On the Economics of Eco-Labeling
by
Charles F. Mason
Department of Economics & Finance
University of Wyoming
Laramie WY 82071-3985
February 2002
ABSTRACT
Firms would like to capitalize on consumers’ willingness to pay a premium for goods
that use environmentally friendly (green) production techniques, but they face a
problem of asymmetric information: consumers cannot determine if a firm is
environmentally friendly. Since the green technique is generally more costly than
environmentally unfriendly (brown) techniques, firms would be disinclined to choose a
green technique, yielding larger pollution flows. One possible remedy is “ecolabeling,” where a third party certifies firms’ products. I model certification as a noisy
test, with green firms more likely to pass than brown firms. I consider two models, one
where firms’ production techniques are exogenous (a short-run setting), and one where
they can choose their technique (a long-run setting). I show that eco-labeling can either
raise or lower the volume of green units, and raise or lower net social surplus; increases
in social surplus are more likely in the long-run.
Keywords:
JEL Areas:
Environmental Economics, Eco-Labeling, Asymmetric Information, Testing
Q2, D82, L15
1.
Introduction
There is abundant evidence that consumers are willing to pay a premium to “protect the
environment” (Cairncross, 1992; Cason and Gangadharan, 2001; Kirchhoff, 2000; Levin, 1990;
Wasik, 1996). Firms that use environmentally friendly production techniques would like to
capitalize on this demand, but they face a problem of asymmetric information. Consumers
cannot typically tell the type of production process a particular firm has used, so they can’t
determine when it is environmentally friendly. Since the environmentally friendly technique is
generally more costly, firms would be disinclined to choose such a technique, with larger
pollution flows resulting. One possible remedy for this informational asymmetry is for firms to
make use of “eco-labeling.” With eco-labeling, a third party certifies a vendor’s product as the
result of an environmentally friendly process.
In the last decade or so, eco-labels have emerged in a wide range of countries (Karl and
Orwatt, 2000; OECD, 1998; Vossenaar, 1997). Some of these certification programs have
become quite popular, as with the German “Blue Angel,” Japanese “Eco-Mark,” Swedish
“Environmental Choice,” and “Nordic Swan” programs (OECD, 1998). These eco-labels are
often applied to products where consumers would generally be individually unable to determine
the environmental friendliness of the product, for example the biodegrability of a paper product,
or of the production process itself. Many of the eco-labeling programs currently in operation
consider production-related criteria in their assessments of firms that seek certification.
It is tempting to regard this certification as absolute, as in Karl and Orwatt [2000],
Mattoo and Singh [1994], Robertson [2000] or Swallow and Sedjo [2000]. But this only makes
sense if the third party can perfectly identify compliance with the eco-label’s avowed standards
at a reasonable cost. Indeed, in many of the current eco-labeling programs provided by a third
1
party, the firm’s compliance with the environmentally friendly process is gauged by random
monitoring. But when monitoring is random, certification must be viewed as noisy. The
certifying party cannot be certain that the firm always uses environmentally friendly techniques,
nor that the monitoring scheme is able to perfectly detect any violations. Even if the certifying
process is perfectly able to evaluate a product’s compliance with the test’s standards, there is
considerable doubt that the standards are perfectly correlated with “environmental friendliness”
(Arda, 1997; Morriss, 1997). Nevertheless, it seems reasonable that environmentally friendly
firms would be more likely to obtain certification. I model this effect by assuming firms are
either environmentally friendly (green) or not (brown), and that the certification process yields a
positive report with some probability. Green firms are more likely to pass the certification test
than are brown firms; all firms must pay the same fee if they wish to pursue certification.
I consider two variants of the eco-labeling process. In the first version, each firm is
endowed with a production technique; in the second version, firms choose their technology at the
outset. In both versions, firms decide whether or not to seek certification and select an output
level based on the production process they will use and the associated costs of production. There
are two issues of interest here. First, when compared to the equilibrium without third party
certification, which I refer to as the “no-information” equilibrium, under what conditions does
eco-labeling expand the volume of green units? Second, what are the welfare implications of the
introduction of an eco-labeling option? Of related interest are the comparative static effects of
increases in test cost or test accuracy (as measured by increased pass rates for green units or
decreased pass rates for brown units).
In the first version, firms sort themselves according to their production costs. Firms with
relatively low production costs are most likely to seek an eco-label, while firms with relatively
2
large costs choose to remain unlabeled. I find that eco-labeling will often lead to an expansion of
the volume of green units on offer, depending on key parameters associated with the certification
test and the degree of flexibility accorded to firms. Putting aside any production externalities
associated with the two techniques, the socially efficient level of production for green
(respectively, brown) products equates supply with full-information price PG (respectively, PB).
In the original no-information equilibrium, it is apparent that an inefficiently large quantity of
brown products is produced, and an inefficiently small quantity of green products. Evidently,
any change that lowers the quantity of brown units while raising the quantity of green units
would reduce deadweight loss. Even if the quantity of green units is smaller in the testing
equilibrium than in the no-information equilibrium, eco-labeling can still be socially beneficial,
since the posited decrease in the volume of green units must lower the volume of brown units.
