3 Garg Shenoy AJAE 2022

14678276, 2021, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12193 by University Of Delhi, Wiley Online Library on [11/02/2023].
See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
T HE E COLOGICAL I MPACT OF P LACE -B ASED
E CONOMIC P OLICIES
T EEVRAT G ARG AND A JAY S HENOY
Does economic development have an unavoidable ecological cost? We examine the impacts on forest
cover of one of India’s signature place-based economic policies involving massive tax benefits for new
industrial and infrastructure development following the creation of the new state of Uttarakhand.
Using a spatial difference-in-discontinuities design, we show that the policy, which explicitly excluded
environmentally damaging industries, resulted in no meaningful change in local forest cover. Our
results suggest that even in settings with low levels of enforcement, place-based economic policies that
deliver transformative economic expansion can be implemented with minimal ecological costs.
Key words: Agglomeration, deforestation, place-based economic policies.
JEL codes: H54, O40, Q53, Q56.
The central challenge of sustainable develop- means to close the rising gaps between regions
ment is bridging the gap between rich and within their borders (Felkner and Town-
poor regions without lasting damage to the send 2011; Busso, Gregory, and Kline 2013;
environment that could in turn undermine Kline and Moretti 2014; Shenoy 2018). Yet,
the goal of poverty alleviation (United even as these policies become ubiquitous, rel-
Nations 2015). Indeed, there has been a long- atively little is known about their environmen-
standing debate in both the conservation and tal impacts, particularly in developing
economics literature on the effects of eco- countries (Greenstone and Jack 2015).
nomic development and policies that encour- We focus on a principal concern about such
age such development on the environment targeted development, the risk that forests will
(Arrow et al. 1995; Grossman and Krue- be cleared in the wake of infrastructure invest-
ger 1995; Stern, Common, and Barbier 1996; ments (Asher, Garg, and Novosad 2020) and ris-
Andreoni and Levinson 2001; Foster and ing incomes (Alix-Garcia et al. 2013). In the
Rosenzweig 2003; Dasgupta 2007; Alix-Garcia context of place-based economic policies, such
et al. 2013; Asher, Garg, and Novosad 2020). land-use change is particularly relevant because
Increasingly, governments around the world these policies often target remote and previously
are using place-based policies—policies that underdeveloped regions with native vegetation.
target tax breaks or infrastructure develop- Furthermore, forest cover loss is an urgent con-
ment to an underdeveloped region—as a cern, generating global greenhouse emissions
(IPCC 2014; Jayachandran et al. 2017), local
Teevrat Garg is an assistant professor of Economics, School of
health externalities (Bauch et al. 2015;
Global Policy and Strategy, University of California – San Diego, Garg 2019; Masuda et al. 2019), deteriorating
La Jolla, California 92093, USA. Ajay Shenoy is an assistant pro- ecosystem services (Masuda et al. 2020), and
fessor in the Department of Economics, University of California
– Santa Cruz, Santa Cruz, California 95064, USA. We are grateful resulting in loss of biodiversity (Gibson
Sam Asher and Paul Novosad for generously sharing data. We et al. 2011). The most recent report by the Inter-
thank Editor Amy Ando, three anonymous reviewers, Jonah governmental Panel on Climate Change (IPCC)
Busch, Francisco Costa, Francisco Garfias, Gaurav Khanna, Kate
Sims, Gregor Singer, Dimitri Szerman, seminar participants at suggests that restoring and protecting forests
UCSD, and conference participants at the 2019 AERE Summer could yield almost a sixth of the emissions miti-
Conference for feedback and comments on earlier versions of this
paper. All errors are our own. gation required to prevent runaway climate
Correspondence to be sent to: [email protected] change by 2030 (IPCC 2019).
Amer. J. Agr. Econ. 103(4): 1239–1250; doi:10.1111/ajae.12193

Published online February 1, 2021
© 2021 Agricultural and Applied Economics Association
14678276, 2021, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12193 by University Of Delhi, Wiley Online Library on [11/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1240 August 2021 Amer. J. Agr. Econ.
We exploit a spatial discontinuity in the and environmental quality—often characterized

