Why are firms that export cleaner?
International trade and CO2 emissions∗
Rikard Forslid†, Toshihiro Okubo‡, and Karen Helene Ulltveit-Moe§
November x2011
Revised version of CEPR DP 8583
Abstract
This paper develops a model of trade with heterogenous firms, where firms make abatement investments and thereby have an impact on their level of emissions. The model shows
how firm productivity and firm exports are both positively related to investments in abatement. Emission intensity is, however, negatively related to firms’ productivity and exports.
The basic reason for these results is that a larger production scale supports more fixed
investments in abatement and, in turn, lower emissions per output. In contrast to the standard models of heterogeneous firms, firms’ productivity, and thus export performance, is
not exogenous, but endogeneously determined by firms’ investment in abatement. We derive
closed form solutions for firm-level abatement investments and emissions per output, and
test the empirical implications of the model using detailed Swedish firm-level data. The
empirical results strongly support the model.
JEL Classification: D21, F12, F15
Keywords:heterogeneous firms, CO2-emissions, international trade
1
Introduction
(test)There is no consensus on the effect of international trade on the environment, in particular
on the effect of trade on global emissions. Neither the theoretical nor the empirical literature
provides a clean cut answer to the link between trade and CO2 emissions. We do not know if
international trade increases or decreases the emissions of greenhouse gases, thereby contributing
∗
We are grateful for comments from Andrew Bernard, Adrian Wood, Tony Venables, Beata Javorcik and
Peter Neary, and thank participants at the GIST conference in Stockholm May 2011, the ETSG conference 2011
and seminar participants in Oxford for valuable discussion. Financial support from Grant-in-Aid for Scientific
Research (JSPS) and Research Institute of Economy, Trade and Industry (RIETI) is gratefully acknowledged by
Okubo.
†
Stockholm University and CEPR; email: [email protected]
‡
Keio university, e-mail: [email protected]
§
University of Oslo and CEPR, e-mail: [email protected]
1
to global warming. But this paper sets out to shed some further light on the forces at work
by focusing on inter-firm productivity differentials and interdependence among productivity,
abatement and emissions.
In theoretical neoclassical models, there are opposing effects. On the one hand, trade increases income, which will tend to increase the demand for a clean environment and therefore
increase investments in clean technology and abatement. However, trade liberalisation may
also imply an overall expansion of dirty production, because trade allows countries with low
emission standards to become ”pollution havens”. Copeland and Taylor (1995) show how trade
liberalisation increases global emissions if the income differences between the liberalising countries are large, as dirty industry expands strongly in the poor country with low environmental
standards.
The empirical literature on the link between trade in goods and emissions is also inconclusive.
Using macro data, Antweiler et al. (2001) and Frankel and Rose (2005) find that trade decreases
emissions. Using sector-level data for the U.S., Ederington et al. (2004) do not find any evidence
that pollution intensive industries have been disproportionately affected by tariff changes. On
the other hand, also using sector-level trade data, Levinson and Taylor (2008) find evidence
that higher environmental standards in the US have increased the imports from Mexico in dirty
industries. Holladay (2009) analyses firm-level data for the US, and find that exporters pollute
less per output, while on the other hand, also using firm level data, Batrakova and Davies (2010)
find that, for low fuel intensity firms, fuel expenditures are positively correlated with exporting.
Finally, Rodrigue and Soumonni (2011) employ Indonesian firm-level data to investigate the
impact of environmental investment on productivity dynamics and exports. While productivity
dynamics do not appear to be affected, growth in exports does.
It has been pointed out (see e.g. Levinson, 2009) that trade may have reduced pollution
as trade liberalization may encourage technological upgrading. Our paper is closely related
to this idea. We present a model of trade with differentiated goods and heterogenous firms
(see Melitz, 2003) where firms make abatement investments and thereby affect their level of
emissions. We show that more productive firms find it profitable to invest in abatement as their
bigger volumes allow them to spread the fixed costs of abatement investment across more units.
Exporting increases investment in abatement, since exports sales boost the scale of production.
Hence, firm productivity and firm exports are both positively related to abatement investments, while emission intensity is negatively related to firms’ productivity and exports.
In contrast to the standard models of heterogeneous firms, firms’ productivity and thus, also
export performance, is not exogenous, but endogenously determined by firms’ investment in
abatement
We derive closed form solutions for firm-level abatement investments and firm-level emission
intensities (emissions per unit produced). These equations are tested on detailed Swedish firmlevel data. Our dataset contains firm-level emissions and firm-level abatement expenditures as
2
well as firm exports.1 We focus on CO2 emissions, which constitute about 80 percent of the
greenhouse gases emitted by Swedish firms (Ministry of the Environment 2009). The empirical
results are overall strongly supportive of the results derived in the theoretical model; more
productive firms abate more and emit less, and exports are associated with more abatement
and lower emission intensity.
2
The Model
To analyse the impact of trade on global CO2 emission, we use an augmented version of Melitz’
seminal model on monopolistic competition, heterogeneous firms and trade (see Melitz, 2003).
More specifically, we introduce emissions, CO2 taxes and abatement investments in the standard
model. We assume there to be two countries with production in two sectors, agriculture (A)
and manufacturing (M ). Each country j has a single primary factor of production, labour, Lj ,
used in both the A and the M sector. The A sector produces homogenous goods subject to
constant returns to scale. The M sector is characterized by increasing returns, monopolistic
competition and heterogeneous firms.
2.1
Demand
Consumers preferences are given by a two-tier utility function with the upper tier (CobbDouglas) determining the representative consumer’s division of expenditure between agricultural
and manufactured goods (sectors A and M ), and the second tier (CES), giving the consumer’s
preferences over the continuum of differentiated varieties produced within the manufacturing
sector. Hence, all individuals in country j have the utility function
1−µ
µ
Uj = CM
j CAj ,
(1)
where µ ∈ (0, 1) and CAj is consumption of the homogenous good. Agricultural (A) goods
can be costlessly traded internationally and are produced under constant returns to scale and
perfect competition. The A-good is chosen as the numeraire, so that the world market price of
the agricultural good, pA , is equal to unity. By choice of scale, the unit labour requirement in
the A-sector is one, which gives
p A = wj = w = 1
(2)
and thus, wages are normalized to one across both countries and sectors. This holds provided
that µ < 0.5, which implies that the demand for agricultural goods is sufficiently large to
guarantee that the agricultural sector is active in both countries irrespective of the location of
manufacturing. The consumption of final goods from the manufacturing sector is defined as an
1
There are very few studies using firm-level data on CO2 emissions. One exception is Cole et al. (2011).