Likewise, a combination of increases in both green and brown production could generate
sufficient gains in the green segment to offset the increased deadweight loss in the brown
segment. On the other hand, it is possible that the reduced deadweight loss in one segment could
be more than offset in the other segment. It seems clear that the ultimate welfare implications
depend upon the various attributes of the test, along with the supply elasticities for the two types
of sellers. Moreover, since the test is costly, any putative welfare gains from moving the
volumes of green and brown units towards their first-best (full information) levels must be
compared against aggregated testing costs. With an unfortunate combination of parameters, the
combination of certification costs and losses from additional brown production are large enough
to generate negative net benefits.1
In the second version, firms that would have green costs that are only moderately larger
than brown costs choose the green technology and pursue eco-labeling. Firms that would have
3
low costs if they chose the brown technology, but moderate to high green costs, choose the
brown technology and pursue eco-labeling.
Firms with sufficiently high costs under both
techniques elect to remain unlabeled, and choose the brown technology. Because all firms would
face higher costs under the green approach, no firms would choose the green technique in the noinformation equilibrium. Since some firms would choose the green technology and then pursue
eco-labeling, were it available, the opportunity to pursue certification raises the volume of green
products (above zero). As in the preceding version, the net effect of eco-labeling on net social
surplus is ambiguous. However, it is more likely to be positive here than in the earlier (short
run) model.
These welfare remarks are focused on the market imperfection associated with
asymmetric information. But there is an additional issue. Since pollution levels are larger with
the brown technology, it will generally cause larger production externalities. This effect is not
fully captured by a divergence between prices for green and brown units, which are more the
result of consumer preferences than any explicit recognition of externalities. Moreover, firms’
private information about their production and abatement costs makes environmental regulation
more complicated. Whether society opts for a command-and-control approach or a market-based
approach, there will still be a welfare loss associated with the informational asymmetries
(Weitzman, 1974). Appealing to outside interests to reduce the informational asymmetries, as
with third party certification, therefore provides an intriguing alternative method of
environmental regulation. Indeed, Tietenberg [1998] refers to this possibility as the “third wave”
of pollution control. The results I discuss below indicate that there are also likely to be welfare
costs associated with third party certification. Before society decides to abandon the more
4
traditional regulatory approaches in favor of this third wave, a careful comparison of the welfare
effects of each strategy is in order.
2.
The Certifying Model
Consider a competitive market for a product that can either be produced using an
environmentally friendly (green) technology, or by a relatively dirty (brown) technology. Some
consumers would be willing to pay extra for green products, so the demand curve for green
products lies above the demand curve for brown products. While these demand curves might be
expected to slope downward, for expositional simplicity I assume that they are perfectly elastic,
with prices fixed at PG and PB for green and brown products, respectively. This assumption
allows a sharper focus on the incentives to pursue eco-labeling, without materially affecting the
results.
Supply curves for green and brown products are upward sloping, reflecting increasing
marginal costs for each technique. For expositional concreteness, I assume that production costs
are quadratic in output. Each firm i is therefore described by a pair of parameters (Bi,Gi);
production costs are cki(q) = Giq2, k = B or G. Each firm’s cost parameters are private
knowledge, as is its output. The latter precludes consumers from drawing inferences about a
firm’s technology on the basis of its output; this greatly simplifies the discussion that follows.
The distribution over parameter combinations is assumed to be common knowledge.
Accordingly, all agents can calculate the equilibrium expected outputs of green and brown
products, and the associated rational expectations prices. In the first version of my model, I
assume exogenously fixed numbers of potential brown and green firms, NB and NG. The
probability distribution functions for cost parameters are fB(B) and fG(G), with the associated
5
cumulative distribution functions FB(B) and FG(G), respectively.
These probability
distributions are defined on the intervals [  k ,  k ], k = B or G, where  B   G and  G   B .
To capture the notion that it is generally more costly to use the green technology, all else equal, I
assume that FB first-order stochastically dominates FG: for any , FB()  FG(), with strict
inequality arising when FB > 0 and FG < 1. In the second version of the model, there are N (= NB
+ NG) potential firms, all of whom can select either technology.
In this variant, the cost
parameters are assumed to follow the joint probability distribution function g(B,G).
Before describing the mechanics of the testing equilibrium, I first discuss the outcome in
the no-information equilibrium. In the absence of third-party information about production
techniques, consumers cannot distinguish a given product’s type. Accordingly, market price is a
weighted average of the price consumers would pay for a green product and the price they would
pay for a brown product, if they were perfectly informed regarding product type. Let Q G0 and
QB0 represent the quantities of green and brown products in the no-information equilibrium,
respectively.
These quantities are identified from the supply curves for the two types of
producer, based on the price P0.
The supply curves, in turn, depend on the underlying
assumptions regarding firms’ flexibility. Market price is then
P0 = 0PG + (1 - 0)PB, where
(1)
0 = QG0/(QG0 + QB0)
(2)
equals the fraction of green markets on the market in the no-information equilibrium.
Now suppose that a third party offers to provide information about a firm’s product, at a
specified cost.
To this end, the third party employs a certification test.
For expositional
concreteness, I suppose that the test involves monitoring of some attribute of the production
process, such as emissions, that is correlated with the production technology.
6
Since it is
prohibitively costly to monitor continuously, the third party monitoring is conducted in a fashion
analogous to random monitoring of emissions by a government agency.
