Firm Age, Investment Opportunities, and Job Creation
Transcript of Firm Age, Investment Opportunities, and Job Creation
Firm Age, Investment Opportunities, and Job Creation ∗
Manuel AdelinoFuqua School of Business
Duke University
Song MaFuqua School of Business
Duke University
David T. RobinsonFuqua School of Business
Duke University and NBER
October 16, 2015
Abstract
New firms are an important source of job creation, but the underlying economicmechanisms responsible for this are not well understood. To explore one importantmechanism, this paper focuses on employment creation that results from local demandshocks, and asks whether these new investment opportunities are seized more throughthe creation of new firms or through the expansion of existing firms. We find thatnew firm entry is responsible for the bulk of net employment creation in responseto changing local economic conditions than growth by established firms. Moreover,their responsiveness is larger in areas with better access to small business finance.Although we focus on the non-tradable sector for identification, our results extend tothe construction sector and the economy as a whole, indicating that the mechanismswe uncover are economically pervasive.
∗Please address correspondence to Manuel Adelino, Fuqua School of Business, 100 Fuqua Drive, Durham,NC 27708. We are grateful to Andrew Abel, Hengjie Ai, Simon Gervais, John Haltiwanger, Javier Miranda,Manju Puri, Michael Roberts, Martin Schmalz, Vish Viswanathan, and seminar participants at ChicagoBooth, Darden, Duke, Emory, Harvard Business School, HEC Paris, Minnesota Corporate Finance Con-ference, NBER Productivity, NBER Entrepreneurship, Nova SBE, Red Rock Conference, Stanford, UCLA,UNC Charlotte, University of Miami, UT Austin, and Wharton for providing helpful feedback. The usualdisclaimer applies.
1 Introduction
Startups create the majority of new jobs in the American economy (Haltiwanger, Jarmin, and
Miranda (2013)), and job creation at young firms exhibits strong comovement with business
cycles.1 However, the precise economic mechanisms underlying the central role that startups
play in the job creation process are not well understood. Identifying and understanding these
mechanisms is critically important, especially in light of recent evidence on the structural
shifts in employment from young to mature firms and the reduction in dynamism of the U.S.
economy.2
In this paper we identify one channel of the complex chain that links new firms, employ-
ment, and growth. We develop an empirical strategy in which cross-sectional variation in
local demand shocks gives rise to variation in sector-specific investment opportunities, which
in turn induces firms to respond by increasing employment. Comparing new firm formation
to the expansion of existing firms allows us to shed new light on mechanisms behind the role
of firms along the age distribution in creating new jobs.
Specifically, we focus on a region’s non-tradable sector and consider how firm entry and
expansion respond to changes in local income. Following Bartik (1991), Blanchard and Katz
(1992), and Autor, Dorn, and Hanson (2013), we identify exogenous shocks to local income by
interacting the preexisting composition of a region’s manufacturing sector with the national
growth in employment in that sector. The idea is that when a national shock hits a specific
manufacturing industry, some regions are hit harder than others because their preexisting
economic structure leaves them more exposed to the underlying shock. Regional variation
in income gives rise to variation in investment opportunities for firms that are especially
dependent on local demand. Importantly, this variation is exogenous to any opportunities
created by the firms in this sector, which resolves the reverse causality problem in analyzing
the link between firm creation and economic growth.
This approach rests on two key features of the non-tradable sector. First, conditions in
1See Fort et al. (2013), Pugsley and Sahin (2014), and Sedlacek and Sterk (2014).2Decker, Haltiwanger, Jarmin, and Miranda (2014) and Pugsley and Sahin (2014) provide evidence on
the structural shifts in employment across the firm age distribution.
1
this sector depend primarily on local demand (Mian and Sufi (2014), Basker and Miranda
(2014), Stroebel and Vavra (2014)), easing concerns that the net job creation we measure
could be confounded by unmeasured changes in fundamentals affecting both national income
and the demand for jobs. Second, in this sector startups do not have an inherent innovation
advantage relative to older firms that would make them especially well suited to generate
opportunities for growth. This is not to say that innovation is unimportant in the non-
tradable sector; on the contrary, in the non-tradable sector existing large firms are also
important innovators.3
We find that firm entry responds strongly to local demand shocks, much more so than
expansion by existing firms. Moving from the 25th to the 75th percentile of two-year income
growth raises a region’s job creation in the non-tradable industries by about 1.5% of the 2000
employment level. Startups account for approximately 90% percent of a region’s total net
employment creation in the non-tradable sector as a result of income shocks. Firms six and
more years old account for the remaining net employment response, and firms between two
and five years old generally shed jobs as a result of these shocks, highlighting the importance
of churning for the job creation process. The magnitude of this response is especially striking
given the patterns of overall employment across the firm age distribution. Firms over five
years old account for more than 84% of total employment on average in non-tradables in
each commuting zone, while employment in startup firms accounts for only 6% of total
employment. Our results suggest that some combination of size, the flexibility of the entry
decision, organizational arrangements, and the incentives provided by entry itself allows
startups to seize on opportunities that older, more established firms are not able to act on.
These results are robust to a number of concerns. One is that organizational arrangements
specific to the non-tradable sector (such as franchising) are driving our results. Our findings
are similar when we look at the construction sector’s response to the same shocks—like the
restaurant and retail sectors, construction is especially sensitive to local demand conditions,
3Wal-Mart is perhaps the most prominent recent example of innovation in retail, which, in turn, causeddramatic displacement of smaller, less competitive, firms (Basker (2005), Foster, Haltiwanger, and Krizan(2006), Neumark et al. (2008)).
2
but unlike the restaurant and retail sectors, franchising is unimportant. We also find that
jobs created by startups as a result of local income shocks are not especially short-lived,
suggesting that they are not the result of overreaction by entrants. Our results are also
robust to using the change in the penetration of imports as the regional economic shock,
employing alternative geographical units of analysis, and to using alternative measurement
periods. The results also do not reflect changes in the value of local real estate or other
banking channels, such as the level of local bank deposits.4
When we consider gross flows (both job creation and job destruction), we find large flows
in both directions from firms across the age distribution as a result of local income shocks.
The small net changes in total employment for older firms mask large absolute changes in
both job creation and destruction, which effectively cancel each other out. Because startups
are such a small portion of the overall share of employment in the economy, the gross job
creation that we witness among startups still amounts to a much larger proportional response
than what we see from older firms.
The recent work of Hurst and Pugsley (2011), Moscarini and Postel-Vinay (2012), Fort,
Haltiwanger, Jarmin, and Miranda (2013), and Basker and Miranda (2014) naturally leads
us to ask how startup responsiveness differs across firm size as well as firm age. Hurst
and Pugsley (2011) show that many so-called entrepreneurs are small business owners who
have no desire to expand their businesses. To explore this issue, we use data from the U.S.
Census Business Dynamics Statistics (BDS), which allow us to examine the joint distribution
of firm size and firm age. We find that the net employment creation by older firms is heavily
concentrated among large older firms: older small firms show no net employment creation
as a result of these shocks. This indicates that we identify a channel through new firms,
not small ones. It also speaks against the idea that increased bureaucratic constraints in
older firms prevent them from seizing the opportunities presented by economic shocks: larger
firms presumably have more complex bureaucracies than small businesses, and yet among
4Given that local income can also impact local house prices, our findings could reflect the fact that realestate is an important source of collateral for new small firms; see, e.g., Adelino, Schoar and Severino (2013),Kleiner (2013), and Schmalz, Sraer, and Thesmar (2013). We re-run our tests in areas with high and lowhouse price appreciation and find that our responsiveness results do not reflect changing house prices.
3
older businesses it is the large, more organizationally complex firms that contribute to net
employment growth as a result of economic shocks.
Our work is related to a number of distinct literatures in finance and economics. Financial
economists have long been concerned with understanding how credit market frictions and
organizational design help shape a firm’s ability to respond to investment opportunities. A
vast literature examining internal capital markets and firm valuation is predicated on the idea
that different organizational arrangements inside the firm impede or promote the adoption
and execution of investment opportunities.5 Relatedly, our work asks how the organizational
flux associated with new firm creation acts as a response to local demand shocks that create
opportunities to grow.
Our research is also related to work connecting financing constraints and new firm cre-
ation. Many have argued that financing constraints impede new business creation. To
explore how this might affect our findings we build on Petersen and Rajan (1994, 2002),
Craig and Hardee (2007), and Robb and Robinson (2012) and develop variation in local
bank market concentration as a measure of access to capital for new firms. Areas with high
levels of local bank market penetration are areas with higher concentrations of startup firms,
and we find that the responsiveness of new firm creation is about one and a half times as
strong in these regions. We find the reverse effect for firms over 6 years of age, where more
local banks mute the response to changing local income. Thus, it appears that financing
constraints are indeed important on the margin for affecting rates of new job and new firm
creation.
The fact that startups are responsible for such a large fraction of net employment creation
as a result of shocks connects our work to a large literature in macroeconomics exploring
the dynamics of aggregate investment and its connection to firm-level investment activity
(see Chirinko (1993) and Caballero (1999) for reviews). Doms and Dunne (1998), Cooper,
Haltiwanger, and Power (1999), Gourio and Kashyap (2007) and other papers show that firm-
level investment activity displays strong spikes followed by long lulls. Caballero, Engel, and
5Lamont (1997), Erickson and Whited (2000), Kaplan and Zingales (2000), Rauh (2006), and Benmelechet al. (2014).
4
Haltiwanger (1997) show similar patterns for employment adjustment at the establishment
level. The connection between this lumpy, micro-level investment behavior and the broader
time-series of investment in the economy is a matter of debate.6 Gourio and Kashyap (2007)
and others highlight the importance of the extensive margin (a change in the number of
firms engaged in investment) relative to the intensive margin (a change in the amount of
investment of firms already previously engaged) for explaining this lumpiness. Most of this
work, however, focuses on the investment behavior of preexisting establishments. Although
a full analysis of this phenomenon is beyond the scope of this paper, our results suggest that
a quantitatively important part of the dynamics of investment activity also stems from the
creation of new firms as a result of shocks.
The remainder of the paper proceeds as follows. We begin in Section 2 by describing the
data, our approach for identifying localized economic regions, and our estimation strategy.
In Section 3 we present our main findings on the link between firm age and responsiveness
to economic shocks, including questions of job resilience, gross job flows, and alternative
sectors. Section 4 explores the role of access to capital. We turn to BDS data to explore the
link between size and age in Section 5. Section 6 concludes.
2 Data and empirical methodology
2.1 Data
The empirical analysis uses publicly available data from the U.S. Census Quarterly Work-
force Indicators (QWI) to compute total employment by firm age and by county for the
non-tradable and construction sectors. The QWI is derived from the Longitudinal Employer-
Household Dynamics (LEHD) program at the Census Bureau and it provides total employ-
ment in the private sector tabulated for five firm age categories—startups (0-1 year-olds),
2-3 year-olds, 4-5 year-olds, 6-10 year-olds, and firms 11 years old or older. The totals are
6Thomas (2002) argues that in general equilibrium these investment shocks become quantitatively unim-portant in the face of endogenously adjusting prices. See also Prescott (2003).
5
provided by county, quarter, and industry, where industry is defined at the two-digit Na-
tional American Industry Classification System (NAICS) level.7 We aggregate county-level
observations in each age category to the Commuting Zone (CZ) level using a county-to-
CZ bridge provided by the Economic Research Service of the United States Department of
Agriculture.8
Net job creation data is constructed by exploiting the transition of firms across firm age
categories over time. For most of our analysis, we calculate the net job creation in the
non-tradable sector over two-year intervals for four firm age categories—startups (0-1 year-
olds), 2-3 year-olds, 4-5 year-olds, and 6+ year-olds. Specifically, the firms in the “startup”
category (0-1 year-olds) in year t − 2 are the same firms in the 2-3 year-old category at
t, conditional on having survived that far. Net employment creation is calculated as the
difference in the stock between these two bins. The difference in the total number of jobs
in these categories at t− 2 and t represents the net job creation by these firms over the two
years (including the effect of firms that disappear). Firms in the “2-3 year-old” category at
t− 2 move into the “4-5 year-old” category at t. The net employment creation for the oldest
firm category (“6+ years old”) is calculated as the difference in the stock of employment in
this category in year t minus the stock of employment in the “4-5 year-old” and the “6+
year-old” categories as of t − 2. For the “0-1 year-old” category we simply take the stock
of employment at time t as our measure of job creation by newly formed firms over the two
year period, since these firms did not exist as of t−2.9 All net employment creation numbers
are scaled by the total non-tradable sector employment as of 2000 in that CZ. In Section 3.2,
we calculate job creation for firms aged 0-5 years and 6 or more years old. Given the age
7The QWI data coverage increases through time. The data set covers 18 states in 1995, 42 states in 2000(the first year in our analysis), and 50 states (including the District of Columbia) in 2007 (the last year weconsider). Massachusetts is not covered by the LEHD data.
8The file is available at http://www.ers.usda.gov/datafiles/Commuting_Zones_and_Labor_Market_Areas/cz00_eqv_v1.xls.
9In this context it is natural to ask what constitutes a new firm as opposed to a newly formed establishmentof an existing firm. The data classify subsidiaries of existing firms as startups whenever they are formedas separate legal entities. For example, a new McDonald’s franchisee opening his or her first McDonald’slocation is classified as a startup, whereas a new location opened by an existing franchisee or by corporateheadquarters is an expansion.
6
bins provided in the QWI, these are the only two horizons that can be used in the analysis.
The LEHD data are quarterly, so we compute the December year t− 2 to December year t
change in the stock of employment in each age group in order to create bi-annual measures of
net employment creation (or December of t− 6 to December of year t for the longer horizon
analysis).
To analyze gross job flows, we use variables available directly from the LEHD Quarterly
Workforce Indicators on the quarterly number of workers who started or separated from a
job in each CZ-sector-firm age category. The gross number of jobs created during a two-year
period is calculated as the sum of new hires (the variable hiras in the QWI) in the past
eight quarters in firms in each age category (0-1 years old, 2-3, etc.). Gross job destruction
is calculated in the same way using the number of separations (seps). These numbers are
scaled by the total non-tradable employment in the CZ as of 2000 to follow the construction
of the net employment creation variables described above. One advantage of using the
variables available in the QWI is that they only include organic employment growth within
establishments, so they are not contaminated by mergers and acquisitions and other types
of reorganization activity. The caveat, though, is that the set of firms in each age bin is
not “fixed”. For example, the 0-1 year-old firms are not always the same set of firms as we
move from one quarter to the next, because new firms are created and enter the 0-1 year-
old category, while others “graduate” to become 2-3 year-old firms. The net employment
creation variable that we describe above and use as our main variable of interest in the rest
of the analysis always compares the same cohort of firms over time (for example, it is the
same set of firms in the 0-1 year-old category and in the 2-3 year-old category two years
later.
Section 5 uses an alternative data source on employment by firm age from the Business
Dynamics Statistics (BDS) to examine finer age breakdowns, as well as the joint distribution
of firm age and firm size. The BDS provides a breakdown of employment by firm age and
firm size for the country as a whole (including the states omitted from the LEHD), but it
does not report sector information at a level of geographic detail that is fine enough for our
7
purposes. Also, unlike the LEHD, the BDS uses Metropolitan Statistical Areas (MSAs) as
its geographic unit of analysis.10
We supplement the QWI and BDS with data from several other sources. Income data
at the county level comes from the Internal Revenue Service (IRS) Statistics of Income and
is measured in calendar years (i.e., January to December of each year). Income in a CZ is
defined as the total CZ wages and salaries deflated to 2007 dollars. To compute the predicted
changes in manufacturing employment, we use the County Business Patterns (CBP) data set
published by the U.S. Census Bureau. Employment in the CBP is measured as of March of
each year, so there is an overlap of seven quarters over each eight-quarter (two-year) period
with the measurement in the LEHD employment data. The measurement in the LEHD
matches the measurement period of the income variable from the IRS (which is calendar
years) and has a mismatch of one quarter relative to the CBP. We use the county-level
employment at the four-digit NAICS level for all subindustries in the manufacturing sector
(NAICS 31-33) to construct the preexisting manufacturing industry structure, as well as the
national changes in employment in each subindustry. We obtain county-level information
from the Census Bureau Summary Files for 2000 on the total population, the number of
households, and the percentage of individuals over age 25 with high school and bachelor’s
degrees. Total labor force at the county level is obtained from the Bureau of Labor Statistics.
These variables are all aggregated to the CZ level for our regressions.
Banking sector variables are calculated from the Federal Deposit Insurance Corporation
(FDIC) Summary of Deposits. HHI is the CZ-level Herfindahl index of the banking sector,
calculated using each bank’s share of total deposits in the CZ. We classify banks as “large” if
they are within the top 30 largest U.S. banks by deposits, and they are defined as “local” to
a CZ if they have 75% or more deposits concentrated in that CZ (following Cortes (2015)).
10This data set provides detailed age data for firms aged 0, 1, 2, 3, 4, 5, 6-10, 11-15, 16-20, 21-25, and26+ years. Firm size is categorized by the number of employees, and the bins used in this data set are1-4, 5-9, 10-19, 20-49, 50-99,100-249, 250-499, 500-999, 1,000-2,499, 2,500-4,999, 5,000-9,999, and 10,000+employees. For consistency with the analysis using the QWI data, we aggregate firms into the same four agecategories, 0-1, 2-3, 4-5, and 6+, and we aggregate size categories into firms with fewer than 20 employees,20-100 employees, and more than 100 employees.
8
The housing prices used in robustness tests come from the Federal Housing Finance
Agency (FHFA) House Price Index (HPI) data at the MSA level. The FHFA HPI is a
weighted, repeat-sales index, and it measures average price changes in repeat sales or refi-
nancings on the same properties. We use data on the MSA-level index between 1999 and
2007. As an alternative to using the change in housing prices during the period, we also use
the housing supply elasticity measure developed by Saiz (2010). This measure varies at the
MSA level, and it is constructed using geographical and local regulatory constraints to new
construction. This measure is available for 269 MSAs that we first match to 776 counties
using the correspondence between MSAs and counties for the year 1999, as provided by the
Census Bureau, and then aggregate up to the CZ level. Finally, we obtain import, export,
and total shipments data at the four-digit NAICS level from Peter Schott’s webpage.11
2.2 Summary statistics
Table I reports summary statistics for commuting zones included in the QWI data. We
report pooled averages for all CZs and years in the sample, yielding 4,018 observations. The
average CZ in our data consists of 5 counties, with total population of 501,000 people, and
a labor force of about 254,000. Total income increased at a real growth rate of 3% over
each two-year period, with no growth in the 25th percentile of CZ-year observations, and 5%
growth in the 75th percentile. The predicted change in manufacturing employment by CZ
is, on average, -1% and it ranges from -1% in the 25th percentile to 0 in the 75th percentile.
This is consistent with the overall downward trend in the manufacturing sector in the U.S.
during this time period. About 79% of individuals over age 25 have a high school diploma,
and 19% have a bachelor’s degree.
Our main analysis focuses on the non-tradable sector, namely Retail Trade (two-digit
NAICS 44-45), and Accommodation and Food Services (two-digit NAICS 72). Our defi-
nition of non-tradable industries matches the definition in Mian and Sufi (2014) as closely
as possible, given that the LEHD data is not broken down by four-digit NAICS industries.
11http://faculty.som.yale.edu/peterschott/sub_international.htm.
9
Around 42,000 workers are employed in the non-tradable sector in the average CZ. Firms
over 6 years of age account for the overwhelming majority of employment (about 84%) in
this sector, with just 6% coming from newly formed firms (0-1 years old). The remaining 9%
of employment is in firms between 2 and 5 years old. It is useful to keep these proportions
in mind when we discuss the net employment creation by firm age category in the sections
below. In particular, the regressions in the next sections are set up such that the coefficients
add up to the total aggregate response of the sector. However, the same net creation of
employment represents a much bigger effect as a proportion of the size of the 0-1 year-old
category than it does for the category of firms that are more than 6 years old.
Table I also shows average employment numbers for the construction industry (NAICS
23). There are about 12,000 employees in the construction sector in each CZ-year observa-
tion, with 78% of those in firms over 6 years old and 7.5% in startups.
The average HHI of deposits in a CZ is 0.13 (or the equivalent of about 7.7 equally-sized
banks). Large banks hold about 38% of all deposits, and local banks hold another 30%.
Table II summarizes the two-year net job creation in the non-tradable sector for each
firm age category in the QWI data set. Panel A shows that, on average, 668 jobs are created
in each CZ’s non-tradable sector over each two-year period. Startup firms (0-1 year-olds)
create 2,503 jobs on average, while job losses occur on average in all other age categories.
