Lecture 4:

Firm Heterogeneity, Endogenous

Markups, and Factor Endowments

Thomas Chaney
Econ 357 - International Trade

1 Introduction
We have seen in the previous lecture how to add …rm heterogeneity to
the classical Krugman model of trade. We have seen in the Melitz paper
how taking into account …rm heterogeneity may explain why exposure to
trade will induce some increase in aggregate productivity, even without
a change in the actual technology of production, through a more e¢ -
cient reallocation of factors of production. We have seen in Chaney that
taking into account …rm heterogeneity will change substantially some
predictions for the patterns of international trade. Among others, …rm
heterogeneity will introduce a new margin of adjustment of trade barri-
ers, the extensive, and this barrier behaves di¤erently from the intensive
margin traditionally studied. One prediction is that the elasticity of sub-
stitution between goods will no longer increase the sensitivity of trade
‡ows to trade barriers, it may actually dampen it.
In this lecture, we will try and re…ne our approach to international
trade with heterogeneous …rms to account for some stylized facts that
these models were missing. The …rst important caveat of models based
on CES preferences and monopolistic competition is that mark-ups are
constant. We will see in Melitz and Ottaviano (2003), as well as in
Bernard, Eaton, Jensen and Kortum (2003) how more elaborate mod-
els can account for the endogenous determination of mark-ups. These
models will give us a better understanding of the adjustments that take
place when a country is exposed to trade. In Bernard, Redding and
Schott (2004), we will see how the Melitz model can be augmented to
include di¤erent factors of production, and how we can then relate this
model with heterogeneous …rms to the Ricardian comparative advantage
model.

1
2 Melitz and Ottaviano (2005)
Melitz and Ottaviano keep the monopolistic competition assumption,
but they move away from CES preferences. This will allow them to
generate endogenous mark-ups, and derive interesting properties related
to market size. Mark-ups respond to the "toughness" of competition
(which we will de…ne precisely). In larger markets, competition will be
tougher, so that …rms charge lower mark-ups, and aggregate productivity
is higher. Integration through costly trade will not entirely kill this e¤ect,
so that larger countries, even if they are open to trade at some cost, will
still be characterized by tougher competition than others. They are
also able to describe the e¤ect of some stylized trade policies. Trade
liberalization increases import competition and therefore reduces mark-
ups, and increases aggregate productivity (as in Melitz).

Main assumptions:

Firms are heterogeneous in terms of productivity, as in Melitz

(their productivity is revealed after they pay a sunk entry cost).
Preferences give a linear demand system (developed in Ottaviano,
Tabuchi and Thisse [2002]), which will generate endogenous mark-
ups.
As in Chaney (2006), there is a homogenous good produced with
constant returns to scale so that wages in every country are pinned
down.
As in Chaney (2006), the world is made of asymmetric countries of
di¤erent size, separated by potentially asymmetric trade barriers.
There are only variable trade barriers, and no …xed cost of export-
ing. Given the linear demand system, no …xed cost is needed to
get an endogenous selection of …rms into the export market.

Closed economy equilibrium

Preferences and consumer behavior:

Each consumer c derives utility from a homogenous good (used as

the numeraire, as in Chaney), and a continuum of di¤erentiated
varieties, ! 2 ,
Z Z Z 2
c c c 1 c 2 1
U = qo + q (!) d! q (!) d! q c (!) d!
!2 2 !2 2 !2
(1)

2
 – (> 0) indexes the degree of product di¤erentiation between
varieties. In the limit of = 0, goods become perfectly
homogenous, so that consumers only care about the total
amount of di¤erentiated goods they consume, not which spe-
ci…c variety. As increases, consumers care more and more
about the distribution of consumption over all varieties, so
that goods become more and more di¤erentiated.
– and (> 0) index the substitution between the di¤eren-
tiated varieties and the homogenous good. Both parameters
shift out the demand for di¤erentiated varieties relative to the
homogenous good.

Note that the marginal utility of all goods is bounded from above,
so that consumers may not consume all goods, even in the absence
of …xed costs. We will use this condition to derive in equilibrium
the set of goods that are produced, and de…ne the extensive margin
of trade. We assume though that the income is always large enough
so that consumers have a positive demand for the numeraire.

Given these preferences, consumers have a linear demand for each

good. If consumers consume a positive quantity of good !, then
the inverse demand function from each consumer is,

p (!) = q c (!) Qc (2)

R
with Qc = !2 q c (!) d! the aggregate demand for di¤erentiated
varieties from consumer c.

With L identical consumers, we can integrate this expression over

all consumers c, rewrite Qc as a function of the prices p (!)’s,
rearrange and derive the total demand for good ! as,
L L N L
q (!) Lq c (!) = p (!) + p; 8! 2 (3)
N+ N+
where is the set of goods that are consumed, N is the mea-
sure of consumed varieties,
R and p is the average price of consumed
varieties, p = (1=N ) !2 p (!) d!.