With random
monitoring, it is conceivable that the third party could mistakenly certify some brown firms as
environmentally friendly. On the other hand, if the test is only imperfectly correlated with the
“green-ness” of the technology, there is the possibility of a false negative as well – that some
green firms will be mistakenly failed. Even so, it stands to reason that the probability that a
green firm would pass the test, G, is larger than the probability that a brown firm would pass the
test, B: 1 > G > B > 0. I assume that seeking certification costs A > 0 for both types of firms.2
Three possible classifications can result from the certifying process. A firm can be
certified, and thereby receive the price Pc; it can seek certification but fail, and then receive the
price Pf; or it can elect not to pursue certification, and thereby receive the price P un. All prices
are formed endogenously, via rational expectations. Accordingly, the values of the three prices
depend on consumers’ predictions of the conditional probability that a randomly selected unit is
green, given that it is labeled as c, f or un. Under plausible conditions, consumer expectations
would be such that failed units and untested units were lumped together as “unlableled.” 3 In
such a scenario, only two prices prevail: Pc and Pun.
In the discussion that follows, I denote the total supply of green (respectively, brown)
units by QG (respectively, QB). Likewise, the quantity of green (brown) units that are tested is
QGt (QBt). Prior to observing any labels, the ex ante probability that a randomly selected unit is
green equals
 = QG/(QG + QB).
(3)
This probability is associated with the ex ante expected price,
P0 = PG + (1 - )PB.
7
(4)
I denote the probability that a randomly selected unit is green, conditional on it being
eco-labeled, by . The probability that a randomly selected unit is green, conditional on it being
unlabeled, is . Using Bayes’ law, these posterior probabilities may be calculated as
 = pr(G|c) = pr(c|G)/pr(c),
 = pr(G|un) = pr(un|G)/pr(un),
where pr(c|G) is the probability that a unit will be certified, conditional on its seller being green,
pr(c) is the marginal probability of observing a certified unit, pr(un|G) is the probability that a
unit will be unlabeled, conditional on its seller being green, and pr(un) is the marginal
probability of observing an unlabeled unit. It is easy to see that pr(c|G) = GQGt/QG and pr(c) =
(GQGt+BQBt)/(QG+QB); it follows that
 = GQGt/(GQGt + BQBt);
 = (QG - GQGt)/(QG - GQGt + QB - BQBt).
(5)
(6)
Moreover, there are only two possibilities: a unit is either certified or it is unlabeled; therefore,
pr(un) = 1 - pr(c) and pr(un|G) = 1 - pr(c|G). Combined with eqs. (5) and (6), these remarks
imply
p(c) + (1 – p(c)) = .
(7)
As G > B by assumption, it follows that  > , and hence  >  > .
Calculation of equilibrium requires the determination of rational expectations prices.
These prices are based on the conditional probabilities  and  according to:
Pc = PG + (1 - )PB;
(8)
Pun = PG + (1 -)PB.
(9)
The expected price a type k seller receives from pursuing an eco-label is the weighted average of
these two prices, where the weights are the probability of passing the test and its complement:
8
k = kPc + (1 - k)Pun, k = B or G.
(10)
As Pc > Pun in the generic equilibrium considered below, and G > B by assumption, it is easy to
see that G > B. Since k is the expected price, conditional on the unit being type k, the ex ante
price must satisfy P0 = G + (1 - )B. It then follows that G > P0 > B.
Since  >  > , the information produced by the test leads to a higher price for ecolabeled units and a lower price for unlabeled units than the ex ante price. The information from
the test is therefore useful, in the sense that it moves expected prices towards the full-information
prices. That said, a determination of the impact of eco-labeling upon the various quantities
requires a comparison of the equilibrium with no-information against the testing equilibrium.
This comparison depends on the degree of flexibility that firms posses.
3.
Model 1: exogenous technology choice, flexible output
In this version of the model, I assume that number of potential sellers of type k products
is exogenously set at Nk. Each firm chooses its output level to maximize expected profits. If the
firm is planning to have its products tested, the privately optimal output level is
qki* = Pk/2ki,
(11)
where Pk is the price the firm anticipates receiving. If the firm enters the untested segment of the
market, Pk = Pun; if it enters the tested segment then Pk = k. Accordingly, the payoff from
testing equals
ki*  ki(qki*) = k 2/4ki – A,
(12)
while the payoff from not testing equals
ki*  ki(qki**) = Pun2/4ki.
9
(13)
Notice that the maximized profit from opting for the unlabeled segment of the market is
strictly positive. Accordingly, both types of sellers strictly prefer participating in the unlabeled
segment to exiting. It follows that all sellers will produce something. The issue is therefore not
one of inducing additional sellers with green technologies to commence production; rather, it is
one of inducing some green sellers to switch from the unlabeled segment to the tested segment of
the market. Since G > Pun, a larger volume of green products would then obtain.
In equilibrium, prices are based on rational expectations, with underlying probabilities
derived using Bayes’ law. Calculation of equilibrium requires the determination of cutoff values
of the cost parameters for green and brown sellers. At these cutoff values, the marginal seller is
indifferent between participation in the two segments. To identify these cutoff values, I compare
expected profits from seeking certification with profits in the unlabeled segment. The gain from
seeking certification, given the anticipation of optimal actions thereafter, is:
Wki = ki* - ki* = (k2 – Pun2)/4ki – A.
(14)
Accordingly, the cutoff value for a type k seller is4
~ = (k2 – Pun2)/4A.

k
(15)
Because G > B, the cutoff value is larger for green units.
~ pursue certification, while those with k > 
~ elect to enter
Type k sellers with k  
k
k
the unlabeled segment. Thus, the number of type k sellers that pursue certification is
~ ),
Nkt = NkFk( 
k
(16)
~ )] sellers preferring to enter the unlabeled segment of the market. Aggregate
with Nk[1 - Fk( 
k
testing costs are A[NGt + NBt].
