Lecture 13:

Heterogeneous Firms and

Aggregate Prices

Instructor: Thomas Chaney

Econ 357 - International Trade (Ph.D.)

We have seen previously in the class several models of international

trade with heterogenous …rms. A key features of these models, that was
strongly in the data, is the existence of an extensive margin of trade. As
trade barriers move around, the set of …rms that are able to overcome
the hustle of entering foreign markets evolves. This endogenous selection
of …rms into the export market plays a key role in explaining long term
‡uctuations in aggregate trade ‡ows.
In this lecture, we move towards the study of much shorter term
‡uctuations. We will look at the behavior of open economies along the
business cycle. This is the stu¤ of international macroeconomics, or
open economy macroeconomics. We will study these ‡uctuations with
the tools developed in the previous trade models. Bridging the gap
between international trade and international macro is a promising area
of research. Powerful tools developed in trade theory give precise micro-
foundations to these international macroeconomics models.

This is however a nascent …eld, with relatively few contributions. We

will focus mainly on two papers: Ghironi and Melitz (2005), and Atke-
son and Burstein (2004). Ghironi and Melitz extend the Melitz (2003)
model to embed it in a dynamic framework. This model gives precise
microfoundations to the Harrod-Balassa-Samuelson e¤ect. Atkeson and
Burstein also build on the Melitz model. They develop a model of im-
perfect competition that endogenously delivers pricing to market. This
allows them to describe qualitatively and quantitatively patterns of price
‡uctuations.

1
1 Ghironi and Melitz (2005)

1.1 Harrod-Balassa-Samuelson e¤ect: a theory of

the Gold rush in California
The Harrod-Balassa-Samuelson e¤ect is initially an empirical …nding:
more advanced economies (higher GDP per capita) have higher prices,
even in PPP. Balassa (1964) and Samuelson (1964) develop a simple
model of international trade with a non traded sector to explain this fact.
If the tradeable sector in a country experiences productivity growth, then
the relative price of non tradeable goods will rise, so that the aggregate
price index in this economy will rise.
A simple way to think about this e¤ect is California during the gold
rush. On January 24th, 1848, James Marshall discovered gold in Cali-
fornia. Gold is in essence a tradeable good. The trade costs associated
with shipping gold is basically negligible compared to the value of gold.
So almost by construction, the price of gold will be equalized in every
country. Finding gold is equivalent to a (massive) increase in the produc-
tivity in the tradeable sector. Such an increase in the productivity of a
sector will drive labor into that sector. This is basically what happened
in California. At that time, John Sutter, a Swiss immigrant, was in the
process of building an agricultural empire in California. Soon after the
…rst discovery of gold, most agricultural workers quit their job and went
into the mountains to …nd gold. The relative price of agricultural goods
(and any good for that matter, except gold itself) went through the roof,
and the wages earned in agriculture gradually increased. An equilibrium
would be reached where agricultural goods would be so expensive that
the relative wages in agriculture and in gold mining would eventually be
equalized. The most evident e¤ect of the technological improvement in
gold mining in California (a positive shock on the productivity of the
tradeable sector) was therefore a massive increase in prices in California.
This is in essence the Harrod-Balassa-Samuelson e¤ect.
There is an interesting anecdote that exempli…es the Harrod-Balassa-
Samuelson e¤ect, and in a sense goes far beyond. Initially, Marshall
and Sutter tried to keep their discovery of gold secret, so as to avoid a
massive and uncontrolled in‡ux of migrants into California. The man
that "made" the gold rush is Sam Brannan. Brannan …rst bought every
pick axe, pan and shovel in the region. Then only, he went through
the streets of San Francisco with a bottle full of gold dust, and shouted
about the discovery of gold in the mountains. He made $36,000 in 9
weeks, and quickly became the richest man in California.
The California story does not exactly …t the traditional models of

2
Balassa (1964) and Samuelson (1964), since there was also a massive
migration of workers into California. This migration of workers arguably
dampened the e¤ect of a rise in prices. However, the productivity in gold
mining in California was so high that the e¤ect still persisted.
Ironically, the initial dream of John Sutter, that came to a quick end
with the discovery of gold, was partially similar to the gold rush. Sutter
had bet on the high productivity of agriculture in California, much higher
than in most of the world. Since agricultural goods are, to some extent,
a tradeable good, this high productivity must have initially increased
the prices in California. For some reason, California keeps …nding new
tradeable sectors where the Californian productivity is much higher than
anywhere else (agriculture, gold, movies, information technologies, sandy
beaches), so that Californian prices basically have stayed pretty high
ever since the arrival of John Sutton. Palo Alto, Orange County or
San Francisco are arguably still among the places where life is most
expensive.
The French writer Blaise Cendrars wrote a marvelous novel in 1925,
L’Or (Sutter’s Gold, 1926, for the English translation), on the tragic
character of John Sutter, the Swiss immigrant who built an empire which
was destroyed overnight when it was discovered that this empire stood
on a pile of gold.

1.2 Endogenous non-tradedness

The Balassa-Samuelson model however takes a whole bunch of para-
meters as exogenously given. The non traded sector is non traded by
de…nition. Trade barriers in the tradeable sector are zero, so that every
tradeable goods are indeed traded. Moreover, this Balassa-Samuelson
e¤ect relied on the assumption that productivity shocks only a¤ected
the non tradeable sector. An economy wide technological shock, in their
model, would have no e¤ect on relative prices in di¤erent countries.
Ghironi and Melitz (2005) on the other hand assume all goods are
potentially tradeable. However, due to the existence of trade barriers,
not all goods, only an endogenously determined set of goods will be
actually traded. The endogenous determination of the tradedness of
goods is a key contribution of this paper.

Main assumptions (same as in Melitz (2003) basically):

Fixed costs of exporting.

Heterogeneity in productivity between …rms.
Endogenous entry into both domestic and export markets.

