superstar firms in international trade j. peter neary university of oxford and cepr 12 september...

Superstar Firms in International Trade

J. Peter Neary

University of Oxford and CEPR

12 September 2011

2

1. Introduction

“Two and a Half Theories of Trade”:1.0: Perfect Competition: Comparative Advantage

2.0: Monopolistic Competition: Product Differentiation and Increasing Returns

2.5: Oligopoly

Despite growing empirical evidence: large firms matter for trade• The “2nd Wave” of micro data on firms & trade

• 1st wave (1995-): Exporting firms are exceptional:• Larger, older, more productive, pay higher wages, do more R&D

• 2nd wave: Even within exporters, large firms dominate:[Bernard et al. (JEP 2007), Mayer and Ottaviano (2007)]

• Distribution of exporters is bimodal

• The firms that matter (for most questions) are different: larger, multi-product, multi-destination

• Possible to model this using a Pareto distribution with high dispersion• “Granularity”: Gabaix (2005), di Giovanni and Levchenko (2009)

• I prefer to try “putting the grains into granularity”

3

5+ products:

Bernard et al. (JEP 2007):• Data on U.S. exporting firms 2000• By # of products & export destinations

83.3% of employment

25.9% of firms

98.0% of export value

7.0% of employment

40.4% of firms

0.20% of export value

1 product, 1 destination only:

& 5+ dests.:

11.9% of firms

68.8% of employment

92.2% of export value

4

Why does oligopoly give only half a theory of trade?In I.O. it gets equal billing: “Two faces”

1. Take-home messages less than overwhelming?– Cross-hauling of identical goods [Brander JIE 1980] ...

• Empirically important?Friberg-Ganslandt (JIE 2006)

– Competition effect of trade [Brander] ...• Also in monopolistic competition if is variable

Krugman (JIE 1999), Behrens-Murata (JET 2007), Melitz-Ottaviano (REStud 2008)

– Strategic trade policy [Brander-Spencer JIE 1984] ...• Non-robust to assumptions about factor mobility and firm behaviour

Dixit-Grossman (1986), Eaton-Grossman (1986)

1. Introduction (cont.)

5

2. Not embedded in general equilibrium?– This can be done: General Oligopolistic Equilibrium [“GOLE”]Neary (JEEA 2003, REStud 2007), Bastos and Krieckemeier (JIE 2009), Grossman and Rossi-Hansberg (QJE 2010), Bastos and

Straume (2010)

– Trade with oligopoly interesting because there is less of it, not more

– Competition effects and comparative advantage interact: profit share may rise

– Suggests an explanation for non-price effects of foreign competition on wages

– Allows consideration of effects of trade on market structure itselfNeary (2002, RIE 2002, REStud 2007)

3. Hard to combine with entry and exit?– Here: Some potential approaches• Based on work in progress with Carsten Eckel and with Kevin Roberts

• … combining oligopoly with entry

• … and with heterogeneous firms

1. Introduction (cont.)

6

Plan

1. Introduction

2. Oligopoly plus Free Entry

3. Heterogeneous Industries

4. Natural Oligopoly

5. Superstar Firms

6. Conclusion

7

Plan

1. Introduction

2. Oligopoly plus Free Entry

3. Heterogeneous Industries

4. Natural Oligopoly

5. Superstar Firms

6. Conclusion

8

2. Oligopoly plus Free Entry

The “Integer Problem”:

• Ignoring it comes close to assuming monopolistic or perfect competition:– Brander and Krugman (1983): Trade liberalization cannot lower welfare

– Markusen and Venables (1988): No role for strategic trade policy

– Head, Mayer and Ries (2002): No home-market effect.

• Taking account of it is hard

• Or is it?

• Promising approach: impose restrictions on profit functions:– Bergstrom-Varian (REStud 1985): Profits depend on own costs and average of all firms’

e.g., Cournot with identical goods

– Acemoglu-Jensen (CEPR WP 2009): “Aggregative games”: Profits depend on own costs and a generalised mean of all firms’:

e.g., Cournot or Bertrand with CES differentiated products

],)(,[

nccii

Proposition:

• The number of firms in any heterogeneous-firm free-entry equilibrium of an aggregative game is the same as the integer number of firms in the corresponding “lean” homogeneous-firm non-integer equilibrium.