Clearly, net employment creation by 0-1 year-old firms cannot be negative (as these firms
did not exist before this period), while net job creation can be (and is) negative for the other
age categories. These net flows mask large gross flows in both directions (as shown in Davis
and Haltiwanger (1992), among others). We return to the issue of gross flows in Section 3.7
below. Old firms shed 1,193 jobs every two years on average, and they were hit particularly
hard in the recession in the early 2000s. These patterns are consistent with the results in
Haltiwanger, Jarmin, and Miranda (2013), who show that all of the net employment creation
over the last 30 years in the U.S. has come from new firms. Panel B reports the two-year
job creation scaled by the CZ’s non-tradable sector labor force in 2000 and shows similar
patterns. On average, new jobs in startups represent about 4.7% of the level of employment
10
in a CZ in non-tradables in 2000, and the net employment loss in the oldest firm category
is about 2.5% of the number of employees. When we aggregate over all firms in the sector,
net employment creation is about 1% on average over two-year periods.
2.3 Empirical strategy
Our primary empirical tests measure how shocks to income in a region affect employment
growth in the non-tradable sector for firms in different age categories. We use the real growth
rate of total income in a CZ as our measure of a demand shock for the local non-tradable
sector. Given that firms in this sector depend primarily on local demand (Mian and Sufi,
2014), higher local income creates more opportunities for those businesses. We are interested
in estimating regressions of the following form:
∆τeait = α + β × ∆τIit + γ ×Xi,2000 + εit (1)
where ∆eait is the net employment creation in firms in the non-tradable sector in each age
category a over the previous τ years in CZ i. We scale all employment numbers by the total
non-tradable sector employment as of 2000 in that CZ, but our results are not sensitive to
the choice of scaling.12 We perform the above empirical strategy on both the totals by CZ
and separately for the subsamples of startups (0-1 year-olds), 2-3 year-old firms, 4-5 year-old
firms, and 6+ year-old firms (age measured at the end of t). The parameter τ is two years
in all specifications except when we consider longer-term shocks (Table IV), where τ is six
years instead (as we discuss in Section 2.1, these are the only two horizons for which we
can measure net employment growth for well-defined firm age categories). ∆τIit is the CZ-
level income growth over the same time period. The time-invariant CZ-level controls Xi,2000
include the logarithm of the total number of residents in the labor force, the percentage of
the population with at least a high school degree, and the logarithm of total income in the
12In the Online Appendix, we report our main findings using two alternative scaling schemes: scalingnet employment creation by the lagged total employment (i.e., as of t − 2 because we always use two-yearchanges) in the non-tradable sector in each CZ, and scaling by the pooled average employment in the sectorin the CZ over the years in the sample.
11
county as of 2000.
Our main findings rely on comparing the β estimates from these age-sorted subsamples.
A higher β indicates a higher sensitivity to the shocks to investment opportunities.
We use the strategy in Bartik (1991) and Blanchard and Katz (1992) to instrument for
CZ-level income growth. This strategy is widely used in economics (see, e.g., Bound and
Holzer (2000), Gallin (2004), Saks and Wozniak (2011), and Charles et al. (2013); Imai and
Takarabe (2011) use this approach on Japanese data). Formally, the instrument is given by:
∆τemit =∑j
ωij(t−τ) × ∆τejt,−i (2)
where ∆τemit is the predicted growth rate in total manufacturing employment in CZ i between
t − τ and t. This is calculated as the percentage change in the nationwide (excluding CZ
i) number of jobs in each four-digit NAICS manufacturing sector j between t − τ and t,
denoted ∆τejt,−i, weighted by region i’s ratio of jobs in that manufacturing subsector j to
overall employment across all sectors as of time t− τ , ωij(t−τ). We instrument the growth in
income ∆τIit with this Bartik instrument, which leads to the following first-stage regression:
∆τIi,t = π0 + π1 × ∆τemit + π3 ×Xi,2000 + ηit (3)
The idea for the identification strategy rests on the fact that at a point in time, differ-
ent geographical regions have different concentrations of employment across manufacturing
industries. These concentrations reflect an accumulation of economic activity over long pe-
riods and are difficult to adjust in the short run. When a national shock to a sector hits
the economy, this regional variation in the preexisting share of each industry causes some
regions to be hit harder than others. The variable construction described in Equation (2)
picks up exactly this logic by interacting a region’s employment weights by manufacturing
industry (at the four-digit NAICS level, scaled by the total employment across all sectors
in a CZ) with the national change in employment in that industry. For example, the area
12
around Detroit is heavily tilted toward the automobile industry, whereas the region around
Billings, Montana, is not. These differences in preexisting conditions mean that a national
change in employment in the automobile industry is much more concentrated in Detroit
than Montana, regardless of the actions taken by any individual firms in either region. This
example also illustrates why we remove Detroit from the computation of the national shock
when computing the Bartik instrument for Detroit, as otherwise the region itself could be
driving the national variation.
Figures 1 and 2 provide a graphical description of our first-stage estimation strategy.
Figure 1 maps quintiles of the six-year change in income at the CZ level between 2000 and
2006. Dark red regions are those that experienced the largest average negative shock, and
the regions grow lighter in shading as the income quintile increases. (Light gray regions are
those with no data availability.) The income map paints a picture of decline throughout
the Rust Belt and in the portion of the Southeast that was heavily tilted toward textile
manufacturing. By comparison, Figure 2 maps quintiles of the six-year Bartik variable, the
main independent variable in Equation (3), using the same coloring scheme. In some sense,
Equation (3) is a regression of Figure 1 on Figure 2; regions where the colors coincide are
those in which, loosely speaking, the Bartik variable predicts the income shock because the
quantiles of the income shock line up with quantiles of the Bartik shock to manufacturing.
Regions where the colors do not coincide are data points that are not well explained in the
first-stage.
The instrumental variable used in the second stage is the portion of the change in income
that is predicted by the Bartik shock—the part of the variation depicted in Figure 1 that is
captured by variation in Figure 2. This predicted variation is then used to explain changes
in employment for startups and firms of different ages.
As we point out above, the 0-1 year-old category cannot, mechanically, have negative
net job creation, while the other categories do exhibit net negative average creation rates.
This does not, however, in any way affect the main results shown below. The average net
job creation rates in each category are accounted for by the constant term, the year fixed
13
effects, and the cross-sectional controls. The estimate on the income growth variable captures
deviations from the average rate in each category due to the instrumented income shocks.
We focus primarily on the sample period between 2000 and 2007 to avoid confounding
the estimates with the effect of the 2008 financial crisis.13 We start in 2000 because of the
limited geographic coverage of the QWI data set before that (as discussed above). Given the
structure of the QWI data (described in detail in Section 2.1), our main sample is a “non-
overlapping” sample that uses observations every two years (2001, 2003, 2005, and 2007).
This ensures that we minimize the potential correlation within a region in consecutive years.
As a robustness check, we also perform the analysis on an “overlapping” sample, where we
keep all years between 2000 and 2007 (available in the Online Appendix). All standard errors
are clustered at the CZ level.
3 Entry, expansion and job creation
This section presents our main findings linking job creation to local demand shocks. We start
by examining the connection between firm age and net job creation in the short and medium
terms (i.e., at the two-year and six-year horizons). Then we examine the construction sector
as a robustness check to guard against the possibility that organizational arrangements
peculiar to the non-tradable sector are driving our results. Finally, we examine a number of
empirical angles, including gross job flows, the role of housing markets, and the duration of
jobs created at new establishments. The Online Appendix contains a number of robustness
checks and additional analyses.
3.1 Short-term effect
The main test of the net employment creation behavior of startups and established firms in
the non-tradable sector as a result of local income shocks is shown in Table III. We run a
13Our results are robust to including data from the 2008-2012 period, but we exclude these results fromour main analysis because this time period is less proximal to the fundamental shocks to manufacturing thatdrive the identification strategy. These results are available on request.
14
two-stage least-squares (2SLS) regression (Equation (1)) of the scaled two-year job creation
on two-year regional income growth and demographic characteristics as of 2000, the first
year of the sample. This table reports results from the “non-overlapping” sample described
above. Column (1) of Table III shows the first-stage result (specified in Equation (3)), where
we regress regional income growth on the Bartik instrument of manufacturing employment
growth. The coefficient of 1.146 means that a 1% increase in the instrument is associated
with an increase in two-year income growth of 1.146 percentage points. In other words, the
point estimate indicates that a 1 % increase in the predicted number of manufacturing jobs
in a CZ translates into a similar increase in total wages and salaries in the CZ, indicating
that the jobs in question are in some sense “middle-income” jobs. The F -statistic of this
first-stage regression is 70.43, well above the conventional threshold for weak instruments
(Stock and Yogo (2005)).
In column (2), we regress job creation on income growth using ordinary least squares
(OLS). The OLS regression shows that the raw, conditional correlation between net em-
ployment growth and local income is strong. This regression suffers from significant reverse
causality problems, however, as employment growth mechanically increases total income of
an area. Column (3) is the 2SLS version of column (2), where income growth is instru-
mented using the Bartik instrument for the same period. The causal effect of income growth
on job creation in the non-tradable sector is strongly positive and very similar to the OLS
estimate. A priori, the direction of the OLS bias in the context of this analysis is unclear.
We implement the instrumental variables strategy because of reverse causality concerns –
we are interested in employment creation due to changes in income, but total income in
an area is itself given by employment multiplied by wages. The Bartik strategy allows us
to shock local income through manufacturing employment and look at the response in the
non-tradable sector. Because the non-tradable sector makes up only approximately 20% of
total employment, the OLS results are generally not very different from the IV, and the
direction of the bias is not obvious, as most of the variation in income comes not from the
sector itself, but rather from the other 80% of employment.
15
The strong effect of increased income on non-tradable employment is an important feature
of the empirical strategy. It shows (indirectly) that higher income does indeed lead to a
demand shock for the non-tradable sector, which then leads to higher total employment.
This result is the basis of our whole empirical strategy, which is also consistent with the
results shown in Mian and Sufi (2014) and Stroebel and Vavra (2014).
The remaining columns identify the firms that are mostly responsible for this strong
positive relation between jobs in the non-tradable sector and local income growth. Columns
(4) and (5) estimate the same regressions as (2) and (3), but examine job creation only among
startups (firms aged 0-1 years old). The coefficient of 0.274 in column (5) means that a one
standard deviation change in the local income growth leads to 593 more jobs per CZ created
because of new firm formation, or around 420,000 jobs nationwide.14 Comparing 0.274 with
the point estimate in column (3) of 0.306 tells us that startup firms are responsible for 90%
of the net employment creation in response to changing investment opportunities at a CZ
level, even though startups represent only 6% of the total non-tradable employment in the
average CZ (as reported in Table I).
Columns (6) through (11) consider the net employment creation of 2-3, 4-5, and 6+ year-
old firms as a result of income shocks. Columns (6) through (9) show that firms between 2
and 5 years old in the non-tradable sector generally do not create net employment as a result
of shocks to local income. If anything, these firms shed employees in response to income
shocks. The point estimates show an economically small negative coefficient (a one standard
deviation change in income leads to a drop in employment of 75 and 23 employees per CZ
for 2-3 year-old and 4-5 year-old firms, respectively). These negative estimates do, however,
suggest that churn is an important component of the overall response by the sector to the
local income shocks. We will show additional evidence of this churn when we discuss gross
flows (job creation and job destuction). Columns (10) and (11) complete the picture by
showing the positive response of mature firms (6+ year-olds) to local economic conditions.
The insignificant coefficient of 0.079 translates to the creation of 171 jobs per CZ, or around
14There are 709 commuting zones in the U.S., and our results are estimated at the CZ level.
16
121,000 jobs nationwide (29% of the net responsiveness of new firms).
These magnitudes should be understood in light of the proportion that each firm age
category makes up of the total sector employment. In particular, the net employment cre-
ation by new entrants is striking given that these firms represent 6% of total employment,
while the oldest age category comprises over 80% of total employment in the average CZ
(see Table I). This means that the response represents a large variation around the average
entry rate for startups and a very small proportional net response for the oldest category.
The regressions in Table III do not include geographic fixed effects. Given the cross-
sectional nature of the shock, as well as its slow-moving nature (the regions suffering negative
shocks in 2001 were likely to be the same ones suffering those shocks in 2005), we cannot
add CZ fixed effects to a specification that already measures net changes in employment.
We do, however, show that the results are robust to including Census Division fixed effects,
which accounts for the possibility that the results are driven primarily by trends across large
geographic regions (the results are shown in Table A.IV).
3.2 Medium-term effects
Although the preceding table indicates that startups generate more net employment than
established firms as a result of shocks, the Bartik instrument may be more suited for longer-
term analysis given that local manufacturing employment may take more than two years to
adjust fully to nationwide shocks to manufacturing. Table IV reports results from a longer
time window of six years and compares the responsiveness of all firms created over a six-year
period to that of all other firms that already existed in a CZ.15 In this test we use the cross-
section of all CZs as of 2007 and consider the net job creation in non-tradable firms during
the period between 2001 and 2007. We instrument for the growth in income with the 6-year
predicted change in employment in the manufacturing sector for the CZ.
The first-stage regression in Table IV (at the six-year horizon) is even stronger than the
15As we discuss in Section 2.1, our choice of a six year period is driven by the need to match the agebins provided by the QWI. Two years and six years are the only two time windows that allow us to cleanlymeasure employment creation of firms of different ages.
17
results in Table III at the two-year level. The F -statistic on the first-stage is over 100, and the
point estimate for the Bartik-weighted manufacturing income implies a much larger impact
on local income than in the shorter horizon analysis. The second-stage results in Table IV
show a similar pattern to what we observe in Table III. The aggregate 2SLS estimate of job
creation implies that a 10% increase in local income at the six-year horizon translates into
about 3% growth in employment (the standard deviation of six-year income growth is 9.6%).
At the six-year horizon, startups are responsible for the entire aggregate net employment
creation, which is similar to what we obtain in Panel A of Table III. The fact that results
are largely stable across different sampling periods reinforces the main message of the paper
that new firms account for the majority of the net employment creation of the non-tradable
sector in response to local demand shocks.
3.3 Job creation in the construction sector
Although the non-tradable sector provides the cleanest setting for identification purposes,
the construction sector (NAICS 23) provides an important robustness test for several rea-
sons. First, the construction sector is also largely driven by local demand, especially at the
geographic scale of Commuting Zones. Second, some of the features of the non-tradable
sector, such as the presence of franchisee firms (and thus well proven business models that
are simply replicated by new firms), do not apply. Third, the construction sector is respon-
sible for a significant fraction of the variation in employment in booms and busts (see, e.g.,
Charles et al. (2013)).
We repeat our main regressions from Table III for the construction sector in Table V. The
findings are similar to those of the non-tradable sector—the total net employment creation of
the construction sector is about 0.84% for every one percentage point increase in local income,
and firms 0-1 years old are responsible for approximately 70% of the total effect. For this
sector, we see an economically and statistically significant response by firms that are more
than 6 years old, but, as before, proportionally to their contribution to total employment in
the sector (approximately 80% of all employees are in this age bin), the response in terms of
18
net employment creation is much smaller than that of new firms. Also, firms between 2 and
5 years old do not generate net employment on average.
3.4 Creating jobs in good times versus destroying jobs in bad
One question that arises from our findings is whether the results for new and existing firms
are symmetric for positive and negative local income shocks. In particular, the results could
indicate that startups fail to create jobs in bad times, rather than creating more jobs in
good times, or vice versa.16 To explore how the responsiveness of startups varies in different
economic conditions, we first split the sample into areas at the median income growth over
the previous two years. Panel A of Table VI shows the results when we restrict the sample to
CZ-years with above-median income growth. The results show that the aggregate response
in the non-tradable sector due to the Bartik shocks is strongly positive and that the net
creation of employment is concentrated in the startup category, as in Table III. We find a
positive coefficient for firms over 6 years old in this specification, but the estimate is still
statistically insignificant.
In Panel B of Table VI we report reduced-form regressions of job creation by age cate-
gory on the Bartik instrument split into terciles (Table B.I of the Online Appendix shows
the reduced form regressions without breaking out the instrument by terciles). The omitted
category is the lowest tercile of the Bartik instrument. Net change in employment is higher
by 0.7 percentage points in CZs that experience median Bartik shocks, and it is 0.9 percent-
age points higher in CZs with the highest Bartik shocks relative to the lowest tercile. This
variation comes mostly from the startup category, where median shocks are associated with
0.5 percentage points higher net employment creation (as a proportion of total non-tradable
employment in the CZ as of 2000), and the CZs with the most positive predicted manufactur-
ing shocks create 0.8 percentage points more jobs in the non-tradable sector. Employment
creation in the other three categories of firms is much flatter across the distribution of Bartik
16The work of Duygan-Bump et al. (2010), Fort et al. (2013), and Fairlie (2013) argues that young firmswere particularly hit during the recent recession.
19
shocks, mirroring the results we showed in the previous tables. Overall, this table shows that
startups create more jobs in good times, and that the sensitivity does not come just from
the downside.
3.5 Are collateral effects driving our results?
A prominent feature of the time period we consider is the nationwide increase in house prices
between 2000 and 2007. Given that our instrument may also affect demand for housing
(e.g., through migration), it is important to explore the implications of changing house
prices for our results. A shock to demand for housing (and higher prices) could affect our
analysis through two channels. First, previous work argues that the increase in house prices
had implications for demand in the non-tradable sector (Mian and Sufi, 2011, 2014). This
mechanism by itself fits into our empirical strategy, as it amplifies the fact that non-tradable
businesses faced higher demand in places with higher values of the Bartik instrument. The
second channel by which housing could affect our results is emphasized in recent work by
Adelino et al. (2013) and Schmalz et al. (2013), who argue that the increase in house prices
also led to easier access to collateral for entrepreneurs, which in turn led to an increase
in employment in firms with fewer than 20 employees. This implies that our results could
reflect differentially easier financing on the part of firms in different age categories, and not
differential ability to pursue investment opportunities.
We first note that, in examining the non-tradable sector, our empirical design minimizes
the relative contribution of the collateral channel. There are two main reasons. First, the
non-tradable sector faces demand shocks that stem mostly from changing local conditions.
Thus, the relative contribution of the collateral channel is minimized in this sector. Second,
the startup capital requirements in restaurants and retail establishments are large (Adelino et
al, 2013): retail inventory requirements and the kitchen up-fit costs associated with starting
a restaurant place the non-tradable sector above the median in terms of startup capital
requirements. These types of firms are therefore harder to start using a house as collateral.
Nevertheless, despite these mitigating factors, it is possible that the response we observe on
20
the part of startups might be significantly affected by the value of residential or commercial
real estate collateral and that removing this effect would alter our conclusions.
To test directly the impact of changing house prices on the responsiveness of firms to the
Bartik shock, we split the sample of CZs into areas that experience high and low house price
appreciation during the sample period. We define high and low house price appreciation areas
using the pooled median of two-year house price growth between 2000 and 2007. Because
house prices could themselves be endogenous to employment growth, we also split the sample
into high and low elasticity areas as defined by the Saiz (2010) housing elasticity measure
using the median of this measure.17
Results are shown in Table VII.18 Column (2) shows that, consistent with our interpreta-
tion of the main results, the effect on new firms of shocks to investment opportunities is very
similar across CZs that experience high and low house price appreciation during this time
period (the coefficient on the interaction between income and above median house price ap-
preciation is small and insignificant). This suggests that our instrument affects employment
creation in firms of this age through shocks to local income, and not due to shocks to house
prices. We also find that the results for startup responsiveness are large, statistically signif-
icant, and unchanged in high and low elasticity areas (shown in column (4)). In sum, these
results suggest that our results are being driven primarily by demand-side considerations
rather than through a collateral channel.
3.6 How permanent are the jobs created by new firms?
One of the most natural questions to arise in the context of job creation is whether the
jobs being created by new firms are jobs that last. Do these jobs persist or are they short-
17The Saiz (2010) housing supply elasticity is a cross-sectional MSA-level measure and it includes a geo-graphic and regulatory component that are meant to capture the relative ease with which the housing stockin an area can adjust to a positive shift in the demand for housing. Areas where is it relatively easy tobuild tend to see more construction (and smaller house price increases) when demand for housing increases,whereas low elasticity areas (those where it is hard to build) tend to see higher prices and lower levels ofnew construction. This measure is available for 269 MSAs in the U.S.