3
 Unlike the initial Melitz (2003) model, this model is fully dynamic
(not only steady state dynamics).

Main …nding, response to a positive productivity shock:

Assume that a positive productivity shock a¤ects all …rms in one

country. Unlike the initial Balassa-Samuelson e¤ect, this produc-
tivity shock does not a¤ect only the tradeable sector. Not that em-
pirically, there is little and unconvincing evidence that the di¤er-
ential in productivity between the non tradeable and the tradeable
sector are correlated with the price level. The empirical evidence
is really that more productive economies (more productive across
all sectors) have higher prices. See Rogo¤ (2006) for empirical
evidence.

The domestic real exchange rate will appreciate for 3 reasons:

1. The home market being more productive, more …rms want

to locate here (the traditional "home market e¤ect" we saw
in the Krugman model). If wages relative wages did not ad-
just, all …rms would locate at home. To keep foreign labor
employed, domestic wages have to rise. This rise in domestic
wages induces an increase in the price of home non traded
goods relative to foreign non traded goods. That is the …rst
channel through which the domestic real exchange rate ap-
preciates.
2. Because domestic labor is more expensive, domestic …rms are
less pro…table in the export market (the domestic terms of
trade deteriorate). Hence, the productivity cuto¤ for export-
ing goes up, only the most productive …rms can keep export-
ing. Symmetrically, it becomes easier for foreign …rms to ex-
port, and more foreign …rms start exporting. Less productive
…rms charge a higher price than more productive …rms. So
domestic imports become more expensive, whereas domestic
exports become cheaper. This is the second channel through
which the domestic real exchange rate appreciates.
3. Finally, as more …rms enter the domestic market, there are
more domestic varieties available for domestic consumers. This
induces domestic consumers to switch their expenditure to-
wards home produced goods. This is the third channel through
which the domestic real exchange rate appreciates.

4
 Note that in the absence of endogenous entry (due here to the ex-
istence of …xed costs of entry), none of these channels would exist.
Absent endogenous entry, home (e¤ective) labor would actually
depreciate. If the number of …rms were …xed, the higher produc-
tivity at home would lead to more demand for foreign goods, hence
an excess demand for (e¤ective) foreign labor, and therefore an ap-
preciation of foreign labor.

Note also that in this model, strictly speaking, all goods are po-
tentially tradeable. In equilibrium, only a subset of goods will
be actually traded though. So this model is able to explain why
the price of tradeable goods increases following an increase in the
productivity in the tradeable sector, which we observe in the data
(empirically, it is not only the price of non tradeable goods that
increases following an increase in productivity). If one were to add
an exogenously non tradeable sector in this economy, the prices in
that sector would obviously increase (this is the traditional Bal-
assa/Samuelson model). This would complement the increase in
the price of tradeable goods described here.

Additional assumptions to the Melitz (2003) model:

Ghironi and Melitz consider an economy similar to the one consid-

ered in Melitz (2003). They make however 2 important changes.

The two economies we consider are no longer symmetric. This in-

troduces possible deviations of the relative wages in the two coun-
tries from 1. This ‡uctuations in the relative wage following asym-
metric shocks will be a key feature of the model. Formally, this
adds a layer of complexity. One now has to solve for all the vari-
ables in the Melitz model, plus the relative wage. Since Ghironi
and Melitz only look at either log-linear approximations around the
steady state, or numerical solutions for the transitional dynamics,
this is not a fundamental di¢ culty.

An important technical assumption that is added to the original

Melitz model is that there are no …xed overhead costs for domestic
production. This means that any …rm, no matter how unproduc-
tive it is, will always produce domestically, and will only die of
exogenous death. There is no endogenous exit from the domestic
market. The only endogenous exit happens on the export market.
Let me brie‡y explain why this assumption matters (and greatly
simpli…es the authors’life).

5
 – With …xed overhead cost of for domestic production, not all
domestic …rms survive. Assume that following a reduction in
overall productivity, the threshold for survival (domestically)
will …rst jump, and then gradually go down, to a higher level
than before the productivity shock.
– At each point along the transition, the distribution of surviv-
ing …rms is not simply a truncation of the underlying produc-
tivity distribution at birth. The lower tail of the distribution
of survivor has been shaped by the entire history of entry
since the initial productivity shock.
– When an entrepreneur contemplate whether or not to enter,
she has to form expectations of her future pro…ts. The pro…ts
she would earn at each point in time depends on whom she
may have to compete against. This depends both on who will
enter the market in the future (forward looking expectation),
but also on the entire history of who has entered the market
(which determines the distribution of …rms at each point in
time). So when making an entry decision, an entrepreneur is
both looking forward, and looking backward.
– Even numerically, this makes this problem intractable (or
computationally to heavy). The theorist has to guess not
only a path for future variables (prices, so simple numbers),
but also a path for the expectations of each cohort, which
itself depends (in a complicated fashion) on the guess for the
path of prices. Recursive methods do not work here.

The model:
Let us go rather quickly over the model. The model basically fol-
lows Melitz (2003) very closely, using the same trick of de…ning special
averages which simplify the analysis greatly.

Preferences: Consumers maximize their intertemporal utility,

"1 #
X 1
s t Ct
Ut = Et
s=t
1
0 1 1
Z
1
with Ct = @ ct (!) d! A
!2 t

is the inverse of intertemporal elasticity of substitution, and is

the elasticity of substitution between goods. The set t of goods

6
actually consumed (accessible) in period t will be determined in
equilibrium.

Demand functions: We can de…ne a price index, and demand

functions for each good !,

pt (!)
ct (!) = Ct
Pt
0 111
Z
with Pt = @ pt (!)1 d! A
!2 t

Size and wages: L (L ) is the number of domestic (foreign)

workers. W (W ) is the domestic (foreign) nominal wage, and
wt W Pt
t
the real wage. Note that there is no cash in this economy.
We could normalize the wage in one country to 1. It turns out to
be more convenient to normalize price indices (rather than pick up
one numeraire), and solve for all nominal variables.