introduction of one of the world’s most gener- as the “Environmental Kuznets Curve”—to the
ous place-based policies. In 2002, the govern- best of our knowledge, none have considered
ment of India provided tax breaks and the effects of place-based economic policies on
infrastructure investments worth nearly $34 forest cover.2 Unlike other development policies,
billion to the recently formed state of Uttarak- place-based economic policies target an underde-
hand. The policy had an important additional veloped region rather than a segment of the pop-
feature, an explicit environmental rider that ulation (e.g. the rural poor). One aim of these
excluded certain environmentally detrimental policies is to concentrate development in a region
industries from receiving any subsidies or to generate a new center of agglomeration. These
tax-exemptions while favoring industries gen- agglomerations could damage the environment
erally considered environmentally friendly.1 by fostering industries that clear land and con-
Our setting is particularly important because sume timber, or they could preserve it by concen-
Uttarakhand contains one of the only large trating people and economic activity within a few
contiguous tracts of forest in Northern India, cities while leaving forests to regenerate. By tar-
with over 63% of the area in the state under getingfirmsratherthan individuals,aplace-based
forest cover. Furthermore, Uttarakhand’s for- policy has the potential to shift production away
ests are important due to their rich biodiver- from environmentally intensive industries. The
sity (Gaur 2007). The region has also environmental damage done in developing coun-
historically identified with the environmental tries by rapid industrialization continues to be a
conservation movement as the birthplace of major source of controversy, and therefore, it is
the Chipko Movement that encouraged local crucial to understand whether a carefully
residents to hug trees in order to dissuade log- designed place-based policy can achieve major
ging efforts (Shiva and Bandyopadhyay 1986). economic development without causing major
The introduction of large scale regional forest loss.
investment in infrastructure and production Using a difference-in-discontinuities design,
subsidies can have ambiguous effects on forest we find that the introduction of these subsidies
cover. Forested land may be cleared for the had a small, statistically insignificant effect on
development of industrial estates and associ- forest cover, even ten years after the introduc-
ated physical infrastructure. Additionally, tim- tion of the policy. By contrast, the same policy
ber demand can increase either because rising increased economic activity by at least 70%
incomes induce demand for land-intensive and as much as 300% (Shenoy 2018). We find
goods (Alix-Garcia et al. 2013) or highways no evidence to suggest that the null effects
and other infrastructure expand the scope for are driven by spillovers across the border or
wood-using industry (Asher, Garg, and Novo- within borders. Ten years after the introduc-
sad 2020). At the same time, increased indus- tion of the policy, we show that the absolute
trial activity could be associated with exits increase in employment in wood-using firms
from agriculture that could lower demands is modest relative to the overall expansion in
on forested land from the agricultural sector employment. Given the cost of transportation
(Assunç~ ao et al. 2017; Abman and Car- and general uncommonness of cross-state
ney 2020) or poverty reduction could reduce investment, any potential spillover would most
demand for forest goods (Ferraro and Simor- likely be near the border within the control
angkir 2020). Yet, other interventions such as state. But we find no evidence of such—if any-
alternative energy sources, even while ex-ante thing we find a general increase in forest cover
promising, have failed to reduce forest loss on both sides of the border. Together, our
except when accompanied by complementary results demonstrate that at least in terms of
policies (Meeks, Sims, and Thompson 2019).
Overall, the effect of directed, geographically
concentrated economic growth on forest cover 2
Other papers have considered cash transfers (Alix-Garcia
is ambiguous. et al. 2013; Wilebore et al. 2019), rural credit (Assunç~ ao
et al. 2019), agriculture (Assunç~
ao et al. 2017; Abman and Car-
Although a broad literature has documented ney 2020), and trade (Antweiler, Copeland, and Taylor 2001;
the relationship between economic development Copeland and Taylor 2004). There is also an extensive literature
documenting the relationship between economic development
and the environment. For a non-exhaustive list, see: Den Butter
and Verbruggen (1994); Arrow et al. (1995); Grossman and Krue-
ger (1995); Stern, Common, and Barbier (1996); Andreoni and
1
In the appendix, we provide both the “positive” or encouraged Levinson (2001); Dasgupta et al. (2002); Foster and Rosenz-
environmentally friendly list and the “negative” or environmen- weig (2003); Stern (2004). For a thorough review on drivers of
tally unfriendly list of industries. deforestation, see Busch and Ferretti-Gallon (2017).
Garg and Shenoy The Ecological Impact 1241
forest cover, place-based economic policies the next five years); a complete exemption
that explicitly exclude “worst offender” indus- from excise taxes for ten years; and a 15%
tries can achieve transformative economic investment subsidy for new or expanded facto-
expansion with relatively minimal environ- ries. For comparison, in 2003 the two exemp-
mental costs.3 tions bought relief from a statutory corporate
The rest of the paper is organized in the fol- tax rate of 36.75% and an excise tax of 16%.6
lowing sections. In Section 1, we provide back- Firms can only exploit the investment sub-
ground on the policy and describe our data sidy and excise tax exemption if they build
sources. In Section 2, we outline the research and produce within Uttarakhand, giving firms
design, and in Section 3, we discuss the corre- an incentive to move factories rather than just
sponding results. In Section 4, we offer conclud- their nominal headquarters. Figure A.1, which
ing remarks. shows the change in the number of factories,
makes it clear that firms were responding in
part to the tax incentives. Only factories regis-
Background and Data
tered by 2010 could claim the excise tax
exemption. After the deadline, the rate of
The Policy new registrations drops sharply, suggesting
In 2002, the federal government initiated of a that firms pushed forward their investment to
series of separate initiatives targeting the state exploit the policy.
of Uttarakhand (Shenoy 2018).4 These included The tax exemptions were designed to attract
spending for new infrastructure, better access to certain industries at the expense of others. The
existing infrastructure, and business tax exemp- government published a “positive” list of indus-
tions. Though some of these funds were avail- tries that it considered “environmentally
able ever since the state was formed in late friendly” (Government of India 2003). These
2000, it was only in 2002 that it began concen- include floriculture, honey, and goods related to
trating the funds in a handful of industrial estates tourism (especially “eco-tourism”). Unlike most
along the border between Uttarakhand and the firms, which got tax exemptions at establishments
state of Uttar Pradesh to the south. These within approved industrial estates, firms in the
estates play a key role in the raft of tax exemp- positive industries were eligible throughout the
tions that were specifically designed to spur state. There was likewise a “negative” list of
growth without harming the environment. industries denied any tax benefits regardless of
These exemptions, titled the “Special Pack- their location. The negative list includes coal
age Scheme for Himachal Pradesh and and oil-based power plants, wood pulp, and most
Uttarakhand,” were first announced in March paper products. The complete positive and nega-
of 2002 with an effective date of 2003.5 The tive lists are provided in the Appendix.
most generous include a complete exemption The explicit environmental focus of the policy
from federal income taxes for the first five - is in part a consequence of Uttarakhand’s his-
years of production (and a 30% reduction for tory. The movement that ultimately led to its
creation had its roots in environmentalist pro-
tests triggered by timber concessions many
3
It is worth noting that we are unable to attribute our findings to decades ago (Tillin 2013). The policy was a cali-
any particular feature of this place-based policy but only to the pol- brated attempt by the central government to win
icy as a whole. As such we provide an existence result that large
economic expansion can be coupled with relatively minor ecolog- political support in the new state by promoting
ical footprint.
4
economic development without alienating the
Uttarakhand was previously a set of administrative districts
that formed the northeast corner of Uttar Pradesh. The new state
still-potent environmentalist movement.
was formed along the boundaries of several existing districts. The firms ultimately attracted to the industrial
Because it followed these existing administrative divisions the estates produce goods across all industries.
new state border was not completely arbitrary, but it also was
not custom designed to include an underdeveloped area (and in Aside from information technology firms specif-
some areas it actually bisected existing urban agglomerations). ically courted by the IT Park at Dehradun’s
Although districts can play a role in administering state and central estate, nearly all registrants at the estates are in
government programs, they do not have leeway to pass laws or set
policy. manufacturing. They produce everything from
5
Uttarakhand and Himachal Pradesh received this favorable processed food to processed metals, Ayurvedic
treatment because the Indian government considers their moun-
tainous terrain a significant barrier to development. India’s Plan-
medicine to automobile parts, plastics to
ning Commission has long allocated extra funds to rugged and
sparsely populated areas. The tax exemptions, which would later
be extended to Jammu and Kashmir (a Himalayan state in north-
6
ern India), were at least on paper intended as extensions of this As explained in Shenoy (2018), the effective rate is somewhat
long-standing practice. lower but still far from trivial.
pharmaceuticals. Though paper products are (Asher et al. 2019). Our regressions thus give
supposedly excluded from the tax subsidies, the impact on employment in the average census
there are still a non-trivial number of firms that location.
produce boxes and packaging (possibly to sup-
ply the other firms). Given their presence, it is Borders. We measure the discontinuous
not a foregone conclusion that the program change in outcomes at the state boundaries
caused little deforestation. That is an empirical by linking the forest cover and firm-level data
question to which we devote the rest of the to shapefiles of administrative boundaries cre-
paper. ated by ML Infomap, a commercial mapping
firm.10 These data give the border between
Uttar Pradesh (control state) and Uttarak-
Data hand (treated state) as well as subdistricts,
which we use as clusters in calculating stan-
We rely on three main types of data: satellite-
dard errors.11
imagerybasedforestcover,dataonsector-specific
firm composition, and geographic boundaries.
Forest cover. Detailed and reliable administra- Research Design