9

3. Oligopoly plus Free Entry (cont.)

Heterogeneous-firm equilibrium in an aggregative game:

nincc ni ...,10]},{,[ )(

Symmetric equilibrium:

Proposition:

• The number of firms in any free-entry equilibrium of an aggregative game is the same as the integer number of firms in the non-integer lean symmetric equilibrium

Lean symmetric equilibrium:

(i) Firms’ costs a random draw:

],)(,[),(

nccncs

0]1},,{,[ )( nccc n

(ii) Incumbents do not make losses:

(iii) A new “lean” entrant would make a loss:

0)1,(0),( ncandnc ss

Non-integer lean symmetric equilibrium:

0),( ncs

i.e., assume own effect dominates cross effect

ci drawn from g(ci) with positive support over [c,)

10

Example: The Linear Cournot Model

• Solution in symmetric (homogeneous-firm) case:

fxcp iii )(

Xsap 1

sn

cax

1

fsn

cafxs

2

21

1

11

12

13

Plan

1. Introduction

2. Oligopoly plus Free Entry

3. Heterogeneous Industries

4. Natural Oligopoly

5. Superstar Firms

6. Conclusion

14

3. Heterogeneous Industries

Firm heterogeneity in monopolistic competition:[Melitz (Em 2003)]

• Firms pay a sunk cost to reveal their unit costDraw c from g(c) with positive support over [c , )

• Given c, they calculate their expected profits and choose to produce or exit

Exit if c < ce where (ce) = 0 or r(ce) = f . • If exporting requires an additional fixed cost, only low-cost firms will

engage in it

Extend to a continuum of industries/sectors:• Firms draw a unit cost and a sector: {c, z}• Sectors differ in their fixed costs: f(z), f ´ > 0• In equilibrium, sectors have different expected firm numbers

15

Equilibrium Expected Market Structure(Vertical scale is ln2n)

n

z 1

Monopoly

0

Oligopoly

Monopolistic Competition

2

4

8

16

1

16

Plan

1. Introduction

2. Oligopoly plus Free Entry

3. Heterogeneous Industries

4. Natural Oligopoly

5. Superstar Firms

6. Conclusion

17

4. Natural Oligopoly

So far: Fixed costs exogenous

Now, allow them to be chosen endogenously Dasgupta-Stiglitz (EJ 1980), Gabszewicz-Thisse (JET 1980), Shaked-Sutton (Em 1983), Sutton (1991, 1998)

Free entry condition:[Equilibrium operating profits a reduced-form function of firm numbers and market size]

(n,s) = 0Define: Market-Size Elasticity of Market Structure:

• Stability requires n < 0 for given s

• Presumption (?) that s > 0

So, E presumptively positive?• Clearly so in simple entry games: Cournot, Bertrand

• Not necessarily positive in multi-stage games

• “Natural Oligopoly” the case where E 0

s

n

ss dnE

n ds n

18

Example: The Linear-Quadratic R&D+Cournot Case

• Investments are unfriendly

• Hence, E can be very small or negative.

• Solution in symmetric non-strategic case:

fkxcp iiii 2

21)(

iii kcc 0Xsap 1

xkandssn

cax

1

0

2

fscasn

sfkxs

202

21

22121 )(

)1(

1

[“Relative efficiency of R&D” for a unit market size]

19

20

21

Plan

1. Introduction

2. Oligopoly plus Free Entry

3. Heterogeneous Industries

4. Natural Oligopoly

5. Superstar Firms

6. Conclusion

22

5. Superstar Firms

Distribution of ex ante identical firms:

Stage 1: Pay a fixed cost f to discover their unit cost c

Stage 2: Choose to either:

• Remain as a “small” firm ; or:

• Pay a further cost fL to invest (in capacity/R&D/technology adoption/ product range) and become a “large” firm

Stage 3: Competition in quantities or prices:

• Strategic competition between large firms

• …… facing a (monopolistically) competitive fringe

23

G(c)

cc

G(c): Cumulative distribution of unit costs

Crucial assumption:

G(c)

'( ) 0G c

(f )

Threshold for entry

(fL )

Threshold for acquisition of“superstar” technology

Entry condition for “superstar” firms? Same as Section 2 only!