Productivity heterogeneity: Firms draw their productivity z

from a distribution G (z) on [zmin ; 1). The productivity of all
units of labor is scaled by Zt (Zt ), the aggregate productivity at
home (abroad).

Entry cost and death: Potential entrants have to pay an entry

cost fE;t units of e¤ective labor (wt fE;t =Zt in units of the home
consumption good). All …rms, irrespective of their productivity,
or their age, are subject to a death shock with probability .

Export costs: exporters have to overcome a variable cost ( t

iceberg transportation cost), and fX;t , a …xed cost of exporting (in
units of e¤ective labor).

Prices: Price setting monopolist have the following simple pricing

rule: a …rm with productivity z sets a price pD;t (z) at home, and
pX;t abroad, with,
(
pD;t (z)
D; t Pt
= 1 Zwttz
pX;t
X;t P
= Qt 1 t D;t
t

P
with Qt "t Ptt the real exchange rate, and "t the nominal exchange
rate.

7
Pro…ts: A …rm with productivity z earn pro…ts, dt (z), that it
redistributes to the owners of the …rm as dividends. The total
pro…ts come from two potential sources: pro…ts on the domestic
market dD;t , and pro…ts from exporting, dX;t (z),
8
>
> dt (z) = dD;t (z) + dX;t (z)
>
< dD;t = 1 1
(D;t (z) Ct
Qt 1 wt fX;t
>
> (z) Ct if …rm exports,
> X;t
: dX;t (z) = 0 otherwise Zt

Export decision: All …rms that earn positive pro…ts export.

That is all …rms with a productivity above the productivity cuto¤
zX;t export, with,

zX;t = inf fz : dX;t (z) > 0g export cuto¤

Special averages: As in Melitz (2003), we can de…ne special

averages,
8 0 1 1 11
>
> Z
>
>
>
>
> z~D @ z 1 dG (z)A
>
< zmin
0 1 11
>
> Z1
>
> B C
>
> z~X;t @ 1 G 1z z 1 dG (z)A
>
> ( X;t )
:
zX;t

So that the economy behaves as if there were ND;t domestic pro-

ducers with productivity z~D who charge a price p~D;t pD;t (~
zD ),
NX;t domestic exporters with productivity z~X;t who charge a price
zX;t ). The price index is therefore,
p~X;t pX;t (~
1
1 1 1
Pt = ND;t (~
pD;t ) + NX;t p~X;t

and the average total dividends are,

d~t dD;t (~
zD ) + [1 G (zX;t )] dX;t (~
zX;t )

Free entry: Households when considering whether or not to start

a new …rm compare the net present value of future pro…ts,
" 1 #
X Cs
v~t = Et [ (1 )]s t d~t
s=t+1
C t

8
and compare it to the cost of entry, wt fE;t =Zt . The free entry
condition implies that, provided that at time t, some …rms enter,
then,
v~t = wt fE;t =Zt
The total number of home …rms at time t is the sum of the …rms
that have survived the exogenous death shock from the previous
period, and the new entrants,

NH;t ND;t + NE;t

with ND;t = (1 ) NH;t 1

Financial markets: In Melitz (2003), we saw that the ownership

structure of the economy was irrelevant in the steady state. Here,
since we want to look at the transitional dynamics of our economy,
we must de…ne precisely the ownership structure.
Assume that all workers in each economy invest in a mutual fund
that owns all the domestic …rms. We assume …nancial autarky
(Ghironi and Melitz consider later the case of international …nan-
cial markets, and …nd similar qualitative results for the transitional
dynamics). From a "no arbitrage" condition, the price of a claim
to the future pro…ts stream of the mutual fund of NH;t …rms must
equal to the average nominal price of claims to future pro…ts of
home …rms, Pt v~t .
In addition, we introduce a riskless bond that delivers a net in-
terest rate rt (in zero net supply, but it will allow us to solve for
the interest rate). The riskless bond will be in zero net supply
(since all consumer in a country are identical, and diversi…es its
risk perfectly). The purpose of the riskless bond is only to de…ne
an equilibrium interest rate that will simplify the solution of the
dynamic equilibrium.
We will assume to start with that each country is in …nancial au-
tarky (the general case is developed in the appendix of Ghironi
and Melitz’s paper). In equilibrium, because we have moved away
from a symmetric case, the interest rate may di¤er between the two
countries. Among other, the uncovered interest rate parity may not
hold in equilibrium. This means that consumers in a country, if
they were allowed to, would be willing to borrow or lend to con-
sumers in the other country. We prevent this by assumption to
simplify the equilibrium, but Ghironi and Melitz prove that open-
ing up international …nancial markets would not change any of the
qualitative predictions of the model.

9
Budget constraint and Euler equations: Consumers earn in-
come from their labor (inelastically supplied), from the shares they
own in the mutual fund, and from the bonds they hold. They spend
their income on consumption, buy shares in the mutual fund, and
buy riskless bonds,

Bt+1 + v~t NH;t xt+1 + Ct = (1 + rt ) Bt + d~t + v~t ND;t xt + wt L

with xt the fraction of all shares in the mutual fund held by house-
holds. The …rst order conditions for consumption give the following
Euler equations,

Ct = (1 ) Et Ct+1
" #
Ct+1
and v~t = (1 ) Et v~t+1 + d~t+1
Ct

Iterating the second equation forward gives the no arbitrage pricing

of a share in the mutual fund.