To facilitate the description of the expected volumes of type k output in the certified and
unlabeled segments, I define the function
10
ℱk() =


( f k ( k ) / 2 k )d k .
(17)
k
The volume of type k units that pursue the eco-label is kQkt, where
~ )Nk.
Qkt = kℱk( 
k
(18)
Of these, kQkt will be certified while (1 - k)Qkt will wind up in the unlabeled segment of the
market, on average. All told, the total number of type k units is
~ ) + Punℱk(  )]Nk.
Qk = [(k – Pun) ℱk( 
k
k
(19)
Rational expectations prices are based on these quantities, using eqs. (5), (6), (8) and (9).
Assessing the impact of eco-labeling upon market quantities requires the calculation of
outputs that would obtain in the no-information equilibrium. In such a setting, each firm i
receives the price P0, and therefore produce the output qik0 = P0/2ik. The resultant aggregate
output of type k sellers in the no-information market is
Qk0 = P0ℱk(  k )Nk.
(20)
Comparing eqs. (19) and (20), the difference between aggregate output for type k sellers in the
testing equilibrium and the no-information equilibrium, Qk, is
~ ) + (Pun - P0)[ℱk(  ) - ℱk( 
~ )] + (P0 - P0) ℱk(  )]Nk.
Qk = [(k – P0)ℱk( 
k
k
k
k
(21)
From the discussion above, the first term is positive for green units and negative for brown units
while the second term is negative for both types. The sign of the third term depends on the
impact of eco-labeling on the percentage of available units that are green. In light of the first
observation, it would seem probable that the combined relative impact of the first two terms was
substantially more negative for sellers of brown units.
If so, one would expect that the
equilibrium in the testing market would have a larger fraction of green units than the no-
11
information equilibrium. As such a change would likely lead to lower pollution levels, it could
be socially beneficial irrespective of consumers’ tastes for environmentally friendlier products.
The next step is to determine the impact that this configuration has upon social surplus.
Because I have assumed that demand is perfectly elastic, social surplus corresponds to producer
surplus. To calculate producer surplus, I first determine aggregate market production costs.
With the assumed quadratic form, the production costs borne by type k firm i in equilibrium are
cki(qki*) = ki(Pk/2ki)2, k = B or G,
(22)
where Pk is the price the firm anticipates receiving. Accordingly, aggregate market production
~ )Nk/2.
costs for those type k sellers that seek the eco-label are readily determined as k2ℱk( 
k
Likewise, market production costs for those type k sellers that do not seek the eco-label may be
~ )]Nk/2. Using these values, producer surplus for each type
determined as Pun2[ℱk(  k ) - ℱk( 
k
of seller is readily determined.
Figure 1 illustrates the market with eco-labeling when production technologies are
exogenous. Full-information prices are marked by the solid horizontal lines at PG (for green) and
PB (for brown). The no-information price is denoted by the solid horizontal line at P0; to avoid
clutter I label this line on the right-hand side of the graph. The equilibrium price an unlabeled
unit receives is shown by the dashed horizontal line Pun, while expected prices with testing are
denoted by the dashed horizontal lines at G and B. The plots Skt may be used to determine the
quantity of type k units whose sellers pursue certification; similarly, the lines S kun show the
volume of type k units whose sellers do not seek the eco-label.5 From the above discussion, the
~ )Nk, while the slopes of the Skun lines are given by [ℱk(  ) slopes of the Skt lines are ℱk( 
k
k
~ )]Nk. If each cohort of sellers faces the same price, as in the no-information market, then
ℱk ( 
k
12
the total amount of type k products supplied is given by the horizontal summation of the Skt and
Skun lines, for both k = G and B. To evaluate the net impact of testing, I separately consider each
of the four cohorts:  G  G > ~G ; ~G  G  G;  B  B > ~B ; and ~B  B  B. With the
availability of the eco-label, production by sellers of untested green units – those in the first
cohort – falls to QGun, read off the SGun curve at Pun. This decrease in volume of green units
induces a reduction in social surplus equal to the trapezoidal area between PG and SGun, from the
points of intersection of SGun with Pun and P0. I label this as area I. On the other hand, there is an
increase of output from green sellers in the second cohort, those that now pursue the eco-label.
Output from this cohort increases to QGt, read off the SGt curve at G. This increase causes social
surplus to rise, by an amount equal to the trapezoidal area between PG and SGt, from the points of
intersection of SGt with G and P0. I label this as area II. There are also two effects for sellers of
brown units. Brown sellers that choose not to pursue certification – those in the third cohort reduce their output to QBun, read off the SBun curve at Pun. As these units are priced above their
value to consumers, PB, this output reduction yields a welfare gain. This gain can be measured
by the trapezoidal area III below the SBun curve and above PB, from the points of intersection of
SBun with P0 and Pun. Finally, brown sellers that pursue the eco-label – sellers in the fourth
cohort – increase their production to QBt, read off the SBt curve at B.
This increase in
production exacerbates the over-production of brown units found in the original no-information
equilibrium, lowering social surplus by an amount equal to the trapezoidal area IV below S Bt and
above PB. The combined effect on social surplus is the sum of areas II and III, less areas I and
IV. In this figure, it is apparent that the gains (areas II and III) dominate the losses (areas I and
IV), so that the combined effect is to raise social surplus.