Since there is a "choke price" for each variety (given the bound on
the marginal utility of each variety), we can de…ne the set of goods
actually consumed, , as the largest subset of such that,
+ Np
p (!) (4)
N+

3
 We can easily see that any price p (!) must be below , the mar-
ginal utility of the numeraire good (which we have assumed is
consumed), so that p .

With this setting, the demand elasticity that each producer ! is

facing is not constant (not equal to ). As the average price p goes
down, or as the number of competitor N increases, the environment
gets "tougher", and the price elasticity of demand increases. This
will be the driving for behind the endogenous adjustment of mark-
ups of …rms.

We can derive a simple expression for the indirect utility of each

consumer c,
1 1 1N 2
U c = Ic + + ( p)2 + p (5)
2 N 2
R
with I c the income of consumer c, and 2
p = (1=N ) !2 (p (!) p)2 d!
the variance of prices.

– Welfare rises with a decrease in average prices, p.

– Welfare also rises with an increase in the variance of price, 2p ,
as consumers reoptimize their consumption across varieties.
– Finally, as in the CES case, consumers exhibit "love for va-
riety", as welfare increase with the number of varieties avail-
able, N (holding the distribution of prices, p and 2p , con-
stant).

Production and …rm behavior:

Labor is the only factor of production. The unit labor require-

ment in the homogenous sector is one. This good is used as the
numeraire. We assume that all countries will be producing at least
some of the homogenous good so that the wages in all countries is
equalized to one, as in Helpman and Krugman, and in Chaney.

There is a sunk cost (in labor units) of starting production, fE .

Once this cost is paid, each …rm receives a random productivity
draw. Production is constant returns to scale, with unit labor cost
c, drawn from a (known) distribution G (c) over [0; cM ].

4
Monopolistic competition means that …rm maximize pro…ts choos-
ing price or quantity, taking as given the residual demand for their
good (i.e. the prices set by their competitors, p and N ). Using the
expression for demand in Eq. (3), total pro…ts earned by a …rm
with cost c selling quantity q is

(q) = (p (q) c) q
N
= + p q c q
N+ N+ L

The optimal quantity to maximize this pro…t must be such that,

N
2 q= + p c
L N+ N+
+ Np
Plugging back the price from Eq. (3), which gives N+
= p+ L q,
we get the optimal pricing strategy,
L
q (c) = (p (c) c) (6)

Firms with too high a cost, i.e. a cost c above the threshold cD =
+ Np
N+
have zero demand, and exit immediately. A …rm with a
cost cD is exactly indi¤erent between staying in business or exiting,
p (cD ) = cD . We assume that the upper bound on cost, cM , is
always have enough so that in equilibrium there are some …rms in
the di¤erentiated sector (cD < cM ). The threshold cD summarizes
all the information that is needed to describe the behavior of the
…rms that stay in business.
1
p (c) = (cD + c) (price)
2
1
(c) = (cD c) (mark-up)
2
L
q (c) = (cD c) (quantity)
2
L 2
r (c) = cD c2 (revenue)
4
L
(c) = (cD c)2 (pro…ts)
4
We get the nice following properties:

– Lower cost …rms charge lower prices.

5
 – Lower cost …rms earn higher revenues and pro…ts.
– Lower cost …rms set higher mark-ups. So unlike the CES, not
the entire productivity gain is passed on to consumers, part
of it is retained as higher mark-ups.

Free entry condition:

To determine the general equilibrium of this economy, we need

to solve for the total number of entrants, and the cost threshold.
To do so, we use the free Rentry condition. The expected pro…ts
c
of a potential entrant are 0 D (c) dG (c) fE . These expected
pro…ts must be driven down to zero from free entry. On top of it,
we know that the threshold for survival is given by cD = N++N p ,
which we can rearrange to get the zero cuto¤ pro…t condition. So
the equilibrium is given by two conditions, the free entry condition,
(F E), and the zero cuto¤ pro…ts condition, (ZCP ):
Z
L cD
(cD c)2 dG (c) = fE (F E)
4 0
2 a cD
N= (ZCP )
cD c
Rc
with c = 0 D cdG (c) =G (cD ) the average cost conditional on sur-
vival.

We can already describe a few properties of the equilibrium:

– Average productivity, c, is higher when sunk costs are lower.

– Average productivity, c, is higher when varieties are closer
substitute (low ).
– Average productivity, c, is higher in larger markets (high L).
– The demand parameters and do not a¤ect the selection
of …rms, it only a¤ects the total number of …rms.
– Competition is "tougher" in larger markets, so that average
prices in such markets are lower, and all …rms respond by
charging lower mark-ups.