Equilibrium conditions: In equilibrium, the riskless bond is in

zero net supply in each period, and all the shares in the mutual
fund are held by domestic consumers, so that we can rewrite the
budget constraint,

Bt+1 = Bt = 0
xt+1 = xt = 1
) Ct = wt L + ND;t d~t NE;t v~t

Note that …nancial autarky implies trade balance. This is where

the arbitrary decision of denominating all prices relative to the
domestic price index comes in handy. It turns out to be simpler to
solve the model with that convention than to de…ne a numeraire
good (labor in one country for instance), and impose trade balance.
This is a purely technical note, but an important one you should
think about if you were to write your own international macro
model...

Solving for the equilibrium: Ghironi and Melitz list all the
equilibrium conditions, and all the unknown variables. This is in
essence a dynamic problem, agents are forward looking, and one
ought to solve for the entire path of future prices and quantities
along the equilibrium path (whether stationary steady state or
not). The stationary equilibrium is rather "simple" to solve for, it

10
is merely an extension of Melitz (2003) with asymmetric countries.
The non stationary equilibrium has to be solved numerically. All
numerical simulations are shown in the paper.

Harrod-Balassa-Samuelson e¤ect: In the data, one doesn’t

have access to ideal price indices, statisticians only report (weighted)
average prices. In this model, the counterpart of the Consumption
Price Index is simply the weighted average price of all domestically
consumed goods (both domestically produced goods, and imported
1=(1 ) ~
goods): Pt = Nt Pt , with P~t the CPI. So the "real exchange
rate" observed in the data is the ratio of these domestic CPI’s,
normalized by the nominal exchange rate,
~
~ t = "t Pt "real exchange rate"
Q
P~t
Ghironi and Melitz log-linearize the economy in the vicinity of the
steady state equilibrium to look at the impact of an increase in
overall productivity at home (Zt goes up, and Zt unchanged). In
log deviations, one gets the following expression for movements in
the "real exchange rate",

~ 1t
Q = (2sD 1) TOLt
(1 sD ) ~zX;t ~zX;t (tt tt )
1 ND
+ sD ND;t NX;t ND;t NX;t
1 ND + NX

with T OL "t (Wt =Zt ) (Wt =Zt ) the "terms of labor", and sD;t
1
ND;t ~D;t the share of spending on domestic goods.
There are 3 e¤ects through which an increase in productivity at
home will lead to an appreciation of the real exchange rate.

1. Because domestic …rms are more productive, and due to the

presence of trade barriers, a traditional "home market e¤ect"
comes into play. Domestic …rms are more productive, so in
the absence of any adjustment, there would only be entry of
new …rms at home, and non abroad. To keep foreign labor
employed, relative wages will adjust. Wages per e¤ective units
of labor at home will go up relative to foreign wages. This is
an appreciation in the "terms of labor".
This induces the price of domestic goods (in units of wage
per e¤ective labor) to go up. Because of the presence of trade

11
 barriers ,consumers spend more income on domestically pro-
duced goods than on foreign goods. The prices of non traded
goods at home relative to abroad go up, so that the real ex-
change rate appreciates.
Note that up to here, we have not looked at the endoge-
nous change in the set of actually traded good. Unlike in
the original Balassa-Samuelson model, there is however an
appreciation of the real exchange rate despite the fact that
productivity of both traded and non traded goods went up.
In the Balassa-Samuelson model, if the productivity of both
sectors goes up, there will be no e¤ect on the real exchange
rate. The di¤erence here is that we have assumed free en-
try of …rms. Because …rms are allowed to enter, an increase
in domestic productivity will put some upward pressure on
domestic wages, and because some goods are non traded in
equilibrium, domestic prices go up.
2. This increase in the domestic wages will also change endoge-
nously the set of domestic exporters. Because domestic wages
have gone up, it becomes harder for domestic …rms to export,
and symmetrically it becomes easier for foreign …rms to ex-
port. The productivity cuto¤ for exports at home goes up,
the cuto¤ abroad goes down. Hence, domestic consumers now
consume on average more expensive imports (they start im-
porting expensive goods from low productivity new foreign
exporters), whereas foreign consumers now consume on av-
erage cheaper imports (they stop importing expensive goods
from low productivity domestic …rms that stop exporting).
Note that this is not a statement about the total volume of
exports: there are more …rms at home, so potentially home
exports more. It is a statement about the composition of ex-
porters. Among those (more numerous) domestic …rms, only
the most productive ones export. This increase in the average
price of imports, and reduction in the average price of exports
will further increase the real exchange rate.
This e¤ect was altogether absent in the Balassa-Samuelson
model, where the tradedness of goods was exogenously pos-
tulated. In this model with endogenous tradedness of goods,
changes in
3. The last e¤ect is a corollary to the increase in the number of
domestic …rms. Because consumers value variety, as the va-
riety of goods available at home relative to the variety avail-

12
 able abroad increases, domestic consumers will switch their
expenditure towards domestic goods. Because domestic va-
rieties are more expensive (due to the increased labor cost),
this further increases the price of the consumption basket of
domestic consumers. This is that last e¤ect through which
the real exchange rate appreciates.

Note that the endogenous entry of …rms is key to derive those pre-
dictions. Absent entry of domestic …rms which drive the domestic
wages up, a productivity gain at home would actually depreciate
e¤ective domestic labor: workers become more e¢ cient at produc-
ing goods, they ‡ood the world market with domestic goods, this
drives down the relative wage of domestic goods (provided that we
are not considering a small open economy that would not have any
impact on world prices). In units of e¤ective labor, wages would
go down.

Ghironi and Melitz also make the important remark that the be-
havior of "real exchange rates" does not re‡ect movements in rel-
ative ideal price indices. If one were to take into account the
impact of variety on ideal price indices, the increased availability
of domestic varieties at home would unambiguously dominate the
increase in average prices, so that the domestic ideal price index
would decrease relative to the foreign one.