13
Even if the combined effects from introducing eco-labeling thus described led to an
increase in social surplus, there is no guarantee that the combined effect would offset the
aggregate testing cost. For example, with the parameters that this diagram is based on, the total
test costs outweigh the combined increase in social surplus. In such an event, society would
ultimately be worse off with eco-labeling.6
4.
Model 2: endogenous technology choice
While firms’ technologies can sensibly be regarded as exogenous in a short run setting,
one would ultimately want to allow firms to select the technology that seemed most profitable.
Since testing is relatively disadvantageous to brown units, it seems plausible that some erstwhile
brown firms would switch to the green technology. With this switch, the over-production of
brown units and the under-production of green units would simultaneously be mitigated. Might
it be the case that these combined effects are sufficiently large to outweigh the aggregate test
costs?
To address these issues, I consider an expanded version of the model where firms select
their technology at the first stage.
While it was sufficient to model the two marginal
distributions over the cost parameters in the preceding section, here it will be necessary to
consider the joint distribution over cost parameters. Accordingly, I write the joint probability
distribution over cost parameters as g(B, G). Because the green technology is not cheaper than
the brown technology for the typical firm, I also assume that Bi  Gi for each firm i. There are
N total firms, each of whom can select either technology. All other aspects of the model are as
in the preceding section.
14
Since the green technology is generally costlier than the brown technology, no firm
would choose to be green in the no-information equilibrium. This result is an example of
Akerlof’s [1970] classic market for lemons. In light of this grim outcome, the presence of any
green firms in a testing equilibrium would represent an increase, relative to the no-information
equilibrium. Such a result would require that green firms could realize some advantage from
testing, relative to brown firms, in a fashion analogous to Spence’s [1973,1974] signaling model.
In selecting a technology, the firm needs to predict its optimal actions that would ensue.
The firm’s optimal output depends on the combination of technology and testing decision, as
described in eq. (11) above. Based on these optimal actions, the firm’s expected profits are ki*
if it tests and ki* if it doesn’t, as given by eqs. (12) and (13). Thus, there are four combinations
to be considered. I label these combinations (G,t), (G,un), (B,t) and (B,un). In these pairs, the
“G” (B) stands for the choice of the green (brown) technology, while the “t” (un) represents the
decision to seek certification (enter the unlabeled segment).
I note first that any firm that chooses the green technology would pursue eco-labeling.7
Therefore, it will be sufficient to restrict attention to the three combinations (G,t), (B,t) and
(B,un). Consider first the decision to test, given that the unit is brown. As in the preceding
section, the firm earns larger profits from opting for the unlabeled segment of the market when
~ given in eq. (15). The comparison between (G,t) and (B,t) is
Bi exceeds the cutoff value 
B
also straightforward. In light of eq. (12), firm i would prefer (G,t) to (B,t) when its cost
parameters satisfy Gi  ˆ G ( B ) , where
ˆ G ( B ) = (G/B)2Bi.
15
(23)
The final pair-wise comparison of interest is between choosing the green technology and testing
or choosing the brown technology and not testing. Firm i will prefer (G,t) to (B,un) if Gi* is no
smaller than Bi*. Referring to eqs. (12) and (13), this occurs when Gi  ˆˆ G ( B ) , where
ˆˆ G ( B ) = G2Bi/(4ABi + Pun2).
(24)
~ and Gi
Combining these observations, the firm selects the green technology if Bi  
B
~ and Gi  ˆˆ ( ) . In either event, the firm has its product
 ˆ G ( B ) or when Bi > 
G
B
B
~ and Gi > ˆ ( ) or Bi > 
~ and Gi > ˆˆ ( ) , the firm chooses the
tested. If Bi  
G
B
G
B
B
B
brown technology.8 In the first case it has its product tested, while in the second it opts for the
unlabeled segment of the market. Armed with these facts, I calculate the quantities of green and
brown units as
QG = N{ 
~

B
B
ˆ G (  B )

( G / 2 G )g( B ,  G )d G d B 
B
(25)
B
ˆˆ G (  B )
~ 
B
( G / 2 G )g( B ,  G )d G d B } ,
B
QBt = N 
~

B
B
G
ˆ
G
( B )
B
QB = QBt + N  ~
B
( B / 2 B )g( B , G )dG d B ,
G
ˆˆ
G
( B )
( Pun / 2 B )g( B , G )dG d B .
(26)
(27)
Rational expectations prices may be computed by combining these expressions with eqs. (5), (6),
(8) and (9).
Calculation of welfare impacts from the introduction of eco-labeling in this context is
similar to the approach described in the preceding sub-section. Because the no-information
16
equilibrium contains only brown units, eco-labeling expands the production of green units.
While the influx of green units is unambiguously beneficial, there are two offsetting effects.
First, this influx occurs because some sellers of brown units convert to the green technology;
accordingly, the brown supply curve shifts in, which lowers net surplus in the market for brown
products. In addition, some of the green units will fail, and so the unlabeled price must exceed
PB, the equilibrium price that brown units receive in the no-information equilibrium. Combined
with the larger certified price, this increase in Pun increases the volume of brown units above the
(new) efficient level, which exacerbates the reduction in net surplus in the market for brown
products. Since consumers place a larger value on green units, one might hope that the benefits
from expanded green production outweigh the losses from brown production, so that the
combined effect yields an increase in social surplus. Even if this is so, it is not clear that the
combined effect will outweigh the aggregate testing cost.