18Because we run the tests only for CZs for which we have house price and elasticity data, Columns 1 and3 show the first-stage results for all subsamples, and in all cases we obtain a strong first-stage.
21
lived? This question has both a normative and a positive dimension to it: perhaps it is
undesirable from a public policy standpoint to promote job creation by new firms if the jobs
themselves are short-lived, and therefore the normative question asks whether these jobs
are somehow better or worse than ones created by established firms. At the same time,
this question reveals alternative mechanisms that may be ultimately driving the result. For
example, perhaps the greater responsiveness of new firms reflects misjudgments about the
magnitude of the economic shocks. Under this view, new firm creation is a mechanism for
seizing on short-lived opportunities and it is possible that a form of irreversibility makes
hiring employees in established businesses inherently difficult.
Table VIII examines these issues. It groups newly created firms by cohort and compares
the magnitude of their initial job creation to the magnitude of later job creation or destruc-
tion. Because the LEHD groups firms into 24-month age buckets until firms are greater than
60 months old, the total net employment in firms that are 2-3 years old t+2 years later (and
4-5 years old t + 4 years later) corresponds to jobs among the same cohort of firms started
at time t. Using this feature of the data, the table examines employment over time for each
cohort based on whether the geographic region in question experienced a high, medium or
low (Bartik) total income shock. To maintain consistency with previous tables, we express
total net jobs as a fraction of year 2000 total population in the commuting zone.
Panel A groups each cohort by terciles of the Bartik instrument. Pooling across the
four cohorts for which complete data are available, startups in regions that experienced
manufacturing shocks in the top tercile added jobs totaling 5.06% of the year 2000 local
employment. The middle column indicates that about 85% of these jobs remain after two
years. The right-most column indicates that around 75% of these jobs remain after four
years. Commuting zones in the lowest tercile of manufacturing shocks witnessed startups
creating fewer jobs per capita (3.93% as opposed to 5.06%, a highly statistically significant
difference), and about two-thirds of these jobs remain after four years in both high and low
shock regions. In Table B.II of the Online Appendix we show that similar patterns hold
across each cohort.
22
Panel B repeats the analysis of Panel A but groups commuting zones according to the
magnitude of their income growth rather than according to the magnitude of the Bartik
shock. The results are similar. In each cohort, the proportion of new jobs created that
remain after four years is higher in the high income growth area than in the low income
growth area.
The analysis presented in Table VIII does not support the idea that the jobs created by
startups as a result of local investment opportunities are particularly short-lived. There is no
evidence that the extra jobs created in high job creation regions are less likely to persist than
the ones in low creation regions—in general, about 7 out of 10 jobs created are still there after
4 years. This evidence speaks against the idea that net job creation among startups results
from misjudging the magnitude of the economic opportunity. It also speaks against the idea
that new firm creation is primarily driven by the desire to organize temporary employment.
3.7 Job creation and destruction: gross flows
All of the preceding analysis uses net employment creation by firms in individual age cat-
egories as the measure of responsiveness to local shocks. The net employment creation
variable, however, masks large underlying gross flows in both directions (both job creation
and job destruction). In this section we address the responsiveness of gross flows across
the firm age distribution using variables available directly from the QWI. As we discuss in
more detail in Section 2.1, the publicly available QWI data do not allow us to measure job
creation and destruction by the same set of firms for a period longer than one quarter (be-
cause firms enter and exit each age bin as time passes). However, there is yearly variation in
the dependent variables happens. As such, in order to measure gross flows and to keep the
analysis consistent with all the other tables, we create two-year job creation and destruction
variables by firms in each bin by adding up the quarterly numbers over a period of 8 quarters
(scaled, as for the other variables, by the employment in the non-tradable sector as of 2000).
One advantage of these variables is that they are within-establishment measures of creation
and destruction, which eliminates concerns that reorganizations or mergers and acquisitions
23
could influence the results.
We run regressions at two-year intervals, as in our main specifications. Importantly, the
average gross job creation and destruction rates in each firm age bin are absorbed by the
constant term, the year fixed effects, and the cross-sectional controls. The coefficient on
income growth captures deviations from the average rate in each category due to the Bartik
shocks.
Table IX shows the results for the gross flow measures. Panel A shows that a one
percentage point shock in income increases gross job creation by almost 2 percentage points
(as a percentage of 2000 employment in the sector), of which approximately 0.28% come from
new firms and 1.2% come from firms over 6 years old. Panel B shows that job destruction
is of a similar magnitude, namely 1.9% of jobs (again as a percentage of 2000 non-tradable
employment in the CZ) are destroyed, with a split that is similar to the job creation numbers.
Importantly, the new firms are the only category where job destruction is significantly smaller
than job creation, which then produces the only positive and significant effect that we see
in Panel C for the difference between job creation and destruction creation.
It is important to highlight two features in particular about Table IX. First, and as in
the rest of the tables, the magnitudes should be read in the context of the share of total
employment in each category. So it is not surprising that the gross flows are much larger
in firms that are over 6 years old than in the 0-1 year-old category, as these make up 84%
of total employment, whereas the youngest category represents only 6%. As before, the
responsiveness relative to each category’s size in the economy is much larger for startups
than for other firms.
Second, these tables highlight the importance of churn for economic growth. In fact, even
when a region is hit with good shocks, there is substantial job destruction that accompanies
similar flows in the opposite direction. This means that some firms are taking advantage of
the positive shocks in all categories, whereas other firms are being displaced.
24
4 Access to capital and startup job creation
The evidence thus far clearly demonstrates that startups create net employment in response
to local income shocks. However, it is far from obvious that this should be the case in the
non-tradable sector, where startups’ technological advantage over established firms is likely
not present and young firms are likely to face more severe financing constraints than older,
more established firms. Indeed, in the retail sector Wal-Mart is a prominent illustration of
innovation by large, established firms. This induced dramatic displacement of smaller, less
competitive, firms (Basker (2005), Foster, Haltiwanger, and Krizan (2006), Neumark et al.
(2008)). Even though startups create more net employment, it could still be the case that
financing constraints create economically meaningful barriers preventing them from taking
advantage of changing opportunities (Hellman and Puri, 2000; Cagetti and De Nardi, 2006;
Kerr and Nanda, 2009; Lelarge et al., 2010; Chemmanur et al., 2011; Kerr et al., 2011;
Chemmanur and Fulghieri, 2014).
To study how access to finance interacts with firms’ ability to pursue investment oppor-
tunities, we use the share of local banks in a CZ as a measure of local access to finance. A
“local” bank is defined as one that has 75% or more deposits concentrated in one CZ (fol-
lowing Cortes (2015)). We then construct the local bank share of a CZ, defined as the share
of all deposits in a CZ that are held by banks local to that CZ. The identifying assumption
is that, as shown by Petersen and Rajan (1994, 2002), small (local) banks are more likely to
be able to lend to small firms, especially to more opaque firms. Lending to old (established)
firms is likely to require less screening and monitoring than lending to new firms in an area,
so potential entrepreneurs in CZs with a higher proportion of local banks are likely to have
better access to financing. In order to mitigate the effect of labor market dynamics on the
evolution of the local banking sector, we use a time-invariant CZ-level measure by calculating
the time-series median of the deposit concentration in local banks for each CZ. As shown
in Table I, the share of deposits held by local banks account, on average, for 30% of all CZ
deposits.
We start by confirming that local banking is important for firm creation. Table X per-
25
forms a cross-sectional OLS regression of employment in young and old firms on the share
of local banks. The dependent variable is the time-series median of the share of CZ em-
ployment in 0-1 year-old firms (column (1)) and in 6+ year-old firms (column (2)). The
independent variables are the time-series median of the local bank share and demographic
covariates. Consistent with the literature (e.g., Guiso et al., 2004), we find that the strength
of local banks in an area is positively correlated with the share of employment in startup
firms (parameter estimate is 0.016) and negatively correlated with the share of employment
in existing older firms (estimate is -0.041). The long term average of the share of employment
in startups is 6.1%, and a one standard deviation change in the share of local banks yields
a change of 0.3 percentage points in the employment in those types of firms. This result
is consistent with the share of deposits held by local banks capturing the ease of access to
finance by startup firms.
To identify the effect of access to bank financing on firms’ ability to capture local invest-
ment opportunities, we introduce the local bank share into the specifications by adding the
main effect of this measure and its interaction with the instrumented income growth. For
interpretation purposes, we incorporate this measure as an indicator variable, where IHigh LB
is equal to 1 if commuting zone i’s long-term median of the share of local banks is higher
than the median share of all CZs. Specifically, we estimate a modified version of Equation
(1) that includes the additional dummy and its interaction by 2SLS:
∆τeait = α + β × ∆τIit
+ γ′ × IHigh LB,i
+ β′ × ∆τIit × IHigh LB,i
+ γ × Controlsi,2000 + εit.
(4)
The interaction term β′ can be interpreted as the “additional” responsiveness to local
investment opportunities of firms in areas with easier access to finance relative to those in
areas with worse access to bank finance. We instrument the income growth and the inter-
action of income growth and the high local bank share dummy with the Bartik instrument
26
and its interaction with the same dummy.
Table XI reports the estimation results of Equation (4) for firms of different ages. Column
(3) shows the regression for startups. The responsiveness to income growth is 0.217 for
startups. If the CZ is an area with a high share of deposits in local banks, the magnitude
for the startups increases to 0.320 (0.217+0.103)—about one and a half times the initial
effect. Interestingly, the effect for established firms decreases in areas with a high share
of local banks. This suggests that, in areas with easier access to credit for new firms,
the responsiveness of firm entry to new opportunities is strongly increased and that the
heightened effect on firm creation may even crowd out employment creation by existing
firms.
A separate but related question is whether the effect we find is the result of improved
access to finance, rather than a local demand shock. This would be the case if increased local
income improved the balance sheet of banks and these, in turn, extended more (and easier)
credit to startups. We cannot rule out that this channel may increase the net employment
response in these areas, nor do we want to exclude this possibility. However, Table A.VI
in the Online Appendix suggests that this is not a first-order effect in our setting. In this
table, we control for the local two-year growth in total deposits (from the FDIC) in Panel A,
and for the total loans extended to businesses with less than one million dollars in revenue
(from the Community Reinvestment Act data set) in Panel B. We find that the estimates
are almost unchanged relative to our main specification in Table III.
5 Firm age and firm size
The empirical strategy in the preceding sections establishes a causal link between economic
shocks and job creation across the firm age distribution, which is consistent with a few
different mechanisms. For one, startups are often thought to be nimbler than older firms,
especially in terms of their ability to seize on disruptive innovations. Perhaps the nimbleness
stems from organizational flexibility. Perhaps their small size, as much as their age, allows
27
them to react more quickly, or perhaps the additional layers of bureaucracy in older firms
create an organizational or geographic barrier between the decision-maker in a mature firm
and the economic opportunity.
In this section we address these possibilities by exploring the separate roles of firm size
and firm age in explaining our results. Specifically, we compare the net employment respon-
siveness not just of new and existing firms, but also of large and small firms to local income
shocks.
The QWI data we have used thus far does not allow us to compare firms of different
sizes, so we turn to data from the U.S. Census Business Dynamics Statistics, described
in detail in Section 2. As we discuss before, this data set differs in some important ways
from the one constructed for the previous analysis. In particular, it contains Metropolitan
Statistical Area (MSA)-level data instead of county-level data, and it also does not break
down employment by sector. We cannot, therefore, consider the effect of the shock on
different sectors in isolation. It does, however, contain age and size breakdowns instead of
just age classifications.
We proceed exactly as we have above, using the Bartik manufacturing instrument for
local income shocks, to confirm that our findings extend to job creation in all sectors (not
just non-tradables). We should emphasize that, by construction, our experiment is most
applicable to the “purely” non-tradable industries (NAICS sectors 44-45 and 72), but the
other sectors in the economy should also respond to changes in local income.19 The results
are reported in Table XII. We first show descriptive statistics for the number of employees
(Panel A), as well as the number of firms (Panel B), in each size and age bin. As in the
QWI, the majority of employees are in older firms, and over 60 percent of jobs are in firms
with more than 100 employees. Of the employees in large firms, almost all are in firms that
are over 6 years old, whereas the smaller firms dominate the other three categories (0-1, 2-3
and 4-5 year-old firms). In Panel B it becomes apparent that large, new firms are very rare,
and that, as one would expect, smaller firms are the most numerous across all age categories.
19Table B.III in the Online Appendix shows that the results are very consistent when we perform ouranalysis using the QWI data for all sectors (rather than just the non-tradable industries) and the BDS data.
28
Panel C shows the regression results. For brevity, we have reported only the point
estimates from the second stage regression on the main variable of interest, instrumented
income growth (the first-stage regressions, as well as the regressions including all controls,
are in Table B.IV of the Online Appendix). The table shows the results of the breakdown of
the age categories into three size bins: firms with fewer than 20 employees, those with more
than 20 and fewer than 100 employees, and those with more than 100 employees.
We find that the responsiveness of new firms comes almost exclusively from small firms
(with fewer than 20 employees), whereas the responsiveness of older firms comes from those
with more than 100 employees. This suggests that geographic proximity is unlikely to be
an important reason for the effects we observe in the previous sections. Indeed, while it
is new small firms that respond to higher income, the fact that we see no net effects from
small established firms suggests that there are other mechanism at play behind the patterns
in the previous sections. These results are consistent with those shown for small and new
firms, as well as the mature and large firms in Moscarini and Postel-Vinay (2012) and
Fort, Haltiwanger, Jarmin, and Miranda (2013). The result on large older firms is also
consistent with the view in Gromb and Scharfstein (2002) that such firms may provide a
safe environment for pursuing investment opportunities because of the ability to redeploy
individuals through internal labor markets if projects fail (which is not the case in small
mature firms). The lack of a net effect of local income shocks on firms aged 2 to 5 years is
present across all size categories. Interestingly, small firms aged 2 years or more all seem to
lose jobs when income rises, potentially pointing to a crowding out effect of startups relative
to older ones. As before, the positive and significant result for firms aged six years or more
is much smaller relative to the proportion that these firms make up of the economy than the
effect we find for new firms.
Although the point estimates are not immediately comparable to those from the preceding
sections, these results using all sectors of the economy reinforce and amplify our previous
results. In particular, the findings in this table support the notion that some unobserved
firm characteristics that are proxied by firm age correlate with job creation: new firms that
29
possess these characteristics grow and thrive, becoming larger, older firms that continue
to respond to changing economic conditions. Firms that lack these characteristics languish.
This perspective is consistent both with Puri and Zarutskie (2012), whose focus is on venture-
versus non-venture-backed firms but who show that a tiny fraction of new firm starts are
responsible for a large fraction of overall employment, and Hurst and Pugsley (2011), who
conversely show that a large number of small businesses simply have no desire to grow.
6 Conclusion
Understanding the mechanics of job creation has become a central objective for academic
researchers, politicians and policy makers alike, especially in the wake of the financial crisis
and ensuing economic recession of 2007-2009. Recent evidence tells us that startups are
responsible for most job creation, and that their ablity to generate employmet comoves with
aggregate shocks. This paper explores why this is the case.
One explanation for the connection between startups and job creation is that new firms
generate opportunities for growth for a variety of potential reasons having to do with the
interaction between innovation and the structure of capital and labor markets. A second
(separate) direction of causality is that startups have more flexible organizational structures
than older firms and create more jobs (also) because they are quicker to respond to economic
shocks. Both channels are undeniably important for understanding the full picture.
Unfortunately for empiricists, these channels are difficult to disentangle. Our empirical
strategy aimes at isolating the second channel. For that reason, we focus primarily on firms
in the non-tradable sector, which allows us to test a geographically segmented version of
q-theory. The thought experiment is as follows: When income from the local manufacturing
sector changes for reasons unrelated to the performance of a given region itself, this ripples
through the local economy, causing retail stores, restaurants, and local service organizations
to expand or contract in response to the shock. Who responds, new or existing firms? We
find strong evidence that startups are much more responsive to these shocks in terms of net
30
employment creation.
In this sense, this paper adds a new dimension to the entrepreneurial spawning literature
found in Bhide (2000), Klepper (1996), or Klepper and Sleeper (2005) and others, which
observes that many new businesses are started by people who already work in existing firms.
Our entrepreneurial responsiveness findings suggest that this mechanism does not simply
arise because old businesses are ill-suited to gamble on the new ideas generated by their
employees: they instead either fail to recognize or act upon the opportunities in their midst
(see also Chatterji (2009)). Our finding that new firm formation responds strongly to new
investment opportunities is also consistent with Glaeser, Kerr, and Kerr (2013), who show
that the amount of local entrepreneurial human capital leads to future city growth.
Why do startups account for such a large fraction of the net response to local economic
shocks? While a complete answer to this question is beyond the scope of any single paper,
factors such as bureaucratic inflexibility, unobserved characteristics tied to the entrepreneur
(Gompers et al. (2010)), and the strength of incentives are thought to play an important
role.. This is consistent with the results we find on gross job creation and destruction, as
well as when we consider the age and size of firms. We see firms in all age buckets respond
to local shocks, but this induces a substantial amount of churn and large job destruction
flows, resulting in net flows that are close to zero for all age groups except startups. Further
understanding the reasons for these differential flows is an important and ongoing research
agenda that connects to questions in macroeconomics, financial economics, productivity, and
the economics of organizations.
31
References
Adelino, Manuel, Antoinette Schoar, and Felipe Severino, 2013, House prices, collateral andself-employment, National Bureau of Economic Research Working Paper .
Autor, David H, David Dorn, and Gordon H Hanson, 2013, The china syndrome: Local labormarket effects of import competition in the united states., American Economic Review 103,2121–2168.
Bartik, Timothy J, 1991, Who benefits from state and local economic development policies?,Books from Upjohn Press .
Basker, Emek, 2005, Job creation or destruction? labor market effects of wal-mart expansion,Review of Economics and Statistics 87, 174–183.
Basker, Emek, and Javier Miranda, 2014, Taken by storm: Business survival in the aftermathof hurricane katrina, US Census Bureau Center for Economic Studies Paper No. CES-WP-14-20 .
Benmelech, Effi, Nittai Bergman, and Amit Seru, 2014, Financing labor, Kellogg WorkingPaper .
Bhide, Amar, 2000, The origin and evolution of new businesses (Oxford University Press).
Blanchard, Olivier Jean, and Lawrence F Katz, 1992, Regional evolutions, Brookings paperson economic activity 1992, 1–75.
Bound, John, and Harry J Holzer, 2000, Demand shifts, population adjustments, and labormarket outcomes during the 1980s, Journal of labor Economics 18, 20–54.
Caballero, Ricardo J, 1999, Aggregate investment, Handbook of macroeconomics 1, 813–862.
Caballero, Ricardo J, Eduardo MRA Engel, and John Haltiwanger, 1997, Aggregate employ-ment dynamics: Building from microeconomic evidence, The American Economic Review115–137.
Cagetti, Marco, and Mariacristina De Nardi, 2006, Entrepreneurship, frictions, and wealth,Journal of Political Economy 114, 835–870.
Charles, Kerwin Kofi, Erik Hurst, and Matthew J Notowidigdo, 2013, Manufacturing decline,housing booms, and non-employment, National Bureau of Economic Research WorkingPaper .
Chatterji, Aaron K, 2009, Spawned with a silver spoon? entrepreneurial performance andinnovation in the medical device industry, Strategic Management Journal 30, 185–206.
32
Chemmanur, Thomas J, and Paolo Fulghieri, 2014, Entrepreneurial finance and innovation:An introduction and agenda for future research, Review of Financial Studies 27, 1–19.
Chemmanur, Thomas J, Karthik Krishnan, and Debarshi K Nandy, 2011, How does venturecapital financing improve efficiency in private firms? a look beneath the surface, Reviewof financial studies 24, 4037–4090.
Chirinko, Robert S, 1993, Business fixed investment spending: Modeling strategies, empiricalresults, and policy implications, Journal of Economic literature 1875–1911.
Cooper, Russell, John Haltiwanger, and Laura Power, 1999, Machine replacement and thebusiness cycle: Lumps and bumps, American Economic Review 921–946.
Cortes, Kristle Romero, 2015, Rebuilding after disaster strikes: How local lenders aid in therecovery, Federal Reserve Bank of Cleveland Working Paper .
Craig, Steven G, and Pauline Hardee, 2007, The impact of bank consolidation on smallbusiness credit availability, Journal of Banking & Finance 31, 1237–1263.
Davis, Steven J, and John Haltiwanger, 1992, Gross job creation, gross job destruction, andemployment reallocation, The Quarterly Journal of Economics 107, 819–863.