3
aggregate CM j ,
σ/(σ−1)

Z
c (i)(σ−1)/σ di
CM j = 
,
(3)
i∈I
where c(i) represents consumption of each variety with elasticity of substitution between any
pair of differentiated goods being σ > 1. The measure of set I represents the mass of varieties
consumed in country j. Each consumer spends a share µ of his income on manufactures, and
demand for each single variety produced in country k and consumed in country j is therefore
given by
xj =
p−σ
jk
Pj1−σ
µYj ,
where pjk is the consumer price, Yj is income, and Pj ≡
(4)
N
Rj
!
p (i)1−σ di
1
1−σ
the price index of
0
manufacturing goods in country j.
2.2
Entry, exit and costs in the manufacturing sector
To enter the manufacturing sector in country j, a firm bears the fixed costs of entry fE measured
in labour units. After having sunk fE , an entrant then draws a labour-per-unit-output coefficient
a from a cumulative distribution G(a). We follow Helpman et al. (2004) in assuming the
probability distribution to be a Pareto distribution,2 i.e. G(a) =
ak
,
ak0
where we normalise the
scale parameter to unity, a0 ≡ 1. Upon observing this draw, a firm may decide to exit and not
produce. If it chooses to stay, it bears the additional fixed overhead labour costs, fD . If the firm
does not only want to serve the domestic market but also wants to export, it has to bear the
additional fixed labour costs, fX . Hence, firm technology is represented by a cost function that
exhibits a variable cost and a fixed overhead cost. In the absence of emissions and abatement
investment, labour is used as a linear function of output according to
l = f + xa
(5)
with f = fD for firms only serving the domestic market and f = fD + fX for exporters. Let us
now also account for the fact that manufacturing activity entails pollution in terms of emission of
CO2. We follow Copeland and Taylor (2003) and assume that each firm produces two outputs:
a manufactured good (x) and emissions (e). Abatement is possible, but this requires diverting
a fraction θ of the primary factor, labour, away from the production of x. Firms pay the fixed
overhead costs, and thereafter the joint production is given by
x = (1 − θ)
2
l
a
This assumption is consistent with the empirical findings by e.g. Axtell (2001).
4
(6)
e = ϕ(θ)
l
a
(7)
with 0 ≤ θ < 1. Emission intensity (e/x) is thus determined by the abatement function
ϕ(θ) =
(1 − θ)1/α
β (fA )
(8)
which is characterised by ϕ(0) = 1, ϕ(1) = 0, ϕ0 (.) < 0 and 0 < α < 1. Firms take θ as given,
but we depart from the standard formulation of the abatement function by assuming that firms
can have an impact on emission intensity through abatement investments (fA ). The higher the
investment in abatement the lesser the emission intensity, i.e. β 0 (fA ) > 0. Below, we will use
the functional form: β (fA ) = fAρ , with ρ > 0.
We use (??) and (??) to substitute for θ in (??), which gives us an integrated expression
for the joint production technology and exploit the fact that although pollution is an output,
it can equivalently also be treated as an input:3
1−α
l
.
x = (β (fA ) e)
a
α
(9)
Hence, production implies the use of labor as well as emission. Firms minimize costs taking
wages (w = 1) and emission taxes (t > 0) as given and subject to (??). This allows us to
derive their variable costs functions. Disregarding the sunk entry cost (fE ) while adding fixed
overhead costs related to domestic and possibly export activity as well as abatement costs, we
get the total cost function
C = f + fA + κ
t
β (fA )
α
a(1−α) x
(10)
with κ ≡ α−α (1 − α)α−1 , where f = fD for firms only serving the domestic market, and
f = fD + fX for exporters, i.e. firms serving both the domestic and the foreign market. The
cost function makes clear that emissions are not for free, rather they incur a tax t, but through
making larger investments in abatement, emissions can be reduced as can the tax bill. Note
that while firms are heterogeneous with respect to labour productivity, a, they are identical in
all other respects, i.e. they share the same cost function and face the same tax rate.
Our analysis exclusively focuses on steady-state equilibria and intertemporal discounting is
ignored. The present value of firms is kept finite by assuming that firms face a constant Poisson
hazard rate δ of ‘death’ independently of productivity. An entering firm with productivity a
will immediately exit if its profit level π (a) is negative, or will produce and earn π (a) ≥ 0 in
every period until it is hit by a bad shock and forced to exit.
3
See Copeland and Taylor (2003) for a discussion of this feature of the model.
5
2.3
Investing in abatement and maximizing profits
Firms follow a two-step decision process. First, they invest in abatement, and second, taking
abatement as given they maximize profits with respect to ouput. As usual we proceed by
solving the problem backwards.
Each producer operates under increasing returns to scale at the plant level and in line with
Dixit and Stiglitz (1977), we assume there to be large group monopolistic competition between
manufacturers. Thus, the perceived elasticity of demand equals the elasticity of substitution
between any pair of differentiated goods and is equal to σ. Regardless of its productivity,
each firm then chooses the same profit maximizing markup over marginal costs (M C) equal to
σ/(σ − 1). This yields a pricing rule
pjk =
σ
τjk M C
σ−1
(11)
for each producer in country j, and reflects that shipping manufactured goods involves a frictional trade cost of the “iceberg” form. For each unit of a good from country j to arrive in
country k, τjk > 1 units must be shipped. It is assumed that trade costs are equal in both
directions and that τjj = 1.Firms maximize profits with respect to volume and investment in
abatement. They choose investment in abatement efficiency in order to maximize profits
π=
px
− f − fA ,
σ
(12)
which may be written as
αρ(σ−1)
π = tα(1−σ) a(1−α)(1−σ) fA
using (??), (??), and where B ≡
κσ−1 σ σ (σ−1)1−σ µL
.
P 1−σ
B − f − fA ,
(13)
Equation (??) reveals that an internal
solution to the profitmaximising choice of fA requires that αρ(σ − 1) < 1. We will assume this
condition to hold in the remainder of the paper. The condition implies a decreasing marginal
effect of abatement on firm profit.