tive records on forest cover and deforestation
rarely exist, especially in developing countries. Forest Cover
Instead, we obtain high resolution time series
estimates of forest cover using a standardized OurdesigncloselymatchesthatofShenoy(2018),
publicly available satellite-based dataset. Veg- which is based on the assumption that there are
etation Continuous Fields (VCF) is available parallel trends at the border in all unobservable
at 250 m resolution and provides annual tree factors that could affect economic growth for rea-
cover from 2000–2014 in the form of the persons unrelated to the place-based policy. Our
centage of each pixel under forest cover assumption is that any unobservable factors (for
(Townshend et al. 2011).7 For our primary example, geography, climate, local demand,
specification, we define forest cover as the etc.) that might affect forest cover or employment
average percentage of forest cover in a pixel. are likewise growing in parallel at the border.
We define 0.1x0.1 degree latitude and longi- This assumption is weaker than the assumption
tude cells, and calculate the average pixel behind difference in differences, a more common
value within each cell. These cells are the unit approach in this literature, which requires paral-
of analysis in the specifications below.8 lel trends throughout the treated and control
states. The stronger assumption is suspect in this
Firm level data. We obtain data on firms and context because governments target place-based
employment from the 1998 and 2013 Economic policies to regions precisely because they are
Census.9 These data were merged to the Socio- growing more slowly than other areas.
economic High-resolution Rural–Urban Geo- Shenoy (2018) shows that the assumption
graphic Dataset on India (SHRUG) and needed for difference in differences fails, whereas
collapsed to a SHRUG location, which is the low- that needed for difference in discontinuities
est identifiable census unit, either village or town appears to hold. There are clear differential
trends in several variables between Uttarakhand
(the treated state) and Uttar Pradesh (the control
7
Some previous studies have used Global Forest Cover (GFC) state), but these differences become statistically
dataset that describes baseline forest cover in the year 2000 and
a binary indicator for the year of deforestation for each and economically insignificant at the border
30mX30m pixel. As noted in Asher, Garg, and Novosad (2020), (see Appendix Table A.1 for a reproduction of
GFC is less useful for the study of forest cover in India because these results).12
GFC does not capture forest gains in areas with positive baseline
forest cover or partial forest loss. Although GFC is an excellent
source for other contexts such as Brazil and Indonesia, it is less
10
suitable in the Indian context, which saw overall increases in forest See: https://www.mlinfomap.com/
cover during our study period (for example, see figure 1). For 11
Subdistricts are the third tier of administrative divisions in
more information on the comparability of different forest cover India, preceded by states and districts.
12
datasets in India, see Asher, Garg, and Novosad (2020). One difference in trends (population) remains marginally sig-
8
Any cell that intersects the border is split in half along the bor- nificant. Given the number of variables tested (and the fact that we
der, with halves assigned to the appropriate state. do not adjust for multiple inference in this table), the difference is
9
Although there was an economic census conducted in 2005, likely to have arisen by chance. Regardless, the point estimate
employment figures for logging firms were combined with those implies there was slightly faster population growth at the border,
engaged in afforestation practices and hence are unsuitable for which if anything would bias the results towards finding greater
the analysis in this paper. deforestation in the treated state.
Figure 1. Comparison of nighttime luminosity and deforestation within ten kilometers of border
we plot the mean of each outcome for cells that lie within ten kilometers of the border
We measure the impact of the policy on third-order polynomial in latitude and longitude.
deforestation and other outcomes using This polynomial acts like a two-dimensional ver-
three difference-in-discontinuities specifica- sion of the running variable in a traditional
tions. Though all three estimate the discon- regression discontinuity design. It will remove
tinuous change in the growth (or loss) of from the estimated treatment coefficient any
forest cover from 2000 to each subsequent changes in deforestation that are smooth in
year through 2014, they differ in the space. Let i index cells and t index years, and
approach they take to control for the unob- let P3 be a third-order polynomial in the latitude
servables that are assumed to vary smoothly and longitude of the centroid of each cell.13 We
around the border. estimate
X
2014
ð1Þ ½Tree Coveri,t = ½Fixed Effect i + κt ½Year Dummyt
t = 2001
X
2014
+ ½Year Dummyt × P3t ½Lat i , ½Loni
t = 2001
X
2014
+ βSt ½Year Dummyt × ½Targetedi + ½Errori,t
t = 2001
The spatial polynomial specification assumes