Decker, Ryan, John Haltiwanger, Ron S Jarmin, and Javier Miranda, 2014, The role ofentrepreneurship in us job creation and economic dynamism, Journal of Economic Per-spectives 28, 3–24.
Doms, Mark, and Timothy Dunne, 1998, Capital adjustment patterns in manufacturingplants, Review of economic dynamics 1, 409–429.
Duygan-Bump, Burcu, Alexey Levkov, and Judit Montoriol-Garriga, 2010, Financing con-straints and unemployment: evidence from the great recession, Federal Reserve Bank ofBoston Working Paper No. QAU10-6 .
Erickson, Timothy, and Toni M Whited, 2000, Measurement error and the relationshipbetween investment and q, Journal of Political Economy 108, 1027–1057.
Fairlie, Robert W, 2013, Entrepreneurship, economic conditions, and the great recession,Journal of Economics & Management Strategy 22, 207–231.
Fort, Teresa, John Haltiwanger, Ron S Jarmin, and Javier Miranda, 2013, How firms respondto business cycles: The role of firm age and firm size, IMF Economic Review .
Foster, Lucia, John Haltiwanger, and Cornell J Krizan, 2006, Market selection, reallocation,and restructuring in the us retail trade sector in the 1990s, The Review of Economics andStatistics 88, 748–758.
33
Gallin, Joshua Hojvat, 2004, Net migration and state labor market dynamics, Journal ofLabor Economics 22, pp. 1–21.
Glaeser, Edward L, Sari Pekkala Kerr, and William R Kerr, 2013, Entrepreneurship and ur-ban growth: An empirical assessment with historical mines, National Bureau of EconomicResearch .
Gompers, Paul, Anna Kovner, Josh Lerner, and David Scharfstein, 2010, Performance per-sistence in entrepreneurship, Journal of Financial Economics 96, 18–32.
Gourio, Francois, and Anil K Kashyap, 2007, Investment spikes: New facts and a generalequilibrium exploration, Journal of Monetary Economics 54, 1–22.
Gromb, Denis, and David Scharfstein, 2002, Entrepreneurship in equilibrium, National bu-reau of economic research .
Guiso, Luigi, Paola Sapienza, and Luigi Zingales, 2004, Does local financial developmentmatter?, The Quarterly Journal of Economics 119, pp. 929–969.
Haltiwanger, John, Ron S Jarmin, and Javier Miranda, 2013, Who creates jobs? small versuslarge versus young, Review of Economics and Statistics 95, 347–361.
Hellman, Thomas, and Manju Puri, 2000, The interaction between product market andfinancing strategy: The role of venture capital, Review of Financial Studies 13, 959–984.
Hurst, Erik, and Benjamin Wild Pugsley, 2011, What do small businesses do?, NationalBureau of Economic Research .
Imai, Masmai, and Seitaro Takarabe, 2011, Bank integration and transmission of financialshocks: Evidence from japan, American Economic Journal: Macroeconomics 3, 155–183.
Kaplan, Steven N, and Luigi Zingales, 2000, Investment-cash flow sensitivities are not validmeasures of financing constraints, The Quarterly Journal of Economics 115, 707–712.
Kerr, William, and Ramana Nanda, 2009, Financing constraints and entrepreneurship, Na-tional Bureau of Economic Research .
Kerr, William R, Josh Lerner, and Antoinette Schoar, 2011, The consequences of en-trepreneurial finance: Evidence from angel financings, Review of Financial Studies .
Kleiner, Kristoph, 2013, How real estate drives the economy: An investigation of small firmbalance sheet shocks on employment .
Klepper, Steven, 1996, Entry, exit, growth, and innovation over the product life cycle, TheAmerican economic review 562–583.
Klepper, Steven, and Sally Sleeper, 2005, Entry by spinoffs, Management science 51, 1291–1306.
34
Lamont, Owen, 1997, Cash flow and investment: Evidence from internal capital markets,The Journal of Finance 52, 83–109.
Lelarge, Claire, David Sraer, and David Thesmar, 2010, Entrepreneurship and credit con-straints: Evidence from a french loan guarantee program, in International differences inentrepreneurship, 243–273 (University of Chicago Press).
Mian, Atif, and Amir Sufi, 2011, House prices, home equity-based borrowing, and the ushousehold leverage crisis, The American Economic Review 101, 2132–56.
Mian, Atif R, and Amir Sufi, 2014, What explains high unemployment?—the aggregatedemand channel, National Bureau of Economic Research Working Paper .
Moscarini, Giuseppe, and Fabien Postel-Vinay, 2012, The contribution of large and smallemployers to job creation in times of high and low unemployment, The American EconomicReview 102, 2509–2539.
Neumark, David, Junfu Zhang, and Stephen Ciccarella, 2008, The effects of wal-mart onlocal labor markets, Journal of Urban Economics 63, 405–430.
Petersen, Mitchell A, and Raghuram G Rajan, 1994, The benefits of lending relationships:Evidence from small business data, The Journal of Finance 49, 3–37.
Petersen, Mitchell A, and Raghuram G Rajan, 2002, Does distance still matter? the infor-mation revolution in small business lending, The Journal of Finance 57, 2533–2570.
Pierce, Justin R, and Peter K Schott, 2012, The surprisingly swift decline of us manufacturingemployment, Technical report, National Bureau of Economic Research.
Prescott, Edward C, 2003, Non-convexities in quantitative general equilibrium studies ofbusiness cycles , volume 312 (Citeseer).
Pugsley, Benjamin, and Aysegul Sahin, 2014, Grown-up business cycles, Working paper,Federal Reserve Bank of New York .
Puri, Manju, and Rebecca Zarutskie, 2012, On the lifecycle dynamics of venture-capital- andnon-venture-capital-financed firms, Journal of Finance 67, 2247–2293.
Rauh, Joshua D, 2006, Investment and financing constraints: Evidence from the funding ofcorporate pension plans, The Journal of Finance 61, 33–71.
Robb, Alicia M, and David T Robinson, 2012, The capital structure decisions of new firms,Review of Financial Studies Forthcoming.
Saiz, Albert, 2010, The geographic determinants of housing supply, The Quarterly Journalof Economics 125, 1253–1296.
35
Saks, Raven E, and Abigail Wozniak, 2011, Labor reallocation over the business cycle: newevidence from internal migration, Journal of Labor Economics 29, 697–739.
Schmalz, Martin C, David A Sraer, and David Thesmar, 2013, Housing collateral and en-trepreneurship, Working Paper .
Sedlacek, Petr, and Vincent Sterk, 2014, The growth potential of startups over the businesscycle .
Stock, James H, and Motohiro Yogo, 2005, Testing for weak instruments in linear iv re-gression, Identification and inference for econometric models: Essays in honor of ThomasRothenberg .
Stroebel, Johannes, and Joseph Vavra, 2014, House prices, local demand, and retail prices,Technical report, National Bureau of Economic Research.
Thomas, Julia K, 2002, Is lumpy investment relevant for the business cycle?, Journal ofPolitical Economy 110, 508–534.
36
Fig
ure
1:C
omm
uti
ng
Zon
eIn
com
eQ
uin
tile
s
The
map
plo
tsquin
tile
sof
loca
lin
com
egr
owth
atth
eco
mm
uti
ng
zone
leve
lb
etw
een
2000
and
2006
.R
egio
ns
range
from
ligh
tta
n(l
arge
stp
osit
ive
shock
quin
tile
)to
dar
kre
d(l
arge
stneg
ativ
esh
ock
quin
tile
).G
rey
regi
ons
are
ones
lack
ing
dat
a.
37
Fig
ure
2:C
omm
uti
ng
Zon
eB
arti
kQ
uin
tile
s
The
map
plo
tsquin
tile
sof
the
Bar
tik
shock
bet
wee
n20
00an
d20
06,
whic
his
the
chan
gein
man
ufa
cturi
ng
atth
enat
ional
leve
lin
tera
cted
wit
hth
ere
gion
alco
mp
osit
ion
ofth
em
anufa
cturi
ng
sect
orat
the
com
muti
ng
zone
leve
l.G
rey
regi
ons
are
ones
lack
ing
dat
a.
38
Table I: Summary statistics for commuting zones (2000 to 2007)This table reports the summary statistics for all commuting zone-year observations in our sample, from 2000to 2007. For each variable, we show the pooled average, standard deviation, 25th, 50th and 75th percentiles.We use the 2000 Department of Agriculture definition of Commuting Zones (CZs). Income in a CZ is definedas the total CZ wages and salaries, extracted from the county-level IRS Statistics of Income. Population,number of households, and the percentage of 25yr+ with a high school (bachelor’s) degree are all obtainedfrom the 2000 Census and aggregated from the county-level to the CZ-level. Total Labor Force is obtainedfrom the Bureau of Labor Statistics. Banking Sector variables are calculated from the FDIC Summary ofDeposits. HHI is the CZ-level Herfindahl index of the banking sector, calculated using the shares of deposits,% of Large Banks is the percentage of the CZ deposits concentrated in Top 30 largest U.S. banks. % ofLocal Banks is the percentage of deposits concentrated in “local” banks (defined in detail in the Section 2).Employment is calculated from the QWI data published by the LEHD program in Census. Non-tradablesector includes two-digit NAICS 44-45 (Retail Trade) and two-digit NAICS 72 (Accommodation and FoodServices); Construction sector is two-digit NAICS 23.
N Mean Std.Dev p25 p50 p75Number of Counties in the Commuting Zone 4,018 4.95 2.55 3 5 62yr Income Growth (Total Wages and Salaries) 4,018 0.03 0.05 0 0.02 0.05Manuf. Employment Bartik 4,018 -0.01 0.01 -0.01 -0.01 0Import Bartik 4,018 -0.01 0.01 -0.01 0 0Population as of 2000 4,018 501,277 1,160,715 82,320 164,116 438,578Total Labor Force 4,018 254,496 586,588 38,346 81,254 212,212Household as of 2000 4,018 187,540 410,958 30,691 62,954 166,610% of 25yr+ with High School Degree 4,018 79.12 7.20 74.51 80.85 84.15% of 25yr+ with Bachelor’s Degree 4,018 18.99 6.46 14.29 17.56 22.32Non-tradable Employment (Aggregate) 4,018 41,849 102,828 3,677 10,781 32,759Non-tradable Employment (Startups) 4,018 2,503 6,530 222 644 1,972Non-tradable Employment (2-3 year-olds) 4,018 2,153 5,561 197 553 1,684Non-tradable Employment (4-5 year-olds) 4,018 1,778 4,498 168 472 1,424Non-tradable Employment (6+ year-olds) 4,018 35,416 86,563 2,982 9,072 27,372Construction Employment (Aggregate) 3,660 11,774 27,671 960 2,866 9,688Construction Employment (Startups) 3,660 883 2,142 76 224 714Construction Employment (2-3 year-olds) 3,660 896 2,202 73 213 716Construction Employment (4-5 year-olds) 3,660 810 1,966 65 192 633Construction Employment (6+ year-olds) 3,660 9,185 21,539 731 2,186 7,388Banking Sector HHI 4,018 0.13 0.07 0.09 0.12 0.15% of Deposit from Large Banks 4,018 38% 23% 19% 35% 57%% of Local Banks 4,018 30% 17% 17% 29% 43%
39
Table II: Job creation and firm age (Non-tradable sector)
This table summarizes the two-year job creation in the non-tradable sector (NAICS2 = 44, 45, 72)in a commuting zone (CZ), sorted by firm age. The data is extracted from the QWI data publishedby the LEHD program in Census, and is calculated by exploiting the mechanical transition of firmsacross firm age categories (details in Section 2). Panel A reports the average number of jobs createdby CZ in each age category and year, and Panel B shows the average net job creation by CZ and yearscaled by the total employment in the non-tradable sector in the CZ as of 2000.
Panel A: Commuting Zone Level, Raw Job CreationYear N 0-1 yrs 2-3 yrs 4-5 yrs 6+ yrs Total2000 403 2,777.16 -152.27 -245.22 -502.68 1,876.972001 420 2,274.64 -415.00 -344.61 -2,373.11 -857.952002 482 2,344.32 -602.12 -509.53 -2,378.77 -1,146.152003 520 2,381.61 -299.94 -467.26 -430.83 1,183.542004 535 2,491.71 -242.06 -286.83 -746.37 1,216.432005 549 2,465.05 -342.49 -213.63 -1,086.20 822.552006 554 2,736.48 -326.86 -261.71 -1,109.27 1,038.652007 555 2,546.62 -195.00 -262.79 -1,106.57 982.43Pooled Average 2,503.41 -320.73 -321.98 -1,193.18 667.52
Panel B: Commuting Zone Level, Scaled by 2000 EmploymentYear N 0-1 yrs 2-3 yrs 4-5 yrs 6+ yrs Total2000 403 5.19% -0.48% -0.57% -1.85% 2.28%2001 420 4.52% -0.78% -0.67% -4.92% -1.85%2002 482 4.33% -0.96% -0.79% -4.27% -1.68%2003 520 4.50% -0.57% -0.87% -1.14% 1.91%2004 535 4.66% -0.49% -0.50% -1.90% 1.77%2005 549 4.89% -0.69% -0.49% -2.43% 1.28%2006 554 5.22% -0.73% -0.49% -2.24% 1.77%2007 555 4.64% -0.56% -0.51% -1.88% 1.68%Pooled Average 4.74% -0.66% -0.60% -2.51% 0.97%
40
Tab
leII
I:Job
crea
tion
and
loca
lin
com
esh
ock
s:non
-tra
dab
lese
ctor
Th
ista
ble
show
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
loca
lin
com
egro
wth
.W
eru
nre
gre
ssio
ns
for
the
aggre
gate
chan
ge
inem
plo
ym
ent
an
dfo
rth
ech
an
ge
inem
plo
ym
ent
in4
age
cate
gori
es(s
tart
up
s,2-3
,4-5
,an
d6+
yea
rsold
).O
bse
rvati
on
sare
at
the
CZ
-yea
r-fi
rmage
level
.T
he
dep
end
ent
vari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inth
en
on
-tra
dab
lese
ctor
(NA
ICS
2=
44,
45,
72)
over
the
pre
vio
us
two
yea
rscr
eate
din
firm
sof
each
age
cate
gory
,an
dth
isvari
ab
leis
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eC
Zas
of
2000.
Inco
me
gro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
.W
ein
stru
men
tfo
rth
isvari
able
usi
ng
the
Bart
ikm
anu
fact
uri
ng
shock
,w
hic
hin
tera
cts
chan
ges
inn
ati
onw
ide
emp
loym
ent
inth
em
anu
fact
uri
ng
sect
or
wit
hth
ep
reex
isti
ng
manu
fact
uri
ng
com
posi
tion
ina
CZ
.T
he
an
aly
sis
isp
erfo
rmed
on
a“n
on
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Colu
mn
(1)
rep
ort
sth
efi
rst-
stage
regre
ssio
nof
inco
me
gro
wth
on
the
Bart
ikin
stru
men
t.C
olu
mn
(2)
isth
eO
LS
regre
ssio
nof
net
emp
loym
ent
chan
ge
inth
eC
Zon
loca
lin
com
egro
wth
,an
dco
lum
n(3
)is
the
2S
LS
regre
ssio
nw
ith
inst
rum
ente
din
com
egro
wth
.C
olu
mn
s(4
)to
(11)
per
form
sim
ilar
regre
ssio
ns
as
colu
mn
s(2
)an
d(3
)fo
rfi
rms
of
diff
eren
tages
.C
ontr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ud
eyea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
eses
an
dst
an
dard
erro
rsare
clu
ster
edby
CZ
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
Non
-over
lap
pin
gS
am
ple
(01,
03,
05,
07)
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.1
46***
(8.3
92)
Inco
me
Gro
wth
0.3
00***
0.3
06***
0.1
19***
0.2
74***
0.0
06
-0.0
35
0.0
00
-0.0
11
0.1
75**
0.0
79
(2.9
53)
(2.8
74)
(7.3
33)
(5.0
19)
(0.5
57)
(-1.2
23)
(0.0
54)
(-0.5
61)
(2.1
47)
(0.7
41)
ln(T
ota
lL
ab
orf
orc
e)-0
.119***
-0.0
31*
-0.0
30
0.0
11
0.0
30***
-0.0
04
-0.0
09*
-0.0
01
-0.0
02
-0.0
37***
-0.0
49***
(-6.7
86)
(-1.9
07)
(-1.6
06)
(1.3
85)
(2.9
15)
(-1.1
95)
(-1.9
25)
(-0.3
07)
(-0.6
71)
(-2.6
04)
(-2.6
54)
%H
igh
sch
ool
Ed
u-0
.657***
-0.0
83
-0.0
82
0.7
30***
0.7
74***
-0.0
89*
-0.1
01**
-0.0
11
-0.0
14
-0.7
14***
-0.7
41***
(-3.1
81)
(-0.5
16)
(-0.5
17)
(5.2
05)
(5.7
30)
(-1.9
04)
(-2.0
92)
(-0.3
18)
(-0.4
10)
(-5.3
81)
(-5.5
33)
ln(T
ota
lC
ZW
ages
)0.1
06***
0.0
30**
0.0
29*
-0.0
07
-0.0
25***
0.0
03
0.0
08*
0.0
00
0.0
01
0.0
34***
0.0
44***
(7.0
47)
(2.0
48)
(1.7
12)
(-1.0
36)
(-2.6
36)
(1.1
32)
(1.8
44)
(0.0
42)
(0.4
75)
(2.6
55)
(2.6
54)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
R-s
qu
are
d0.2
69
0.1
35
0.1
35
0.1
41
0.0
46
0.0
09
0.0
35
0.0
31
0.0
95
0.0
89
F-S
tati
stic
s70.4
3
41
Tab
leIV
:Job
crea
tion
and
loca
lin
com
esh
ock
s:m
ediu
m-t
erm
anal
ysi
s
Th
ista
ble
show
sre
gre
ssio
ns
of
lon
g-t
erm
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
lon
ger
-ter
mm
easu
res
of
loca
lin
com
egro
wth
.O
bse
rvati
on
sare
at
the
CZ
-firm
age
level
.T
he
dep
end
ent
vari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inth
en
on
-tra
dab
lese
ctor
(NA
ICS
2=
44,
45,
72)
bet
wee
n2001
an
d2007
infi
rms
of
each
age
cate
gory
,an
dth
isvari
ab
leis
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eC
Zas
of
2000.
Inco
me
gro
wth
isth
egro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
bet
wee
n2001
an
d2007.
We
inst
rum
ent
for
this
vari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
,w
hic
hin
tera
cts
chan
ges
inn
ati
onw
ide
emp
loym
ent
inth
em
anu
fact
uri
ng
sect
or
wit
hth
ep
reex
isti
ng
manu
fact
uri
ng
com
posi
tion
ina
CZ
.C
olu
mn
(1)
rep
ort
sth
efi
rst-
stage
regre
ssio
nof
inco
me
gro
wth
on
the
Bart
ikin
stru
men
t.C
olu
mn
(2)
isth
eO
LS
regre
ssio
nof
net
emp
loym
ent
chan
ge
inth
eC
Zon
loca
lin
com
egro
wth
,an
dco
lum
n(3
)is
the
2S
LS
regre
ssio
nw
ith
inst
rum
ente
din
com
egro
wth
.C
olu
mn
s(4
)to
(7)
per
form
sim
ilar
regre
ssio
ns
as
colu
mn
s(2
)an
d(3
)fo
rfi
rms
of
diff
eren
tages
.C
ontr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
Het
erosk
edast
icit
y-r
ob
ust
t-st
ati
stic
sare
show
nin
pare
nth
esis
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
Job
Cre
ati
on
from
2001
to2007
Aggre
gate
0-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.5
34***
(10.3
76)
Inco
me
Gro
wth
0.4
47***
0.3
00**
0.2
22***
0.3
39***
0.2
25***
-0.0
38
(8.0
65)
(2.2
69)
(5.7
20)
(3.9
99)
(4.7
89)
(-0.3
04)
ln(T
ota
lL
ab
orf
orc
e)-0
.386***
-0.0
43
-0.1
08
0.0
15
0.0
66
-0.0
58
-0.1
74**
(-10.1
31)
(-1.1
55)
(-1.5
18)
(0.5
43)
(1.4
80)
(-1.5
39)
(-2.3
96)
%H
igh
sch
ool
Ed
u-2
.461***
-0.5
14
-0.7
14
1.6
01***
1.7
59***
-2.1
15***
-2.4
72***
(-4.6
28)
(-0.9
50)
(-1.2
97)
(4.3
16)
(4.5
37)
(-4.7
15)
(-5.4
04)
ln(T
ota
lC
ZW
ages
)0.3
42***
0.0
48
0.1
06*
-0.0
07
-0.0
53
0.0
55
0.1
58**
(10.1
24)
(1.4
55)
(1.6
67)
(-0.2
82)
(-1.3
30)
(1.6
37)
(2.4
40)
Ob
serv
ati
on
s543
543
543
543
543
543
543
R-s
qu
are
d0.4
25
0.2
53
0.2
38
0.1
87
0.1
64
0.1
17
0.0
55
F-S
tati
stic
s107.6
6
42
Tab
leV
:Job
crea
tion
and
loca
lin
com
esh
ock
s:co
nst
ruct
ion
sect
or
Th
ista
ble
show
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
loca
lin
com
egro
wth
,u
sin
gd
ata
from
the
con
stru
ctio
nse
ctor
(NA
ICS
2=
23).