We assume for simplicity that the cost is drawn from a Pareto

distribution with a scaling parameter k 1,
k
c
G (c) = ; c 2 [0; cM ] (7)
cM

6
 This is exactly the same distribution as in Chaney (2006). There,
the labor productivity, ' = 1=c, was drawn from a Pareto over
[1; +1) with a scaling parameter , so it’s exactly the same as
here replacing k = and cM = 1. As in Chaney (2006), k is an
inverse measure of the dispersion of labor productivity, or labor
cost. A high k means that most of the cost draws are concentrated
around cM , whereas k = 1 corresponds to a uniform distribution
over [0; cM ]. With this speci…c functional form, we get the simple
closed form solutions,
1
k+2
cD = (F E)
L
2 (k + 1) cD
N= (ZCP )
cD

with = 2 (k + 1) (k + 2) ckM fE . We can describe with speci…c

functional assumption several properties of the equilibrium.

Size matters:

Larger markets are characterized by:

– a higher average …rm size, and average pro…ts (the market

size e¤ect dominates the indirect e¤ect of lower prices and
mark-ups).
– higher average mark-ups (the direct e¤ect of tougher compe-
tition outweighs the selection e¤ect on more productive …rms
with lower mark-ups).
– invariant average pro…tability ( =r) (both average pro…ts and
average sales increase proportionally).
– a lower variance of cost, prices and mark-ups (the support of
the distribution of cost draws shrinks with tougher competi-
tion).
– a higher variance of …rm size (the direct market size mag-
ni…cation e¤ect dominates the reduction in the support of
productivity draws).

These results are consistent with stylized facts from the IO lit-
erature, mainly Campbell and Hopenhayn (2002) and Syverson
(2004, 2005). Campbell and Hopenhayn look at the retail sector,
and …nd that larger markets have higher average size (measured

7
 in sales or employment), as well as more dispersed sizes. Syverson
looks at sectors where real output (quantities) is measurable (ce-
ment and concrete), so that he can recover unit prices (this is a
unique example of reliable unit price data at the …rm level!). He
…nds larger plants, higher productivity, and tougher competition
(less dispersed productivity as well as a higher lower bound for
productivity) in larger markets.

Open economy equilibrium

We now consider opening up trade between two of these economies,
home and foreign (denominated with a *), with respective size L
and L (potentially asymmetric), and separated by some trade
barriers, modelled as iceberg transport cost, . We could have
asymmetric transport costs, and more than two countries too, the
model would still be (almost) as tractable. This tractability mainly
comes from the assumption of a freely tradable homogenous sector
with constant returns to scale.
The price that a …rm charges in a given market (domestic or ex-
port) depends on the local demand. The unit cost for a …rm with
cost draw c is c for the domestic market, c for the export mar-
ket. Because of the constant returns to scale assumption, …rms
decide separately how much to produce on each market, home and
foreign. As in the autarky case, we can easily derive the "choke"
price on each market, that is the price below which a …rm would
have zero demand, and therefore the cost threshold for domestic
production,
+ Np
pD = = cD (8)
N+
+ N p
pD = = cD
N +
as well as the "choke" price for export and the cost threshold for
exports,
pD
pX = = cX (9)
pD
pX = = cX

When considering whether or not to enter, a …rm compare the

cost of entry, fE , to the bene…t from entering, i.e. the expected

8
discounted sum of future pro…ts. Those pro…ts include both pro…ts
earned from domestic sales, and potentially pro…ts earned from
foreign sales. The free entry condition imposes that these two are
equalized,
Z cD Z cX
D (c) dG (c) + X (c) dG (c) = fE (10)
0 0
Z cD Z cX
D (c) dG (c) + X (c) dG (c) = fE
0 0

If we use the speci…c functional form of Pareto distributed cost

shocks, after rearranging, we get the simple closed solutions for
the open economy equilibrium,
1
1 k+2
cD = k
< cAutarky
D (F E)
1+ L
1
1 k+2
cD = (F E )
1+ k L
2 (k + 1) cD
N= > N Autarky (ZCP )
cD
2 (k + 1) cD
N = (ZCP )
cD

Note that in that case of an open economy, N represents the total

number of varieties consumed in the home market. It is made of the
sum of domestic producers, and exporters from the foreign country:
if NE (NE ) is the number of entrants in the home country (foreign),
of which only a subset survive on the domestic market/export, then
N = G (cD ) NE + G (cX ) NE . We can solve for the total number of
entrants in each country (see the paper). In the special case of the
Pareto distribution, we can see that the total number of varieties
consumed increases.

We can immediately see that opening up to trade reduces the cost

cuto¤ cD . Only the most productive …rms still survive in the open
economy equilibrium, and labor is reallocated towards the most
productive …rms. As in Melitz (2003), the least productive …rms
disappear, and their workers are reallocated towards the more pro-
ductive survivors. On top of that, only the most productive among
survivors export to the foreign market, inducing a further reallo-
cation of workers towards more productive …rms. In this model,
all variables of interest depend only on the cuto¤ cost cD .

9
The impact of trade on prices, mark-ups, sizes, welfare:

We can describe precisely the impact of opening up to trade. All

the e¤ects are qualitatively similar to increasing the size of the
economy in autarky.