that any unobservable correlates of deforesta-
tion vary smoothly over the surface of the earth 13
This polynomial is simply the variables that arise from a triple
and can thus be absorbed into a time-varying, interaction of latitude and longitude with themselves and one
another.
Figure 2. Regression discontinuity at the border in 2000 and 2014 we plot average tree cover
against distance to the boundary (positive values are in the targeted state).
Note: Each dot shows average tree cover within a bin, where the bins are chosen by the variance evenly-spaced method estimated using code from Calonico,
Cattaneo, and Titiunik (2014)
where [Targeted] is an indicator for whether the by the term Lt([Distance]i, [Targeted]i) = ω1,t[Dis-
cell is inside the targeted region. The direct term tance]i + ω2,t[Distance]i × [Targeted]i. This speci-
for the polynomial P3() and the dummy [Tar- fication assumes that the unobservables vary
geted] areabsorbed into the fixed effect. The smoothly and (approximately) linearly with dis-
terms βSt measure the impact on forest cover tance to the border, though the slope of that line
in t relative to the level in 2000. may differ on the treated versus control side of
The distance to border specification estimates the border and the entire relationship may
a local linear regression, as is standard for a uni- change from year to year. We estimate
X
2014
ð2Þ ½Tree Coveri,t = ½Fixed Effect i + κ t ½Year Dummyt
t = 2001
X
2014
+ ½Year Dummyt × Lt ½Distancei , ½Targetedi
t = 2001
X
2014
+ βD
t ½Year Dummyt × ½Targetedi + ½Error i,t
t = 2001

variate regression discontinuity design. We calcu- The coefficients βD t measure the impact
late the distance from each cell’s centroid to the of the border on the change in forest cover.
new border and assume that any unobservable The comparison of means specification simply
covariate of deforestation will be controlled for calculates the difference in forest cover (relative
to 2000) in a 4-kilometer neighborhood around must correct for arbitrary correlation in the
the border. We estimate error terms along those dimensions. We take
X
2014
t = 2001
X
2014
+ βC
t ½Year Dummyt × ½Targetedi + ½Error i,t
t = 2001
Finally, we estimate average program a nonparametric approach by clustering the

impacts by pooling pre- and post-program standard errors within geographic administra-
years in all three specifications: tive units. Though there are several levels of
X
2014
ð4Þ ½Tree Coveri,t = ½Fixed Effect i + κ t ½Year Dummyt
t = 2001

+ ½Post t × P3t ½Lat i , ½Loni + βS ½Post t × ½Targetedi + ½Errori,t
X
2014
t = 2001

+ ½Post t × Lt ½Distancei , ½Targetedi
+ βD ½Post t × ½Targetedi + ½Errori,t
X
2014
t = 2001
+ βC ½Post t × ½Targetedi + ½Errori,t
We follow Shenoy (2018) in setting the administrative unit that might serve, we follow
bandwidth of the first two specifications at Shenoy (2018), who shows that clustering by
thirty kilometers and using a rectangular ker- subdistrict gives standard errors that reject a
nel.14 As noted above, the comparison of placebo null hypothesis at roughly the correct
means specification is restricted to a band- rate. Because the number of subdistricts avail-
width of four kilometers. We deliberately able for the comparison of means specification
choose specifications as similar as possible to is relatively small, we also confirm in appendix
those of Shenoy (2018) to let us compare the table A.3 that bootstrapped standard errors
estimated impact on forest cover to the esti- yield similar results to asymptotic errors for
mated impact on aggregate output. this specification.
Because our research design exploits varia-
tion across both time and space, our inference
Employment and Firm Growth
14
Because the 2005 Economic Census did not
The bandwidth of a regression discontinuity design is the set of
observations given positive weight in the estimation—in this case, separate logging and tree-felling from other
those within thirty kilometers of the border. forestry industries (e.g. forest conservation),
Figure 3. Difference-in-discontinuities estimate of effect of place-based policy on deforesation.

