − λ)(1 + g)σ(1−γ)−1 1 − (1 − λ)(1 + g)(σ−1)(1−γ)

where ∆ is independent of g. The magnitude of the decrease in Y through factor misallocation from

an increase in γ is clearly higher when g is higher. It follows that the lower g induced by γ dampens

the impact of factor misallocation.

66

You might also like