– the increased competition from foreign exporters induces a

reduction in mark-ups. Note that only the most productive
…rms survive, which charge higher mark-ups than less produc-
tive …rms. But since they face increased competition, they
have to reduce their mark-ups, and this e¤ect dominates.
– prices go down, both because only the most productive …rms
stay, and because all …rms reduce their mark-ups.
– average …rm size, …rm pro…ts, and product variety increase.

We also see that if trade is costly ( > 1), trade does not entirely
integrate markets. This is obvious from the fact that size still
matters:

– the larger country has a lower cost cuto¤, higher average pro-
ductivity and product variety, and lower mark-ups and prices
(consumers bene…t from all these combined e¤ects).
– with that speci…c functional form, the size of one’s trading
partner does not a¤ect domestic variables. Even though a
larger trading partner represents increased export opportuni-
ties, this is o¤set by increased competition. Similarly, even
though a larger trading partner represents an increased im-
port competition, exit in the long run reduces the number of
entrants and o¤sets the competition e¤ect.

Aggregate exports:

Finally, from this model, we can derive gravity type predictions

for bilateral trade ‡ows. With NE the total number of entrants in
the domestic country, we get total exports from home to foreign,

cMk
X= NE L cDk+2 k
2 (k + 1)

As in the Melitz/Chaney or the Eaton and Kortum models, total

exports result from both the extensive and the intensive margins
of trade. As in Eaton and Kortum (2002) and Chaney (2006), the

10
 parameter driving the substitutability between varieties ( here)
does not a¤ect the sensitivity of aggregate exports to trade barri-
ers. Only the distribution of productivity shocks (indexed by k)
matters.

3 Bernard, Eaton, Jensen and Kortum (2003)

We have seen with the Melitz and Ottaviano (2004) paper how we can
remove the assumption of CES preferences, but keep the monopolistic
competition framework, in a model with heterogeneous …rms. We keep
most of the predictions of the initial Melitz (2003) model, and in addi-
tion, we derive prediction for the endogenous determination of mark-ups.
Among other, the model predicts that larger countries will be character-
ized by a "tougher competition", and therefore lower mark-ups. Con-
sumers bene…t from opening up to trade for several reasons then: …rst,
as in Krugman, they get access to a wider range of di¤erentiated goods;
second, as in Melitz, opening up to trade induces in increase in aggregate
productivity; last, and unlike the previous models we saw, they bene…t
from a reduction in prices, part of it due to lower mark-ups charged by
…rms.
We will see now how we can instead keep the CES preferences as-
sumption, but instead remove the monopolistic competition assumption.
Eaton and Kortum (2002) consider the case of perfect competition. Per-
fect competition actually gives results that are almost identical to mo-
nopolistic competition: instead of all …rms charging exactly their mar-
ginal cost, they charge a constant mark-up (Dixit-Stiglitz mark-up) over
their marginal cost. BEJK instead consider Bertrand competition. We
will see that once heterogeneous …rms compete in prices, since the price
they charge depends on the price of their direct competitors, mark-ups
will endogenously respond to changes in the "toughness" of competition,
which will be the case when countries open up to trade.

Set-up

Preferences:

As in Krugman, we assume that consumers maximize CES prefer-

ences. Unlike the Melitz/Chaney model, we assume a …xed set of
di¤erentiated goods of mass 1. Consumers in country n consume

11
 a quantity qn (!) of each variety !, and derive a utility,
Z 1 1
1
Un qn (!) d! (11)
0

This gives a simple isoelastic demand system. If good ! has a

price pn (!) in country n, total expenditure by consumers in n is
Xn , then the demand for each good ! is,
1
pn (!)
xn (!) = Xn (12)
Pn
Z 1 1
1

with Pn = pn (!)1 d!
0

Fréchet distributions:

In each country, there are many di¤erent …rms. Each of these …rms
get a random productivity draw. In each country, only the best
technology will be used, so that only the minimal cost is used. The
distribution of the lowest cost of producing good ! in country i,
zi1 (!) (note that the superscript 1 denotes the best draw), is drawn
from a Fréchet distribution, Fi :

Fi (z1 ) = Pr zi1 (!) z1 (13)

= exp Ti z1

Because of the nature of competition, we will need the distrib-

ution not only of the highest productivity, but also on the joint
distribution of the highest and the second highest productivity.
As proved in Appendix A.1, if the joint distribution of the highest
productivity and of the second highest productivity, is,

Fi (z1 ; z2 ) = Pr zi1 (!) z1 ; zi2 (!) z2 j 0 z2 z1 (14)

= 1 + Ti z2 z1 exp Ti z2

then the distribution of the highest productivity is exactly as given

in Eq. (13). This proposition can be proven by looking at the
di¤erent possible orders of zi1 (!) ; zi2 (!) ; z1 and z2 , and the
respective probabilities of these di¤erent orders. Note that setting
z1 = z2 returns the initial Fréchet distribution from Eq. (13).