We then proceed to the first stage. Since firms’ profits depend on whether thet are exporters
or non-exporters, abatement investments will differ between the to groups of firms. Maximizing
domestic firms’ profits w.r.t. abatement investments fA using (??) and (??) gives
fAD
where B ≡
=
1 1−α α σ−1
1
a
t
B
αρ(σ − 1)
κσ−1 σ σ (σ−1)1−σ µLj
Pj1−σ
1
αρ(σ−1)−1
(1−α)(σ−1)
= ΩD a αρ(σ−1)−1 ,
∀ j is exogenous to the firm and ΩD ≡
n
1
B
(tα )σ−1
(14)
κσ−1
αρ(σ−1)
o
1
αρ(σ−1)−1
while the optimal fixed cost investment in abatement for exporters is
fAx
=
σ−1
1
1
a1−α tα
∗
(B + φB )
αρ(σ − 1)
6
1
αρ(σ−1)−1
(1−α)(σ−1)
= ΩX a αρ(σ−1)−1 ,
(15)
,
with
B∗
≡
κσ−1 σ σ (σ−1)1−σ µLk
Pk1−σ
∀ k 6= j and φ =
τ 1−σ
and where ΩX ≡
n
1
(B+φB ∗ )
(tα )σ−1
1
αρ(σ−1)
o
1
αρ(σ−1)−1
Proposition 1 More productive firms invest more in abatement given that αρ(σ − 1) < 1.
Proof: Investment in abatement rises when firm productivity (1/a) rises since ∂fAD /∂a < 0
and ∂fAX /∂a < 0. These inequalities will always hold as long as the second-order condition for
profit maximization is satisfied, as this requires that ∂ 2 πD /∂fA2 < 0 which, in turn, requires
that (σ − 1) αρ < 1 (see the Appendix).
The logic behind this result is that more productive firms have higher sales. Hence, the
exploiting of scale economies makes it profitable for them to make a higher fixed cost investment
in order to reduce the marginal costs.
Proposition 2 For any given level of productivity, exporters invest more in abatement than
non-exporters, and trade liberalization increases exporting firms’ investments in abatement given
that αρ(σ − 1) < 1.
Proof: Since B < (B + φB ∗ ), then ΩX > ΩD , from which it follows that fAx > fAD for any
given productivity level (1/a). If trade is liberalized (a higher φ), then (B + φB ∗ ) rises as does
ΩX , and thus, fAx .
The intuition for this result is that the production volume increases with exports, and the
exploiting of scale economies warrants a higher fixed cost investment in abatement.
Proposition 3 A higher tax rate on emission leads to lower investments in abatement given
that αρ(σ − 1) < 1.
x
∂fA
∂t
Proof : It follows from (??) and (??) that
1
n
o
(1−ρ)α(σ−1)+1
αρ(σ−1)−1
α(σ−1)
1
1
(1−α)(σ−1)
αρ(σ−1)−1
= αρ(σ−1)−1
t
< 0.
a
∗
(B+φB )
αρ(σ−1)
This result may seem counterintuitive. But the logic behind it is well known from the theory
of production: say that a firm is considering investing in a new machine. Using the machine
requires labour. If the price of labour increases, this decreases the incentive to invest in the
machine. In our case, a higher tax rates implies that the price on emission, i.e. the cost of
emission, rises. As a consequence, firms want to emit less. As firms substitute away from
emissions, it becomes less profitable to invest in abatement.
2.4
Equilibrium conditions
Equilibrium in the model is determined by the zero profit conditions for firms only serving the
domestic market and exporters, respectively
πD = a1−α tα
πX = a1−α tα
1−σ
1−σ
B − fD − fA = 0,
(16)
(B + φB ∗ ) − fD − fX − fA = 0.
(17)
7
.
Since a is unit labour requirement, 1/a depicts labour productivity and, with σ > 1, it follows
that a(1−α)(1−σ) increases along with productivity and can thus be used as a productivity index.
From (??) and (??), we see that profits are increasing in firms’ productivity. The least productive firms expect negative operating profits and therefore exit the industry. This applies to all
(1−α)(1−σ)
firms with a productivity below aD
which is the cutoff at which profits from domestic
sales equal zero, and is determined by
1−α
fD = aD
t
β(fA )
α 1−σ
B − fAD (aD ).
(18)
In order to break even as an exporter, firms have to cover a higher fixed cost, and the cutoff
productivity level at which producers serving the domestic as well as the foreign market just
break even is determined by
fX + fD =
a1−α
X
t
β(fA )
α 1−σ
(φB ∗ + B) − fAX (aX ),
(19)
where φ ≡ τ 1−σ ∈ [0, 1] depicts freeness of trade.
The model is closed by the free-entry condition
fE = na0X {πX (a) − fX − fD − fA (a)} dG(a)
(20)
+naaD
{πD (a) − fD − fA (a)} dG(a).
X
Together, the zero cutoff profit conditions and the free entry condition ensure the existence
and uniqueness of the equilibrium productivity and profit levels. Based on these conditions,
we are also able to derive some results on the relative productivity and the relative abatement
investment of domestic versus exporting firms.
First, the relative cut-off productivities of exporters and non-exporters can be calculated
using the cutoff conditions (??) and (??), and the expressions for optimal abatement investment
(??) and (??):
aX
aD
(1−α)(σ−1)
αρ(σ−1)−1
ΩD
=
ΩX
fX
1+
fD
=
B∗
1+φ
B
1
αρ(σ−1)−1
fX
1+
.
fD
(21)
Endogenous exporter status
Since abatement investments have an impact on firms’ marginal costs, and on the profitability of being a domestic versus an exporting firm, this implies that exporting status is not only
determined by firms’ labour productivity, but also by their abatement investments.We see that
exporters are more productive than non-exporters when trade costs are high (φ close to zero).
This follows from the assumption that the entry cost in foreign markets fX is larger than the
entry cost in the domestic market fD . From (??), it can also be seen that higher foreign market
8
entry costs fX increase the relative productivity of exporters.