Figure 2 illustrates the testing equilibrium with endogenous technology choice. The
green supply curve is labeled as SG; the supply curve for brown units that are placed in the
unlabeled segment is SBun, while the supply curve for brown units that seek the eco-label is SBt.9
Equilibrium values are QG, QBu, and QBt, respectively. For comparison, I also depict the supply
curve for green units in the first best (full information) equilibrium, which for the parameters
used to generate this example contains green units only. The efficient production levels are Q G*
green units and no brown units. In the testing equilibrium, welfare gains (gross of test costs)
from delivery of green units are measured by the area between S G and PG, from the y-axis out to
QG. While smaller than the full information welfare gains from green units, the gains in the
testing equilibrium are substantial, particularly in light of the observation that no green units
would be available without testing in the context of this model. Net welfare gains from the sale
17
of brown units are linked to the two sub-markets. For those brown units placed in the unlabeled
segment, net gains are the area below PB, less the area below SBun, out to QB. In the context of
Figure 2, these net gains are quite small, owing to the continued overproduction of brown units.
For those brown units that seek certification, net gains are the area below PB, less the area below
SBt, out to QBt. This net area is also small (though positive), again because an inefficiently large
volume of brown units is made available. Even with the combined overproduction of brown
units, combined social surplus in this example, net of testing costs, is nearly 90% of social
surplus in the full information equilibrium, which is considerably larger than social surplus in the
no-information equilibrium.10 For these parameters, at least, eco-labeling leads to an increase in
social surplus that is sufficient to cover the aggregate testing costs in the long-run.
Finally, I observe that increases in test cost are likely to be socially attractive. The point
~ . All else
is based on three observations. First, an increase in the test cost surely lowers 
B
~ implies a decrease in QBt. From eqs. (5) and (6), one then infers that
equal, the reduction in 
B
 rises and  falls, leading to an increase in Pc and a decrease in Pun. These collateral effects
would tend to induce an increase in the volume of brown units that are tested, but not so large an
impact as to completely offset the original changes in the two prices. So the second observation
is that the ratio G/B must increase. Accordingly, the function ˆ G ( B ) shifts up. But since
~ ) = ˆˆ (
~ ) , the third observation follows: that the ˆˆ ( ) curve must shift up as
ˆ G (
B
G
B
G
B
well. With the ˆ G ( B ) curve shifting up, some brown firms that were planning to test switch to
green; with the ˆˆ G ( B ) curve shifting up, other brown firms shift out of the unlabeled segment
of the market into the tested section, and change to the green technology. All those firms that
were originally green increase their output, since Pc goes up. Ultimately, an increase in green
18
production results. At the same time, those firms that remain in the unlabeled segment of the
market produce less as Pun fell. The result is a mitigation of the excessive production by brown
units in the original testing equilibrium. Between these welfare impacts, net social surplus rises
with the increase in test cost.11
Very similar effects emerge from reductions in B. Similarly, increases in G would raise
~ . Accordingly, increases in test accuracy
the ratio G/B, though there is no direct impact on 
B
would be expected to screen additional brown units from the testing segment into the unlabeled
segment, thereby increasing Pc and lowering Pun. Combining the comparative static remarks for
increases in test cost and test accuracy, it seems plausible that more precise and more expensive
tests would be socially attractive.
5.
Discussion
Putting aside any production externalities associated with the two techniques, the socially
efficient level of production for green (respectively, brown) products equates supply with fullinformation price PG (respectively, PB). In the original no-information equilibrium, it is apparent
that an inefficiently large quantity of brown products is produced, and an inefficiently small
amount of green products. Evidently, any change that lowers the quantity of brown units below
QB0 while raising the quantity of green units above QG0 would reduce deadweight loss. Even if
the quantity of green units is smaller in the testing equilibrium than in the no-information
equilibrium, eco-labeling can still be socially beneficial, since the posited decrease in the volume
of green units must lower the volume of brown units. Likewise, a combination of increases in
both green and brown production could generate sufficient gains in the green segment to offset
the increased deadweight loss in the brown segment. On the other hand, it is possible that the
19
reduced deadweight loss in one segment could be more than offset in the other segment. In any
event, the combined impacts on the markets for green and brown products must be compared
against aggregate test cost. It seems clear that the ultimate welfare implications depend upon the
various attributes of the test, along with the supply elasticities for the two types of sellers. For
the parameters upon which figures 1 and 2 are based, it turns out that eco-labeling yields a
combined increase in social surplus, but that this effect is not large enough to offset the
aggregate testing costs in the short run. By contrast, the combined increase in the long run is
markedly larger than the aggregate testing costs. Accordingly, the strength of the case to be
made in favor of eco-labeling is likely to depend on the degree of flexibility firms possess in
choosing production technologies, which in turn depends on the time frame one adopts.
The welfare remarks I discussed above focused on the market imperfection associated
with asymmetric information. But there is an additional issue, namely that the brown technology
is more likely to generate production externalities.
This effect is not fully captured by a
divergence between prices for green and brown units, which are more the result of consumer
preferences than any explicit recognition of externalities. When firms are privately informed
about production and abatement costs, as in the context of my model, environmental regulation is
notoriously difficult.