1.3 Welfare di¤erentials and labor mobility: Krug-

man’s economic geography
A key …nding of models with trade barriers is that welfare may di¤er
between countries (ideal price indices are di¤erent between countries).
In all those models, we have always assumed that labor is immobile
across country. That means that even if welfare di¤ers across countries,
we rule out the possibility of workers moving from the low welfare to the
high welfare country.
This is only an approximation of real economies. Labor indeed is
far less mobile between countries than goods (both …nal goods and in-
termediate goods) or capital. However, one may want to think about
what happens if we were to allow labor to move across countries (or
between di¤erent locations within a country for that matter). This is
the starting point of the literature on economic geography, pioneered by
Paul Krugman in the early 1990’s (Krugman 1991). The main …nding
of this literature is that once we allow for factors of production to move

13
over space, we will typically observe catastrophic agglomeration. Fur-
thermore, history matters, in the sense that agglomerations of workers
are typically self-sustainable, so that historical accidents can explain the
observed patterns of the geographic distribution of production facilities.
I will not go into the details of these models, but only sketch Krug-
man’s argument for agglomeration forces, developed among others in
Krugman 1991. Consider a simple model with two potential locations,
and with some initial distribution of workers between the two locations.
Those two locations are separated by trade barriers. In the location
with more workers, more goods are produced. Because of the existence
of trade barriers, welfare will be higher in the location with more goods.
That means that real wages are higher in the location with more work-
ers. Absent any force that keeps the workers in the unfavored location,
workers will emigrate towards the better location. As more and more
workers move towards the crowded location, real wage di¤erentials keep
increasing, so that workers keep moving. Eventually, we converge to a
"core-periphery" system, where all workers are in the core, and no one is
left in the periphery. One interesting pattern is that if initially the two
locations are exactly identical, any small shock will trigger a catastrophic
agglomeration.
Since that pioneering paper, a whole literature has developed that
tries to explain the patterns of specialization of countries/region, and
more generally the patterns of the geographic distribution of production
across space. The key insight of this literature is to take into account
the fact that in the presence of trade barriers, real wages will generically
di¤er across country, which gives rise to some incentive for workers to
change location.

2 Atkeson and Burstein (2005)

2.1 The PPP puzzle

Obstfeld and Rogo¤ (2001) list 6 major puzzles in international trade.
In their paper, they try to o¤er a simple integrated explanation for all
these puzzles. They argue that the presence of trade barriers in trade in
goods may explain a big chunk of these so called puzzles. I won’t talk
about their explanation here. One of the major puzzles in the so called
"PPP puzzle". TO BE CONTINUED...

14
2.2 Stylized facts about international relative prices
Beyond the PPP puzzle, Atkeson and Burstein (2005) list some
stylized facts of international aggregate prices time series. Those
stylized facts are not trivially understood with a simple model of
open economies. They develop a model with endogenous pricing
to market to explain qualitatively those facts. They then go one
step further, calibrate their model and try to match quantitatively
those facts. They do a pretty good job at matching those facts
quantitatively, and in addition are able to tell which assumption
crucially matters for getting each fact right.

The main two stylized facts (in addition to departures from even
relative PPP) that Atkeson and Burstein focus on are the following:
First, the terns if trade in the manufacturing sector are less volatile
than the relative prices of manufacturing goods.
Second, the ‡uctuations of the relative prices of tradeable con-
sumption goods are almost as volatile as the ‡uctuations of the
relative prices of all consumption goods.

Let’s go in more detail over each stylized fact, and why they may be
surprising, or not trivial:

1. The …rst …nding is related to the relative price of manufacturing

goods (tradeable goods in essence). The manufacturing price in
the data corresponds to the production price index (P P I). We
can simply decompose the relative manufacturing prices between
two country into 3 components,

PPI EP I PPI IP I
=
eP P I IP I EP I eP P I
| {z } | {z } | {z } | {z }
relative prices manuf. manuf. over import over foreign
of manuf. TOT export prices manuf. prices

The …rst component, EP I

IP I
, corresponds to the terms of trade in
manufacturing. That is the ratio of the price of exported man-
ufacturing goods to imported manufacturing goods. The second
PPI
term, EP I
, corresponds to the relative price of domestically pro-
duced goods sold at home, versus sold abroad. The last term,
IP I
eP P I
, corresponds to the relative price of goods produced abroad,
those imported, and those sold abroad.

15
 Atkeson and Burstein …nd that the relative price of manufacturing
goods is much more volatile than the manufacturing terms of trade.
By construction, it must therefore mean that the relative price of
domestically produced goods exported and sold domestically also
move, and/or that the relative price of foreign produced goods
imported and sold abroad move as well.
Atkeson and Burstein take this fact as an indirect evidence of "pric-
ing to market". As shocks hit the home and the foreign economy,
both domestic and foreign exporters adjust the price they charge
PPI
abroad relative to the price they charge at home, so that both EP I
IP I
and eP P I move.
Pricing to market: Pricing to market corresponds to the fact
that exporters set di¤erent prices for the same good sold at home
or abroad, and that they adjust the relative price that they charge
in each market depending on the condition on each market. Dorn-
busch (1987) and Krugman (1987) developed early models of pric-
ing to market.
Obstfeld and Rogo¤ (2001) describe 2 extreme scenarios for the
pricing strategy of …rms. If exporters set the nominal price of
their exports …xed in the domestic currency (that is an price equal
or at least proportional to the price they set domestically), then
PPI
neither EP I
nor ePIPPII will move, and the manufacturing terms
of trade will move exactly one-for-one with the relative prices of
manufacturing goods. The other extreme corresponds to a case
where exporters set the nominal prices of their exports …xed in the
PPI
foreign currency. In that case, then both EP I
and ePIPPII will move
one-for-one with the relative prices in manufacturing. In that case,
the terms of trade will move one-for-one with the relative prices of
manufacturing, but in the exact opposite direction. The reality is
somewhere in between those two extreme cases.
It is not trivial to derive a model that would have such character-
istics (the simple Krugman model of trade wouldn’t for instance).
However, composition e¤ects in the Melitz (2003) model would give
similar predictions at the aggregate, and so would the Dornbusch,
Fischer and Samuelson model with trade barriers (maybe, I’m not
sure though...).