A well known feature of the models with heterogeneous firms is that cut-off productivities
of exporters and non-exporters converge as trade is liberalised (φ increased). While our model
shares this feature, it differs from the standard models as we may get aD = aX before trade is
completely liberalized (φ = 1).4 The reason for this is that in our model, firms’ marginal costs
are endogenous and depend on investment in abatement. As a consequence, and unlike what we
find in the standard models of heterogeneous firms and trade, firms’ export status is not purely
given by their randomly drawn productivity, but also by their deliberate choice of abatement
investment. If the foreign market is sufficiently large, all firms will find it optimal to invest in
abatement to lower their marginal cost and thereby become exporters.
Finally, we note that the tax on emissions affects exporters and non-exporters in the same
fashion and it will therefore not affect the relative cut-off productivities. However, since the
tax on emissions will have an impact on abatement investment, everything else equal, the tax
rate will have an impact on the profitability of domestic versus exporting firms and thus, on the
choice of becoming an exporter, and then, in turn, on the number of exporters in the economy.
2.5
Emissions
Taking abatement investment as given, firms decide on the use of labour as well as on emissions.
We focus on emissions and use Shepard’s lemma on the cost function (??) to derive optimal
emissions and emission intensity (emissions per output):
eD
−ρα 1−α
i
= ακtα−1 fAi
ai .
xi
(22)
After having substituted for optimal fA using (??) and (??), respectively, we state that the
emission intensity of non-exporters is given by
1−α α−η
eD
− ρα
i
= ai η t η B η Ψ,
xi
where η ≡ 1−ρα(σ −1) > 0 and Ψ ≡
ρα
− η
α(ρα(σ−1))
η
(23)
κ − ρα(σ − 1)2 . We assume that ρ is small
enough so that Ψ > 0. This limits the effect of the abatement investments, which prevents firms
from completely substituting away from the use of emissions. For exporters, the expression is
given by:
1−α α−η
eX
− ρα
i
= ai η t η (B + φB ∗ ) η Ψ.
xi
Several results emerge directly from equations (??) and (??).
Proposition 4 More productive firms have a lower emission intensity.
4
The critical trade cost is determined by φ =
B
B∗
fD
fX +fD
9
αδ(σ−1)−1
−1 .
(24)
Proof: Since profit maximization
with respect to abatement investment requires that (σ − 1) αρ <
1 (see the Appendix), then ∂
eD
i
xi
/∂a > 0.
Proposition 5 For any given level of productivity, exporters have a lower emission intensity than non-exporters, and trade liberalisation (higher φ) reduces the emission intensity of
exporters.
Proof: Since B < (B + φB ∗ ), then
eD
i
xi
>
eX
i
xi
must be true for any given level of a, exporters
will have a lower emission intensity than non-exporters.
If trade is liberalized (a higher φ), then
(B + φB ∗ ) rises, while
eX
i
xi
declines, i.e. ∂
eD
i
xi
/∂φ < 0.
These two propositions are closely related to those on abatement investments. Firms with
higher sales and thus a higher production volume will invest more in abatement, and they will
have lower emission intensities.
Proposition 6 Higher emission taxes lead to lower emission intensity when η > α.
Proof:
This can be seen directly from equations (??) and (??).
Note that aggregate emissions may decrease even in the case when the firms’ emission
intensity increases in the emission tax (when η < α), since a higher tax rate weeds out some of
the least productive firms.5
3
Testable implications
We focus our empirical work on the model’s predictions of the relationship among (i) productivity and abatement investments, (ii) productivity and emission intensity, (iii) exporting and
abatement investments and (iv) exporting and emission intensity. The reduced form solutions
for emission intensity, see equations (??) and (??), suggest a log linear specification of the
following type:
log emission intensityi = α0 + α1 log productivityi + α2 exportdummyi + εi ,
(25)
where the model predicts that β1 , β2 > 0. The tax rate on emissions (t in the model) does not
vary between firms and it is therefore controlled for by time fixed effects.
As for the reduced form solutions for abatement investments, see equations (??) and (??),
they suggest log linear specifications as:
log abatement investmentsi = β0 + β1 log productivityi + β2 exportdummyi + εi ,
(26)
where theory predicts α1 , α2 < 0.
5
The second-order condition for optimal investments in abatement is that αρ(σ − 1) < 1. The condition that
η = 1 − αρ(σ − 1) > α therefore implies a restriction on α, which must be sufficiently small.
10
However, according to our model, we encounter an endogeneity problem when analysing the
impact of export status on abatement. The reason for this is that in our model, firms’ marginal
costs are endogenous and depend on the investment in abatement. As pointed out above, unlike
what we find in the standard models of heterogeneous firms and trade, firms’ export status is not
purely given by their randomly drawn productivity, but is also determined by firms’ abatement
investments. When estimating (??), we must therefore instrument for firms’ export status.
Details on choice and construction of instruments are given below.
4
Empirical implementation
4.1
Data
We use manufacturing census data from Statistics Sweden.6 The dataset contains information
on a large number of variables such as export, employment, capital stock (tSEK), use of raw
materials, value added and production value, at the firm level from 2000-2007. We construct
CO2 emissions using data on all types of fuel use at the plant level together with the relevant fuel
coefficients provided by Statistics Sweden. Data on fuel use is collected for all manufacturing
plants with more than 10 employees from 2004-2007. Plant emissions are aggregated to the
firm level. This gives CO2 emission data (kg/mioSEK) for about 5 000 firms. We also use firm
level data on abatement investments per firm (tSEK) over the whole period 2000-2007. The
abatement data is a semi-random sample of manufacturing firms (all firms with more than 250
employees, 50 percent of the firms with 100-249 employees, and 20 percent of the firms with
50-99 employees). Thus, the abatement data is collected for fewer and larger firms and the
data set covers about 1 500 firms. This implies that almost 95 percent of the firms in the latter
sample are exporters, which makes identification of the export dummy harder.
Table ?? and table ?? illustrate that there are significant differences across firms in our
datasets. It can be seen that exporters, on average, are larger and more capital intensive than
non-exporters. Exporters also, on average, invest more in abatement and emit less CO2 per
output.
Table 1: Descriptives for abatement data
Abatement inv.
Capital/labour
Employment
6
obs.
5450
6971
6997
All
mean std.dev.
2801
15310
337.4
1100
446.0
1157
obs.
5226
6598
6619
The sector classification is shown in the appendix.
11
Exporters
mean std.dev.
