The Impact of Trade on Intra-Industry Reallocations

and Aggregate Industry Productivity

Marc Melitz, Econometrica (2003).

Introduction
Empirical evidence1 shows some stylized fact on …rms heterogeneity, productivity and trade.
We have in particular evidence of the following regularities:

Turnover status: …rms entering and exiting the market are smaller than other …rms, con-
trary to the idea of a representative …rm; exiting …rms are less productive on average.

Export status: exporters have higher k=l ratio, are larger, are more productive.

These evidences cannot be explained through classical general equilibrium trade models
(e.g. Krugman, 1979 or 1980), because they rely on a representative …rm framework.

The paper from Melitz published on Econometrica in the 2003 opens the doors to a new
strand of literature in international economics, by formally introducing …rm heterogene-
ity in trade models: …rms are characterized by di¤erent productivity levels. The Melitz’s
model builds on the setting of Krugman (1980) and builds a dynamic industry model
with heterogeneous …rms that explains why international trade induces a reallocation of
resources among …rms in an industry. Once economic integration progresses, only the
most productive …rms enter the export market, the middle productive …rms serve only the
local market, while the least productive …rms exit. The latter selection e¤ect pushes up
aggregate productivity as induced by a process of economic integration, thus stimulating
long-run growth.

1 Set up of the model

The set up of Melitz (2003) is the same of Krugman (1980), so we will not repeat all the passages
to derive the fundamental equations here. The closed economy of Melitz (2003) represents a
continuum version of Krugman (1980). The notation used in these notes is the same. When a
new notation is introduced, it will be explained.

1.1 Endowments

As in Krugman (1980) the model starts by assuming that there is full employment and that the
number of workers equals the number of consumers.
L: number of workers = number of consumers

1
See James Tybout (2002) and Bernard, Eaton, Jensen and Kortum (2000) for a review of the preliminary
empirical evidence on …rm heterogeneity and trade.

1
1.2 Preferences

The functional form is again a function with Constant Elasticity of Substitution (CES).
2 3
Z 1
1
U =4 q d! 5 =Q (1) utility function
!2

Where:
! = single variety
= set of varieties available
Q = a ‘quantity’index, i.e. an aggregator of all the quantities q(!) consumed of each variety
!
The utility of the individuals increases with the number of varieties they consume: love for
variety principle.
The term > 1 is the elasticity of substitution among di¤erent varieties.

1.3 Technology of production and entry/exit decision

The production side is characterized by monopolistic competition. The "free entry" condition
applies and leads to have as many …rms as varieties ) n of varieties = n of …rms. Each variety
is produced by a single …rm. It is assumed that there is only one production factor: labour.
All goods will be produced with a cost function that is formed by a …xed cost and a constant
marginal cost (recall Krugman, 1980: T C = w li = w + w xi )
Di¤erently from Krugman (1980), this model modi…es the production technology in order to
allow for heterogeneity across …rms. In particular, the …xed cost w is assumed to be identical
across …rms, while the marginal cost w di¤ers across …rms, since it depends (negatively) on
each …rm’s productivity (from this moment on denoted by '). The higher is ', the lower is
the amount of labour necessary to produce one unit of the di¤erentiated good. In the notation
of Melitz (2003), we normalize w = 1 and then write a total cost function depending on the
(heterogeneous) productivity level of each …rm, i.e. …rm-speci…c:

1
T C(') = f + q(!) (2) total cost function
'

1.3.1 Firm entry and exit

Contrary to Krugman (1980), the setup by Melitz also models both the decision of entering
into the market and the decision of exiting from it. Two sequential decisions are taken by
…rms. First, they decide whether entering or not, second they decide whether to produce
or exit.

More speci…cally, at the beginning there is a large pool of potential entrants in the industry
that are ex ante identical. They have two options: to enter the market by paying a sunk

2
 cost equal to fE > 0; or not to enter, in that case their outside option (pro…t) is 0. This
is the “free entry” condition.

The sunk cost to entry the market can be thought as an investment in R&D which is
necessary to generate a new product to sell. After having paid this entry "sunk" cost,
each …rm learns about its productivity ', drawing it from a common distribution g(').

From this moment on, …rms are not identical anymore, since they are now characterized
by di¤erent productivity levels. After having learned about its productivity level, each
…rm takes the second decision: deciding whether to produce or exit from the market2 .