Whether society opts for a command-and-control approach, using
standards, or a market-based approach, using effluent taxes or tradable permits, there is generally
a welfare loss associated with the informational asymmetries. Appealing to outside interests, as
with third party certification, to reduce the informational asymmetries therefore provides an
intriguing alternative. Indeed, Tietenberg [1998] refers to this as the “third wave” of pollution
control. The results I obtain above indicate that there can be welfare costs associated with third
party certification, to the extent that welfare gains from improving information are smaller than
20
the costs of providing that information. That being said, it is conceivable that the inclusion of an
eco-labeling option with a more traditional form of environmental regulation would yield an
outcome that is socially preferable to the second-best outcome typically found in models of
environmental regulation. Identifying conditions where such an improvement could be expected
to occur would have important implications for public policy towards environmental regulation.
While the presence of an eco-labeling option may be socially attractive under certain
conditions, it is not likely that society would benefit by requiring firms to pursue certification.
Such mandatory testing schemes have been suggested in a number of contexts of asymmetric
information, including the Federal Trade Commission’s recent consideration of so-called lemons
laws. Economists are generally unimpressed by such schemes, largely because of the belief that
it is better to let firms choose from self-interest. In both versions of my model, some brown
sellers eschew testing. By requiring all sellers submit to the certification test, this outlet is
eliminated; all firms must take the certification test or exit. Since profits from the unlabeled
segment are strictly positive, these sellers are likely to remain in the market even if forced to
pursue certification. But this influx of brown output into the cohort of tested units serves to
depress both certified and unlabeled prices, which will lead to a reduction in green output.
Worse, the difference between the certified and unlabeled prices would then tend to shrink,
which in turn reduces the incentive for some sellers to select the green technology. Altogether,
these effects serve to lower the percentage of green units in the market. For the set of parameters
that Figure 2 is based on, the effect is so dramatic that green products disappear altogether; the
result is a lemons market. In this case, the imposition of mandatory testing is ultimately selfdefeating, in that it eliminates any motive for testing to occur at all.
21
REFERENCES
Akerlof, G. A. [1970], “The Market for Lemons: Qualitative Uncertainty and the Market
Mechanism,” Quarterly Journal of Economics 84, 488-500.
Arda, M. [1997], “Environmentally Preferable Commodities,” in Eco-Labelling and International
Trade, S. Zarrilli, V. Jha and R. Vossenaar eds. (New York, NY: Macmillan Press).
Banks, J. S. and J. Sobel [1987], “Equilibrium Selection in Signaling Games,” Econometrica 55,
647-661.
Cairncross, F. [1992], Costing the Earth, (Cambridge, Mass.: Harvard University Press).
Cason, T. N. and L. Gangadharan [2001], “Environmental Labeling and Incomplete Consumer
Information in Laboratory Markets,” Journal of Environmental Economics and
Management, forthcoming.
Cho, I.-K. and D. M. Kreps [1987], “Signalling Games and Stable Equilibria,” Quarterly Journal
of Economics 102, 179-222.
Karl, H. and C. Orwatt [2000], “Economic Aspects of Environmental Labelling,” in
International Yearbook of Environmental and Resource Economics, T. Tietenberg ed.
(Cheltenham, UK: Edward Elgar).
Kirchhoff, S. [2000], “Green Business and Blue Angels,” Environmental and Resource
Economics 15, 403-420.
Levin, G. [1990], “Consumers Turning Green: JWT Survey,” Advertising Age 61, 74.
Mason, C. and F. Sterbenz [1994], “Imperfect Product Testing and Market Size,” International
Economic Review 35, 61-85.
Mattoo, A. and H. V. Singh [1994], “Eco-Labelling: Policy Considerations,” Kyklos 47, 53-65.
22
Morris, J. [1997], Green Goods? Consumers, Product Labels and the Environment, (London:
IEA publications).
OECD [1997], “Eco-Labelling: Actual Effects of Selected Programs,” OECD Working Paper
vol. 5, no. 44, Paris.
Robertson, P. E. [2000], “Eco-labels and Easy Riding,” mimeo, University of New South Wales.
Spence, A.M. [1973], “Job Market Signaling,” Quarterly Journal of Economics 87, 355-374.
Spence, A.M. [1974], Market Signaling: Informational Transfer in Hiring and Related Processes
(Cambridge, Mass.: Harvard University Press).
Stiglitz, J. E. [1975], “The Theory of ‘Screening,’ Education, and the Distribution of Income,”
American Economic Review 65, 283-300.
Swallow, S. and R. Sedjo [2000], “Eco-Labelling Consequences in General Equilibrium: A
Graphical Assessment,” Land Economics 76, 28-36.
Tietenberg, T. [1998], “Disclosure Strategies for Pollution Control,” Environmental and
Resource Economics 11, 587-602.
Vossenaar, R. [1997], “Eco-Labelling and International Trade: The Main Issues,” in EcoLabelling and International Trade, S. Zarrilli, V. Jha and R. Vossenaar eds. (New York,
NY: Macmillan Press).
Wasik, J. F. [1996] Green Marketing and Management: A Global Perspective (Cambridge,
Mass.: Blackwell Publishers, Inc.).
Weitzman, M. [1974], “Prices vs. Quantities,” Review of Economic Studies 41, 477-491.
23
24
25
ENDNOTES
1. The results in the model I am proposing differ substantially from the results in the context of a
perfect (but costly) signal. There, the signal itself provides no useful information: all relevant
information is conveyed by the identity of agents who purchase the signal (Stiglitz, 1975). The
result is a first-best combination of green and brown outputs, with welfare gains that are typically
larger than the aggregate signaling cost. In my model, since some brown firms succeed in
obtaining certification, the net increase in social surplus from the signal is smaller, while
aggregated signaling costs can be larger. On balance, the net effect on social surplus is less
likely to be positive.