2. The second …nding is related to the movements of the tradeable

part of consumption goods prices, and the overall consumption
goods prices (both tradeable and non tradeable). The consump-
tion based real exchange rate is the ratio of the consumption price

16
 indices at home and abroad (normalized for the nominal interest
rate). If we call the CPI at home P , and abroad P , and if we call
the consumption price index of only tradeable goods P T at home
and P T abroad, we can decompose the consumption based real
exchange rate into 2 components,

P PT P=P T
RER =
eP eP T P =P T
= RERT RERN

Atkeson and Burstein …nd that relative prices of the tradeable part
of consumption goods is almost as volatile as the relative prices of
overall consumption goods. This is true despite the fact that more
than half of consumption goods are non tradeable.
The fact that the prices of tradeable goods moves as much as the
prices of non tradeable goods seems to suggest that at the ag-
gregate level, trade arbitrages only play a minimal role. If trade
arbitrages were large, the relative prices of tradeable goods would
not move as much. They may still move if there are some trade
barriers that prevent full arbitrage of price di¤erentials, but unless
trade barriers are large, they shouldn’t move too much. In the
end, the theoretical answer put forward by Atkeson and Burstein
to that "puzzle" is indeed that trade barriers are large, so that
for most tradeable goods, arbitraging price di¤erentials between
countries is not worth it.
Note however that reality is micro and not macro. For some goods,
trade arbitrages do take place (to some extent). But since most
goods are not actually traded, those arbitrages do not have a large
impact.

2.3 A model of endogenous "pricing to market"

Atkeson and Burstein develop a very neat and simple partial equi-
librium model to explain "pricing to market". Their model is based
on the assumption of imperfect competition. They keep a simple
CES preferences framework, and add a very simple extension to
get endogenous mark-ups.

Aggregate consumption: Atkeson and Burstein assume exoge-

nously that there are two sectors, a tradeable sector, and a non
tradeable sector. Consumers have Cobb-Douglas preferences over
these two sectors. If consumers in country i consume a composite

17
of tradeable goods cTi , and a composite of non tradeable goods cN
i ,
they derive a utility ci ,
1
ci = cTi cN
i
1
Pi = PiT PiN

where Pi is the ideal aggregate price index for consumption.

Aggregate tradeable and non tradeable consumption: Each

sector (tradeable and non tradeable) consist of a continuum of
di¤erentiated sectors, with a constant elasticity of substitution .
Looking at the non tradeable sector, the composite consumption
aggregate, the corresponding price index for non tradeable goods,
and the demand for goods of each sector are given by,
Z 1 1 1

cN
i = yijN dj
0
Z 1 1
1
1
PiN = PijN dj
0
!
yijN PijN
=
cN
i PiN

Aggregate consumption for each sector: Each sector itself is

composed of di¤erentiated goods. The important assumption here
is that there is only an exogenously …xed …nite (small) number
of K di¤erentiated goods. The composite consumption bundle of
goods in a given sector is a CES aggregate of those K goods, with
a constant elasticity of substitution . The composite index, the
corresponding price index, and the demand for each good are given
by,
!
X
K
1
1

yijN = N
qijk
k=1
!11
X
K
1
N N
Pijk = Pijk
k=1
!
N N
qijk Pijk
=
yijN PijN

18
Assumption:
1< <
This means that goods are more substitutable within sectors than
they are between sectors (and then, the elasticity of substitution
between tradeable and non tradeable goods is simply equal to 1).

Optimal pricing strategy, Bertrand versus Cournot com-

petition: Atkeson and Burstein assume that within each sector,
there is imperfect competition. Within each sector, …rms either
compete Bertrand or Cournot. That means that …rms simulta-
neously set prices (Bertrand), or quantities (Cournot), taking as
given the strategy of the other …rms in the sector. However, be-
cause there is only a …nite (and small) number of …rms within each
sector, …rms will take into account the impact they have on the ag-
gregate prices in the sector. Therefore, we depart from the simpler
Dixit-Stiglitz case with CES preferences and monopolistic compe-
tition that delivered constant mark-up for all …rms. Here, …rms
are going to charge a mark-up over marginal cost that depends on
their market share in the sector,

N " (s) W
Pijk (s) = N
" (s) 1 zijk
"Bertrand (s) = s + (1 s)
with 1
"Cournot (s) = (s 1 + (1 s) 1
)
1
!1
N N N N
Pijk qijk Pijk Pijk
and sN
ijk = PK = PK 1 =
N N
l=1 Pijl qijl
N PijN
l=1 Pijl

Atkeson and Burstein very cleverly use the properties of the CES
preferences. It turns out that a …rm with a market share s will
charge a constant mark-up over its marginal cost. The mark-up
that this …rm charges depends on some theoretical elasticity of
substitution. If …rms compete Bertrand, this elasticity is exactly
the demand elasticity that this …rm faces (which indeed depends
on its market share). If …rms compete Cournot, it’s a similar
elasticity.