The …rm decides to produce if its pro…ts > 0, in this case she will have to pay the
additional …xed cost f (Eq. 2), while facing every year a constant probability of exiting
(a negative productivity shock forcing it out from the market). On the other side, a …rm
decides to exit if, after having observed its productivity, its pro…ts are < 0, in which
case she will loose the sunk cost fE paid to play the “productivity” lottery. This is the
“free exit” condition. Firms whose productivity is such to make them indi¤erent between
producing or not are called the “cuto¤ ”…rms, and their productivity level will correspond
to (' ) = 0:

2 Closed economy
In the above setting, knowing the number of …rms operating is not enought to characterize the
economy: we also need to know the "cuto¤" productivity ' : this is exactly what we are going
to derive in this section.

2.1 The consumer problem

Z
Let’s de…ne by r(!) the expenditure on variety !; then r(!)d! = R is the total expenditure,
!2
that is the budget constraint in a continuum of goods.
From the two stage budgeting we know that the price index is, as in Krugman (1980):

n
!
X 1

P = pi1 (3) price index

i=1

Solving the maximization problem of the consumer for one individual variety we obtain the
usual demand function (see the class notes on Krugman if you still have doubts):

2
In a di¤erent setup, you can think at fE > 0 as the (sunk) cost of taking your GRE/GMAT test. Once
you have paid the fee, you can sit the exam and learn how good you are (your estimated productivity). At
this moment, you can decide whether to drop the application to your Ph.D. (outside option = 0) or to continue
competing for it.

3
 p(!)
q(!) = R
P1
Given our Quantity index Q, we can also write P Q = R , leading to the following, equivalent
characterization of the demand function:

p(!)
q(!) = Q (4) demand function
P
We can then rewrite the expenditure on an individual variety ! as follows:

1
p(!)
r(!) = p(!)q(!) = R (5) expenditure on variety !
P

2.2 The …rm problem

We have already analyzed the technology used by …rms. Here we will focus on their pro…t
maximization problem.
From the maximization of …rms’ pro…t function (follow the same steps explained in Krug-
man’s class notes) we get the usual pricing rule telling us that prices are a constant markup over
marginal costs:
1
p(') = (6) pricing rule
1 '
1
1 is the mark up, while ' is the marginal cost that negatively depends on ' (remember
that here we suppose w = 1). In this model, also the optimal pricing rule is …rm–speci…c, since
it depends on the speci…c …rm productivity.
Now we want to demonstrate that also the …rm pro…t is a (positive) function of the
productivity ': Let’s start by writing in the following form:

1
(') = p(') q(') (f + q('))
'

then, collecting q(') we get

1
(') = p(') q(') f
'

using the pricing rule (Eq. 6) we can replace p(') with its function obtaining:

1 1 1 q(')
(') = q(') f= f
1' ' 1 '

p(')
Now, multiplying the above equation by p(') , substituting in the previous denominator
1
p(') = 1 ' (Eq. 6), and de…ning r(') = p(')q('), we arrive to the following notation for
(') :
r(')
(') = f

4
where the …rm-speci…c pro…t depends positively on revenues, and negatively on elasticity and
…xed costs. Nevertheless we still have to demonstrate that pro…ts depends directly on produc-
tivity '. Let’s rewrite equation (5) substituting the terms in ! with the the equivalent in '.
Replacing p(!) with its function (equation 6):

1
1 1
r(!) = P R
1'

reshu- ing a bit we obtain:

1 1
r(') = R(P ')

hence replacing r(') into equation (6) we get our …nal version of the pro…t as a function of
productivity:
R 1 1
(') = (P ') f (7) pro…ts as a function of '

Firm pro…ts are thus a function of the following variables:

R: total expenditure, hence the size of the market (+)

: the elasticity of substitution
P : the price index, i.e. the average price of all the …rms. From the optimal pricing rule we
know that more productive …rms will charge a lower price. Thus the price index is inversely
related to the TFP of all other …rms (your …rm has an atomistic size in monopolistic compe-
tition). Hence the pro…ts of a given …rms depend negatively on the productivity of the other
…rms, thus positively on P : if all other …rms are very productive, they charge a lower P which
lowers your pro…ts (+)
f : the …xed cost, obviously (-)
': its own productivity level (+)
This is the …rst important result of Melitz’s model. Even if it could seem quite intuitive, (
the fact that …rms with higher productivity have higher pro…ts) it is an innovative one obtained
by allowing heterogeneity among …rms and formalizing it in an elegant way.