2838
15501
344.6
1122
459.3
1179
Non-Exporters
obs. mean std.dev.
224 1919
9820
373 209.8
559.0
378 213.2
620.5
Table 2: Descriptives for CO2 data
CO2 per output
Capital/labour
Employment
4.2
obs.
14325
14262
14384
All
mean
13.70
226.1
128.3
std.dev.
249.5
671.3
681.4
obs.
11026
10980
11067
Exporters
mean std.dev.
11.65
160.2
248.1
746.2
158.3
770.9
Non-Exporters
obs. mean std.dev.
3299 20.56
429.6
3282 152.5
296.8
3317 28.28
134.5
Abatement and export
We start by estimating equation (??) with OLS. Based on out theoretical model, we expect
more productive firms to abate more. We also expect exporting firms to have higher abatement
investments. In Table ?? we report the results with the log of total abatement investments as the
dependent variable. Firms’ productivity is measured as TFP (estimated using the LevinsohnPetrin method), while we use an export dummy to identify exporters.7 We also employ the
capital labour ratio as a control variable. We do not control for size, since the left-hand side
variable is given by the absolute level of investment. To account for changes in the CO2 tax rate
over time, we use year fixed effects, but it turns out that our results are actually little affected
by this. All variables are in logs, and errors are clustered at the firm level.
Table 3: Abatement investments, productivity and exporting (OLS)
(1)
ln TFP
Export dummy
1.486a
(.335)
ln capitalintensity (K/L)
Industry dummies
Year dummies
R-squared
Number of obs.
No
Yes
.01
5450
Dependent variable: ln abatement investments
(2)
(3)
(4)
(5)
(6)
(7)
.258a
.247a
.151a
.970a
.970a
(.061) (.061) (.050)
(.173) (.173)
1.354a
.519c
.998a
1.041a
(.335) (.282) (.300)
(.296)
1.183a
(.085)
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
.02
.03
.17
.13
.14
.14
5338
5338
5336
5450
5338
5338
(8)
.987a
(.149)
.241
(-.267)
1.019a
(.085)
Yes
Yes
.22
5336
Note: Estimates are based on the panel 2000-2007. Errors are clustered at the firm level.
a significant at 1% level, b significant at 5% level, c significant at 10% level.
The coefficient for productivity is positive and significant at the one percent level in all
regressions. The export dummy is positive and significant regardless of whether we control
for productivity differentials, but becomes insignificant in the within sector estimation when
controlling for capital intensity. The results are overall supportive of the theoretical model.
7
Exporters are defined as firms exporting more than 10 000 SEK (about 1 000 euros) per year.
12
But as pointed out above, estimating the impact of exporting on abatement investments we
encounter an endogeneity problem regarding the export variable. Our model predicts that exporters and high-productivity firms have higher investments in abatement. However, according
to theory, firms’ exporting status depends on their abatement investments. We aim at correcting for this by instrumenting for the export dummy using firm-specific exchange rates. We
construct the instrument variables using the USD/SEK exchange rate times firms’ export share
in the year 2000 (the first year of our panel). Firms that do not export that year are assigned
the industry (nace 2) average export share,8 the idea being that firms with a relatively high
export share are relatively more affected by the exchange rate. First stage estimation results
are reported in Table ??. The instrument is strongly correlated with the endogenous regressor
with a t-value of 9.0 and 3.9 respectively in regression 1 and 2 (thus passing the F-stat.¿10 test
for weak instruments).
Table 4: First stage estimation
expshare*exchrate
ln TFP
ln capitalintensity (K/L)
Industry dummies
Year dummies
R-squared
Number of obs.
(1)
0.017a
(0.0019)
0.0075a
(0.0024)
0.025a
(0.0049)
No
Yes
0.072
6817
Dependent variable: export dummy
(2)
0.0068a
(0.0017)
-0.0065
(0.0074)
a
0.033
(0.0053)
Yes
Yes
0.15
6817
Note: Estimates are based on the panel 2000-2007. Errors are clustered at the firm level.
a significant at 1% level, b significant at 5% level, c significant at 10% level.
The results from the second stage estimation are reported in Table ??.
8
Industry 10 (Coke) consists of 10 firms that do not export. This means that there is no industry average
export share and these observations are therefore dropped.
13
Table 5: Abatement investment, productivity and exporting (IV)
ln TFP
Export dummy
ln capitalintensity (K/L)
Industry dummies
Year dummies
Number of obs.
Dependent variable: ln abatement investments
(1)
(2)
(3)
(4)
(5)
(6)
.186a
.130b
.995a
.999a
(.067) (.056)
(.207)
(.167)
1.604a 1.334a
.563b
1.802a 1.676a
.860
(.258) (.267) (.228) (.591) (.564)
(.633)
1.079a
.786a
(.102)
(.213)
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
5434
5322
5320
5434
5322
5320
Note: Estimates are based on the panel 2000-2007. Errors are clustered at the firm level.
a significant at 1% level, b significant at 5% level, c significant at 10% level.
Comparing the results with and without instruments (see Tables ?? and ??), we observe
that export status has a positive significant impact on abatement also after instrumenting.
Eyeballing our data on abatement, we realise that abatement investments are, nevertheless,
quite volatile for individual firms over the years. Therefore, we calculate average abatement
investments for our period of observation 2000-2007, and reestimate (??) based on the resulting
cross-section data set. The OLS results are reported in Table ??. We use explanatory variables
from the sample mid-point, 2004, but robustness checks prove that the choice of year is of
less importance. As before, abatement investments turn out to be positively and significantly
affected by firms’ productivity and export status. This also holds when controlling for capital
intensity.
Table 6: Abatement investments, productivity and export (OLS), III
ln TFP
Export dummy
ln capitalintensity (K/L)
Industry dummies
R-squared
Number of obs.
Dependent variable: ln average abatement 2000-2007
(1)
(2)
(3)
.311b
.335b
(.131) (.125)
.823a
.828b
.639b
(.343) (.338) (.322)
.514a
(.061)
Yes
Yes
Yes
.23
.23
.31
655
642
641
Note: Errors are clustered at the firm level.
a significant at 1% level, b significant at 5% level, c significant at 10% level.