Small versus large …rms: in the extreme case of a single …rm in

the sector, this …rm only competes against other …rms from other
sectors. Since there is a continuum of sectors, the price charged
by this one individual …rm will not a¤ect aggregate prices, and
therefore this …rm will charge a constant mark-up over marginal

19
cost. This mark-up depends on how substitutable goods between
industries are. It is simply the Dixit-Stiglitz mark-up, 1 .
In the other extreme case of in…nitely many …rms within a given
sector on the other hand, each individual …rm does not a¤ect the
price index in that sector. It will therefore charge a constant mark-
up over marginal cost. This mark-up depends on how substitutable
the goods are within this sector. It will therefore charge the Dixit-
Stliglitz mark-up 1 .
In the intermediate case of a …rm with only some market share,
the …rm will have to take into account the impact it has on aggre-
gate prices. It will charge a mark-up in between the two extreme
cases. A larger …rm takes into account the impact it has on sec-
toral prices. It mainly has to compete against other sectors, and
therefore charges a mark-up close to 1 . A smaller …rm on the
other hand mainly has to compete against other …rms in the sec-
tor. The fraction of income that consumers spend on goods from
that sector depends on the aggregate price index of goods in that
sector. A small …rm has only a negligible impact on this aggregate
price. Taking the prices set by its competitors as given, it there-
fore takes as given the share of income consumers spend on goods
from that sector. The only impact of changing the price of its
own variety will be that the …rm loses market shares against other
…rms in the same sector, not that the entire sector loses market
shares against other sectors in the economy. Since goods are more
substitutable within the sector than between sectors, it will charge
a lower mark-up, close to 1 .
The very cute …nding of Atkeson and Burstein is that the mark-up
charged by a …rm exactly corresponds to the mark-up a …rm would
charge if it faced a demand elasticity equal to a weighted average
of the within sector and the between sector elasticity.
Atkeson and Burstein versus BEJK: BEJK considered an-
other case of imperfect competition. Their …nding was a much
cruder strategy for the …rm: either competition from other …rms
in the sector binds, and a …rm charges a cost equal to the marginal
cost of the second lowest cost in the sector; or the competition of
…rms within the sector doesn’t bind, and the …rm charges a mark-
up equal to the Dixit-Stiglitz mark-up. In Atkeson and Burstein,
we have an intermediate case where both the competition of …rms
within the sector and of …rms in other sectors always matters. It’s
not one or the other. Whether the competition of …rms within the
sector of …rms in other sectors matters more depends on the size

20
of the …rm. Large …rms care more about the competition from
other sectors, small …rms from the competition within the sector.
Formally, BEJK simply corresponds to the extreme case where
= +1.

Trade: In the tradeable sector, things are almost equivalent. How-

ever, there are costs associated with trading:

a …xed cost of trade F

an iceberg cost of trade D

As in Melitz (2003), potentially only a subset of …rms will export.

However, in this case with a …nite number of …rms, and a more
elaborate market structure, there are potentially many Nash equi-
libria with entry of exporters. Atkeson and Burstein restrict their
analysis to equilibria such that only the most productive …rms
export (they rule out the possibility that a …rm with a given pro-
ductivity exports, but some more productive …rms do not).

Numerical solution: At this stage, they stop working out formal

solutions to the model, and they only compute numerical solutions.
They use a simple iterative procedure to sequentially solve for the
entry in each sector. To do so, they solve for the equilibrium prices
if only the most productive foreign …rm exports. They check that
under those prices, the net pro…ts from exporting for this …rm are
positive. If not, no …rm enters, and the procedure stops. If yes,
they then compute the equilibrium prices if both the most produc-
tive and the second most productive …rm enter. If at those prices,
the second most productive …rm doesn’t make positive pro…ts from
exporting, only the most productive …rm enters, and the procedure
stops. Otherwise, they consider the entry of the 3rd most produc-
tive …rm,..., and so on potentially until the K th most productive
…rm.

Pricing to market and incomplete pass through: The main

…nding of Atkeson and Burstein is that …rms will only pass through
in prices part of the shocks that may hit them (cost shocks), and
that …rms endogenously price to market. Pricing to market means
that …rms will charge di¤erent prices in di¤erent markets, and that
they will when hit by shocks, they will change di¤erently the price
they charge in di¤erent markets.

21
Incomplete pass through: The fact that …rms pass on to foreign
consumers only part of their productivity shocks (or exchange rate
shocks) comes from the endogenous mark-ups charged by …rms.
If …rms in a country are hit by a negative productivity shock (in
partial equilibrium, this is equivalent to an appreciation of the
exchange rate), they will increase their prices by less than their
productivity shock. The reason is that …rms adjust their mark-up
in response to a negative productivity shock. When a …rm is hit
by a negative productivity shock, it faces higher per unit costs.
In a pure Dixit-Stiglitz world, such a …rm would raise its price
one for one with the increase in the unit cost of production. In
this setting with endogenous mark-ups though, as the …rm loses
some market share, it will try to alleviate the shock by reducing
its mark-up. This is because it takes into account the fact that
by increasing the price it sets, not only does it lose market shares
against other competitors in the same sector, but it also loses mar-
ket shares against …rms in other sectors. Taking that into account,
the price set abroad increases by less than the increase in the cost
of production. The …rm bears part of the shock by reducing its
mark-up.
So after an appreciation of a country’s exchange rate, exporters will
not increase the price they charge abroad by as much as the in-
crease in the exchange rate. In order to protect their market share
abroad, they will reduce their mark-up, so that prices increase by
less than the exchange rate.

Pricing to market: Pricing to market corresponds to the fact

that when hit by a shock, a …rm will change the price it sets in
di¤erent markets di¤erently. In the context of this model, the main
prediction is that when hit by a negative productivity shock (or a
nominal exchange rate shock in this partial equilibrium model), an
exporting …rm will reduce the price it sets in the foreign market
more than it reduces the price it sets in the domestic market.
The reason is the following. Because of the existence of both …xed
costs, only a subset of …rms actually export. Note that if the trade
barriers are low enough though, all …rms will export, and we lose
pricing to market. This means that when a country is hit by a
negative productivity shock (equivalent here to an appreciation
of the exchange rate), an exporter will lose more competitiveness
abroad than it does at home. This is because abroad, it competes
against all foreign …rms, and only against a few other domestic
exporters. So only a small (less than half on average) share of

22
competitors are hit by the same shock. At home on the other
hand, all domestic …rms are hit by the same negative productivity
shock, but only a small share of foreign exporters are not.
Because domestic …rms are hit by this negative productivity shock,
but foreign …rms aren’t, exporters will lose some market shares
both at home and abroad, and in response to that will reduce the
mark-up they charge both at home and abroad. But because only a
subset of …rms actually export, an exporter will lose more market
share abroad than it does at home, and it will therefore reduce
the mark-up it charges abroad more than it reduces the mark-up
it charges domestically. This prediction of the model corresponds
to the (aggregate) fact that over the business cycle, the price of
exports relative to goods sold domestically moves up and down
PPI
( EP I
moves).
Symmetrically, when the domestic …rms are hit by a negative pro-
ductivity shock, foreign exporters will increase their mark-up for
exports more than for goods sold domestically ( ePIPPII moves).