2.3 Productivity levels, revenues and production quantities

Following the model it can also be demonstrated that the ratios of any two …rms outputs and
revenues only depend on the ratio of their productivity levels. Let’s start by the outputs of two
…rms with di¤erent productivity levels:

p('1 ) p('1 ) p('2 ) p('2 )

'1 : q('1 ) = Q= R and '1 : q('2 ) = Q= R
P P1 P P1

Then, doing some simpli…cations and replacing p(') with its function, we get the ratio

5
between the outputs of the two …rms:
" 1
#
q('1 ) p('1 ) 1 '1 '1
= = 1 = (8) ratio between outputs (VI)
q('2 ) p('2 ) 1 '2
'2

Note that the ratio between the two outputs only depends on the ratio between the produc-
tivity levels of the two …rms.
Similarly we can demonstrate that also the ratio between the revenues of two …rms only
depends on the ratio between the productivity levels of them, obtaining:

1
r('1 ) '1
= (9) ratio between revenues (VI)
r('2 ) '2

These results imply that more productive …rms ('1 > '2 ) will be bigger in terms of both
quantities and revenues. The latter actually matches the empirical evidence. Moreover, the
result on revenues is less obvious than the one on quantities, and it is useful since in many
datasets we have only data on revenues and not data on actual production (quantities).

2.4 The equilibrium in closed economy

We have an economy that is characterized by a probability density function of …rms, g('); similar
to that showed in the …gure below. Firms are distributed according to their productivity level.
Firm with ' ' will stay in the market and produce. On the other side …rms with ' < ' will
exit the market. We will denote with G(') the cumulative distribution function.

EXIT
g(φ)
PRODUCE

φ*

Notation 1 Note how this distribution di¤ ers from a normal one, with very productive …rms
being relatively less common than low productive ones. This is not an accident: the above distrib-
ution of …rms in terms of productivity actually …ts rather well the one suggested by the empirical
evidence.

6
 The equilibrium of the economy will then be characterized by a mass M of …rms (and thus
varieties) with their associated distribution of productivity over the subset ' 2 (0; 1).
In order to solve the equilibrium of the model in the closed economy, we need to use the free
entry and free exit condition to solve for ' . Let us start from the latter.

2.4.1 Free exit condition

Exit ultimately depends on ' : (' ) = 0 i.e. the "cuto¤ " productivity.
The cuto¤ also determines the distribution of the productivity of …rms in equilibrium: the
latter is di¤erent from g(') (the distribution of productivity from which each …rms draws its
'), since in equilibrium only those …rms whose ' ' will be operating. Given the exit rule,
the distribution of …rms’productivity in equilibrium, denoted ('), it is thus conditional upon
the productivity ' ' and it is distributed according to the following (Bayes) rule:

g(')
(') = if ' ' (10) g(') (VIII)
1 G(' )
(') = 0 otherwise

Knowing the density function (') we can then write a measure of the weighted average
productivity in equilibrium, de…ned in general as:

Z 1
1
1
1
e=
' ' (')d' (11) weighted average productivity (VII)
0

In equilibrium, the latter is a function of ' since …rms with ' < ' do not produce (all the
…rms that are in the "exit" part of the graphic above), thus their productivity is not oberved.
We can thus concentrate only on the operating …rms, writing the average productivity level
e as a function of the cuto¤ level ' just by replacing (') with its expression conditional on
'
' ' and integrating from '
1
Z 1 1
1 g(')
e (' ) =
' ' d'
' 1 G(' )

1
bringing out of the integral 1 G(' ) that is a constant, we obtain the …nal version of the average
e as a function of the cuto¤ level '
productivity level '
1
Z 1 1
1 1
e (' ) =
' ' g(')d' (12) aggregate (average) productivity (IX)
1 G(' ) '

We can already make two comments:

The average productivity in the market is uniquely identi…ed by the cuto¤ level ' and the
shape of the distribution g, exogenous (empirical evidence should tell us what the most
likely distribution looks like, and thus derive the ensuing consequences).

7
 Based on these parameters, the model determines endogenously the range of the produc-
tivity levels of the …rms.

Now note that an industry characterized by M …rms whose productivity is distributed as

e , is equivalent to an industry characterized by M …rms
('), yielding an average productivity '
e . Hence, in this setting the average
who share the same uniform productivity distribution ' = '
e is calculated as above. More in
revenue is = to the revenue of the …rm whose productivity '
e it is possible to characterize all the aggregate variables (aggregate
general, note that through '
pro…ts, price and quantity indexes and revenues) of this economy (see Melitz, 2003, p. 1700).
Recalling equation (9), remember that the ratios of any two …rms revenues only depend on
the ratio of their productivity levels. Hence, taking the ratio between the revenue of the average
…rm r(e
') with the revenue of the cuto¤ …rm r(' ) and using Eq. (9), we obtain that

1
e (' )
'
r = r(e
') = r(' ) (13) average revenue
'

e (the weighted average of productivities) cannot be smaller than ' (the lower
Note that '
bound of productivity) by de…nition.
Let us now write the expression for the average pro…t in the economy, recalling our expression
of pro…ts in terms of r(') we derived above. The latter is again equal to the pro…t of the average
…rm r(e
') and, by using Eq. (13) depends on the cuto¤ productivity level ' :