As a final robustness check, we look at how abatement investments are affected by firms
14
switching from being non-exporters to exporters. We would expect non-exporters to carry
out particularly high investments in abatement as they switch from being non-exporters to
exporters. Table ?? shows the average abatement investments for non-exporters, exporters,
and switchers for the period 2000-2007.
.
Table 7: Export status, productivity, size and capital intensity
Switchers
Exporters
Non-exporters
Abatement
investment
27.78
27.38
3.34
TFP
Size (employees)
1.18
2.80
0.96
172.26
206.15
61.33
Capital
intensity
93.99
176.71
58.51
Note: All number are averages for 200-2007
All variables are in logs.
According to what we would expect, Table 4 shows that exporters, which are the largest,
most productive, and most capital intensive firms, have much higher abatement investments
than non-exporters, while the highest abatement investments are found among the switching
firms, i.e. the firms that switch from being non-exporters to exporters during our period of
observation. We further investigate the relationship between abatement and switching export
status by running a probit panel regression with the switch from non-exporter to exporter as the
dependent variable. We only consider pure switches while dropping cases where firms alternate
back and forth between being non-exporters and exporters. Table ?? shows the results.
Table 8: Switching to exporting: the role of abatement investments (probit)
Dependent variable: Probablity of switching to exporting
(1)
ln abatement investment, t
.035b
ln abatement investment, t-1
(.016)
.047a
(.016)
ln abatement investment, t-2
(2)
(3)
(4)
(5)
(6)
.037
(.030)
.048c
(.027)
.044c
(.025)
.036b
.025
(.020)
.037b
(.0.17)
.136a
(.035)
.035
(.030)
.052c
(.030)
.045c
(.027)
.212a
(.055)
No
Yes
3391
Yes
Yes
2360
.025
(.030)
.038
(.030)
.031
(.030)
.208a
(.058)
.175b
(.084)
Yes
Yes
2359
ln TFP
(.017)
.048a
(.017)
ln capitalintensity (K/L)
Industry dummies
Year dummies
Number of obs.
No
Yes
3452
No
Yes
2395
.129a
(.036)
.152a
(.057)
Yes
Yes
3389
Note: Estimates are based on the panel 2000-2007.
a significant at 1% level, b significant at 5% level, c significant at 10% level.
15
We consider the impact of abatement investment in the period where the switch takes place,
and the impact of abatement investment taking place one and two periods before the switch
on firms’ probability to become exporters.9 In particular, abatement investments lagged one
time period turn out to be a significant predictor of firms’ switching to exporting. This is
consistent with firms investing in abatement to become exporters. A weakness with our data
is that almost all firms are exporters since abatement data is primarily sampled from larger
firms.10 This means that there are very few switchers; just somewhat more than ten per year
on average. This leaves us with little variation, and reduces the potential for identification.
Adding lags further diminishes the sample and in the last regression with all variables included,
we do not get any significant coefficients for abatement, even if the estimated coefficients are of
the expected sign and very similar to the significant ones in terms of magnitude.
4.3
CO2-emission intensity
Next we move on to estimate equation (??). Our model predicts that the firm-level emission
intensity should be negatively affected by productivity and export status. Emission intensity is
measured as firm-level CO2 emissions per output. We have emission data for the years 20042007. Emission tax rates do not vary between firms, and are therefore not controlled for.11
We control for emission tax changes by including time fixed effects. However, the exclusion of
year dummies does not affect the estimates in any significant way. We report regression results
where errors are clustered at the firm level. Clustering at the sector level gives very similar
results. Table ?? shows the OLS regression results.
9
We have also run the regressions with productivity and capital labour ratio lagged one period with almost
identical results.
10
The abatement data is a random sample of all manufacturing firms with more than 250 employees, 50 percent
of the firms with 100-249 employees, and 20 percent of the firms with 50-99 employees.
11
Manufacturing firms enjoy a rebate on CO2-taxes in Sweden. The same rebate (and tax rate) applies to all
firms in our sample.
16
Table 9: CO2 emissions, productivity and exporting (OLS)
Dependent variable: ln CO2 emission intensity
(1)
ln TFP
(2)
(3)
(4)
(5)
-.060a
(.023)
-.060b
(.024)
-.460a
(.050)
-.055b
(.022)
-.550a
(.050)
.180a
(.023)
No
Yes
.02
14038
No
Yes
.03
14027
-.060b
(.023)
-.250a
(.052)
.250a
(.023)
-.330a
(.029)
No
Yes
.07
14027
-.450a
(.050)
Export dummy
ln capitalintensity (K/L)
ln employment
Industry dummies
Year dummies
R-squared
Number of obs.
No
Yes
.01
14325
No
Yes
.01
14038
(6)
(7)
(8)
-1.020a
(.072)
-.970a
(.073)
-.250a
(.047)
Yes
Yes
.21
14038
Yes
Yes
.21
14038
-.420a
(.047)
Yes
Yes
.17
14325
(9)
-.980a
(.073)
-.270a
(.048)
.040c
(.021)
Yes
Yes
.21
14027
Note: Estimates are based on the panel 2000-2007. Errors are clustered at the firm level.
a significant at 1% level, b significant at 5% level, c significant at 10% level.
The coefficients for productivity and the export dummy are negative and strongly significant
in all regressions. The capital-labour ratio is positively significant at the one percent level in all
regressions as is firm size (employment). In terms of magnitudes, export has a relatively small
effect on emissions, maybe some 0.3-0.5 percent decrease in emission intensity, after controlling
for productivity. Productivity appears to be more important. In the within sector estimates,
we find a negative emission intensity elasticity w.r.t. productivity of around one percent.
The estimates can be used to back out some of the structural parameters of the model. It is
seen from (??) and (??) that the ratio of the estimated coefficients for productivity in table ??
and table ?? equals 1 − σ. Likewise, is it possible to calculate ρ from these estimated coefficients
if a value for α is assumed. 1 − α is the labour share in the variable production cost, and in the
table below we let this share vary between 0.5 and 0.75.
Table ?? shows the imputed values of σ and αρ,where αρ is the elasticity of the emisssion
intensity to abatement investments, as seen from (??).
.