n So n Do n Co
Note: We plot the estimates β^t , β^t , and β^t from estimating Equations 1–3. Each estimate gives discontinuous change in tree cover at the boundary
relative to the discontinuity in the year 2000. The red dashed line shows the first year of the policy
we must rely on only the 1998 and 2013 expansion of employment. Finally, we dis-
rounds. Because there are only two periods cuss potential threats to our research
(pre and post), the specifications in the empir- design, most notably the risk that forest
ical section are not identified. We instead take loss is displaced from treatment to control
the location-level change from 1998 to 2013 areas.
and run a local linear regression with a trian-
gular kernel.15 Because the difference in dis-
continuities is now essentially a standard
regression discontinuity design (but taking a Effect on Forest Cover
difference as the outcome), we apply the opti-
mal bandwidth selection method of Calonico, Figure 1 compares raw average night time
Cattaneo, and Titiunik (2014).]16 We estimate luminosity (left panel) to average forest cover
(right panel) within ten kilometers on either
ð7Þ Δ½Outcomei = π 0 + π 1 ½Distancei side of the discontinuity.17 Although average
+ π 2 ½Distancei × ½Targetedi night time luminosity between treatment and
control areas diverges substantially within a
+ ω½Targetedi + ½Errori
few years of the introduction of the policy
(2002), average forest cover in treatment areas
again clustering by subdistrict.
tracks closely with average forest cover in con-
trol areas showing no divergence in trends.
Results Figure 2 shows the discontinuity at the bor-
der in average forest cover in the years 2000
(left panel) and 2014 (right panel). Even twel-
We report two principal findings in this ve years after the introduction of the policy,
paper. First, across a number of specifica- and four years after the end of the policy,
tions and robustness checks, we find that there is no discernible difference in forest
the policy had a small and statistically insig- cover at the border.
nificant effect on forest cover. The effect is Figure 3 shows the year-by-year estimates
especially small relative to the expansion of corresponding to each of the Equations 1–3.
economic activity. Second, we find a pre- In all three figures, each estimate provides
cisely estimated impact on employment in the discontinuous change in tree cover at the
logging and wood-using firms that, though boundary relative to the discontinuity in the
positive, is small relative to the overall year 2000. The red line indicates the year
2002 when the policy came into effect. Across
15
A triangular kernel puts heavier weight on observations close
to the discontinuity, making it less sensitive than a rectangular ker-
17
nel to small changes in bandwidth. This property is especially use- Nighttime luminosity has been used a proxy for economic
ful when dealing with a relatively noisy outcome like the change in activity (Cook and Shah 2020) and development outcomes such
employment. as electrification (Baskaran, Min, and Uppal 2015; Mahade-
16
The optimal bandwidth selection method minimizes weighted van 2020). As in (Shenoy 2018) we use nighttime lights as a proxy
mean-squared error. for local economic development and activity.
Table 1. Difference-in-Discontinuities Estimate of Place-Based Policies on Tree Cover