1
r(e
') e (' )
' r(' )
= (e
') = f= f (14) average pro…t
'

We now use the exit rule: the cuto¤ productivity ' represents the threshold productivity
level to stay in the market. The pro…ts of the "marginal" …rm (the one having the cuto¤
productivity) are by de…nition equal to 0, and consequently all the revenues are absorbed by
the …xed cost of production f :

(' ) = 0 , r(' ) = f (15) zero pro…t (X)

we now replace r(' ) in the average pro…t function (14)

1
e (' )
' f
= f
'

Simply…ng and collecting f we get the following formulation for the "zero cuto¤ pro…t"
(ZCP) condition:
" #
1
e (' )
'
=f 1 (16) "zero cuto¤ pro…t" condition (ZCP) (X)
'

The Zero Cuto¤ Pro…t (ZCP) condition thus shows the equilibrium relationship between the

8
average pro…t made by the …rms operating in the industry and each cuto¤ productivity level
' .

2.4.2 Free entry condition

We now use the "free entry" condition. A …rm will produce only if the expected discounted value
of pro…ts for a potential entrant equal the …xed cost of entry fE .
Let’s de…ne the present value of the average pro…t ‡ows earned by an operating …rm as
follows:
1
X
v= (1 )t
t=0

Notation 2 Note that the probability of exit from the market is exogenous and its the same
for all …rms. Since this probability is uncorrelated with ' the exit process will not a¤ ect the
equilibrium productivity distribution (').

This sum converges to:

1
v= =
1 (1 )
v is also the average value of a producer so it can be rewritten as follows:
Z 1
v= v(') (')d'
'

However, recall that ex-ante a …rm does not know its productivity level, so it will actually
enter only if the expected present value of its discounted pro…ts v is larger than the …xed costs
of playing the productivity lottery. We can thuse de…ne ve to be the net value of entry:

ve = pin v fE

where

pin is the probability of ' > ' i.e. the probability of having a "successful" productivity
draw and thus of being a producer
v is the presented discounted value of being a producer for the average …rm
fE is the …xed cost to entry
We can now explicit the probability of being a producer by using the cumulative function
G(') and thus rewrite the net value of entry in its …nal form

ve = pin v fE = [1 G(' )] fE (17) net value of entry (XI)

In equilibrium, this value cannot be negative, as no …rm would enter. Also, it cannot be
positive, given the unbounded number of …rms who otherwise will keep entering. Thus, setting
ve = 0 we can solve for the equilibrium level in Equation (17). Doing so yields us the second

9
condtition we need in order to derive the equilibrium of the economy: the "free entry" (FE)
condition, that is the minimum level of (expected) average pro…ts at which …rms would decide
to enter and produce in the market.

fE
= (18) free entry condition (FE) (XII)
1 G(' )

2.4.3 Equilibrium cuto¤ and average pro…ts

The free entry (FE) and the zero cuto¤ pro…t (ZCP) conditions represent two di¤erent rela-
tionships linking the average pro…t level with the cuto¤ productivity level ' . Melitz (2003)
shows in the Appendix that in the ( ,') space the two curves are shaped as in the graph below,
and have a single interception, thus ensuring the existence and uniqueness of the equilibrium:
" #
1
e (' )
'
=f 1 ZCP (XII)
'

fE
= FE (XII)
1 G(' )

FE

(ϕ*)

(δfE) ZCP

ϕ* ϕ(cutoff)
Note that (' ) in the graph indicates the average productivity level associated to a given
cuto¤ productivity level ' :
In particular, the two curves are shaped as follows:

FE curve: from the moment that G0 (' ) > 0, then is an increasing function of ' along
the FE schedule and satis…es (0) = fE . The intuition is that, on the entry side, for each
given value of the entry …xed cost fE a higher cuto¤ ' is associated to a lower probability
of success (…rms operating in the market would be more competitive) thus pushing up the
average (expected) pro…t level required for entering.