12
 Ti scales up the technology of all goods in country i. It is a measure
of the absolute advantage of country i. The parameter , which
we assume is the same in all countries, is an inverse measure of
the heterogeneity in productivity between di¤erent sectors. It will
index the strength of comparative advantages between countries.

Why the Fréchet?

Kortum (1997) and Eaton and Kortum (1999) derive the distrib-
ution of the leading-edge e¢ ciencies from a dynamic model with
endogenous innovation. If one randomly draws technologies from
some distribution, and the market only keeps the best draws, the
distribution of the best draws (suitably normalized), if it exists,
will be extremal. There are 3 types of extreme value distributions
(Gumbel, Fréchet and Weibull). The Fréchet is one of the 3 ex-
teme type distribution, and the only one in which heterogeneity
does not vanish in the limit. So we have some reasons to believe
that the distribution of the e¢ ciency frontier in a given country
may be Fréchet.
The second property (which is a direct corrolary) of the Fréchet
is that the maximum of a family of Fréchet distributions is still
a Fréchet. This propoerty will be usefull when we consider inter-
national trade: in a given country, consumers will only consume
goods that are the cheapest available.

Production and trade costs:

Firms face constant returns to scale technology, so that a …rm

with productivity z will produce z units per unit of labor. If a
…rm in country i wants to export goods towards country n, it faces
an iceberg transportation cost, in 1. We impose that those
bilateral trade barriers satisfy the triangular inequality, kn
ki in ; 8k; i; n. So the cost of shipping the cheapest version of
good ! from country i to country n, if the wage in country i is wi ,
is,
in wi
c1in (!) = 1 (15)
zi (!)
The cheapest version of good ! in country n, looking at all poten-
tial source countries, is

c1n (!) = min c1in (!) (16)

i

13
Knowing that the cost depends on the productivity draw, on the
domestic wage, and on the trade barriers, we can derive the distri-
bution of the cheapest cost for good ! from country i in country
n,

G1in (c1 ) = Pr c1in (!) c1 (17)

1 in wi
= Pr zin (!)
c1
=1 exp Ti ( in wi ) c1

The distribution of the cheapest cost for good ! in country n (from

potentially any country in the world),

G1n (c1 ) = Pr c1n (!) c1 (18)

h i
= Pr min c1in (!) c1
i
h i
= 1 Pr min c1in (!) > c1
i
Y
=1 1 G1in (c1 )
i
= 1 exp n c1
X
with n= Ti ( in wi )
i

Note that n is some aggregate measure of the productivity that

country n has access to: it’s some average of the productivity
(both absolute and comparative) of each country in the world (Ti
and wi ), scaled by the trade barriers into country n (the 0in s).

From this, we can derive the probability that country i is the cheap-
est provider of a given good ! in country n. Given that we have
a continuum of goods, the law of large numbers holds, and this
probability is exactly the share of goods that are imported from i

14
 to n,

in = Pr c1in (!) = c1n (!) (19)

= Pr c1n (!) min c1jn (!)

j6=i
Z1 Y
= 1 G1kn (c1 ) dG1in (c1 )
0 k6=i
Z1 Y
= Ti ( in wi ) 1 G1kn (c1 ) c1 1 dc1
0 k
Z1
= Ti ( in wi ) exp n c1 c1 1 dc1
0

Ti ( in wi )
=
n

This share of imported good from i (and we’ll see later that this
is also the share of imports in nominal terms) depends on the
productivity of country i (scaled by the trade barriers between i
and n), relative to the productivity of all other trading partners of
n.

Bertrand competition:

We now assume that in each sector !, …rms compete Bertrand. If

…rms were identical, even with only two …rms in a sector, prices
would be equalized to marginal costs. With …rm heterogeneity,
…rms can charge some mark-up. In equilibrium, …rms either set
a price equal to the second lowest cost, i.e. a price equal to the
cost of their closest competitor, or the Dixit-Stiglitz mark-up if the
second lowest price is not binding (which may happen with some
probability). The price in country n is therefore,

pn (!) = (!) c1n (!) = min c2n (!) ; c1n (!) (20)
1
; >1
with =
1 ; 1

So we can see this paper as an extension of the Melitz model, where

we allow more than one …rm within each sector. The Dixit-Stiglitz
outcome will only only occur with some probability, when the …rst

15
and the second draw are su¢ ciently far apart. Otherwise, the
mark-up depends on the realization of the …rst and second draw
of productivity (or cost).