Spatial polynomial Distance to border Comparison of means
Post-PBP −0.49 −0.31 −0.38
(0.45) (0.54) (0.53)
Cell years 4320 4320 1350
Cells 288 288 90
Subdistricts 38 38 26
Mean at baseline 16.4 16.4 16.4
Estimates of β^ , β^ , β^ from Equations 4–6. The outcome is the average tree cover within each cell. Standard errors are clustered by subdistrict. Significance levels
S D C
denoted at conventional levels.
all three specifications, we observe a small We find there is a marked increase in overall
negative effect of the policy on forest cover. employment. In Column (1) we show that
We formally estimate the effect of the policy on employment increased by 84.26 persons in each
tree cover and report the aggregate results of our census location, and the effect is significant at the
difference-in-discontinuities design in table 1. In 5% level. Compared to a baseline treatment
Column (1) we employ a spatial polynomial esti- group mean of sixty four employed persons per
mator, in Column (2) we use a distance to border census location, this translates to a 32.2% increase
approach, and in Column (3) we calculate a simple in overall employment. By contrast, we see a pre-
difference of means. Across all three specifica- cise but modest increase in employment in logging
tions, we find that the shift in the estimate at the firms. The average censuslocation saw an increase
border before and after the implementation of of0.56workersinthiscategory(Column2,table2).
policy was small and statistically insignificant at There was virtually no employment in this sector
conventional levels. These null effects are unlikely on either side of the discontinuity before the
tobe the result of a lack of statistical power; indeed implementation of the policy. Logging firms rep-
our results onemployment reported subsequently resent 0.66% of total change in employment as a
show that our design has statistical power to pick result of policy. When considering wood-using
up even small changes in forest cover/employ- firms (Column 3, table 2), we find that the policy
ment if they exist. Using our preferred specifica- increased employment in this category by nearly
tion in Column (1), we find a mean reduction of 6.35 workers per census location, or 7.5% of over-
0.49% points or 2.98% of forest cover. Based on all increase in employment.
a95%confidenceinterval,wecanrejectforestloss
in excess of 1.37% points or 8.3% of the mean at
baseline.18 We are able to reject similar increases Does Displacement Explain the Null-Result?
using alternative specifications (Columns 2–3). One reason for our null-estimate could be that
To alleviate concerns that forest cover in levels the effect of the policy led to increased forest
may be too noisy, we show that our results are loss in not only the treatment area but also
robusttousinganinversehyperbolicsinetransfor- the control area.20 Although it is not possible
mation of the dependent variable (appendix fig- to test for displacement explicitly, figure 1
ure A.4, appendix table A.2).19 shows that forest cover in the treatment area
closely tracks forest cover in the control areas,
before, during, and after the policy is in effect.
Effect on Employment There is also no evidence of an overall decline in
Table 2 shows the effects of the policy on employ- control areas. Appendix figure A.2 shows maps
ment in all firms and specifically the subset of firms of forest cover in 2000 and 2014 that confirm
in the logging industry and more generally in
industries where the primary input is raw lumber. 20
There is also the possibility of displacement from the border to
locations in the treated state further away from the border. But the
policy contains stricter environmental protections outside the
18
Our results compare to the effects the national rural roads pro- industrial estates along the border, making it unlikely that forest
gram that had no discernible effect of forest cover and contrast cover loss was displaced from the border to the interior. Moreover,
with a national highway construction program reduced forest reasonable alterations in the bandwidth of our discontinuity
cover by 17% (Asher, Garg, and Novosad 2020). design do not overturn our result, suggesting that there is no rea-
19
We prefer inverse hyperbolic sine transformation over a log son to suspect spillovers to neighboring regions away from the
transformation to account for zero values in forest cover data. border.
Table 2. Regression Discontinuity Estimate of Place-Based Policy on Employment and Firms

Employment
All Logging Wood using
** **
Post-PBP 84.26 0.56 6.35**
(33.81) (0.23) (2.48)
Observations 19236 19236 19236
Subdistricts 29 45 29
Optimal BW 12.9 30.9 15.3
Control mean, 1998 64.0 0.0 1.2
Treated mean, 1998 80.0 0.0 3.2
We estimate Equation 7 for employment and the number of firms within each of the given industries (“all” is all employment measured in the Economic Census).
The unit of observation is a census location (either a town or a village). Standard errors are clustered by subdistrict. Significance levels denoted at conventional
levels.
**p < 0.05.
no evidence of systematic change in the control spatial discontinuity in the policy, even ten -
region (south of the border) after the implemen- years after its introduction and four years after
tation of the policy. In 2013, employment in log- its end we find no effect on forest cover. By
ging is 0.07% of total employment near the contrast, the expansion of economic activity
border of the control region—not much of an was massive. Finally, we find no evidence for
increase from 0% in 1998. Taken together these spillovers across the border from the treat-
results suggest local displacement is unlikely to ment to the control region.
be the source of our null finding. One possible reason for this win–win result is
Is it possible that deforestation was displaced that the policy had an explicit environmental
to regions far from the border in the control state rider that excluded tax breaks to certain envi-
or to states entirely outside our study region? ronmentally detrimental industries such as pulp,
Although we cannot categorically rule out this paper, and mining while explicitly promoting
kind of displacement, it is unlikely because ship- others such as food, pharmaceuticals, and non-
ping logs over long distances across India’s rela- timber forest-based products. In effect, the pol-
tively slow infrastructure is costly. The least icy increased the relative costs of setting up envi-
costly option is to source wood from the closest ronmentally detrimental industries. The policy
point in the control state. The absence of any also created strong incentives to locate produc-
detectable decline in the border region of the con- tion inside a set of compact industrial areas
trol state makes it unlikely that such displacement located within existing cities. By increasing the
is driving our result.21 density of industrial production, the policy may
have minimized the deforestation that arises
Discussion and Conclusion
from sprawling new construction.
An important caveat for our findings is that
The rising concern of increasing, geographi- we focus on one measure of environmental
cally concentrated economic divisions within quality: forest cover. Economic development
national borders has spurred the growth of can also affect air and water quality; however,
place-based economic policies. These policies the lack of detailed data during the relevant
provide incentives for industrial development time period in our study region precludes us
and infrastructure through subsidies and tax- from estimating these effects. A second caveat
breaks and typically target remote areas that to our study is that we are unable to uncover
are more likely to have native vegetation. the exact mechanisms that led to minor forest
Although concern has been expressed over
loss from the policy. Future research should
the short- and long-run ecological footprint
address other such potential external costs of
of such rapid development, the policy we study
policy-driven, geographically concentrated eco-
showed no such ramifications. Exploiting a
nomic development to fully uncover their wel-
fare effects. Moreover, research on similar
21
We also cannot categorically rule out that in the absence of the policies such as special economic zones (SEZs)
policy, firms would counterfactually have sited their factories else- in other contexts is urgently needed to under-
where, and workers might have been drawn out of environmen-
tally unfriendly industries. stand the external validity of our findings.
Supplementary Material Baskaran, Thushyanthan, Brian Min, and