ZCP curve: once …rms have entered the market and compete, their equilibrium average

10
 pro…t as derived from the zero-pro…t condition in ' (ZCP) is downward sloping in the
space ( ; ' ). The intuition in this case is the following. First of all, a higher cuto¤ pro…ts
e for the surviving …rms, as seen in the
' leads to an higher average productivity level '
following derivative of equation (12):
Z
0 '2
g(' )e 1
1 1
e (' ) =
' (' (' ) )g(')d' > 0
( 1)(1 G(' )) '

However, an increase in ' has two opposite e¤ects on the avreage pro…ts along the ZCP.
On one side, we have seen in Eq. 7 that pro…ts tend to increase with …rm’s productivity: hence,
e will have a positive e¤ect on pro…ts (recall that average
an increase in ' leading to higher '
pro…ts are equal to the pro…ts of the average …rm). On the other side, since pro…ts tend to
decrease with the productivity of rival …rms (recall the role of the price index P in Eq. 7), an
e ) will have a negative e¤ect on pro…ts
increase in ' (and thus in ' . Which one of the two
e¤ects dominate depends on the shape of the distribution G('): If G(') has a fat enough right
tail, the second (negative) e¤ect dominates and the ZCP will be downward sloping (as in our
case).
Let’s now try to consider another case: suppose that our distribution of productivity is
shaped as a Pareto distribution ( 'b )k for ' b, whose cumulative distribution is G(') =
dG(') kbk
1 ( 'b )k and density function g(') = d' = 'k+1
. The distribution is shaped as in the graphs
below:

Pareto distribution density function for b=1 Pareto distribution cum. density function for b=1

g(ϕ) G(ϕ)

ϕ ϕ

In this case, substituting in (12) a Pareto distribution with b = 1 and k > 1 (note that
in this case G(') = 1 ( '1 )k =1 ' k, while g(') = k
'k+1
) we get

1 1
k(' ) 1 1 k 1
e (' ) =
' ='
k +1 k +1

e (' )0 and the associated ZCP will be the following

Hence the derivative '

1
0 k 1 f 1
e (' ) =
' which implies = , i.e. the ZCP is ‡at
k +1 k +1

11
Notation 3 Try to verify the calculus by your own by applying the pareto distribution in equa-
tion 12. See Helpman, Melitz and Yeaple (2003) for an application.

2.4.4 A comparative statics exercise: the e¤ects of a subsidy

Through this model we are able to predict the e¤ect of several policies on the aggregate
productivity level of a given economy as on the average pro…ts in the economy. Let’s
suppose that the government decides to provide a subsidy to production. Suppose, in
particular, that rather than subsidizing “pure” R&D (basic research sustained by …rms
before they enter the market, i.e. linked to the …xed cost of entry fE ), the Government
provides a subsidy to the production process, which in the setting of the model can be
seen as decrease of the …xed cost f that every …rm has to sustain in order to produce.

Thanks to the subsidy, the …xed cost f decreases and subsequently …rms that before
would have been forced to exit the market due to their low productivity level will be able
to survive. Graphically, as the following …gure shows, a subsidy makes the ZCP condition
curve shifting to the left. In the new equilibrium we have a lower average productivity
level and a lower average pro…t in the economy. The subsidy in this case has a permanent
negative e¤ect on the economy, since it structurally lowers its aggregate productivity! By
lowering, indeed, the cuto¤ productivity level the subsidy permits to ine¢ cient …rms to
“arti…cially”survive, preventing the resources (labour and capital) from naturally shifting
to more productive …rms.

FE

π (φ*)
ZCP
π (φsub)
ZCPsub
δfe

φsub φ* φ

2.4.5 Number of …rms

In a stationary equilibrium, the aggregate variables must also remain constant over time. Thus,
denoting M as the total number of …rms, the mass of new entrants must equal the mass of …rms

12
who exit:
pin Me = M (19)

The aggregate labour used for production Lp and the aggregate labour used for investment
Le must sum the aggregate labour : L = Lp + Le
Aggregate payments to the production workers must equal the di¤erence between aggregate
revenue and pro…ts: Lp = R
Using (19) and the free entry condition we can rewrite Le as follows:

M
Le = Me fe = fe = M =
pin

using the aggregate revenue and the average pro…t level it is then possible to derive the
number of producing …rms M in equilibrium.

R L
M= = (20) number of …rms (XIII)
r ( + f)

Note how similar this equation is to the one deriving the equilibrium number of …rms in
Krugman (1980). The only di¤erence is that in Melitz (2003). In other words, once e
and '
are determined, this model yields a similar outcome to the one generated by Krugman (1980)
e and pro…ts . The only di¤erence is that
assuming a representative …rms with productivity '
in this model pro…ts and productivity are endogenously determined, not assumed, so they react
to external shocks, such as economic integration.

2.5 Implications of the model in closed economy

The e¤ct of country size: the size of the country in the Melitz model does not in‡uence
all the …rm level variables. L does not appear in the ZCP condition neither in the FE
condition. The mass of …rms increases proportionally with country size, although the
distribution of …rm productivity levels remains unchanged.