Table 10: Implied values for structural parameters
Sector dummies
Yes
No
σ
2 − 2.3
3.2 − 4.7
αρ
(1 − α = 0.5)
1.3 − 1.5
3.0 − 4.3
αρ
(1 − α = 0.75)
1.5 − 1.8
4.3 − 6.2
The imputed values for the elasticity of substitution, σ, are relatively well in line with
standard estimates of this parameter (cite***). When it comes to the effectiveness of abatement
17
(10)
-.760a
(.071)
-.100b
(.049)
.090a
(.021)
-.250a
(.028)
Yes
Yes
.23
14027
investments, the estimates with sector fixed effect probably produces more relevant values with
elasticities somewhat above one.
5
Conclusion
This paper analyses emissions and endogenous abatement investments within a framework with
heterogeneous firms and trade. The model shows how firms’ productivity and firms’ export
status are both positively related to investments in abatement. Emissions per output, in turn,
are negatively related to firm-level productivity and firm exports. The basic reason for these
results is that a larger production scale supports more fixed investments in abatement. As a
consequence, and unlike what we find in the standard models of heterogeneous firms and trade,
firms’ export status is not purely given by their randomly drawn productivity, but also by their
deliberate choice of abatement investment. If the foreign market is sufficiently large, all firms
will find it optimal to invest in abatement to lower their marginal cost and thereby become
exporters.
We derive closed form solutions for firm-level abatement expenditures and firm-level emissions per output, and test these using Swedish firm-level data that contains firm-level abatement
investments and firm-level emissions of CO2. The empirical results show that abatement investment is positively related to export status and productivity. A probit regression for switchers
from non-export to export status shows that lagged abatement investments are also a significant
predictor of a switch to becoming an exporter.
The empirical results also strongly support the notion that the firm-level CO2 emission
intensity (CO2-emissions per output) is negatively related to firm productivity and to being an
exporter. Overall, the estimations provide strong evidence in favour of our theoretical model.
The paper provides evidence of one mechanism whereby international trade can be beneficial
for the environment, since trade promotes investments making production cleaner. This effect
stands in stark contrast to e.g. the pollution haven hypothesis, which suggests that international
trade will decrease the effects of environmental regulations by making it easier for firms to
expand polluting activities in countries with less stringent environmental standards. Our results
therefore suggest one explanation for the mixed empirical evidence on the effects of globalisation
on emissions.
18
References
[1] Antweiler, W., Copeland, B.R. and Taylor, M.S. (2001), ”Is Free Trade Good for the
Environment?” American Economic Review, Vol. 91, no. 4, pp. 877-908 (Sept).
[2] Axtell, R. L. (2001), ”Zipf distribution of u.s. firm sizes”, Science 293.
[3] Batrakova, S. and R. B. Davies, (2010). ”Is there an environmental benefit to being an
exporter? Evidence from firm level data,” Working Papers 201007, School Of Economics,
University College Dublin.
[4] Copeland, B.R. and M.S. Taylor (1995), ”Trade and Transboundary Pollution,” American
Economic Review, 85 : 716-737.
[5] Cole, M.A. and Elliott, R.J.R. (2003), ”Determining the Trade-Environment Composition
Effect: The Role of Capital, Labour and Environmental Regulations”, Journal of Environmental Economics and Management, Vol. 46, 3, pp. 363-83.
[6] Cole, M.A., Elliott, R.J.R. and Fredriksson, P. (2006), ”Endogenous Pollution Havens:
Does FDI Influence Environmental Regulations?” Scandinavian Journal of Economics, Vol.
108, 1, pp. 157-178.
[7] Cole, M.A., Elliott, R.J.R. and Okubo, T (2010), ”Environmental Outsourcing”, RIETI
Discussion Paper Series No.10-E-055
[8] Cole, M.A., Elliott, R.J.R. and Okubo, T (2010), ”Trade, Environmental Regulations and
Industrial Mobility: An Industry-Level Study of Japan” Ecological Economics vol.69 (10),
pp.1995-2002.
[9] Cole, M.A., Elliott, R.J.R., Okubo, T. and Y. Zhou (2011),”Examining the Carbon Dioxide
Emissions of Firms:A Spatial Approach”, mimeo, University of Birmingham.
[10] Cole, M.A., Elliott, R.J.R. and Shimamoto, K. (2005), ”Industrial Characteristics, Environmental Regulations and Air Pollution: An Analysis of the UK Manufacturing Sector”,
Journal of Environmental Economics and Management, Vol. 50, 1, pp.121-43.
[11] Dean, J. M., M.E. Lovely, and H. Wang. ”Are Foreign Investors Attracted to Weak Environmental Regulations?: Evaluating the Evidence from China.” Journal of Development
Economics 90. (2009): 1-13.
[12] Ederington, J., A. Levinson and J. Minier, (2005). ”Footloose and Pollution-Free,” The
Review of Economics and Statistics, MIT Press, vol. 87(1), pages 92-99.
[13] Ederington, J., A. Levinson and J. Minier, (2004). ”Trade Liberalization and Pollution
Havens,” The B.E. Journal of Economic Analysis & Policy, Berkeley Electronic Press, vol.
0(2).
19
[14] Eskeland, G.S. and Harrison, A.E. (2003), ”Moving to Greener Pastures? Multinationals
and the Pollution Haven Hypothesis”, Journal of Development Economics, Vol. 70, pp.
1-23.
[15] Frankel, J. A. and Rose, A. K. (2005), ”Is Trade Good or Bad for the Environment? Sorting
Out the Causality”, The Review of Economics and Statistics, Vol. 87, 1, 85-91.
[16] Greenstone, M., List, J.A. and Syverson, C. (2011), ”The Effects of Environmental Regulation on the Competitiveness of US Manufacturing”, Center for Economic Studies Working
Paper CES 11-03.
[17] Helpman, E., M. J. Melitz, and S. R. Yeaple, (2004). ”Export versus FDI with Heterogeneous Firms”,American Economic Review, 94(1): 300–316.
[18] Holladay, S. ”Are Exporters Mother Nature’s Best Friends?”, mimeo, NYU, 2009.
[19] Levinson, A. (2009). ”Technology, International Trade, and Pollution from US Manufacturing,” American Economic Review, vol. 99(5), pages 2177-92, December.
[20] Levinson, A and M. S. Taylor, (2008). ”Unmasking The Pollution Haven Effect,” International Economic Review, vol. 49(1), pages 223-254.