We
run
regre
ssio
ns
for
the
aggre
gate
chan
ge
inem
plo
ym
ent
an
dfo
rth
ech
an
ge
inem
plo
ym
ent
in4
age
cate
gori
es(s
tart
up
s,2-3
,4-5
,an
d6+
yea
rsold
).O
bse
rvati
on
sare
at
the
CZ
-yea
r-fi
rmage
level
.T
he
dep
end
ent
vari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inth
eco
nst
ruct
ion
sect
or
over
the
pre
vio
us
two
yea
rscr
eate
din
firm
sof
each
age
cate
gory
,an
dth
isvari
ab
leis
scale
dby
the
tota
lco
nst
ruct
ion
emp
loym
ent
inth
eC
Zas
of
2000.
Inco
me
gro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
.W
ein
stru
men
tfo
rth
isvari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
,w
hic
hin
tera
cts
chan
ges
inn
ati
onw
ide
emp
loym
ent
inth
em
anu
fact
uri
ng
sect
or
wit
hth
ep
reex
isti
ng
manu
fact
uri
ng
com
posi
tion
ina
CZ
.T
he
an
aly
sis
isp
erfo
rmed
on
a“n
on
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Colu
mn
(1)
rep
ort
sth
efi
rst-
stage
regre
ssio
nof
inco
me
gro
wth
on
the
Bart
ikin
stru
men
t.C
olu
mn
(2)
isth
eO
LS
regre
ssio
nof
net
emp
loym
ent
change
inth
eC
Zon
loca
lin
com
egro
wth
,an
dco
lum
n(3
)is
the
2S
LS
regre
ssio
nw
ith
inst
rum
ente
din
com
egro
wth
.C
olu
mn
s(4
)to
(11)
per
form
sim
ilar
regre
ssio
ns
as
colu
mn
s(2
)an
d(3
)fo
rfi
rms
of
diff
eren
tages
.C
ontr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ude
yea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
eses
an
dst
an
dard
erro
rsare
clu
ster
edby
CZ
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.2
30***
(8.2
42)
Inco
me
Gro
wth
0.8
13***
0.8
42***
0.2
92***
0.5
92***
0.0
78***
-0.0
61
0.0
37**
-0.0
38
0.4
06***
0.3
51**
(3.5
13)
(4.4
23)
(3.4
16)
(6.4
40)
(3.2
23)
(-0.9
33)
(2.3
51)
(-0.8
90)
(3.3
34)
(2.0
63)
ln(T
ota
lL
ab
orf
orc
e)-0
.117***
0.0
03
0.0
06
-0.0
24
0.0
12
0.0
14*
-0.0
02
0.0
03
-0.0
06
0.0
10
0.0
04
(-6.5
84)
(0.0
57)
(0.1
58)
(-1.3
91)
(0.6
62)
(1.8
77)
(-0.2
50)
(0.5
19)
(-0.9
37)
(0.2
31)
(0.0
91)
%H
igh
sch
ool
Ed
u-0
.878***
0.9
23*
0.9
37**
0.3
26
0.4
71
0.4
89**
0.4
21**
-0.0
63
-0.0
99
0.1
71
0.1
44
(-3.8
89)
(1.8
14)
(1.9
96)
(1.0
19)
(1.6
33)
(2.4
76)
(2.3
55)
(-0.7
57)
(-1.1
95)
(0.3
23)
(0.2
92)
ln(T
ota
lC
ZW
ages
)0.1
04***
-0.0
05
-0.0
08
0.0
21
-0.0
12
-0.0
12*
0.0
03
-0.0
03
0.0
05
-0.0
11
-0.0
05
(6.7
97)
(-0.1
19)
(-0.2
41)
(1.3
04)
(-0.7
64)
(-1.7
84)
(0.3
73)
(-0.6
98)
(0.8
03)
(-0.2
79)
(-0.1
35)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s1,8
60
1,8
60
1,8
60
1,8
60
1,8
60
1,8
60
1,8
60
1,8
60
1,8
60
1,8
60
1,8
60
R-s
qu
are
d0.2
61
0.1
34
0.1
34
0.1
40
0.0
43
0.0
34
-0.0
11
0.0
35
0.0
01
0.0
47
0.0
46
F-S
tati
stic
s67.9
3
43
Table VI: Job growth by firm age in good and bad times
Panel A of this table shows regressions of net employment creation in the non-tradable sector on instrumented localincome growth for areas with above median income growth. All regressions include the same controls as the previoustables. Panel B shows regressions of net employment creation in the non-tradable sector on the manufacturingemployment Bartik variable. The analysis is performed on a categorized version of the instrument divided intoterciles. Observations are at the CZ-year-firm age category level. The dependent variable is the net change inemployment in the non-tradable sector (NAICS2 = 44, 45, 72) over the previous two years created in firms of eachage category, and this variable is scaled by the total non-tradable employment in the CZ as of 2000. Both panelsuse a non-overlapping sample of years 2001, 2003, 2005, and 2007. All regressions include the same controls as theprevious tables. T-statistics in parentheses are based on standard errors clustered at the CZ level. *, **, *** denotestatistical significance at the 10, 5 and 1% levels, respectively.
Panel A: Areas with above-median income growth, non-overlapping sample (01, 03, 05, 07)First Stage Aggregate 0-1 year-olds 2-3 year-olds 4-5 year-olds 6+ year-olds
(1) (2) (3) (4) (5) (6)
Manuf. Employment Bartik 1.320***(6.390)
Income Growth 0.400* 0.245** -0.011 0.016 0.150(1.868) (2.459) (-0.261) (0.455) (0.703)
Controls Yes Yes Yes Yes Yes YesYear FE Yes Yes Yes Yes Yes YesObservations 1,022 1,022 1,022 1,022 1,022 1,022R-squared 0.123 0.067 0.076 0.009 0.002 0.069F-Statistics 40.83
Panel B: Categorized IV Variable, non-overlapping sample (01, 03, 05, 07)Aggregate 0-1 year-olds 2-3 year-olds 4-5 year-olds 6+ year-olds
(1) (2) (3) (4) (5)
DummyMedianBartik 0.007** 0.005*** 0.000 -0.000 0.003(2.277) (3.436) (0.353) (-0.756) (0.886)
DummyHighBartik 0.009*** 0.008*** -0.001 0.000 0.001(2.806) (4.685) (-1.191) (0.207) (0.505)
Controls Yes Yes Yes Yes YesYear FE Yes Yes Yes Yes YesObservations 2,044 2,044 2,044 2,044 2,044R-squared 0.081 0.104 0.010 0.035 0.073
44
Table VII: Job creation and local income shocks in high and low house price appreciationareas
This table shows regressions of net employment creation at the commuting zone (CZ) level on local income growth interactedwith local housing market conditions during the 2000-2007 period. Observations are at the CZ-year-firm age level. Thedependent variable is the net change in employment in the non-tradable sector (NAICS2 = 44, 45, 72) over the previous twoyears created in startup firms (0-1 years old), and this variable is scaled by the total non-tradable employment in the CZ as of2000. Income growth is the two-year growth of total wages and salaries in the CZ. We instrument for income growth and itsinteraction with the local housing market conditions using the Bartik manufacturing shock and its interaction with the housingvariables. We perform the analysis on a “non-overlapping” sample of years 2001, 2003, 2005, and 2007. We use local houseprice growth and local housing supply elasticity as local housing market variables. Local house price growth is the two-yearhouse price index growth provided by the Federal Housing Finance Agency; Saiz elasticity is provided by Saiz (2010) and itmeasures the geographic and regulatory constraints to the supply of housing in 269 MSAs in the U.S. The sample is categorizedinto “high” and “low” by the median of each variable. Control variables are extracted from the 2000 Census and the Bureau ofLabor Statistics but omitted here for brevity. All regressions include year fixed effects. T-statistics are shown in parenthesis.Standard errors are clustered by CZ. *, **, *** denote statistical significance at the 10, 5 and 1% levels, respectively.
Job creation in startups, in high and low house price appreciation (HPA) areas(1) (2) (3) (4)
HPA = High House Price Growth HPA = Low Saiz Elasticity
0-1 yrs 0-1 yrs1st Stage IV 1st Stage IV
Manuf. Employment Bartik 1.268*** 2.017***(7.848) (9.726)
I(High HPA) 0.033*** -0.002 0.015*** 0.003(6.239) (-1.103) (2.881) (1.011)
Income Growth 0.335*** 0.209***(4.447) (3.096)
Income Growth × I(High HPA) -0.052 0.067(-0.773) (1.095)
Controls Yes Yes Yes YesYear FE Yes Yes Yes YesObservations 1,155 1,155 892 892R-squared 0.284 0.239F-Statistics 41.47 48.73
45
Table VIII: Startup job creation and job resilience
This table shows the share of employment of four cohorts of new firms (between 2000 and 2003) and asks how many jobsremain after two or four years in those firms. The table shows the number of employees in these cohorts of firms at the timethat they are started, two years later, and four years later. Employment is scaled by the total employment in each CZ as of2000. The sample includes only firms until 2003 because that is the last year that we can track firms for a full four years.T-statistics for the difference between high and low shock areas are shown in parentheses on the last line of each panel.
Panel A: Bartik Shocks (2000-2003 cohorts)
Job creation from startups Jobs remaining after 2 years Jobs remaining after 4 yearsLow Bartik Area 3.93% 3.34% 2.92%Medium Bartik Area 4.84% 4.12% 3.55%High Bartik Area 5.06% 4.38% 3.87%High Bartik−Low Bartik 1.13%*** 1.04%*** 0.95%***t-statistics (8.72) (8.93) (8.53)
Panel B: Income Shocks (2000-2003 cohorts)
Job creation from startups Jobs remaining after 2 years Jobs remaining after 4 yearsLow ∆Income Area 4.19% 3.64% 3.17%Medium ∆Income Area 4.46% 3.78% 3.29%High ∆Income Area 5.18% 4.42% 3.89%High ∆Income−Low ∆Income 0.98%*** 0.78%*** 0.72%***t-statistics (7.58) (6.86) (6.69)
46
Tab
leIX
:Job
Cre
atio
nan
dJob
Des
truct
ion
Th
ista
ble
show
sre
gre
ssio
ns
of
gro
ssem
plo
ym
ent
crea
tion
an
dd
estr
uct
ion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
loca
lin
com
egro
wth
.W
eco
nst
ruct
gro
ssjo
bcr
eati
on
an
dd
estr
uct
ion
over
two-y
ear
per
iod
sby
ad
din
gquart
erly
hir
esan
dse
para
tion
sfr
om
the
LE
HD
QW
Id
ata
set
for
fou
rfi
rmage
cate
gori
es(s
tart
up
s,2-3
,4-5
,an
d6+
yea
rsold
).O
bse
rvati
on
sare
at
the
CZ
-yea
r-fi
rmage
level
.A
llem
plo
ym
ent
vari
ab
les
are
for
the
non
-tra
dab
lese
ctor
(NA
ICS
2=
44,
45,
72),
an
dth
eyare
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eC
Zas
of
2000.
Inco
me
gro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
.W
ein
stru
men
tfo
rth
isvari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
.T
he
an
aly
sis
isp
erfo
rmed
on
a“n
on
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Colu
mn
(1)
rep
ort
sth
efi
rst-
stage
regre
ssio
nof
inco
me
gro
wth
on
the
Bart
ikin
stru
men
t.C
olu
mn
(2)
isth
eO
LS
regre
ssio
nof
net
emp
loym
ent
chan
ge
inth
eC
Zon
loca
lin
com
egro
wth
,an
dco
lum
n(3
)is
the
2S
LS
regre
ssio
nw
ith
inst
rum
ente
din
com
egro
wth
.C
olu
mn
s(4
)to
(11)
per
form
sim
ilar
regre
ssio
ns
as
colu
mn
s(2
)an
d(3
)fo
rfi
rms
of
diff
eren
tages
.A
llre
gre
ssio
ns
incl
ud
eyea
rfi
xed
effec
tsan
dco
ntr
ol
vari
ab
les
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics
that
mim
icth
ose
inT
ab
leII
I.T
-sta
tist
ics
are
show
nin
pare
nth
eses
an
dst
an
dard
erro
rsare
clu
ster
edby
CZ
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
PanelA
:G
ross
Job
Cre
ati
on
Manu
f.E
mp
loym
ent
Bart
ik1.1
57***
(8.4
70)
Inco
me
Gro
wth
0.4
59***
1.9
85***
0.0
90***
0.2
82***
0.0
58***
0.2
31***
0.0
43***
0.2
70***
0.2
67***
1.1
99***
(6.8
67)
(6.1
76)
(5.4
94)
(4.5
67)
(5.3
60)
(4.8
67)
(5.2
73)
(4.7
54)
(6.8
64)
(6.0
98)
PanelB
:G
ross
Job
Des
tru
ctio
n
Manu
f.E
mp
loym
ent
Bart
ik1.1
57***
(8.4
70)
Inco
me
Gro
wth
0.4
13***
1.8
68***
0.0
66***
0.1
79***
0.0
58***
0.2
16***
0.0
40***
0.2
34***
0.2
48***
1.2
35***
(6.2
63)
(6.2
50)
(5.4
40)
(3.8
04)
(5.3
05)
(4.7
05)
(4.5
45)
(5.1
73)
(6.0
62)
(6.2
40)
PanelC
:N
etJob
Cre
ati
on
Manu
f.E
mp
loym
ent
Bart
ik1.1
57***
(8.4
70)
Inco
me
Gro
wth
0.0
46***
0.1
18**
0.0
24***
0.1
03***
-0.0
00
0.0
15
0.0
03
0.0
36
0.0
19**
-0.0
36
(4.5
19)
(2.1
78)
(3.9
21)
(3.8
07)
(-0.0
20)
(1.0
20)
(0.9
36)
(1.3
24)
(2.5
69)
(-1.3
70)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
32
2,0
32
2,0
32
2,0
32
2,0
32
2,0
32
2,0
32
2,0
32
2,0
32
2,0
32
2,0
32
F-S
tati
stic
s71.7
4
47
Table X: Local bank share and employment in different-aged firms
This table analyzes the relation between the share of deposits held by “local banks” and the employment distribution acrossyoung and old firms. If 75% or more of a bank’s deposits are concentrated in one CZ we define this bank as “local”. The localbank share is the percentage of total deposits in a CZ held by “local” banks. In order to mitigate the effect of labor marketdynamics on the evolution of the local banking sector, we calculate a time-invariant CZ-level measure—Local Bank Share—bycalculating the time-series median of the share of local banks in the CZ. In column (1), the dependent variable is the time-seriesmedian level of the share of employment in firms ≤ 1 year-old in the CZ, and in column (2) the dependent variable is thetime-series median level of employment share in firms ≥ 6 years old. Control variables are extracted from the 2000 Census.Both regressions include state fixed-effects. Heteroskedasticity-robust t-statistics are in parenthesis. *, **, *** denote statisticalsignificance at the 10, 5 and 1% levels, respectively.
(1) (2)% of Firms 5 1 Year Old % of Firms = 6 Years Old
% of Local Banks 0.016** -0.041***(2.344) (-2.852)
ln(Total Laborforce) 0.007 0.003(0.889) (0.147)
% Highschool Edu 0.972*** -2.147***(5.280) (-4.782)
ln(Total CZ Wages) -0.010 0.008(-1.377) (0.459)
State FE Yes YesNumber of CZs 555 555R-squared 0.287 0.311
48
Tab
leX
I:A
cces
sto
finan
cean
dth
ere
spon
seof
emplo
ym
ent
crea
tion
tosh
ock
s
Th
ista
ble
show
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
loca
lin
com
egro
wth
,an
ind
icato
rfo
rab
ove
med
ian
share
of
dep
osi
tsh
eld
by
loca
lb
an
ks,
an
dth
ein
tera
ctio
nof
the
two
vari
ab
les.
Loca
lb
an
ks
are
defi
ned
as
those
wit
h75%
or
more
of
its
dep
osi
tsin
on
eC
Z.
I (H
igh
LB)
isa
du
mm
yvari
ab
leeq
ual
to1
ifth
eC
Z’s
loca
lb
an
ksh
are
ish
igh
erth
an
the
med
ian
of
all
CZ
s.O
bse
rvati
on
sare
at
the
CZ
-yea
r-fi
rmage
level
.T
he
dep
end
ent
vari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inth
en
on
-tra
dab
lese
ctor
(NA
ICS
2=
44,
45,
72)
over
the
pre
vio
us
two
yea
rscr
eate
din
firm
sof
each
age
cate
gory
,an
dth
isvari
ab
leis
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eC
Zas
of
2000.
Inco
me
gro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
.W
ein
stru
men
tfo
rth
isvari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
,w
hic
hin
tera
cts
chan
ges
inn
ati
onw
ide
emp
loym
ent
inth
em
anu
fact
uri
ng
sect
or
wit
hth
ep
reex
isti
ng
manu
fact
uri
ng
com
posi
tion
ina
CZ
.T
he
inte
ract
ion
term
Inco
me
Gro
wth×I (
Hig
hLB)
cap
ture
sth
ero
leof
loca
lb
an
ksh
are
son
the
resp
on
siven
ess
of
firm
sin
crea
tin
gjo
bs.
An
aly
sis
isp
erfo
rmed
on
a“n
on
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Colu
mn
(1)
rep
ort
sth
efi
rst-
stage
regre
ssio
nof
inco
me
gro
wth
on
the
Bart
ikin
stru
men
t.C
olu
mn
(2)
to(6
)are
IVre
gre
ssio
nfo
rd
iffer
ent
firm
age
cate
gori
es.
Contr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ud
eyea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
esis
.S
tan
dard
erro
rsare
clu
ster
edby
CZ
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
1st
Sta
ge
IVIV
IVIV
IV
Manu
f.E
mp
loym
ent
Bart
ik1.2
47***
(6.0
69)
Inco
me
Gro
wth
0.3
05***
0.2
17***
-0.0
32
-0.0
19
0.1
40
(2.5
84)
(3.8
29)
(-1.2
71)
(-0.9
03)
(1.1
85)
Inco
me
Gro
wth×I (
Hig
hLB)
-0.0
24
0.1
03*
-0.0
15
0.0
18
-0.1
31
(-0.1
89)
(1.7
65)
(-0.5
02)
(0.7
96)
(-1.0
30)
I (H
igh
LB)
-0.0
09***
-0.0
03
-0.0
05**
-0.0
01
-0.0
00
0.0
03
(-2.6
59)
(-0.7
88)
(-2.4
79)
(-1.2
93)
(-0.4
82)
(0.6
46)
ln(T
ota
lL
ab
orf
orc
e)-0
.118***
-0.0
31*
0.0
29***
-0.0
09**
-0.0
02
-0.0
48***
(-6.7
76)
(-1.6
55)
(2.7
33)
(-1.9
95)
(-0.7
13)
(-2.5
90)
%H
igh
sch
ool
Ed
u-0
.636***
-0.0
79
0.7
60***
-0.0
99**
-0.0
16
-0.7
24***
(-3.1
74)
(-0.4
95)
(5.6
89)
(-2.0
78)
(-0.4
75)
(-5.3
58)
ln(T
ota
lC
ZW
ages
)0.1
05***
0.0
30*
-0.0
23**
0.0
08*
0.0
01
0.0
43***
(7.0
05)
(1.7
50)
(-2.4
79)
(1.9
00)
(0.5
23)
(2.5
93)
Yea
rF
EY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
R-s
qu
are
d0.2
74
0.1
36
0.0
76
0.0
29
0.0
90
F-S
tati
stic
s36.8
3
49
Table XII: Firm size and firm age
Panel A (B) summarizes the average regional employment (number of firms) tabulated by firm age and firm size, from 2000to 2007. Observations are at the MSA-year-firm age-firm size level. Data is from the Census Business Dynamics Statistics.Panel C summarizes the regressions of net employment creation by firm age and firm size at the MSA level (coefficients for allcontrol variables are shown in Table B.IV). The reported coefficients are from instrumental variables regressions of the changein employment in each of the four age categories and three firm size categories on local income growth. For each age-size pair,the dependent variable is the net change in employment in all the industries over the previous two years created in firms ineach age-size bin, and this variable is scaled by the total employment in the MSA as of 2000. Income growth is the two-yeargrowth of total wages and salaries in the MSA. We instrument for this variable using the Bartik manufacturing shock, whichinteracts changes in nationwide employment in the manufacturing sector with the preexisting manufacturing composition inthe MSA. The analysis is performed on a “non-overlapping” sample of years 2001, 2003, 2005, and 2007. Control variablesare extracted from the 2000 Census and the Bureau of Labor Statistics. All regressions include year fixed effects. T-statisticsare shown in parentheses and standard errors are clustered by MSA. *, **, *** denote statistical significance at the 10, 5 and1% levels, respectively.