Welfare: welfare per worker is de…ned as follows:

1
1
W =P =M 1 pe
' (21) welfare per worker (XIV)

The number of …rms (varieties) M increases with the size of the country, we have thus the
same "variety e¤ect" already seen in the Krugman model. Consequently the welfare per
worker is higher in a larger country due only to increased product variety (remember the
"love for variety" assumed in our utility function).

Aggregate productivity endogenously determined

As already stated, the novelty of this model is that it introduces …rm heterogeneity and thus
is able to treat aggregate productivity level and the average …rm pro…t level as endogenous.

13
3 Open economy
Let’s see how things changes in the model when the economy open up to trade.
The hypotheses of Melitz (2003) in open economy are the followings:

There are n+1 countries who trade among them (the notation n+1 is used for mathematical
convenience). Countries are identical among them. This assumption is important because
will allow us to assume that the factor price equalization condition hold (see Helpman,
Melitz and Rubinstein, 2006, for a generalization to the asymmetric case).

Firms face a per unit trade cost modelled as an "iceberg" cost: > 1 as in Krugman
(1980).

Firms also face a …xed cost before starting to export fx > 0. Note that at the stage when
…rms have to decide whether to export or not they already have knowledge about their
productivity level '.

Note that with no additional costs for trade the e¤ect of trade would be the same as that
of an increase in the size of the country. In this case the existence of …rm heterogeneity
does not a¤ect the impact of trade, with the result being similar to the ones obtained
by Krugman (1980). To allow for …rm heterogeneity to interact with trade, the model
assumes two (realistic) sources of asymmetry: the "iceberg" cost (already introduced by
Krugman, 1980) and the …xed cost fx .

3.1 Pricing rule, revenues and pro…ts in open economy

These di¤erences in costs will lead to di¤erent …rm level variables according to their export
status.

Pricing rule

1
Domestic market: pd (') = 1 ' (equation 6)
Foreign market: px (') = 1 ' = pd ('), once again as in Krugman (1980)

Revenues
1
p(')
Domestic market: rd (') = P R (from Eq. 5 in closed economy)

Foreign market: rx (') = 1 rd (') by substituting the pricing rule for the foreign
market.

The combined revenue of a …rm depends on its export status. If the …rm only sell in the
home market (no export):

r(') = rd (') (22) revenue for "only domestic" …rms (XV)

14
 If the …rm export to all countries its revenue will be the sum of the revenue from the home
market and the revenue from all the export markets:

1
r(') = rd (') + nrx (') = (1 + n )rd (') (23) revenue for exporting …rms (XV)

We can now write the average revenue function (that is the sum of the average revenue from
domestic sales and the average revenue from export sales):

r = rd (e
') + px nrx (e
'x ) (24) average revenue (XVIII)

where px is the probability of exporting.

Pro…ts and …rm value

The pro…ts for …rms that only sell at home are (see the case of closed economy):

rd (')
d (') = f (25) pro…ts for "only domestic" …rms (XVI)

Pro…ts for …rms that also export are:

rx (')
x (') = fx (26) pro…ts for exporting …rms (XVI)

The …rm’s combined pro…t can be written as:

(') = d (') + max f0; n x (')g (27) combined pro…t

The …rm value in open economy is as in closed economy:

(')
v(') = max 0;

We can now write the average pro…t function:

= d (e
') + px n x (e
'x ) (28) average pro…t (XVIII)

3.2 Entry and export cuto¤s

In the open economy model there are two cuto¤s, the entry cuto¤ and the export one.
The entry cuto¤ can be written as usual as:

' = inf f' : v(') > 0g (29) entry cuto¤

i.e. taking into account the value of the …rm in the open economy (i.e. the discounted value
of pro…ts obtained in the open economy).

15
 The export cuto¤ has two conditions: the …rst one is to have a …rm value >0, the second
one is to make positive pro…ts from exporting:

'x = inf f' : v(') > 0 and x (') > 0g (30) export cuto¤

By their de…nition the cuto¤ levels must then satisfy d (' ) = 0 and x ('x ) = 0.
Since the …xed costs of exporting fx > 0 add to the …xed cost of production f , according to
the cuto¤s three cases can arise depending on the productivity level of a …rm.

If ' < ' =) d (') < 0 : the …rm leaves the market.
If ' < ' < 'x =) d (') > 0 but x (') < 0 : the …rm produces exclusively for the local
market, because it does not make enough pro…ts from exporting
If ' > 'x =) d (') > 0 and x (') > 0 : the …rm is productive enough to earn positive
pro…ts from both its domestic and export sales.

See below the graphical intuition of this argument, where pro…ts are a linear function of
productivity according to Eq. 7 (or, more precisely of ' 1 ).3 According to the cuto¤s thus
there exists a selection of …rms into the export status in an open economy. The selection occurs
when 'x > ' , that is true if x (' )< d (' ). In other words, only the most productive …rms
export.