If country i is the cheapest provider of good ! in country n, it

means it has the lowest cost, and therefore we know that the lowest
cost for that good ! in country n must be such that,

c2n (!) = min min c1jn (!) ; c2ni (!) (21)

j6=i

From this result and the distribution of the highest and the second
highest productivities in Eq. (14), we get the joint distribution of
the lowest and the second lowest costs from country i in country
n. It is convenient to work with the complementary distributions
for a moment, Gc = 1 G,

Gcin (c1 ; c2 ) = Pr c1in (!) c1 ; c2in (!) c2 (22)

1 in wi 2 in wi
= Pr zin (!) ; zin (!)
c1 c2
h i
= 1 + Ti ( in wi ) c2 c1 exp Ti ( in wi ) c2

From this, and from Eq. (21), we get the joint distribution of the
lowest and the second lowest cost in country n, unconditional of
the origin country,

Gcn (c1 ; c2 ) = Pr c1n (!) c1 ; c2n (!) c2 (23)

2 0 1 2 0
3
for all i s, "cin (!) c2 and cin (!) c2 in all i s"
= Pr 4 "c1kn (!) c2 and c2kn (!) c2 in all k 6= i 5
or, for all i0 s,
but c1 c1in (!) < c2 and c2in (!) c2 in i"
Y
= [1 Gin (c2 ; c2 )]
i
( )
X Y
+ ([1 Gin (c1 ; c2 )] [1 Gin (c1 ; c2 )]) [1 Gkn (c2 ; c2 )]
i k6=i
Y
Ti ( in wi ) c2
= e
i
( )
X Y
Ti ( in wi ) c2 Tk ( kn wk ) c2
+ Ti ( in wi ) c2 c1 e e
i k6=i

= exp n c2 + n c2 c1 exp n c2

16
 And …nally, we recover the joint distribution of the lowest and
second lowest costs, unconditional on the country of origin,
Gn (c1 ; c2 ) = Pr c1n (!) c1 ; c2n (!) c2 (24)
= 1 Gcn (0; c2 ) Gcn (c1 ; c1 ) + Gcn (c1 ; c2 )
= 1 exp n c1 + n c1 exp n c2

Distribution of mark-ups:

Now that we have the joint distribution of the lowest and second
lowest costs in country n, we can describe the distribution of mark-
ups in country n. For all 0 s such that 1 ,
c2
Pr n (!) jc2n (!) = c2 = Pr c1n (!) c2 jc2n (!) = c(25)
2
R c2 @ 2 Gn
j
c2 = @c1 @c2 c1 ;c2
dc1
= R c2 @ 2 Gn
j
0 @c1 @c2 c1 ;c2
dc1
c2 (c2 = )
=
c2
=1
So the distribution of the mark-ups in country n, conditional on
the second lowest cost being c2 , is a Pareto distribution that does
not depend on c2 . This property is speci…c to the functional form
we assumed for the distribution of costs, and it is quite convenient.
The unconditional distribution will therefore be the same (we just
integrate that probability for all realizations of c2 , which is inte-
grating a constant over the support of c2 , and we get exactly that
same constant). This was the distribution of mark-ups in country
n conditional on the mark-up being below the Dixit-Stiglitz mark-
up. The unconditional distributions of mark-ups in country n is
then this Pareto distribution, truncated from above by m,
1 ; 1 ;
H ( ) = Pr [ n (!) ]= (26)
1 ; > = 1

We have the following properties for the distribution of mark-ups:

– Note that the distribution of mark-ups is the same in all coun-

tries, irrespective of the cost of trading with that country.
This results di¤ers radically from the prediction in Melitz
and Ottaviano (2005).

17
 – The reason for that is that while reducing trade barriers will
increase the number of potential competitors in sector ! and
therefore lower mark-ups. At the same time however, this is
exactly o¤set by the exit of domestic …rms who used to charge
the lowest mark-ups.
– Note also that the distribution of mark-ups only depends on
the heterogeneity parameters, (inverse) heterogeneity in pref-
erences, , and (inverse) heterogeneity in productivity be-
tween …rms, . A higher heterogeneity in productivity be-
tween …rms, lower , will increase the probability of high
mark-ups, as there are relatively more dispersion between
…rms (and therefore more distance between the lowest and
the second lowest cost draw, on average). If agents see goods
as more di¤erentiated, lower , …rms are more likely to charge
a high mark-up, as mark-ups are truncated at a higher point.

Measured productivity:

We now know the distribution of mark-ups in a given country. The

next question is, what is the distribution of mark-ups for a given
…rm with productivity z1 ? This mark-up is exactly the typical
measure of labor productivity that we would observe in the data:
output(z1 )
workers(z1 )
= (z1 ). This mark-up does not only depend on the
…rm’s characteristics, it depends on what the competitors of that
…rm are doing. We can derive the distribution of the mark-up of
a …rm with productivity z1 . For example, let’s look at the mark-
up of a …rm from a country i that is the cheapest provider of a
good in country n, and assume that this …rm is not charging the