Panel A: Total Employee in the Region Age
Aggregate 0-1 yrs 2-3 years 4-5 years 6+ years
Aggregate 263811 13359 11984 10607 227862<20 48810 8347 5789 4724 29950
Size 20-100 47301 3472 3893 3572 36363>100 167701 1540 2302 2310 161549
Panel B: Total Firms in the Region Age
Aggregate 0-1 yrs 2-3 years 4-5 years 6+ years
Aggregate 12390 2119 1460 1156 7654<20 10050 2026 1326 1028 5670
Size 20-100 1322 80 113 104 1025>100 1018 13 21 24 959
Panel C: Regression Coefficients Age
Aggregate 0-1 yrs 2-3 years 4-5 years 6+ years
Aggregate 1.247*** 0.408*** -0.004 -0.005 0.848***<20 0.152*** 0.314*** -0.059*** -0.031*** -0.073***
Size 20-100 0.181*** 0.079*** 0.034*** 0.016 0.053***>100 0.914*** 0.015*** 0.021*** 0.010 0.868***
50
Online Appendix—Not for Publication
Table A.I of the Online Appendix repeats the analysis in Table III using the overlappingsample (2000 to 2007). 20 The results produced from this bigger sample are similar to thoseusing the non-overlapping sample, with a larger difference between the coefficients of theyoungest and oldest firms. The coefficient of local income growth on startup job creation,estimated in column (5) is 0.286, which implies that a one standard deviation increase inthe income growth will bring 619 new jobs in startups in the non-tradable section per CZ.The effect for firms over 6 years old is statistically indistinguishable from zero.
The Online Appendix shows that the results are similar when we weight the regressions byCZ-level population and CZ-level total employment (Table A.II). The Appendix also showsthat the results are unchanged when we use alternative scaling for the net employmentcreation variable, namely scaling by the lagged total employment (i.e., as of t − 2) in thenon-tradable sector in each CZ, shown in Panel A of Table A.III, and scaling by the pooledaverage employment in the sector in the CZ over the years in the sample, Panel B of TableA.III).
We also show in the Online Appendix that the results are robust to using Census Divisionfixed effects (Table A.IV).21 This helps mitigate concerns that we are picking up generalregional trends in startup creation rates or in the dynamism of existing firms.
Finally, the Online Appendix also includes a regression where we use the fact that China’sascension to most-favored nation status in the World Trade Organization in 2000 induced asharp drop in U.S. manufacturing employment, especially in low-skilled, low-wage industries(Pierce and Schott (2012)). This drop, in turn, induced geographic variation in employmentresponses depending on the degree to which a region was exposed to the sectors that werehit hardest by Chinese import penetration (Autor, Dorn and Hanson, 2013).
This suggests an alternative “Import Bartik” that is constructed in the same vein asour main instrument (shown in equation (2)), except that we replace the change in nation-wide employment by industry with the change in import penetration in each industry. Theimport penetration measure is constructed as the net imports (total imports minus totalexports) over the total U.S. shipments for each four-digit NAICS manufacturing sub-sectorin each year. Results from this extension are reported in Table A.V. The F -statistic forthe first-stage regression is approximately 25, indicating that import penetration is also apowerful way to generate meaningful variation in manufacturing employment and to shocklocal income. The remaining columns echo the responsiveness by age category shown in theanalysis, both qualitatively and quantitatively.
20Clustering at the CZ level should largely account for the correlation in standard errors due to theoverlapping nature of the sample (we have to measure employment creation over two-year periods becauseof the way the QWI data are organized). Still, our main sample only uses non-overlapping observations toavoid this problem.
21There are 9 Census Divisions in the United States, all shown in detail at https://www.census.gov/
geo/maps-data/maps/pdfs/reference/us_regdiv.pdf.
51
Appendix A. Robustness Check
52
Tab
leA
.I:
Job
crea
tion
and
loca
lin
com
esh
ock
s:ov
erla
ppin
gsa
mple
from
2000
to20
07
Th
ista
ble
show
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
loca
lin
com
egro
wth
.W
eru
nre
gre
ssio
ns
for
the
aggre
gate
chan
ge
inem
plo
ym
ent
an
dfo
rth
ech
an
ge
inem
plo
ym
ent
in4
age
cate
gori
es(s
tart
up
s,2-3
,4-5
,an
d6+
yea
rsold
).O
bse
rvati
on
sare
at
the
CZ
-yea
r-fi
rmage
level
.T
he
dep
end
ent
vari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inth
en
on
-tra
dab
lese
ctor
(NA
ICS
2=
44,
45,
72)
over
the
pre
vio
us
two
yea
rscr
eate
din
firm
sof
each
age
cate
gory
,an
dth
isvari
ab
leis
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eC
Zas
of
2000.
Inco
me
gro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
.W
ein
stru
men
tfo
rth
isvari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
,w
hic
hin
tera
cts
chan
ges
inn
ati
onw
ide
emp
loym
ent
inth
em
anu
fact
uri
ng
sect
or
wit
hth
ep
reex
isti
ng
manu
fact
uri
ng
com
posi
tion
ina
CZ
.T
he
an
aly
sis
isp
erfo
rmed
on
an
over
lap
pin
gsa
mp
leof
2000
to2007.
Colu
mn
(1)
rep
ort
sth
efi
rst-
stage
regre
ssio
nof
inco
me
gro
wth
on
the
Bart
ikin
stru
men
t.C
olu
mn
(2)
isth
eO
LS
regre
ssio
nof
net
emp
loym
ent
chan
ge
inth
eC
Zon
loca
lin
com
egro
wth
,an
dco
lum
n(3
)is
the
2S
LS
regre
ssio
nw
ith
inst
rum
ente
din
com
egro
wth
.C
olu
mn
s(4
)to
(11)
per
form
sim
ilar
regre
ssio
ns
as
colu
mn
s(2
)an
d(3
)fo
rfi
rms
of
diff
eren
tages
.C
ontr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ud
eyea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
eses
an
dst
an
dard
erro
rsare
clu
ster
edby
CZ
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
Over
lap
pin
gsa
mp
le(2
000
to2007)
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manuf.
Em
plo
ym
ent
Bart
ik1.0
33***
(7.8
25)
Inco
me
Gro
wth
0.2
11**
0.1
76
0.1
12***
0.2
86***
-0.0
02
-0.0
78***
0.0
01
-0.0
11
0.1
00
-0.0
21
(2.1
76)
(1.6
03)
(6.3
35)
(4.8
23)
(-0.2
51)
(-3.5
60)
(0.1
49)
(-0.6
60)
(1.3
41)
(-0.1
87)
ln(T
ota
lL
ab
orf
orc
e)-0
.128***
-0.0
32**
-0.0
37*
0.0
16*
0.0
38***
-0.0
07**
-0.0
17***
-0.0
01
-0.0
02
-0.0
40***
-0.0
56***
(-8.9
13)
(-2.5
17)
(-1.9
03)
(1.7
91)
(3.3
02)
(-2.1
53)
(-4.2
72)
(-0.3
06)
(-0.7
67)
(-3.2
62)
(-2.9
02)
%H
igh
sch
ool
Ed
u-0
.597***
-0.1
81
-0.1
90
0.7
69***
0.8
13***
-0.1
02***
-0.1
22***
-0.0
68**
-0.0
71**
-0.7
80***
-0.8
11***
(-2.7
76)
(-1.1
75)
(-1.2
36)
(5.3
74)
(5.9
82)
(-2.7
58)
(-2.9
94)
(-2.2
84)
(-2.3
85)
(-6.0
51)
(-6.1
95)
ln(T
ota
lC
ZW
ages
)0.1
15***
0.0
31***
0.0
35**
-0.0
12
-0.0
32***
0.0
06**
0.0
15***
0.0
00
0.0
02
0.0
36***
0.0
51***
(9.0
68)
(2.7
20)
(2.0
29)
(-1.4
90)
(-3.0
60)
(2.0
75)
(4.1
50)
(0.0
58)
(0.5
86)
(3.3
44)
(2.9
24)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s4,0
18
4,0
18
4,0
18
4,0
18
4,0
18
4,0
18
4,0
18
4,0
18
4,0
18
4,0
18
4,0
18
R-s
qu
are
d0.2
27
0.1
14
0.1
13
0.1
44
0.0
25
0.0
22
0.0
33
0.0
29
0.0
69
0.0
57
F-S
tati
stic
s61.2
3
53
Tab
leA
.II:
Job
crea
tion
and
loca
lin
com
esh
ock
s:w
eigh
ted
regr
essi
ons
Th
ista
ble
show
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
loca
lin
com
egro
wth
.B
oth
the
OL
San
dth
e2S
LS
regre
ssio
ns
are
wei
ghte
dby
the
tota
lp
op
ula
tion
inth
eC
Zas
of
2000
(Pan
elA
)an
dto
tal
emp
loym
ent
inth
eC
Z(P
an
elB
).R
egre
ssio
ns
mim
icth
ose
inT
ab
leII
Iin
the
pap
er.
Th
ed
epen
den
tvari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inth
en
on
-tra
dab
lese
ctor
(NA
ICS
2=
44,
45,
72)
over
the
pre
vio
us
two
yea
rscr
eate
din
firm
sof
each
age
cate
gory
.In
com
egro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
.W
ein
stru
men
tfo
rth
isvari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
.T
he
an
aly
sis
isp
erfo
rmed
on
a“n
on
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Colu
mn
s(4
)to
(11)
per
form
sim
ilar
regre
ssio
ns
as
colu
mn
s(2
)an
d(3
)fo
rfi
rms
of
diff
eren
tages
.C
ontr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ud
eyea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
eses
an
dst
an
dard
erro
rsare
clu
ster
edby
CZ
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
PanelA
:W
eighte
dby
CZ
pop
ula
tion
in2000
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manuf.
Em
plo
ym
ent
Bart
ik1.2
13***
(8.3
77)
Inco
me
Gro
wth
0.2
90**
0.3
13***
0.1
16***
0.2
76***
0.0
04
-0.0
31
0.0
00
-0.0
14
0.1
70*
0.0
83
(2.5
83)
(3.1
66)
(7.4
01)
(5.4
06)
(0.3
59)
(-1.1
70)
(0.0
07)
(-0.8
12)
(1.8
85)
(0.8
46)
ln(T
ota
lL
ab
orf
orc
e)-0
.113***
-0.0
34*
-0.0
31*
0.0
13*
0.0
32***
-0.0
05
-0.0
09**
-0.0
01
-0.0
03
-0.0
41***
-0.0
51***
(-6.0
12)
(-1.9
62)
(-1.7
27)
(1.6
77)
(3.3
28)
(-1.3
80)
(-2.0
45)
(-0.4
11)
(-0.9
17)
(-2.6
87)
(-2.9
36)
%H
igh
sch
ool
Ed
u-0
.741***
-0.1
28
-0.1
20
0.6
93***
0.7
50***
-0.0
90**
-0.1
02**
-0.0
05
-0.0
10
-0.7
27***
-0.7
58***
(-3.4
70)
(-0.7
78)
(-0.7
53)
(5.1
03)
(5.8
24)
(-1.9
88)
(-2.2
12)
(-0.1
56)
(-0.3
14)
(-5.5
87)
(-5.6
88)
ln(T
ota
lC
ZW
ages
)0.1
01***
0.0
32**
0.0
30*
-0.0
09
-0.0
26***
0.0
04
0.0
08*
0.0
00
0.0
02
0.0
37***
0.0
46***
(6.2
90)
(2.1
00)
(1.8
44)
(-1.2
68)
(-2.9
84)
(1.3
13)
(1.9
55)
(0.1
34)
(0.7
01)
(2.7
25)
(2.9
26)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
R-s
qu
are
d0.2
52
0.1
39
0.1
39
0.1
49
0.0
37
0.0
10
0.0
39
0.0
33
0.0
98
0.0
92
F-S
tati
stic
s70.1
7
54
PanelB
:W
eighte
dby
loca
lem
plo
ym
ent
in2000 A
ggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.2
18***
(8.3
73)
Inco
me
Gro
wth
0.2
89**
0.3
13***
0.1
16***
0.2
76***
0.0
04
-0.0
31
-0.0
00
-0.0
15
0.1
69*
0.0
83
(2.5
61)
(3.1
71)
(7.3
93)
(5.4
30)
(0.3
45)
(-1.1
73)
(-0.0
02)
(-0.8
34)
(1.8
69)
(0.8
47)
ln(T
ota
lL
ab
orf
orc
e)-0
.113***
-0.0
34**
-0.0
31*
0.0
13*
0.0
32***
-0.0
05
-0.0
09**
-0.0
01
-0.0
03
-0.0
41***
-0.0
52***
(-5.9
76)
(-1.9
68)
(-1.7
35)
(1.6
97)
(3.3
46)
(-1.3
96)
(-2.0
61)
(-0.4
19)
(-0.9
35)
(-2.6
93)
(-2.9
46)
%H
igh
sch
ool
Ed
u-0
.747***
-0.1
32
-0.1
23
0.6
91***
0.7
48***
-0.0
89**
-0.1
01**
-0.0
04
-0.0
09
-0.7
30***
-0.7
61***
(-3.4
90)
(-0.8
00)
(-0.7
74)
(5.1
03)
(5.8
27)
(-1.9
73)
(-2.1
98)
(-0.1
29)
(-0.2
91)
(-5.5
96)
(-5.6
97)
ln(T
ota
lC
ZW
ages
)0.1
01***
0.0
32**
0.0
30*
-0.0
09
-0.0
26***
0.0
04
0.0
08**
0.0
00
0.0
02
0.0
37***
0.0
46***
(6.2
53)
(2.1
05)
(1.8
53)
(-1.2
86)
(-2.9
99)
(1.3
29)
(1.9
70)
(0.1
42)
(0.7
19)
(2.7
30)
(2.9
36)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
R-s
qu
are
d0.2
51
0.1
40
0.1
39
0.1
49
0.0
37
0.0
10
-0.0
08
0.0
39
0.0
33
0.0
98
0.0
92
F-S
tati
stic
s70.1
1
55
Tab
leA
.III
:Job
crea
tion
and
loca
lin
com
esh
ock
s:diff
eren
tsc
alin
gof
net
emplo
ym
ent
Th
ista
ble
show
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
loca
lin
com
egro
wth
.R
egre
ssio
ns
mim
icth
ose
inT
ab
leII
Iin
the
pap
er.
Th
ed
epen
den
tvari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inth
en
on
-tra
dab
lese
ctor
(NA
ICS
2=
44,
45,
72)
over
the
pre
vio
us
two
yea
rscr
eate
din
firm
sof
each
age
cate
gory
.T
his
vari
ab
leis
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eC
Zla
gged
by
two
yea
rs(P
an
elA
)an
dby
the
aver
age
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eC
Zb
etw
een
2000
an
d2007
(Pan
elB
).In
com
egro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
.W
ein
stru
men
tfo
rth
isvari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
.T
he
an
aly
sis
isp
erfo
rmed
on
a“n
on
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Contr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ud
eyea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
eses
an
dst
an
dard
erro
rsare
clu
ster
edby
CZ
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
PanelA
:Job
crea
tion
scale
dby
lagged
2-Y
ear
CZ
non
-tra
dab
leem
plo
ym
ent
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.1
46***
(8.3
92)
Inco
me
Gro
wth
0.2
89***
0.2
88***
0.1
09***
0.2
57***
0.0
06
-0.0
34
0.0
01
-0.0
09
0.1
73**
0.0
74
(2.9
31)
(2.7
17)
(7.6
43)
(4.8
23)
(0.6
03)
(-1.1
99)
(0.0
90)
(-0.4
84)
(2.1
51)
(0.7
04)
ln(T
ota
lL
ab
orf
orc
e)-0
.119***
-0.0
28*
-0.0
28
0.0
14*
0.0
33***
-0.0
04
-0.0
09**
-0.0
01
-0.0
02
-0.0
37***
-0.0
49***
(-6.7
86)
(-1.7
40)
(-1.4
89)
(1.9
26)
(3.2
59)
(-1.3
75)
(-2.0
53)
(-0.5
60)
(-0.7
99)
(-2.6
87)
(-2.6
65)
%H
igh
sch
ool
Ed
u-0
.657***
-0.0
49
-0.0
49
0.7
56***
0.7
97***
-0.0
95**
-0.1
06**
-0.0
15
-0.0
18
-0.6
95***
-0.7
22***
(-3.1
81)
(-0.3
17)
(-0.3
26)
(5.8
23)
(6.3
02)
(-2.1
30)
(-2.3
30)
(-0.4
68)
(-0.5
47)
(-5.3
43)
(-5.5
16)
ln(T
ota
lC
ZW
ages
)0.1
06***
0.0
27*
0.0
27
-0.0
11
-0.0
27***
0.0
04
0.0
08**
0.0
01
0.0
02
0.0
33***
0.0
44***
(7.0
47)
(1.8
78)
(1.5
90)
(-1.5
99)
(-3.0
05)
(1.3
05)
(1.9
70)
(0.3
07)
(0.6
14)
(2.7
41)
(2.6
71)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
R-s
qu
are
d0.2
69
0.1
31
0.1
31
0.1
40
0.0
45
0.0
11
0.0
38
0.0
36
0.0
98
0.0
91
F-S
tati
stic
s70.4
3
56
PanelB
:Job
crea
tion
scale
dby
aver
age
CZ
non
-tra
dab
leem
plo
ym
ent
bet
wee
n2000
an
d2007
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.1
46***
(8.3
92)
Inco
me
Gro
wth
0.2
91***
0.2
76***
0.0
80***
0.2
30***
0.0
12
-0.0
25
0.0
05
-0.0
05
0.1
94***
0.0
76
(3.0
16)
(2.7
21)
(3.9
66)
(4.5
41)
(1.5
19)
(-0.9
14)
(1.0
28)
(-0.2
78)
(2.6
46)
(0.7
52)
ln(T
ota
lL
ab
orf
orc
e)-0
.119***
-0.0
28*
-0.0
30*
0.0
10
0.0
29***
-0.0
04
-0.0
08*
-0.0
01
-0.0
02
-0.0
34***
-0.0
49***
(-6.7
86)
(-1.7
91)
(-1.6
50)
(1.3
78)
(2.9
36)
(-1.2
03)
(-1.8
23)
(-0.3
97)
(-0.6
98)
(-2.7
32)
(-2.7
45)
%H
igh
sch
ool
Ed
u-0
.657***
-0.0
54
-0.0
59
0.7
54***
0.7
96***
-0.0
91**
-0.1
02**
-0.0
20
-0.0
23
-0.6
97***
-0.7
30***
(-3.1
81)
(-0.3
63)
(-0.3
96)
(5.9
84)
(6.5
20)
(-2.0
73)
(-2.2
59)
(-0.6
28)
(-0.7
12)
(-5.4
52)
(-5.6
29)
ln(T
ota
lC
ZW
ages
)0.1
06***
0.0
27*
0.0
29*
-0.0
07
-0.0
24***
0.0
03
0.0
07*
0.0
00
0.0
01
0.0
31***
0.0
44***
(7.0
47)
(1.9
24)
(1.7
45)
(-1.0
83)
(-2.7
10)
(1.1
38)
(1.7
53)
(0.1
56)
(0.5
28)
(2.7
85)
(2.7
48)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
R-s
qu
are
d0.2
69
0.1
34
0.1
34
0.1
22
0.0
14
0.0
11
0.0
34
0.0
31
0.1
01
0.0
91
F-S
tati
stic
s70.4
3
57
Tab
leA
.IV
:Job
crea
tion
and
loca
lin
com
esh
ock
s:fixed
effec
tssp
ecifi
cati
on
Th
ista
ble
show
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
loca
lin
com
egro
wth
.T
hey
mim
icth
ere
gre
ssio
ns
inT
ab
leII
Iof
the
pap
er,
bu
tad
dC
ensu
sd
ivis
ion
fixed
effec
ts.