π(ϕ)
πd(ϕ) πx(ϕ)

ϕ∗ ϕx∗ ϕ
-f
-fx

In particular, to derive the precise ranking in cuto¤s, consider the foreign pro…ts evaluated
rx ('x ) 1
at the export cuto¤, that is x ('x ) = fx . Substituting the expression rx (') = rd (')
and recalling Eq. 9 evaluated for 'x and ' , one obtains

1 ('x ) 1 r(' )
x ('x ) = fx
(' ) 1

3
Note that x (') is not parallel to d (') as the former incorporates transport costs in the pricing equation.
This graphical argument is made in Helpman, Melitz and Rubinstein (2003) and already assumes a precise relations
between fx and f where fx > f . The precise hypotheses are clari…ed in the remaining of the paragraph.

16
 Recalling that exporting …rms already satisfy the domestic zero pro…t condition, i.e. they
have a productivity larger than the cuto¤ such that (' ) = 0 , r(' ) = f , hence substituting
r(' ) above and imposing the same zero pro…t condition for the pro…ts obtained in foreign
markets x ('x ) = 0, we obtain that

1= 1
fx
'x = '
f

We will therefore have that 'x > ' as long as 1f > f . If, in other words, fx > f = 1
x

the graph above applies, and exporters will be more productive than purely domestic …rms.

Notation 4 Note that when there are no …xed export costs (fx = 0) no level of variable cost
> 1 can induce this selection. On the other side a large enough …xed export cost fx > f will
induce this partitioning even when there are no variable trade costs!

3.3 Zero cuto¤ pro…t and free entry condition in open economy

ZCP

The zero cuto¤ pro…t condition (as in closed economy) implies a relationship between the
average pro…t per …rm and the cuto¤ productivity level ' :
For the pro…ts earned by selling locally: d (' )=0, d (e
') = f k(' )
For the pro…ts earned by exporting: x ('x ) =0, x (f
'x ) = fx k('x )
h i 1
where k(') = 'e (')
' 1
Using the zero cuto¤ for the pro…ts earned by selling locally and by selling in foreign markets
in the average pro…t equation (28), we get the ZCP condition for the open economy4 :

= f k(' ) + px n fx k('x ) (31) ZCP for open economy (XX)

It is clear that this is the ZCP condition in closed economy (the …rst addendum) plus an
element that derives from the fact that now there is more competition in the market due to
trade.

FE
The free entry condition remains the same as in closed economy: the expected value of a
…rm is = 0 if and only if

fe
= (32) FE in open economy
1 G(' )

The new equilibrium characterized by ' and (' ) is shown in the followin graph. Trade
causes an upward shift of the ZCP curve (while the FE curve remains unchanged) leading to

4
See the paper from Melitz in particular the equation (19) on p. 1711 for the analytical derivation of the ZCP
condition in open economy.

17
a new equilibrium where the average productivity level , i.e. the level (' ) in the graph, is
higher than in autarchy. Most importantly, the cuto¤ productivity to stay in the market is also
higher than in autarchy. Since the cuto¤ pro…t level drives also the average productivity level
e is increased.
in the economy, the main implication of the model is that with trade also the '

FE

π(φ*)
ZCPtrade
π (φaut*)
ZCPaut
δfe

φaut* φ* φ

As for the closed economy case, there is a last equilibrium variable to de…ne, the number
of …rms operating in the market after liberalization. Interestingly, the mass of domestic …rms
operating in the home country after trade liberalization is smaller with respect to the autarchy
situation, as it clear from the followin equation.

R L L
M= = < (33) Number of …rms in OE (XXI)
r ( + f + px nfx ) ( + f)

The number of domestic …rms is smaller because of the "selection" e¤ect that forces the
least productive …rms to leave the market. But the total number of …rms (varieties) in the home
market does not go down (foreign exporters can now enter !), so welfare unambiguously goes up
with trade.

4 Gains from trade in the Melitz’s model

By modelling …rm heterogeneity the Melitz (2003) model is able to reach extremely realistic
conclusions in terms of the e¤ects induced by trade on the economy.

4.1 The selection e¤ect

With trade the ZCP condition curve shifts up. The economic intuition of this shifting is
that the exposure to trade induces an increase in the cuto¤ productivity level ' (the level

18
 of productivity needed in order to stay in the market) and, consequently, in the average
e . This e¤ect occurs because …rms with productivity
productivity level of the economy '
levels between 'aut (the cuto¤ productivity level in autarchy) and ' can no longer earn
positive pro…ts in the new equilibrium and therefore exit from the market. In other words,
in the open economy surviving is harder and only the most productive …rms remain in the
market5 . The positive selection process induces the aggregate productivity of the economy
to raise. Thus, economic integration can act as a powerful device selecting only the ‘…ttest’
…rms (the right tail of the productivity distribution): the best industrial policy to boost
growth and productivity is opening up to trade!