24

5. Superstar Firms (cont.)

So far: Advantage of “large firms” unspecified

One important case: Large firms produce a continuum of products– Consistent with earlier empirical evidence

– Solves a technical problem: large firms are of finite, small ones of zero measure

Different approaches to modelling multi-product firms:

(1) I.O.: Products few and/or fixed; product line / quality competition.

(2) Symmetric demands and costsAllanson and Montagna (IJIO 2005), Feenstra and Ma (2009), Nocke and Yeaple (2006)

(3) Stochastic market-specific costsBernard, Redding and Schott (AER 2009)

(4) “Flexible manufacturing”Eckel and Neary (REStud 2010)

Empirical evidence favours (4):Arkolakis-Muendler (2010)

• (2) predicts equal sales of all varieties - rejected

• (3) predicts sales ranks uncorrelated across markets - rejected

But (2) is more tractable!– Work in progress: Extend model of oligopoly with multi-product firms to allow for

entry by “small” firms

25

Flexible Manufacturing

ic

i

“Core Competence”

Product Range

Output, Price, and Cost Profiles

tc 0

“Core Competence”

i

tic

ip ix

0p

0x

26

27

Intra-Firm Effects of Globalization

ix

i

Outputs rise Outputs fall

“Leaner and Meaner”:•More Competition: Scope falls•Larger Market: Scale rises

28

Plan

1. Introduction

2. Oligopoly plus Free Entry

3. Heterogeneous Industries

4. Natural Oligopoly

5. Superstar Firms

6. Conclusion

29

6. Conclusion

A road-map, not a model; hopefully with implications for:

1. Theory:• Reconcile the “Two faces of IO”

• Large firms matter for more than just reciprocal dumping

2. Policy:• Hosting superstar firms, or not, may matter?

• Fostering entrepreneurship / entry?

• Competition policy in general equilibrium

• Trade liberalization in oligopoly

3. Empirics:• Firm level data: Good news and bad

• Errors at the top more costly

• Ideally, think of the industry at the world level

30

6. Conclusion (cont.)

Summary:

• Time for Trade Theory 3.0 ...

• Payoff to looking at the grains in granularity

31

Thank you!

32

1. Introduction

Inspiration: The Krugman Oral Tradition!

1.0: Perfect Competition: Comparative Advantage

2.0: Monopolistic Competition: Product Differentiation and Increasing Returns

2.5: Oligopoly

33

2. How Many Theories of Trade?

Perfect versus Monopolistic Competition?

Perfect and/or Monopolistic Competition?• Similar assumptions

• Similar implications with constant : e.g., gravity equations

• Both imply production efficiency, so can be represented by a GDP function:Dixit-Norman (1980)

– Dixit and Stiglitz (1977), Helpman (1984), Feenstra and Kee (JIE 2008)

Anyway: 2 full-fledged theoriesSo, why not trade with oligopoly too?

In I.O. it gets equal billing: “Two faces”

34

2. How Many Theories of Trade? (cont.)

Why does oligopoly give only half a theory of trade?• Take-home messages less than overwhelming?

– Cross-hauling of identical goods ...• Empirically important?

Friberg-Ganslandt (JIE 2006)

– Competition effect of trade ...• Also in monopolistic competition if is variable

Krugman (JIE 1999), Behrens-Murata (JET 2007), Melitz-Ottaviano (REStud 2008)

– Strategic trade policy ...• Non-robust to assumptions about factor mobility and firm behaviour

Dixit-Grossman (1986), Eaton-Grossman (1986)

• Not embedded in general equilibrium?– This can be done: General Oligopolistic Equilibrium [“GOLE”]

Neary (JEEA 2003)

– Trade with oligopoly interesting because there is less of it, not more– Competition effects and comparative advantage interact: profit share may rise– Suggests an explanation for non-price effects of foreign competition on wages– Allows consideration of effects of trade on market structure itself

Neary (2002, RIE 2002, REStud 2007)

35

How Many Theories of Trade? (cont.)