Th
ed
epen
den
tvari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inth
en
on
-tra
dab
lese
ctor
(NA
ICS
2=
44,
45,
72)
over
the
pre
vio
us
two
yea
rscr
eate
din
firm
sof
each
age
cate
gory
,an
dth
isvari
ab
leis
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eC
Zas
of
2000.
Inco
me
gro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
.W
ein
stru
men
tfo
rth
isvari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
.T
he
an
aly
sis
isp
erfo
rmed
on
a“n
on
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Contr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ud
eyea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
eses
an
dst
an
dard
erro
rsare
clu
ster
edby
CZ
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
Cen
sus
Div
isio
nF
ixed
Eff
ect
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manuf.
Em
plo
ym
ent
Bart
ik0.6
46***
(4.9
94)
Inco
me
Gro
wth
0.2
54**
0.0
50
0.0
87***
0.1
90**
0.0
07
-0.0
84
0.0
05
0.0
09
0.1
53*
-0.0
81
(2.4
46)
(0.2
44)
(6.8
10)
(1.9
69)
(0.5
69)
(-1.6
39)
(0.5
64)
(0.2
46)
(1.7
57)
(-0.4
10)
ln(T
ota
lL
ab
orf
orc
e)-0
.109***
-0.0
25
-0.0
48
0.0
17*
0.0
28**
-0.0
05
-0.0
15**
0.0
00
0.0
01
-0.0
37**
-0.0
63**
(-6.2
21)
(-1.5
62)
(-1.6
40)
(1.9
43)
(2.0
75)
(-1.3
00)
(-2.1
71)
(0.1
01)
(0.1
59)
(-2.5
33)
(-2.2
83)
%H
igh
sch
ool
Ed
u-0
.521**
0.1
42
0.0
63
0.8
77***
0.9
18***
-0.1
41**
-0.1
77***
-0.0
81**
-0.0
80*
-0.5
35***
-0.6
24***
(-2.3
13)
(0.6
75)
(0.2
66)
(4.9
62)
(5.2
58)
(-2.4
50)
(-2.7
33)
(-1.9
83)
(-1.8
50)
(-3.2
84)
(-3.2
48)
ln(T
ota
lC
ZW
ages
)0.1
01***
0.0
25*
0.0
46*
-0.0
13
-0.0
23*
0.0
04
0.0
14**
-0.0
01
-0.0
01
0.0
34***
0.0
59**
(6.6
92)
(1.7
14)
(1.7
09)
(-1.6
20)
(-1.8
51)
(1.2
35)
(2.1
08)
(-0.3
57)
(-0.2
94)
(2.6
02)
(2.2
77)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esC
ensu
sD
ivF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
R-s
qu
are
d0.3
40
0.1
49
0.1
25
0.2
01
0.1
64
0.0
17
0.0
50
0.0
50
0.1
02
0.0
67
F-S
tati
stic
s24.9
4
58
Tab
leA
.V:
Job
crea
tion
and
loca
lin
com
esh
ock
s:im
por
tp
enet
rati
onas
the
inst
rum
ent
Th
ista
ble
per
form
sth
esa
me
an
aly
sis
as
Tab
leII
Iu
sin
ga
diff
eren
tin
stru
men
tfo
rth
elo
cal
inco
me
gro
wth
—a
mea
sure
of
loca
lim
port
pen
etra
tion
.T
his
tab
lesh
ow
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
loca
lin
com
egro
wth
.R
egre
ssio
ns
are
run
for
aggre
gate
chan
ge
inem
plo
ym
ent
an
dfo
rth
ech
an
ge
inem
plo
ym
ent
inea
chof
the
4d
iffer
ent
age
cate
gori
es.
Ob
serv
ati
on
sare
at
the
CZ
-yea
r-fi
rmage
level
.T
he
dep
end
ent
vari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inth
en
on
-tra
dab
lese
ctor
(NA
ICS
2=
44,
45,
72)
over
the
pre
vio
us
two
yea
rscr
eate
din
firm
sof
each
age
cate
gory
,an
dth
isvari
ab
leis
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eC
Zas
of
2000.
Inco
me
gro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
.W
ein
stru
men
tfo
rth
isvari
ab
leu
sin
gan
inte
ract
ion
of
chan
ges
inim
port
pen
etra
tion
by
fou
r-d
igit
NA
ICS
manu
fact
uri
ng
sect
or
wit
hth
ep
reex
isti
ng
manu
fact
uri
ng
com
posi
tion
ina
CZ
.T
he
an
aly
sis
isp
erfo
rmed
on
a“n
on
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Colu
mn
(1)
rep
ort
sth
efi
rst-
stage
regre
ssio
nof
inco
me
gro
wth
on
the
Imp
ort
Bart
ikin
stru
men
t.C
olu
mn
(2)
isth
eO
LS
regre
ssio
nof
net
emp
loym
ent
chan
ge
inth
eC
Zon
loca
lin
com
egro
wth
,an
dco
lum
n(3
)is
the
2S
LS
regre
ssio
nw
ith
inst
rum
ente
din
com
egro
wth
.C
olu
mn
s(4
)to
(11)
per
form
sim
ilar
regre
ssio
ns
as
colu
mn
(2)
an
d(3
)fo
rth
en
etch
an
ge
inem
plo
ym
ent
infi
rms
of
diff
eren
tages
.C
ontr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ud
eyea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
esis
.S
tan
dard
erro
rsare
clu
ster
edby
CZ
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
Non
-over
lap
pin
gS
am
ple
(01,
03,
05,
07)
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Imp
ort
Bart
ik1.0
57***
(5.5
14)
Inco
me
Gro
wth
0.3
00***
0.3
24**
0.1
19***
0.3
68***
0.0
06
-0.0
80**
0.0
00
-0.0
16
0.1
75**
0.0
52
(2.9
53)
(2.1
06)
(7.3
33)
(4.2
97)
(0.5
57)
(-1.9
80)
(0.0
54)
(-0.4
95)
(2.1
47)
(0.3
57)
ln(T
ota
lL
ab
orf
orc
e)-0
.121***
-0.0
31*
-0.0
28
0.0
11
0.0
42***
-0.0
04
-0.0
15**
-0.0
01
-0.0
03
-0.0
37***
-0.0
53**
(-6.8
33)
(-1.9
07)
(-1.1
77)
(1.3
85)
(3.1
05)
(-1.1
95)
(-2.4
73)
(-0.3
07)
(-0.5
97)
(-2.6
04)
(-2.2
78)
%H
igh
sch
ool
Ed
u-0
.443**
-0.0
83
-0.0
77
0.7
30***
0.8
00***
-0.0
89*
-0.1
13**
-0.0
11
-0.0
15
-0.7
14***
-0.7
49***
(-2.1
30)
(-0.5
16)
(-0.4
84)
(5.2
05)
(5.8
01)
(-1.9
04)
(-2.2
01)
(-0.3
18)
(-0.4
50)
(-5.3
81)
(-5.4
52)
ln(T
ota
lC
ZW
ages
)0.1
08***
0.0
30**
0.0
27
-0.0
07
-0.0
35***
0.0
03
0.0
13**
0.0
00
0.0
02
0.0
34***
0.0
47**
(7.0
90)
(2.0
48)
(1.2
57)
(-1.0
36)
(-2.8
83)
(1.1
32)
(2.4
00)
(0.0
42)
(0.4
64)
(2.6
55)
(2.2
72)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
R-s
qu
are
d0.2
48
0.1
35
0.1
35
0.1
41
0.0
09
0.0
35
0.0
28
0.0
95
0.0
84
F-S
tati
stic
s30.4
59
Tab
leA
.VI:
Job
crea
tion
and
loca
lin
com
esh
ock
s:co
ntr
olling
for
loca
ldep
osit
san
dcr
edit
tosm
all
busi
nes
ses
Th
ista
ble
show
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)le
vel
on
loca
lin
com
egro
wth
.P
an
elA
incl
ud
esco
ntr
ols
for
CZ
-lev
eld
eposi
tgro
wth
an
dP
an
elB
contr
ols
for
gro
wth
inC
Z-l
evel
small
bu
sin
ess
loan
s.W
eru
nre
gre
ssio
ns
for
the
aggre
gate
chan
ge
inem
plo
ym
ent
an
dfo
rth
ech
an
ge
inem
plo
ym
ent
in4
age
cate
gori
es(s
tart
up
s,2-3
,4-5
,an
d6+
yea
rsold
).O
bse
rvati
on
sare
at
the
CZ
-yea
r-fi
rmage
level
.T
he
dep
end
ent
vari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inth
en
on
-tra
dab
lese
ctor
(NA
ICS
2=
44,
45,
72)
over
the
pre
vio
us
two
yea
rscr
eate
din
firm
sof
each
age
cate
gory
,an
dth
isvari
ab
leis
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eC
Zas
of
2000.
Inco
me
gro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
CZ
.W
ein
stru
men
tfo
rth
isvari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
,w
hic
hin
tera
cts
chan
ges
inn
ati
onw
ide
emp
loym
ent
inth
em
anu
fact
uri
ng
sect
or
wit
hth
ep
reex
isti
ng
manu
fact
uri
ng
com
posi
tion
ina
CZ
.T
he
an
aly
sis
isp
erfo
rmed
on
a“non
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Colu
mn
(1)
rep
ort
sth
efi
rst-
stage
regre
ssio
nof
inco
me
gro
wth
on
the
Bart
ikin
stru
men
t.C
olu
mn
(2)
isth
eO
LS
regre
ssio
nof
net
emp
loym
ent
chan
ge
inth
eC
Zon
loca
lin
com
egro
wth
,an
dco
lum
n(3
)is
the
2S
LS
regre
ssio
nw
ith
inst
rum
ente
din
com
egro
wth
.C
olu
mn
s(4
)to
(11)
per
form
sim
ilar
regre
ssio
ns
as
colu
mn
s(2
)an
d(3
)fo
rfi
rms
of
diff
eren
tages
.P
an
elB
per
form
sth
esa
me
an
aly
sis
as
Pan
elA
on
an
over
lap
pin
gsa
mp
leof
2000
to2007.
CZ
-lev
eld
eposi
tgro
wth
com
esfr
om
the
FD
ICan
dgro
wth
insm
all
bu
sin
ess
loan
sis
from
CR
Ad
ata
.C
ontr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
and
the
Bu
reau
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ud
eyea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
eses
an
dst
an
dard
erro
rsare
clu
ster
edby
CZ
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
PanelA
:C
ontr
ollin
gfo
rlo
cal
dep
osi
tgro
wth
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.1
27***
(8.3
61)
Inco
me
Gro
wth
0.2
93***
0.2
96***
0.1
15***
0.2
71***
0.0
06
-0.0
35
0.0
01
-0.0
11
0.1
71**
0.0
70
(2.8
99)
(2.7
97)
(7.4
22)
(4.8
90)
(0.5
85)
(-1.2
04)
(0.1
02)
(-0.5
38)
(2.0
93)
(0.6
66)
ln(T
ota
lL
ab
orf
orc
e)-0
.114***
-0.0
29*
-0.0
28
0.0
12
0.0
31***
-0.0
04
-0.0
09*
-0.0
01
-0.0
02
-0.0
36**
-0.0
48**
(-6.5
66)
(-1.8
22)
(-1.5
05)
(1.5
27)
(2.9
98)
(-1.2
33)
(-1.9
51)
(-0.3
49)
(-0.6
95)
(-2.5
57)
(-2.5
71)
%H
igh
sch
ool
Ed
u-0
.661***
-0.0
92
-0.0
91
0.7
26***
0.7
72***
-0.0
89*
-0.1
01**
-0.0
10
-0.0
13
-0.7
20***
-0.7
49***
(-3.2
71)
(-0.5
73)
(-0.5
81)
(5.2
10)
(5.7
28)
(-1.8
95)
(-2.0
90)
(-0.3
07)
(-0.4
03)
(-5.4
08)
(-5.5
81)
ln(T
ota
lC
ZW
ages
)0.1
01***
0.0
27*
0.0
27
-0.0
09
-0.0
25***
0.0
04
0.0
08*
0.0
00
0.0
01
0.0
32***
0.0
43**
(6.7
86)
(1.9
49)
(1.5
93)
(-1.1
95)
(-2.7
25)
(1.1
74)
(1.8
71)
(0.0
91)
(0.5
00)
(2.5
94)
(2.5
54)
Loca
lD
eposi
tG
row
th0.0
23**
0.0
15*
0.0
15**
0.0
07**
0.0
03
-0.0
01
0.0
00
-0.0
01
-0.0
00
0.0
10
0.0
12**
(2.2
79)
(1.8
93)
(2.2
70)
(2.1
29)
(1.3
11)
(-0.5
80)
(0.0
91)
(-0.8
96)
(-0.4
32)
(1.5
98)
(2.1
52)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
R-s
qu
are
d0.2
77
0.1
38
0.1
38
0.1
45
0.0
50
0.0
09
-0.0
14
0.0
35
0.0
32
0.0
97
0.0
89
F-S
tati
stic
s69.9
1
60
PanelB
:C
ontr
ollin
gfo
rlo
cal
small
bu
sin
ess
loan
gro
wth A
ggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.1
01***
(8.6
80)
Inco
me
Gro
wth
0.3
01***
0.3
10***
0.1
15***
0.2
73***
0.0
07
-0.0
35
0.0
03
-0.0
06
0.1
77**
0.0
77
(2.9
55)
(2.7
96)
(7.3
09)
(4.7
89)
(0.6
35)
(-1.1
66)
(0.4
13)
(-0.3
11)
(2.1
52)
(0.6
96)
ln(T
ota
lL
ab
orf
orc
e)-0
.114***
-0.0
31*
-0.0
30
0.0
11
0.0
30***
-0.0
04
-0.0
09*
-0.0
01
-0.0
02
-0.0
38***
-0.0
49***
(-6.2
65)
(-1.9
06)
(-1.5
97)
(1.4
45)
(2.9
38)
(-1.1
80)
(-1.9
41)
(-0.4
39)
(-0.6
54)
(-2.6
03)
(-2.6
49)
%H
igh
sch
ool
Ed
u-0
.598***
-0.0
81
-0.0
79
0.7
14***
0.7
51***
-0.0
77*
-0.0
86*
-0.0
15
-0.0
17
-0.7
05***
-0.7
28***
(-2.9
95)
(-0.5
02)
(-0.5
02)
(5.2
47)
(5.7
24)
(-1.7
46)
(-1.9
32)
(-0.4
60)
(-0.5
20)
(-5.4
26)
(-5.5
43)
ln(T
ota
lC
ZW
ages
)0.1
02***
0.0
30**
0.0
29*
-0.0
07
-0.0
24***
0.0
03
0.0
08*
0.0
00
0.0
01
0.0
34***
0.0
44***
(6.5
69)
(2.0
45)
(1.6
97)
(-1.0
66)
(-2.6
46)
(1.0
93)
(1.8
45)
(0.1
45)
(0.4
40)
(2.6
51)
(2.6
40)
Loca
lS
mall
Bu
sin
ess
Loan
Gro
wth
0.0
12**
-0.0
01
-0.0
01
0.0
01
-0.0
01
0.0
00
0.0
01
-0.0
02**
-0.0
01**
-0.0
00
0.0
01
(2.1
43)
(-0.2
99)
(-0.2
74)
(0.6
46)
(-0.5
83)
(0.2
15)
(0.7
68)
(-2.3
56)
(-2.1
13)
(-0.1
03)
(0.3
38)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
2,0
44
R-s
qu
are
d0.2
78
0.1
35
0.1
35
0.1
42
0.0
44
0.0
10
-0.0
14
0.0
41
0.0
39
0.0
96
0.0
89
F-S
tati
stic
s75.3
4
61
Appendix B. Supplementary detailed regressions
Table B.I: Job creation and local income shocks: reduced form
This table shows regressions of net employment creation at the commuting zone (CZ) level onthe national change in manufacturing employment at the sub-sector level weighted by the localregion’s exposure to that sub-sector (Manufacturing Employment Bartik). We run regressionsfor the aggregate change in employment and for the change in employment in 4 age categories(startups, 2-3, 4-5, and 6+ years old). Observations are at the CZ-year-firm age level. Thedependent variable is the net change in employment in the non-tradable sector (NAICS2 = 44,45, 72) over the previous two years created in firms of each age category, and this variable isscaled by the total non-tradable employment in the CZ as of 2000. The analysis is performed ona continuous version of the Bartik variable using non-overlapping samples in years 2001, 2003,2005, and 2007. T-statistics in parentheses are based on standard errors clustered at the CZlevel. *, **, *** denote statistical significance at the 10, 5 and 1% levels, respectively.
Continuous IV Variable, Non-overlapping Sample (01, 03, 05, 07)Aggregate 0-1 yrs 2-3 years 4-5 years 6+ years
(1) (2) (3) (4) (5)OLS OLS OLS OLS OLS
Manuf. Employment Bartik 0.351*** 0.313*** -0.040 -0.013 0.090(2.610) (4.205) (-1.236) (-0.559) (0.732)
ln(Total Laborforce) -0.067*** -0.002 -0.005* -0.001 -0.059***(-5.385) (-0.252) (-1.716) (-0.377) (-5.551)
% Highschool Edu -0.283 0.594*** -0.078 -0.007 -0.793***(-1.423) (3.949) (-1.604) (-0.191) (-5.237)
ln(Total CZ Wages) 0.062*** 0.004 0.004 0.000 0.053***(5.534) (0.555) (1.626) (0.096) (5.546)
Year FE Yes Yes Yes Yes YesObservations 2,044 2,044 2,044 2,044 2,044R-squared 0.080 0.099 0.009 0.035 0.073
62
Table B.II: Startup job creation and job resilience—results by year
This table shows the share of employment of four cohorts of new firms (between 2000 and 2003) and asks how many jobsremain after 2 or 4 years in those firms. The table shows the number of employees in these cohorts of firms at the time thatthey are started, 2 years later, and 4 years later. Employment is scaled by the total employment in each CZ as of 2000. Thesample includes only firms until 2003 because that is the last year that we can track firms for a full four years. T-statisticsfor the difference between high and low shock areas are shown in parentheses on the last line of each panel.
Panel A: 2000 cohort
Job creation from startups Jobs remaining after 2 years Jobs remaining after 4 yearsLow Bartik Area 4.81% 3.88% 3.50%Medium Bartik Area 5.50% 4.43% 3.83%High Bartik Area 5.25% 4.37% 3.86%High Bartik−Low Bartik 0.44% 0.49%* 0.36%t-statistics (1.32) (1.68) (1.32)
Job creation from startups Jobs remaining after 2 years Jobs remaining after 4 yearsLow ∆Income Area 4.51% 3.64% 3.21%Medium ∆Income Area 4.98% 4.12% 3.62%High ∆Income Area 6.06% 4.91% 4.36%High ∆Income−Low ∆Income 1.55%*** 1.27%*** 1.15%***t-statistics (5.19) (5.37) (5.31)
Panel B: 2001 cohort
Job creation from startups Jobs remaining after 2 years Jobs remaining after 4 yearsLow Bartik Area 3.94% 3.54% 3.10%Medium Bartik Area 4.63% 4.03% 3.51%High Bartik Area 4.98% 4.34% 3.82%High Bartik−Low Bartik 1.04%*** 0.81%*** 0.71%***t-statistics (4.36) (3.58) (3.27)
Job creation from startups Jobs remaining after 2 years Jobs remaining after 4 yearsLow ∆Income Area 3.99% 3.52% 3.09%Medium ∆Income Area 4.52% 3.91% 3.37%High ∆Income Area 5.06% 4.49% 3.96%High ∆Income−Low ∆Income 1.07%*** 0.97%*** 0.87%***t-statistics (4.33) (4.12) (3.91)
63
Panel C: 2002 cohort
Job creation from startups Jobs remaining after 2 years Jobs remaining after 4 yearsLow Bartik Area 3.48% 3.07% 2.68%Medium Bartik Area 4.53% 4.01% 3.43%High Bartik Area 4.98% 4.47% 3.97%High Bartik−Low Bartik 1.50%*** 1.39%*** 1.29%***t-statistics (6.63) (6.54) (6.13)
Job creation from startups Jobs remaining after 2 years Jobs remaining after 4 yearsLow ∆Income Area 4.03% 3.71% 3.21%Medium ∆Income Area 4.13% 3.66% 3.13%High ∆Income Area 4.83% 4.18% 3.72%High ∆Income−Low ∆Income 0.80%*** 0.47%** 0.51%**t-statistics (3.29) (2.08) (2.30)
Panel D: 2003 cohort
Job creation from startups Jobs remaining after 2 years Jobs remaining after 4 yearsLow Bartik Area 3.66% 3.02% 2.56%Medium Bartik Area 4.77% 4.06% 3.48%High Bartik Area 5.06% 4.34% 3.84%High Bartik−Low Bartik 1.40%*** 1.32%*** 1.28%***t-statistics (5.89) (6.51) (6.70)
Job creation from startups Jobs remaining after 2 years Jobs remaining after 4 yearsLow ∆Income Area 4.27% 3.68% 3.16%Medium ∆Income Area 4.32% 3.53% 3.11%High ∆Income Area 4.91% 4.22% 3.61%High ∆Income−Low ∆Income 0.64%*** 0.54%** 0.45%**t-statistics (2.65) (2.55) (2.32)
64
Tab
leB
.III
:Job
crea
tion
and
loca
lin
com
esh
ock
s:A
llin
dust
ries
–com
par
ison
ofQ
WI
and
BD
SD
ata
Th
ista
ble
show
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
com
mu
tin
gzo
ne
(CZ
)an
dM
etro
polita
nS
tati
stic
al
Are
a(M
SA
)le
vel
on
loca
lin
com
egro
wth
.T
he
dep
end
ent
vari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inall
sect
ors
over
the
pre
vio
us
two
yea
rsin
firm
sof
each
age
cate
gory
.T
his
vari
ab
leis
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
ere
gio
n(C
Zin
Pan
elA
,M
SA
inP
an
els
Ban
dC
)as
of
2000.