Another selection process occurs because only …rms with productivity levels above 'x enter
the export market.

Both selections e¤ect reallocate market shares toward more e¢ cient …rms thus contributing
to an aggregate productivity gain (but we will talk further about the reallocation e¤ect
below). The increase in average productivity generates an increase in welfare because
consumers enjoy lower prices. This is the …rst time that an international trade model
shows how the selection e¤ect can e¤ectively occur. The selection e¤ect in fact was not
present in the Krugman’s model that was predicting only the variety e¤ect as a gain from
trade.

4.2 The variety e¤ect

The classical variety e¤ect is also predict by this model. Consumers after trade, indeed,
can choose between goods produced locally and goods produced by foreign …rms. Even if,
as shown in equation (33) the number of …rms operating in each country is smaller than
in autarchy, the varieties available to the consumers increase.

In particular, since Mtrade = (1 + npx )M > Maut the decrease in the numebr of domestic
…rms following the transition to trade is dominated by the number of new foreign exporters.
There are less …rms operating in each countries but more varieties available: this e¤ect
generates an additional gain from trade through the variety channel (remember "love for
variety" assumption under the CES utility function of the consumer).

4.3 The reallocation e¤ect

The model also predicts that trade causes reallocation of resources from less productive
to more productive …rms within the economy (or within sectors). This impact can be
analyzed using the following inequalities:

rd (') < raut (') < rd (') + nrx (') for every ' '

5
The channel by which …rms …nd it di¢ culty to survive is the domestic factor market: …rms compete for a
common source of labour L that is …xed.

19
 The inequalities must be interpreted as follows:
- since rd (') < raut (') all …rms incur a loss in domestic sales in open economy because of
the entry of foreign competitors in the local market. Those …rms who do not export (' < 'x )
also incur a total revenue loss.
- since raut (') < rd (') + nrx (') …rms that are able to sell also on the export market increase
their share of industry revenues.

On the pro…t side the scenario is more complex:

- Firms who do not export loose pro…ts because their revenues decreases.
- For the …rms who export there are two opposite e¤ects on pro…ts, one (positive) due to the
increase in revenues, the second one (negative) due to the additional export cost. Pro…ts, thus,
will increase for some …rms, among the exporters, and decrease for some others, depending on
their productivity level.
The …gure below shows the changes in revenues and pro…ts driven by trade.

4.3.1 The impact of an increase in the exposure of an economy to trade

Previously we discussed the e¤ect of passing from autarchy to trade. Now we want to analyze
the e¤ects of an increase in the exposure of an economy to trade, so gains from various forms of
deeper economic integration.

Increase in n, number of trading partners.

20
 An increase in the number of trading partners n makes the ZCP curve shift right. As a
result both cuto¤ productivity levels increases. The increase in n forces the least productive
…rms to exit. The e¤ect is similar to the one induced by the passing from autarchy to open
economy. Both market shares and pro…ts are reallocated towards the more e¢ cient …rms. This
reallocation of shares generates an aggregate productivity gain and an increase in welfare.

Decrease in the export market entry …xed cost fx .

A decrease in the …xed cost to entry the export market also makes the ZCP curve shift right
but will produce two di¤erent e¤ects. One is the usual selection of the least productive …rms
who are forced to leave the market. The second e¤ect is that now the entry in the export status
is less costly, thus more …rms will be able to export. Graphically the export cuto¤ 'x is smaller
than before. The e¤ects on the reallocation of pro…ts and revenues are similar to those caused
by passing from autarchy to trade.

4.4 Concluding remarks

The message of the model is that trade generates "winners" and "losers" among …rms depending
on their productivity levels.
For the most productive …rms "open economy" is good because they are good enought to
export and thus gain market share in foreign markets. These …rms will growth and improve
their situation with respect to autarchy.
For the least productive …rms "open economy" is bad, because they are either forced to leave
the market (' < ' ) or sell only in the local market (' < ' < 'x ). These latter …rms will
loose market shares and see their pro…ts decrease. Their situation is worsened with respect to
autarchy.

Nevetheless what is important is that this "Darwinian" selection process is favourable for the
economy as a whole. On one side, indeed, the aggregate productivity level increase (and we know
how important productivity is) because resources are allocated more e¢ ciently among …rms; on
the other side the average welfare increases because consumers can enjoy lower prices and and
increase in varieties.

21