Composition e¤ects, in Atkeson and Burstein and in Melitz

(2003): One important note. The model developed by Atkeson
and Burstein tries to explain the ‡uctuations of mark-ups at home
and abroad for individual …rms. On top of these …rm level ad-
justments of mark-ups, there will also be a composition e¤ect,
that would be present in the initial Melitz model too. In a par-
tial equilibrium view of the Melitz model (which is in essence the
Atkeson and Burstein model, if one forgets about the endogenous
mark-ups), when domestic …rms are hit by a negative productivity
shock, some exporters, because they lose competitiveness, will stop
exporting altogether. These are the least productive exporters.
These guys are charging a higher price than the more productive
exporters. So when a negative productivity shock hits to home
country, the high price exporters pull out of the export market.
This will tend to reduce the average price of exports. Symmet-
rically, some low productivity exporters can now start exporting,
these …rms charge a higher price than existing exporters, and this
will tend to increase the average price of imports. Some domestic
…rms may shut down altogether, and some new foreign …rms may
enter. So average domestic prices may go down also, and foreign
prices may go up too. But because of the presence of …xed costs
(for some distributions of productivity shocks, not any obviously),
there will be more movements in and out of export than in and
out of the domestic market. So this composition e¤ect reinforces

23
the relative movements in mark-ups charged by exporters.
Such an e¤ect would be present in a simpli…ed version of the Melitz
(2003) model even without endogenous mark-ups.
Composition e¤ects, in Atkeson and Burstein and in Ghi-
roni and Melitz (2005): Note however that this composition
e¤ects going in this exact direction crucially depends on the fact
that we consider a partial equilibrium version of the Melitz model,
that is a version without free entry of …rms. The Ghironi and
Melitz model points out that once the free entry of …rms is taken
into account, the composition e¤ect may actually be reversed alto-
gether. We have seen in the Ghironi and Melitz model that when
the home country is hit by a negative productivity shock, there will
be exit of …rms (or not entry, or more precisely no entry of new
…rms at home, so exit by death, and entry of …rms abroad). This
exit of …rms will actually drive down the productivity threshold for
exports. So the average price of exports will actually go up, once
endogenous entry and exit is taken into account. This mechanism
was at the heart of the Harrod-Balassa-Samuelson e¤ect.
Short run versus long run price adjustments: The conclu-
sion of that note is that depending of what time horizon one looks
at, di¤erent e¤ects may play di¤erent roles. In the very short run,
without entry and exit of …rms neither on the domestic market, nor
on the export market, only the mechanism described by Atkeson
and Burstein (endogenous mark-ups, incomplete pass-through and
pricing to market) will be present: in response to a negative pro-
ductivity shock at home, domestic exporters increase their prices
at home more than abroad. Over a slightly longer horizon, with
no entry or exit of …rms on the domestic market, but with entry
and exit of …rms on the export market, on top of the Atkeson and
Burstein paper, there will be a composition e¤ect: in response to
a negative productivity shock, high price domestic exporters stop
exporting, so that average export prices increase. Over an even
longer horizon, the mechanism described in Ghironi and Melitz
starts kicking in: domestic …rms exit, foreign …rms enter, so that
the productivity cuto¤ for exports decreases at home and increases
abroad, so that the average price of exports actually goes up. In
a model with free entry and endogenous mark-ups, whether the
Ghironi and Melitz e¤ect or the Atkeson and Burstein e¤ect dom-
inates depends (I guess) on the speci…c assumptions of the model
(the distribution of productivity shocks, the size of trade barri-
ers, the number of …rms per sector, the elasticities of substitution

24
 within and between sectors). In the end, it is an empirical question,
for which I don’t know the answer.

References
[1] Balassa, B. (1964), “The Purchasing Power Parity Doctrine: A Reap-
praisal”, Journal of Political Economy, 72, 584-96.
[2] Cendrars, Blaise (1925), L’Or.
[3] Dornbusch, Rudiger (1987) “Exchange Rates and Prices” American
Economic Review, vol. 77, March.
[4] Harrod, Roy. (1933), "International Economics". London: Nisbet and
Cambridge University Press.
[5] Krugman, Paul (1987), “Pricing to Market When the Exchange Rate
Changes,” in S. W. Arndt and J. Richardson, eds., Real Financial
Linkages Among Open Economies (London: MIT Press)
[6] Krugman, Paul (1991), "Increasing Returns and Economic Geogra-
phy," Journal of Political Economy, vol. 99(3), pages 483-99, June.
[7] Obstfeld, Maurice, and Kenneth S. Rogo¤ (2001), “The Six Ma-
jor Puzzles in International Macroeconomics: Is There a Common
Cause?”NBER Macroeconomics Annual 2000, Ben S. Bernanke and
Kenneth S. Rogo¤, eds. (Cambridge, MA: MIT Press, 2001), pp.
339–390.
[8] Rogo¤, Kenneth S. (1996), “The Purchasing Power Parity Puzzle,”
Journal of Economic Literature, XXXIV (1996), 647–668.
[9] Samuelson, Paul (1964), “Theoretical Notes on Trade Problems”,
Review of Economics and Statistics, 23, 1-60.

25