[21] Levinson, A. (2010), ”Offshoring Pollution: Is the US Increasingly Importing Pollution
Intensive Production?” Review of Environmental Economics and Policy, Vol. 4(1), Winter
2010, pp. 63-83.
[22] Lovely, M. and Popp, D. (2011), ”Trade, Technology and the Environment: Does Access
to Technology Promote Environmental Regulation?” Journal of Environmental Economics
and Management, 61, pp. 16-35.
[23] M. Mani and D. Wheeler (1998), ”In Search of Pollution Havens? Dirty Industry in the
World Economy, 1960-1995”, Journal of Environment and Development, Fall.
[24] Melitz, M.J. (2003), ”The Impact of Trade on Intra-Industry Reallocations and Aggregate
Industry Productivity”, Econometrica, Vol. 71, 6, pp 1695-1725.
[25] Moulton, B (1990), ”An illustration of a pitfall in estimating the effects of aggregate variables on micro units”, Review of Economics and Statistics, Vol. 72, 2, pp. 334–338.
20
A
Appendix
A.1
Optimal abatement
To maximise profit, firms choose the optimal level of abatement investments. We first derive
the optimal level of abatement for firms only supplying the domestic market:
1−σ D −αρ(1−σ)−1
∂πD
= αρ(σ − 1) a1−α tα
fA
B − 1 = 0.
D
∂fA
(27)
The optimal level of abatement investments is given by
fAD
with ΩD ≡
n
=
(tα )σ−1
1
B
1
1 1−α α σ−1
a
t
B
αρ(σ − 1)
1
αρ(σ−1)
o
1
αρ(σ−1)−1
1
αρ(σ−1)−1
(1−α)(σ−1)
= ΩD a αρ(σ−1)−1
(28)
.
(1−α)(σ−1)
∂fAD
(1 − α)(σ − 1)
−1
=
ΩD a αρ(σ−1)−1
<0
∂a
αρ(σ − 1) − 1
(29)
i.e. abatement investments are increasing in firms’ productivity level, provided that
αρ(σ − 1) − 1 < 0.
(30)
However, the latter condition will always hold, as it is a necessary condition for profit maxmization:
∂ 2 πD
1−α α 1−σ
D −αρ(1−σ)−2
=
(αρ(σ
−
1)
−
1)αρ(σ
−
1)
a
t
f
B < 0 ∇ αρ(σ − 1) − 1 < 0.
A
2
∂ fAD
(31)
We proceed in the same way to derive optimal abatement investments for exporters:
∂πX
1−α α 1−σ
X −αρ(1−σ)−1
=
αρ(σ
−
1)
a
t
f
(B + φB ∗ ) − 1 = 0.
A
X
∂fA
(32)
The optimal level for the exporter is
fAX
where ΩX ≡
n
=
σ−1
1
1
a1−α tα
∗
(B + φB )
αρ(σ − 1)
1
(B+φB ∗ )
(tα )σ−1
1
αρ(σ−1)
o
1
αρ(σ−1)−1
1
αρ(σ−1)−1
(1−α)(σ−1)
= ΩX a αρ(σ−1)−1 ,
(33)
> ΩD . We note that the relationship between
abatement investments and firms’ productivity level is the same as for firms only serving the
domestic market.
(1−α)(σ−1)
∂fAX
(1 − α)(σ − 1)
−1
=
ΩX a αρ(σ−1)−1
< 0,
∂a
αρ(σ − 1) − 1
21
(34)
i.e. abatement investments are increasing in firms’ productivity level, provided that
αρ(σ − 1) − 1 < 0,
(35)
and that the latter condition will always hold, as it is a necessary condition for profit maxmization:
1−σ X −αρ(1−σ)−2
∂ 2 πX
= (αρ(σ − 1) − 1)αρ(σ − 1) a1−α tα
fA
(B + φB ∗ ) < 0.
2
X
∂ fA
(36)
Comparing with D-firms and X-firms, since ΩX > ΩD , aD > aX and αρ(σ − 1) − 1 < 0,
exporters will invest more in abatement.
A.2
Relative cut-offs
From the cut-off conditions, we have
−αρ 1−σ
aD 1−α fAD
B
fAD + fD
=
−αρ 1−σ
fAX + fX
φB ∗
aX 1−α fAX
(37)
while from the conditions for optimal abatement investment
−αρ(1−σ)−1
∂πD
α 1−σ
fAD
B−1=0
= αρ(σ − 1) a1−α
D t
D
∂fA
(38)
∂πX
1−α α 1−σ
X −αρ(1−σ)−1
t
f
φB ∗ − 1 = 0
=
αρ(σ
−
1)
a
A
X
∂fAX
(39)
1−σ D −αρ(1−σ)−1
a1−α
(fA )
B
D
1−σ
a1−α
(fAX )−αρ(1−σ)−1 φB ∗
X
(1−α)(1−σ) D −αρ(1−σ)
aD
(fA )
B
(1−α)(1−σ) X −αρ(1−σ)
aX
(fA )
φB ∗
=
1
(40)
⇔
=
fAD
.
fAX
We can derive one relationship:
fAD + fD
fAD
=
.
fAX + fX
fAX
Thus,
22
(41)
fAX fD
=
fAD fX
(42)
⇔
fAX
fAD
=
fX
>1
fD
(1−α)(σ−1)
ΩX aXαρ(σ−1)−1
(1−α)(σ−1)
αρ(σ−1)−1
ΩD aD
aX
aD
d(aX /aD )
dfX
< 0 and
d(aX /aD )
dφ
=
αρ(σ−1)−1
=
fX
fD
(43)
⇔
(1−α)(σ−1)
(1−α)(σ−1)
aX
aD
d(aX /aD ) dΩX
dΩX
dφ
The cut-off level ratio is given by
=
αρ(σ−1)−1
ΩD fX
.
ΩX fD
=
ΩD fX
ΩX fD .
This is the standard property.
> 0. Trade liberalisation and TBT liberalisation
lead to more selection mechanisms.
Related to environmental policy, tax has no impact on the cut-off ratio. The cutoff ratio is
1
(1−α)(σ−1) αρ(σ−1)−1
αρ(σ−1)−1
a
B
not a function of tax, i.e. aX
= B+B
1 + ffX
.
∗φ
D
D
A.3
Sector classification
dtbpF3.1842in3.5224in0ptFigure
23