18
Dixit-Stiglitz mark-up:
1
Hin ( j z1 ) = Pr in (!) j zin (!) = zn1 (!) = z1 (27)
c2 (!) in wi
= Pr 1n j c1in (!) = c1 =
cn (!) z1
2 1
= Pr c1 cn (!) c1 j cin = c1
R c1 @ 2 G(c1 ;c2 )
@c1 @c2
dc2
= Rc11 @ 2 G(c 1 ;c2 )
c1 @c1 @c2
dc2
h i
exp n c1 exp n ( c1 )
= h i
exp n ( c1 )
=1 exp n 1 c1
h i
=1 exp n 1 ( in wi ) z1

We now have to truncate that distribution to account for the fact

that with some probability, the …rm will charge the Dixit-Stiglitz
mark-up,
1
Hin ( j z1 ) = Pr in(!) j zin (!) = zn1 (!) = z1 (28)
( h i
1 exp n 1 ( w
in i ) z1 ; 1 ;
=
1 ; > = 1

The measured productivity of a …rm has the following properties:

– A plant with a higher e¢ ciency will charge a higher mark-up

on average than another …rm (its distribution of mark-ups
…rst-order stochastically dominates the other’s). This result
is similar to Melitz and Ottaviano (2005). In this model,
it comes from the fact that the distance between the best
e¢ ciency and the second best e¢ ciency increases with the
level of e¢ ciency (in a statistical sense). Therefore, the actual
measure of productivity (realized (z1 )) is (on average) an
increasing of a …rm’s intrinsic productivity.
– The easier it is for a …rm to access a given country ( ij low or
wi low), the higher the mark-up it will charge (on average).
On the other hand, the "tougher" the competition it faces
(the easier it is for its competitors to compete, i.e. the lower
n ), the lower the mark-up it charges.

19
 – The two measures of heterogeneity (heterogeneity in tastes, ,
and heterogeneity in productivity, ) a¤ect the distribution
of the mark-up of a single …rm in the same way that they
a¤ect the whole distribution of mark-ups: more heterogeneity
(lower or ) implies higher mak-ups, on average.

E¢ ciency and exporting:

It is easy to see that the model predicts that exporters will typically
be more productive than non-exporters. The model also predicts
that all exporters will also sell on their domestic market, whereas
only a fraction of domestic …rms are also exporters.
To sell domestically, a domestic producer of good ! must be more
e¢ cient than any of its foreign competitors,
wi
zi1 (!) zk1 (!) ; 8k 6= i
ki wk

To be able to export towards country n on the other hand, this

same producer must be more e¢ cient than any other foreign com-
petitor after incurring the trade barrier in ;
in wi
zi1 (!) zk1 (!) ; 8k 6= i
kn wk

From the triangular inequality that we imposed on trade barriers,

kn ki in , it is harder to export than to sell domestically,

wi in wi
zk1 (!) zk1 (!)
ki wk kn wk

Exporters also have larger domestic sales than non exporters be-
cause exporters tend to be more productive, and more productive
…rms tend to be larger: more productive …rms charge lower mark-
ups, and therefore have larger market shares.

Aggregate exports and welfare:

We have seen that prices of a given commodity ! in a given country

n do not systematically vary with the country of origin. Therefore,
the share of country n’s export from country i is directly given by
the share of goods that country n imports from country i, given in
Eq. (19),
Xin Ti ( in wi )
= in = (29)
Xn n

20
– As in Chaney (2006), or Eaton and Kortum (2002), this share
only depends on the relative trade barriers between i and n
(scaled by i’s productivity) and the trade barriers from all
countries and n.
– As in Eaton and Kortum (2002) and Chaney (2006), the sen-
sitivity of trade ‡ows to trade barriers does not depend on
the elasticity of substitution between goods, but only on the
measure of …rm heterogeneity (equivalent to the parame-
ter of the Pareto distribution in Chaney). The reason why
drops out of the exports expression is somehow similar to the
mechanism described in Chaney (2006). When trade barriers
between i and n ( in ) increase, several things happen.
First, the extensive margin of trade adjusts. Because the
trade barriers are higher, there are some goods for which
country i is not the cheapest producer anymore. How many
of these goods there are does not depend on , it only de-
pends on the distribution of productivity shocks, governed by
. How much market shares each of these exporters had prior
to losing their edge does depend on , but it does not depend
on average.
Second, there are some goods for which i is still the cheapest
provider. However, because the cost for i’s exporters has
increased, the di¤erence between the lowest cost (from i) and
the second lowest (from some other country, una¤ected by
this change in in ) has shrunk. For some fraction of goods for
which the second lowest cost was not binding, and therefore
for which the exporter from i was charging the Dixit-Stiglitz
mark-up, the second highest price is still not binding, so that
i’s exporter is still charging the Dixit-Stiglitz mark-up, and
hence increases its price. How much market share does i’s
exporter lose to other sectors, this depends on the elasticity of
substitution , as consumers substitute towards other sectors:
the bigger , the larger the loss in market share. For some
other i’s exporters, the second lowest cost becomes binding, so
that the …rm switches from the Dixit-Stiglitz mark-up to the
constrained mark-up (therefore reduces its mark-up). These
types of i’s exporters increase their price in n, but less so that
the previous category of exporters. How much they increase
their price depends on (which determins the Dixit-Stiglitz
mark-up). How much market share they lose to other sectors
depends on as well. Finally, for some of i’s exporters, the
second lowest cost was binding and is still binding, so that