One more reason why oligopoly matters for trade:

• Empirics: The “2nd Wave” of micro data on firms & trade• 1st wave (1995-): Exporting firms are exceptional:

• Larger, more productive

• 2nd wave: Bernard et al. (JEP 2007): Even within exporters, big firms dominate:

[Similar results for France: Mayer and Ottaviano (2007)]

• Distribution of exporters is bimodal:– 40.4% of U.S. exporting firms in 2000 exported one product only; accounted for only 0.20% of export value– 25.9% of U.S. exporting firms in 2000 exported 5 or more products; accounted for 98.0% of export value– 11.9% of U.S. exporting firms in 2000 exported 5 + products to 5+ destinations; accounted for 92.2% of export value

• The firms that matter (for most questions) are different: larger, multi-product, multi-destination

• Possible to model this using a Pareto distribution with high dispersion• “Granularity”: Gabaix (2005), di Giovanni and Levchenko (2009)

• I prefer to try “putting the grains into granularity”

36

How Many Theories of Trade? (cont.)

One more theoretical hurdle to be crossed:• How to combine oligopoly with entry and exit?

• Here: Some potential approaches– Based on work in progress with Carsten Eckel and with Kevin Roberts

– … combining oligopoly with entry

– … and with heterogeneous firms

37

3. Oligopoly plus Free Entry (cont.)

For given market size s, assume profits of firm i depend on own costs ci, on the average of rivals’ costs c–i, and on n:[Bergstrom-Varian REStud 1985]

i= (ci , c–i , n)

– – i.e., assume own effect dominates cross effect

Start with symmetric case:

• We can characterise the admissible range of c and n:

(1) s(c,n) 0 [Incumbents do not make losses]

(2) s(c,n1) < 0 [A new entrant would make a loss]

In symmetric equilibrium:

s(c,n) (c, c, n)

– + –

38

n

c

s(c,n) = 0

c

n* n* is the maximum equilibrium number of firms;BUT: A real number, not necessarily an integer

(1) s(c,n) 0 [Incumbents do not make losses]

(2) s(c,n1) < 0 [A new entrant would make a loss]

s(c,n1) = 0

n*–1

So: “Lean” symmetric equilibrium has:ne = int (n*–1, n*]

ne

39

n

c

s(c,n) = 0

c

n*

So: Shaded area is the admissible region for equilibrium n in symmetric equilibriaActual equilibrum values are on the horizontal line corresponding to the integer value:

ne = int (n*–1, n*]

So far: Only “lean” symmetric equilibria;

Now, consider symmetric equilibria with less efficient firms:

[A new lean entrant would make a loss]

s(c,n+1) =0

n*–1

ne

0)1,,( ncc

0)1,,( ncc

c

],[ ccc

40

Finally: Consider asymmetric equilibria:• Start with simplest case: monopoly vs. duopoly

• Assume there is a monopoly equilibrium:• “Lean outsider” condition:

• Consider a candidate duopoly equilibrium with:

],[ cccm

0)2,,( mcc

ccc 21

)2,,( 21 cc

)2,,( cc

)2,,( mcc

Proof can be extended to oligopoly and generalised mean [n] case:

• Assume an equilibrium with n–1 firms

• Replace “2” by n, c2 by n, and cm by n–1 in above

• Conclusion: There cannot be an equilibrium with n firms

So: Equilibrium integer n is a monotonic function of (e.g.) industry size

… and, if we know the population distribution of c, we can calculate the distribution of c in free-entry oligopoly equilibrium

0

)2,,( 22 cc

Possible Cost Configurations in Duopoly

(i) Low s (ii) Medium s (iii) High s

(ii) Lean outsider constraintbinds in symmetric equilibria;

insider constraint in asymmetric ones

(i) Only insider constraints bind (iii) Only lean outsiderconstraint binds

41

c2 c2 c2

c1 c1 c1

]0[ c

42

5. Natural Oligopoly (cont.)

With multi-stage competition:

Presumption that dk/ds is positive:• Seade stability condition

ˆ ˆ( , ) ( , ; , ) when: argmax j

i i i j j

kn s k K n s k j

ˆ ˆ( 1)i is s K

dkn

ds

1ˆ ˆ ˆ( 1)i i ikk kK ks

dkn

ds

• Increase in market size raises investment (?)

Hence, for natural oligopoly:• Necessary condition: investments are “unfriendly”:

ˆ 0ikK

ˆ 0iK

• More likely the more investment responds to market size

• … which is more likely if investments are strategic complements:

+?

Sales Revenue Profiles in Home and (Larger) Foreign Market

*

irir ,

ir

ir

i

43