Yogesh Uppal. 2015. Election Cycles and
Supplementary material are available at Amer- Electricity Provision: Evidence from a
ican Journal of Agricultural Economics online. Quasi-Experiment with Indian Special
Elections. Journal of Public Economics
126: 64–73.
REFERENCES Bauch, Simone C, Anna M Birkenbach,
Subhrendu K Pattanayak, and Erin O
Abman, Ryan, and Conor Carney. 2020. Agri- Sills. 2015. Public Health Impacts of Eco-
cultural Productivity and Deforestation: system Change in the Brazilian Amazon.
Evidence from Input Subsidies and Ethnic Proceedings of the National Academy of
Favoritism in Malawi. Journal of Environ- Sciences 112(24): 7414–9.
mental Economics and Management 103: Busch, Jonah, and Kalifi Ferretti-Gallon.
102342. 2017. What Drives Deforestation and
Alix-Garcia, Jennifer, Craig McIntosh, What Stops It? A Meta-Analysis.
Katharine RE Sims, and Jarrod R Welch. Review of Environmental Economics
2013. The Ecological Footprint of Poverty and Policy 11(1): 3–23.
Alleviation: Evidence from Mexico’s Busso, Matias, Jesse Gregory, and Patrick Kline.
Oportunidades Program. Review of Eco- 2013. Assessing the Incidence and Efficiency
nomics and Statistics 95(2): 417–35. of a Prominent Place Based Policy. Ameri-
Andreoni, James, and Arik Levinson. 2001. can Economic Review 103(2): 897–947.
The Simple Analytics of the Environmen- Calonico, Sebastian, Matias D Cattaneo, and
tal Kuznets Curve. Journal of Public Eco- Rocio Titiunik. 2014. Robust Nonpara-
nomics 80(2): 269–86. metric Confidence Intervals for
Antweiler, Werner, Brian R Copeland, and Regression-Discontinuity Designs. Econ-
Scott M Taylor. 2001. Is Free Trade Good ometrica 82(6): 2295–326.
for the Environment? American Eco- Cook, C Justin, and Manisha Shah. 2020.
nomic Review 91(4): 877–908. Aggregate Effects from Public Works:
Arrow, Kenneth, Bert Bolin, Robert Evidence from India. Review of Econom-
Costanza, Partha Dasgupta, Carl Folke, ics and Statistics: 1–38.
Crawford S Holling, Bengt-Owe Jansson, Copeland, Brian R, and Scott M Taylor. 2004.
Simon Levin, Karl-Göran Mäler, Charles Trade, Growth, and the Environment. Jour-
Perrings, et al. 1995. Economic Growth, nal of Economic Literature 42(1): 7–71.
Carrying Capacity, and the Environment. Dasgupta, Partha. 2007. The Idea of Sustain-
Ecological Economics 15(2): 91–5. able Development. Sustainability Science
Asher, Sam, Ryu Matsuura, Tobias Lunt, and 2(1): 5–11.
Paul Novosad. 2019. The Socioeconomic Dasgupta, Susmita, Benoit Laplante, Hua Wang,
High-resolution Rural-Urban Geographic and David Wheeler. 2002. Confronting the
Dataset on India (SHRUG). Working paper Environmental Kuznets Curve. Journal of
C-89414-INC-1. International Growth Economic Perspectives 16(1): 147–68.
Centre. Den Butter, FAG, and Harmen Verbruggen.
Asher, Sam, Teevrat Garg, and Paul Novosad. 1994. Measuring the Trade-Off between
2020. The Ecological Impact of Transpor- Economic Growth and a Clean Environ-
tation Infrastructure*. Economic Journal ment. Environmental and Resource Eco-
130(629): 1173–99. https://doi.org/10. nomics 4(2): 187–208.
1093/ej/ueaa013. Felkner, John S, and Robert M Townsend. 2011.
Assunç~ ao, Juliano, Clarissa Gandour, Romero The Geographic Concentration of Cnter-
Rocha, and Rudi Rocha. 2019. The Effect prise in Developing Countries. Quarterly
of Rural Credit on Deforestation: Evidence Journal of Economics 126(4): 2005–61.
from the Brazilian Amazon: Effect of Rural Ferraro, Paul J, and Rhita Simorangkir. 2020.
Credit on Deforestation. Economic Journal Conditional Cash Transfers to Alleviate Pov-
130(626): 290–330. https://doi.org/10.1093/ erty Also Reduced Deforestation in Indone-
ej/uez060. sia. Science Advances 6(24): eaaz1298.
Assunç~ ao, Juliano, Molly Lipscomb, Ahmed Foster, Andrew D, and Mark R Rosenzweig.
Mushfiq Mobarak, and Dimitri Szerman. 2003. Economic Growth and the Rise of
2017. Agricultural Productivity and Defor- Forests. Quarterly Journal of Economics
estation in Brazil. Mimeo. 118(2): 601–37.
Garg, Teevrat. 2019. Ecosystems and Human Tropics? Implications for Climate Change
Health: The Local Benefits of Forest Cover Adaptation Policies. Global Environmen-
in Indonesia. Journal of Environmental tal Change 56: 29–40.
Economics and Management 98: 102271. Masuda, Yuta, Teevrat Garg, Ike Anggraeni,
Gaur, RD. 2007. Biodiversity and River Valley Nicholas H Wolff, Kristie L Ebi, Edward
Projects in Uttarakhand. Proceedings of Eddie Game, Jennifer Krenz, and June
the National Academy of Sciences India. Spector. 2020. Heat Exposure from
Section B, Biological Sciences 77(3): 253–62. Tropical Deforestation Decreases Cog-
Gibson, Luke, Tien Ming Lee, Lian Pin Koh, nitive Performance of Rural Workers:
Barry W Brook, Toby A Gardner, Jos An Experimental Study. Environmental
Barlow, Carlos A Peres, Corey JA Research Letters 15(12): 1–12.
Bradshaw, William F Laurance, Thomas Meeks, Robyn, Katharine RE Sims, and Hope
E Lovejoy, et al. 2011. Primary Forests Thompson. 2019. Waste Not: Can House-
Are Irreplaceable for Sustaining Tropical hold Biogas Deliver Sustainable Develop-
Biodiversity. Nature 478(7369): 378–81. ment? Environmental and Resource
Government of India. 2003. Office Memoran- Economics 72(3): 763–94.
dum. Ministry of Commerce & Industry, Shenoy, Ajay. 2018. Regional Development
No. 1 (10)/2001-NER. Through Place-Based Policies: Evidence
Greenstone, Michael, and B Kelsey Jack. from a Spatial Discontinuity. Journal of
2015. Envirodevonomics: A Research Development Economics 130: 173–89.
Agenda for an Emerging Field. Journal Shiva, Vandana, and Jayanta Bandyopadhyay.
of Economic Literature 53(1): 5–42. 1986. The Evolution, Structure, and Impact
Grossman, Gene M, and Alan B Krueger. of the Chipko Movement. Mountain
1995. Economic Growth and the Environ- Research and Development 6(2): 133–142.
ment. Quarterly Journal of Economics 110 Stern, David I. 2004. The Rise and Fall of the
(2): 353–77. Environmental Kuznets Curve. World
Intergovernmental Panel on Climate Change Development 32(8): 1419–39.
(IPCC). 2014. Climate Change 2014: Syn- Stern, David I, Michael S Common, and
thesis Report. Intergovernmental Panel Edward B Barbier. 1996. Economic
on Climate Change. Growth and Environmental Degradation:
. 2019. Global Warming of 1.5 Degree The Environmental Kuznets Curve and
C: Special Report on the Impacts of Global Sustainable Development. World Devel-
Warming. Geneva, Switzerland: IPCC. opment 24(7): 1151–60.
Jayachandran, Seema, Joost de Laat, Eric F Tillin, Louise. 2013. Remapping India: New
Lambin, Charlotte Y Stanton, Robin States and Their Political Origins.
Audy, and Nancy E Thomas. 2017. Cash Oxford, UK: Oxford University Press.
for Carbon: A Randomized Trial of Pay- Townshend, John, Matthew Hansen, Mark
ments for Ecosystem Services to Reduce Carroll, Charlene DiMiceli, Robert
Deforestation. Science 357(6348): 267–73. Sohlberg, and Chengquan Huang. 2011. User
Kline, Patrick, and Enrico Moretti. 2014. Guide for the MODISVegetation Continuous
Local Economic Development, Agglom- Fields product Collection 5 Version 1. College
eration Economies, and the Big Push: Park, MD: University of Maryland.
100 Years of Evidence from the Tennes- United Nations. 2015. Transforming Our World:
see Valley Authority. Quarterly Journal The 2030 Agenda for Sustainable Develop-
of Economics 129(1): 275–331. ment. https://sustainabledevelopment.un.
Mahadevan, Meera. 2020. The Price of Power: org/content/documents/21252030Agenda
Costs of Political Corruption in Indian forSustainableDevelopmentweb.pdf
Electricity. Department of Economics, Wilebore, Beccy, Maarten Voors, Erwin H
University of Michigan. Bulte, David Coomes, and Andreas
Masuda, Yuta J, Brianna Castro, Ike Kontoleon. 2019. Unconditional Transfers
Aggraeni, Nicholas H Wolff, Kristie Ebi, and Tropical Forest Conservation: Evidence
Teevrat Garg, Edward T Game, Jennifer from a Randomized Control Trial in Sierra
Krenz, and June Spector. 2019. How Are Leone. American Journal of Agricultural
Healthy, Working Populations Affected Economics 101(3): 894–918.
by Increasing Temperatures in the

3 Garg Shenoy AJAE 2022

Uploaded by

Copyright:

Available Formats

3 Garg Shenoy AJAE 2022

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

3 Garg Shenoy AJAE 2022

Uploaded by

Copyright:

Available Formats

14678276, 2021, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12193 by University Of Delhi, Wiley Online Library on [11/02/2023].

Key words: Agglomeration, deforestation, place-based economic policies.

JEL codes: H54, O40, Q53, Q56.

Amer. J. Agr. Econ. 103(4): 1239–1250; doi:10.1111/ajae.12193

We exploit a spatial discontinuity in the and environmental quality—often characterized

Forest cover. Detailed and reliable administra- Research Design

The spatial polynomial speciﬁcation assumes

Finally, we estimate average program a nonparametric approach by clustering the

+ βC ½Post t × ½Targetedi + ½Errori,t

Figure 3. Difference-in-discontinuities estimate of effect of place-based policy on deforesation.

Table 1. Difference-in-Discontinuities Estimate of Place-Based Policies on Tree Cover

denoted at conventional levels.

Table 2. Regression Discontinuity Estimate of Place-Based Policy on Employment and Firms

Supplementary Material Baskaran, Thushyanthan, Brian Min, and

You might also like