Inco
me
gro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
regio
n(C
Zin
Pan
elA
,M
SA
inP
an
els
Ban
dC
).W
ein
stru
men
tfo
rth
isvari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
,w
hic
hin
tera
cts
chan
ges
inn
ati
onw
ide
emp
loym
ent
inth
em
anu
fact
uri
ng
sect
or
wit
hth
ep
reex
isti
ng
manu
fact
uri
ng
com
posi
tion
inan
are
a.
All
pan
els
use
a“n
on
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Colu
mn
(1)
rep
ort
sth
efi
rst-
stage
regre
ssio
nof
inco
me
gro
wth
on
the
manu
fact
uri
ng
emp
loym
ent
inst
rum
ent.
Colu
mn
(2)
isth
eO
LS
regre
ssio
nof
net
emp
loym
ent
chan
ge
on
loca
lin
com
egro
wth
,an
dco
lum
n(3
)is
the
2S
LS
regre
ssio
nw
ith
inst
rum
ente
din
com
egro
wth
.C
olu
mn
s(4
)to
(11)
per
form
sim
ilar
regre
ssio
ns
as
colu
mn
s(2
)an
d(3
)fo
rfi
rms
of
diff
eren
tages
.C
ontr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ud
eyea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
eses
an
dst
an
dard
erro
rsare
clu
ster
edby
CZ
inP
an
elA
an
dby
MS
Ain
Pan
elB
,C
.*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
PanelA
:C
Z-l
evel
,Q
WI
Data
,N
on
-over
lap
pin
gS
am
ple
(01,
03,
05,
07)
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.1
24***
(8.5
06)
Inco
me
Gro
wth
0.3
83***
0.4
56***
0.1
62***
0.3
94***
0.0
15
-0.0
61**
0.0
05
-0.0
27
0.2
02***
0.1
49*
(4.1
26)
(4.9
65)
(6.5
59)
(7.6
64)
(1.6
47)
(-2.2
94)
(0.6
84)
(-1.4
31)
(3.2
12)
(1.7
19)
ln(T
ota
lL
ab
orf
orc
e)-0
.120***
-0.0
29*
-0.0
20
0.0
01
0.0
30***
-0.0
01
-0.0
10**
-0.0
02
-0.0
06*
-0.0
28**
-0.0
35**
(-6.8
95)
(-1.8
61)
(-1.2
61)
(0.1
68)
(2.9
34)
(-0.3
03)
(-2.2
34)
(-0.6
98)
(-1.7
75)
(-2.2
34)
(-2.3
07)
%H
igh
sch
ool
Ed
u-0
.715***
0.0
30
0.0
56
0.4
65***
0.5
46***
-0.0
39
-0.0
65
-0.0
17
-0.0
28
-0.3
79***
-0.3
98***
(-3.4
80)
(0.1
98)
(0.3
86)
(3.1
93)
(4.0
83)
(-0.9
71)
(-1.4
34)
(-0.4
82)
(-0.7
62)
(-3.2
71)
(-3.3
63)
ln(T
ota
lC
ZW
ages
)0.1
07***
0.0
26*
0.0
17
-0.0
02
-0.0
28***
0.0
01
0.0
10**
0.0
01
0.0
05*
0.0
25**
0.0
31**
(7.1
68)
(1.8
37)
(1.2
20)
(-0.2
58)
(-3.0
05)
(0.5
29)
(2.3
51)
(0.6
82)
(1.7
58)
(2.2
39)
(2.2
80)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s2,0
79
2,0
79
2,0
79
2,0
79
2,0
79
2,0
79
2,0
79
2,0
79
2,0
79
2,0
79
2,0
79
R-s
qu
are
d0.2
68
0.1
97
0.1
92
0.1
39
0.0
11
-0.0
93
0.0
23
0.1
09
0.1
06
F-S
tati
stic
s72.3
5
65
PanelB
:M
SA
-lev
el,
QW
ID
ata
,N
on
-over
lap
pin
gS
am
ple
(01,
03,
05,
07)
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.6
55***
(8.5
65)
Inco
me
Gro
wth
0.6
44***
0.8
05***
0.1
65***
0.3
42***
0.0
29***
0.0
34
0.0
19*
0.0
12
0.4
31***
0.4
16***
(15.2
53)
(9.9
23)
(5.6
58)
(5.9
27)
(5.4
62)
(1.4
65)
(1.7
62)
(0.7
10)
(11.5
83)
(7.4
12)
ln(T
ota
lL
ab
orf
orc
e)-0
.068***
0.0
01
0.0
11
-0.0
08
0.0
03
-0.0
02
-0.0
02
-0.0
03
-0.0
04
0.0
14
0.0
13
(-4.3
33)
(0.0
61)
(1.1
26)
(-1.0
87)
(0.4
68)
(-0.8
46)
(-0.6
68)
(-1.3
62)
(-1.5
31)
(1.5
48)
(1.3
66)
%H
igh
sch
ool
Ed
u-1
.227***
-0.1
51
0.0
03
-0.3
56**
-0.1
85
0.0
88*
0.0
92
0.0
30
0.0
24
0.0
86
0.0
72
(-5.0
03)
(-0.7
95)
(0.0
17)
(-2.0
00)
(-1.1
19)
(1.7
10)
(1.6
28)
(0.6
79)
(0.4
81)
(0.5
96)
(0.4
77)
ln(T
ota
lC
ZW
ages
)0.0
59***
-0.0
03
-0.0
12
0.0
05
-0.0
05
0.0
02
0.0
02
0.0
03
0.0
03
-0.0
13
-0.0
12
(4.1
53)
(-0.3
16)
(-1.3
93)
(0.8
47)
(-0.7
52)
(0.9
66)
(0.7
66)
(1.3
36)
(1.4
61)
(-1.6
19)
(-1.4
26)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s1,3
86
1,3
86
1,3
86
1,3
86
1,3
86
1,3
86
1,3
86
1,3
86
1,3
86
1,3
86
1,3
86
R-s
qu
are
d0.3
08
0.3
58
0.3
41
0.0
79
0.0
09
0.0
27
0.0
27
0.0
39
0.0
38
0.2
01
0.2
01
F-S
tati
stic
s73.3
6
PanelC
:M
SA
-lev
el,
BD
SD
ata
,N
on
-over
lap
pin
gS
am
ple
(01,
03,
05,
07)
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.7
29***
(8.6
74)
Inco
me
Gro
wth
0.6
45***
1.2
47***
0.2
22***
0.4
08***
0.0
03
-0.0
04
0.0
04
-0.0
05
0.4
15***
0.8
48***
(6.1
64)
(8.0
58)
(6.5
12)
(11.3
39)
(0.6
94)
(-0.2
82)
(1.0
87)
(-0.4
34)
(5.4
20)
(5.5
81)
ln(T
ota
lL
ab
orf
orc
e)-0
.071***
-0.0
42***
-0.0
05
0.0
04
0.0
15**
-0.0
04*
-0.0
04**
0.0
00
-0.0
00
-0.0
42***
-0.0
15
(-4.0
35)
(-2.8
53)
(-0.2
76)
(0.5
27)
(2.1
79)
(-1.8
76)
(-2.0
45)
(0.0
73)
(-0.2
88)
(-3.2
16)
(-0.9
85)
%H
igh
sch
ool
Ed
u-1
.195***
-0.3
24
0.2
35
0.0
13
0.1
87
-0.0
18
-0.0
26
-0.0
20
-0.0
29
-0.2
99
0.1
03
(-4.6
13)
(-0.9
29)
(0.7
42)
(0.0
86)
(1.4
38)
(-0.4
35)
(-0.5
99)
(-0.6
46)
(-0.8
99)
(-1.1
39)
(0.3
82)
ln(T
ota
lC
ZW
ages
)0.0
64***
0.0
39***
0.0
03
-0.0
03
-0.0
14**
0.0
04**
0.0
04**
0.0
00
0.0
01
0.0
38***
0.0
12
(3.9
38)
(2.9
38)
(0.1
84)
(-0.4
95)
(-2.2
16)
(1.9
73)
(2.1
36)
(0.0
23)
(0.3
98)
(3.2
65)
(0.8
91)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
R-s
qu
are
d0.2
99
0.1
62
0.0
83
0.2
71
0.0
96
0.0
42
0.0
41
0.0
34
0.0
29
0.1
10
0.0
64
F-S
tati
stic
s75.2
4
66
Tab
leB
.IV
:Job
crea
tion
and
loca
lin
com
esh
ock
s:F
irm
age
by
firm
size
—det
ail
Th
ista
ble
show
sre
gre
ssio
ns
of
net
emp
loym
ent
crea
tion
at
the
Met
rop
oli
tan
Sta
tist
ical
Are
a(M
SA
)le
vel
on
loca
lin
com
egro
wth
(th
isis
ad
etailed
ver
sion
of
Tab
leX
II).
Reg
ress
ion
sare
run
for
the
aggre
gate
chan
ge
inem
plo
ym
ent
inea
chof
the
4d
iffer
ent
age
cate
gori
esan
d3
size
cate
gori
es.
Pan
elA
focu
ses
on
the
firm
sw
ith
less
than
20
emp
loyee
s,P
an
elB
an
aly
zes
the
firm
sw
ith
20-1
00
emp
loyee
sw
hile
Pan
elC
focu
ses
on
larg
erfi
rms
wit
hm
ore
than
100
emp
loyee
s.T
he
dep
end
ent
vari
ab
leis
the
net
chan
ge
inem
plo
ym
ent
inall
sect
ors
over
the
pre
vio
us
two
yea
rscr
eate
din
firm
sof
each
age-
size
cate
gory
,an
dth
isvari
ab
leis
scale
dby
the
tota
ln
on
-tra
dab
leem
plo
ym
ent
inth
eM
SA
as
of
2000.
Inco
me
gro
wth
isth
etw
o-y
ear
gro
wth
of
tota
lw
ages
an
dsa
lari
esin
the
MS
A.
We
inst
rum
ent
for
this
vari
ab
leu
sin
gth
eB
art
ikm
anu
fact
uri
ng
shock
,w
hic
hin
tera
cts
chan
ges
inn
ati
onw
ide
emp
loym
ent
inth
em
anu
fact
uri
ng
sect
or
wit
hth
ep
reex
isti
ng
manu
fact
uri
ng
com
posi
tion
inth
eM
SA
.T
he
an
aly
sis
isp
erfo
rmed
on
a“n
on
-over
lap
pin
g”
sam
ple
of
yea
rs2001,
2003,
2005,
an
d2007.
Inea
chp
an
el,
Colu
mn
(1)
rep
ort
sth
efi
rst-
stage
regre
ssio
nof
inco
me
gro
wth
on
the
Bart
ikin
stru
men
t.C
olu
mn
(2)
isth
eO
LS
regre
ssio
nof
net
emp
loym
ent
chan
ge
inth
eM
SA
on
loca
lin
com
egro
wth
,an
dco
lum
n(3
)is
the
2S
LS
regre
ssio
nw
ith
inst
rum
ente
din
com
egro
wth
.C
olu
mn
s(4
)to
(11)
per
form
sim
ilar
regre
ssio
ns
as
colu
mn
s(2
)an
d(3
)fo
rfi
rms
of
diff
eren
tages
.C
ontr
ol
vari
ab
les
are
extr
act
edfr
om
the
2000
Cen
sus
an
dth
eB
ure
au
of
Lab
or
Sta
tist
ics.
All
regre
ssio
ns
incl
ud
eyea
rfi
xed
effec
ts.
T-s
tati
stic
sare
show
nin
pare
nth
eses
an
dst
an
dard
erro
rsare
clu
ster
edby
MS
A.
*,
**,
***
den
ote
stati
stic
al
sign
ifica
nce
at
the
10,
5an
d1%
level
s,re
spec
tivel
y.
PanelA
:E
mp
loyee<
20,
Non
-over
lap
pin
gS
am
ple
(01,
03,
05,
07)
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.7
29***
(8.6
74)
Inco
me
Gro
wth
0.1
20***
0.1
52***
0.1
73***
0.3
14***
-0.0
31***
-0.0
59***
-0.0
10***
-0.0
31***
-0.0
11***
-0.0
73***
(6.1
45)
(8.5
19)
(6.5
06)
(10.0
70)
(-5.6
86)
(-5.3
40)
(-4.0
08)
(-4.5
55)
(-2.8
25)
(-5.9
00)
ln(T
ota
lL
ab
orf
orc
e)-0
.071***
-0.0
00
0.0
02
-0.0
02
0.0
06
-0.0
00
-0.0
02
0.0
00
-0.0
01
0.0
02
-0.0
02
(-4.0
35)
(-0.1
40)
(0.7
64)
(-0.4
71)
(1.1
96)
(-0.1
47)
(-1.1
17)
(0.3
88)
(-0.9
08)
(1.0
21)
(-0.8
29)
%H
igh
sch
ool
Ed
u-1
.195***
0.0
99*
0.1
28***
0.1
06
0.2
37**
0.0
12
-0.0
13
0.0
16
-0.0
03
-0.0
35
-0.0
92**
(-4.6
13)
(1.8
62)
(2.6
92)
(0.8
47)
(2.1
99)
(0.3
27)
(-0.3
70)
(0.8
90)
(-0.1
97)
(-0.8
03)
(-2.1
87)
ln(T
ota
lC
ZW
ages
)0.0
64***
-0.0
01
-0.0
03
-0.0
01
-0.0
10**
0.0
01
0.0
02
-0.0
00
0.0
01
-0.0
00
0.0
04*
(3.9
38)
(-0.4
58)
(-1.3
02)
(-0.3
06)
(-2.0
08)
(0.4
73)
(1.4
96)
(-0.1
08)
(1.2
76)
(-0.0
18)
(1.8
14)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
R-s
qu
are
d0.2
99
0.2
68
0.2
52
0.3
12
0.1
47
0.1
08
0.0
56
0.0
66
0.1
20
F-S
tati
stic
s75.2
4
67
PanelB
:E
mp
loyee
bet
wee
n20
an
d100,
Non
-over
lap
pin
gS
am
ple
(01,
03,
05,
07)
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.7
29***
(8.6
74)
Inco
me
Gro
wth
0.1
17***
0.1
81***
0.0
40***
0.0
79***
0.0
25***
0.0
34***
0.0
08***
0.0
16
0.0
43***
0.0
53***
(5.8
29)
(7.1
16)
(5.8
17)
(6.7
59)
(4.9
21)
(3.2
23)
(2.6
23)
(1.4
81)
(4.2
07)
(3.0
33)
ln(T
ota
lL
ab
orf
orc
e)-0
.071***
-0.0
11***
-0.0
07**
0.0
05**
0.0
07***
-0.0
05***
-0.0
05***
0.0
00
0.0
01
-0.0
11***
-0.0
10***
(-4.0
35)
(-3.4
53)
(-2.1
66)
(2.3
14)
(3.3
67)
(-3.3
42)
(-2.9
45)
(0.2
80)
(0.6
40)
(-4.3
99)
(-4.1
49)
%H
igh
sch
ool
Ed
u-1
.195***
-0.1
54**
-0.0
94
-0.0
47
-0.0
12
-0.0
48
-0.0
40
0.0
20
0.0
27
-0.0
79
-0.0
70
(-4.6
13)
(-2.5
01)
(-1.6
04)
(-1.2
31)
(-0.3
20)
(-1.4
97)
(-1.2
55)
(0.6
02)
(0.8
20)
(-1.5
32)
(-1.3
51)
ln(T
ota
lC
ZW
ages
)0.0
64***
0.0
09***
0.0
05
-0.0
02
-0.0
05**
0.0
03**
0.0
03**
-0.0
01
-0.0
01
0.0
08***
0.0
07***
(3.9
38)
(3.0
05)
(1.6
29)
(-1.2
02)
(-2.3
18)
(2.4
46)
(2.0
38)
(-0.5
25)
(-0.8
83)
(3.4
83)
(3.2
01)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
R-s
qu
are
d0.2
99
0.1
45
0.1
12
0.2
09
0.1
41
0.1
12
0.1
08
0.0
33
0.0
29
0.0
72
0.0
70
F-S
tati
stic
s75.2
4
PanelC
:E
mp
loyee>
100,
Non
-over
lap
pin
gS
am
ple
(01,
03,
05,
07)
Aggre
gate
0-1
yea
r-old
s2-3
yea
r-old
s4-5
yea
r-old
s6+
yea
r-old
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
1st
Sta
ge
OL
SIV
OL
SIV
OL
SIV
OL
SIV
OL
SIV
Manu
f.E
mp
loym
ent
Bart
ik1.7
29***
(8.6
74)
Inco
me
Gro
wth
0.4
08***
0.9
14***
0.0
08**
0.0
15***
0.0
09***
0.0
21***
0.0
07**
0.0
10
0.3
84***
0.8
68***
(5.4
47)
(6.3
00)
(2.4
04)
(2.8
07)
(3.0
37)
(2.8
31)
(2.3
59)
(1.5
23)
(5.3
72)
(5.9
76)
ln(T
ota
lL
ab
orf
orc
e)-0
.071***
-0.0
31**
0.0
01
0.0
01
0.0
02*
0.0
02
0.0
02**
-0.0
01
-0.0
00
-0.0
33***
-0.0
03
(-4.0
35)
(-2.4
64)
(0.0
40)
(1.3
64)
(1.8
40)
(1.5
13)
(2.0
53)
(-0.6
48)
(-0.3
87)
(-2.6
63)
(-0.1
93)
%H
igh
sch
ool
Ed
u-1
.195***
-0.2
69
0.2
01
-0.0
45*
-0.0
39*
0.0
17
0.0
27
-0.0
56***
-0.0
53**
-0.1
85
0.2
65
(-4.6
13)
(-0.9
21)
(0.7
08)
(-1.8
98)
(-1.7
85)
(0.6
16)
(1.0
52)
(-2.6
45)
(-2.5
18)
(-0.6
44)
(0.9
36)
ln(T
ota
lC
ZW
ages
)0.0
64***
0.0
31***
0.0
01
0.0
01
0.0
00
-0.0
01
-0.0
01
0.0
01
0.0
00
0.0
30***
0.0
01
(3.9
38)
(2.7
58)
(0.0
50)
(0.8
71)
(0.3
07)
(-0.5
45)
(-1.1
67)
(0.8
29)
(0.5
41)
(2.7
20)
(0.0
85)
Yea
rF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esO
bse
rvati
on
s1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
1,3
84
R-s
qu
are
d0.2
99
0.1
13
0.0
48
0.2
03
0.1
97
0.0
81
0.0
70
0.0
29
0.0
28
0.1
05
0.0
45
F-S
tati
stic
s75.2
4
68