2. This precludes, for example, schemes where eco-labeled firms that are found to be brown are
required to pay a penalty to the certifying company, as in the Canadian Environmental Choice
Program (Wasik, 1996). For an analysis of a model with such fines, see Kirchhoff [2000].
3.
As a practical matter, information on denied applications are generally unavailable
(Vossenaar, 1997). Even if such information were available, if consumers believed that all failed
units were brown, then any seller with a failed unit would (weakly) prefer the untested price. As
a result, no units would be offered for sale at the failed price, so that Bayes’ rule could not be
applied (Mason and Sterbenz, 1994). This awkwardness, which often arises in signaling games,
could be resolved by applying a refinement such as Cho and Kreps [1987] Intuitive Criterion, or
one of Banks and Sobels’ [1987] Divinity Criteria.
In the present case, however, these
refinements have no bite, so that the equilibrium I propose cannot be excluded. Whether
consumers would be inclined to form such pessimistic expectations is of course an empirical
26
matter. It is interesting to consider Indonesia’s Public Disclosure program in this context. Under
the Indonesian scheme, firms are assigned one of five color-coded factors, ranging from black
(factories that have not attempted to control pollution and so cause serious damage) to gold
(plants that are among the cleanest anywhere in the world). As reported in Table 1 of Tietenberg
[1998], the vast majority of plants are in the 2nd or 3rd dirtiest category. One could then regard
the 3rd dirtiest category as those that have passed the test and the 2nd dirtiest category as those
that are unlabeled.
4. If (G2 – Pun2)/4A   G , then ~G   G : all green units are tested.
5. The lines can also be used to determine market production costs. From the above discussion,
~ )Nk, while the slopes of the Skun lines are given by [ℱk(  )
the slopes of the Skt lines are ℱk( 
k
k
~ )]Nk. Accordingly, production costs are given by triangular areas below the lines, out to
- ℱk ( 
k
the equilibrium outputs. These lines are closely related to the supply curves. Strictly speaking,
supply curves would depend on the cutoff values at which the seller is just indifferent between
testing and not testing. Since this cutoff value changes as price changes, the supply curve is
somewhat more complicated. The constructs in the figure are better interpreted as showing the
relation between price and quantity supplied for the cohorts of sellers induced by the equilibrium
cutoff value. Accordingly, they can be used to determine production levels, and associated costs,
in equilibrium.
6. The diagram is based on the parameter values G = B = 1,  G = 2.5,  B = 2, NB = 2NG, PG =
3, PB = 1.5, G = .6, B = .45, A = .1, and k uniformly distributed. From these parameter values,
27
one may calculate the equilibrium cutoff values as ~G = 2.0665 and ~B = 1.5283. The resultant
equilibrium outputs are QG = 1.2664NG and QB = 2.6565NG, with rational expectations prices Pc
= 2.1779 and Pun = 1.8206. The no-information price is 1.9588, with outputs QG0 = 1.1966NG
and QB0 = 2.7155NG. With the increase in output, deadweight loss from under-provision of
green units is reduced by 0.0712NG and deadweight loss from over-provision of brown units is
reduced by 0.0247NG. The combined test cost is 0.1768NG, so that net social surplus falls by
0.0809NG.
7. As Bi  Gi, it follows that Bi*  Gi* for all i. Therefore, any firm that is inclined to enter
the unlabeled segment of the market would prefer the brown technology.
~ ) : at this point, the firm is indifferent between
8. Note that, by construction, ˆ G (~B ) = ˆˆ G (
B
all three options.
9. As I noted in footnote 5, the relation between price and quantity associated with these “supply
curves” is based on the cohorts of sellers induced by the equilibrium cutoff values.
10. The diagram is based on the same parameter values as Figure 1 to the extent possible: P G =
3, PB = 1.5, G = .6, B = .45, and A = .1. For comparability to Figure 1, I interpret N = 3N G.
The joint distribution over cost parameters is likewise similar, in that it is uniformly distributed
over the trapezoidal area delimited by G = B = 1,  G = 2.5,  B = 2, with G B. From this
information, one may calculate the equilibrium cutoff value as ~B = 1.7359. The resultant
equilibrium outputs are QG = 1.38131N, QBt = 1.8290N and QBun = .2165N, with rational
expectations prices Pc = 2.2526 and Pun = 1.8561. The social surplus associated with these
28
outputs is 2.6977N in the market for green products, .1238N in the untested segment of the
market for brown products, and .6627N in the tested segment of the market for brown products.
Aggregate test costs are .3210N, so that net social surplus is 3.1632N. The first best output is
QG* = 2.3850N, with social surplus equal to 3.5775N. The net social surplus from eco-labeling
in this example is therefore approximately 88.4% of the maximal level. By comparison, social
surplus in the no-information equilibrium is equal to 1.7545N.
11. This effect is in stark contrast to the impact of increased test costs in a model where
certification is perfect. In such a model, the only firms that pursue certification are green.
Increases in test costs do not raise the fraction of green units that pursue certification; indeed, it
is conceivable that such cost increases could induce some green sellers to exit the market. Even
without such exit, the higher test costs lower net social surplus, and so are unambiguously
welfare-reducing.
29