21
 their price is unchanged.
Third, there are goods for which the best producer in i either
remains the (potential) second cheapest provider, or was the
second cheapest provider, but no longer is. Because the cost
has changed for those (potential) exporters, the price charged
by the cheapest provider (no matter where this guy comes
from) will change. How much market share gain that price
change induces depends on the elasticity of substitution
(the bigger , the bigger the gains in market share of these
cheapest suppliers).
However, because we have a continuum of goods, and because
from the distribution of mark-ups in Eq. (26) that the distri-
bution of prices in country n is independent of any trade bar-
riers, those impacts of changing in on the intensive margin of
trade exactly cancel out. The only margin of adjustment, on
average, is the extensive margin. This margin only depends
on the distribution of productivity shocks, driven by .
– "Absolute advantages" (captured by Ti ) increase the export
of i to any country in the world, if we do not solve for wages
in general equilibrium at least.
– "Comparative advantages" (captured by ) dampen the im-
pact of trade barriers: more dispersed comparative advan-
tages, i.e. lower, imply that trade barriers do not have
much of an impact on trade ‡ows: no matter how large trade
barriers are, there is always a fraction of goods for which i is
extremely good, so it always exports at least some of these
goods.

We can compute welfare as the inverse of the price index in country

n:
Pn 1 = 1=
n

with a constant1 . We immediately see that the cheaper it is to

import goods from the rest of the world increase welfare ( n lower,
either because trade barriers towards n are lower, or because n’s
trading partners have a better technology/lower wages). Interest-
ingly, welfare in country n does not depend on wages in country n,
and therefore does not depend on the trade barriers that potential
exporters from n face.

h i1=( 1)
1 1+ +( 1) 1+2
= 1+ :

22
Calibration and empirical exercises
BEJK go on to calibrate their model on actual …rm level data (US …rms),
to test some of the predictions of their model, and to do some simulation
exercises.

Parameters:

From Eq. (29), using bilateral trade ‡ows and aggregate output
data between the US and 47 other countries, BEJK infer trade
barriers measures, Ti ( in wi ) = n ’s for all i; n’s,

Xin Ti ( in wi )
=
Xn n

The data on bilateral trade ‡ows come from Feenstra, Lipsey and
Bower (1997). Data on aggregate output come from UNIDO (1999),
completed with data from the World Bank.

From Eaton and Kortum, they use = 8:28 and = 6 (they

experiment with other values too).

They take 500; 000 draws from joint Fréchet distributions (for the
highest and second highest productivity) for each 47 countries (we
have 500,000 sectors). From these productivity draws, they know
which …rm sells in which country, what price it charges, what its
total size and measured productivity is...

Matching the data:

They compare their simulated US economy (they also have a simu-

lated economy for all other countries, that they needed to determin
which US …rm survives, where it exports, and what price it sets),
to actual data on US …rms. These are Census data on 200,000
…rms in 1992.

They overpredict the fraction of exporters (21% in the data,

51% in their simulation). Weighting by size gets them closer to
the actual fraction.

They underpredict the dispersion of productivity between …rms.

They get the exporters’size advantage about right (overpredict

it for large ’s).

23
 They get about right the fraction of revenue from exports:
most exporters export only a small fraction of their output. They
underestimate the fraction of export-oriented …rms when they take
large ’s.

Simulations:

They go on simulating counterfactual worlds. To do so, they need

however to solve for the general equilibrium wages and prices in
these counterfactual worlds. They do so by assuming a homoge-
nous sector freely tradable with constant marginal labor produc-
tivity Wn in country n. They expand the model to account for
labor and intermediate inputs in production.

They submit their simulated economy to 3 types of shocks:

1. (i) Globalization: a reduction of trade barriers of 5%.

(ii) Autarky: prohibitive trade barriers.
(iii) Exchange rate appreciation: a 10% rise in US wages
relative to wages in other countries.

Globalization: reducing trade barriers leads to an increase of

aggregate productivity (4%). As in Melitz, part of the story is re-
allocation of labor towards more productive …rms (8% of US …rms
exit, 4.6% of gross industrial job creation, 7.3% gross industrial
job destruction), another driving force is a fall in the cost of inter-
mediates.

Autarky: a rise in trade barriers reduces aggregate productivity

by 9%, mainly because of an increase in the cost of intermediates,
but also due to the entry of ine¢ cient US …rms.

Exchange rate appreciation: this loss of competitiveness in-

creases US productivity by 4%, as plants substitute intermediate
for workers (the price of intermediate goods rises less than wages),
and as less productive US …rms exit. Those two together induce
an 18% fall in industrial employment. 24% of US exporters stop
exporting.

24