Scribd
Upload
66 views

ECON 357: Lecture 7: The Distribution of RM Sizes

The document summarizes a lecture on the distribution of firm sizes. It discusses how firm size distributions often follow stable patterns like Pareto and Zipf distributions. It then presents different stochastic models that can generate these stable distributions, such as models based on Gibrat's law of proportional growth and social network theory. The key models discussed are those of Gabaix, Simon, and Chatterjee and Rossi-Hansberg.

Uploaded by

Hector Perez Saiz
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
66 views

ECON 357: Lecture 7: The Distribution of RM Sizes

The document summarizes a lecture on the distribution of firm sizes. It discusses how firm size distributions often follow stable patterns like Pareto and Zipf distributions. It then presents different stochastic models that can generate these stable distributions, such as models based on Gibrat's law of proportional growth and social network theory. The key models discussed are those of Gabaix, Simon, and Chatterjee and Rossi-Hansberg.

Uploaded by

Hector Perez Saiz
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 23

ECON 357

Lecture 7: the distribution of …rm sizes

Thomas Chaney

University of Chicago

Thomas Chaney (Chicago) ECON 357: Lecture 7 1 / 23


Motivation

Productivity distribution governs …rm level and aggregate trade


patterns in:
Melitz/Chaney
Eaton-Kortum/BEJK
Melitz-Ottaviano
Distortions of the productivity distribution can have large e¤ects on
e¢ ciency:
Trade in Melitz
Hsieh-Klenow
Well calibrated random mechanical model describes trade patterns
well:
Armenter-Koren (Balls and Bins)

Thomas Chaney (Chicago) ECON 357: Lecture 7 2 / 23


Motivation
Very stable …rm size distribution (source: Axtell, Science 2001):

Figure: R2 = .992, slope 1.059.


Thomas Chaney (Chicago) ECON 357: Lecture 7 3 / 23
Motivation
Pareto distributions and Zipf’s law appear often in social science and
nature (e.g. city sizes, Gabaix, QJE 1999)

Figure: ln(Rank) = 10.53 1.005 ln (Size)


(.010 )
Thomas Chaney (Chicago) ECON 357: Lecture 7 4 / 23
Motivation
Pareto distributions and Zipf’s law holds even for "small" cities
(Rozenfeld, Rybski, Gabaix and Maske, AER 2010)

Figure: Zipf holds for cities above 5,000 inhabitants.


Thomas Chaney (Chicago) ECON 357: Lecture 7 5 / 23
Motivation
Small departures from Pareto/Zipf:

Figure: Luttmer (QJE 2007).

Thomas Chaney (Chicago) ECON 357: Lecture 7 6 / 23


Target

Generate stable Zipf distribution from a simple stochastic process.


Yule (1925): distribution of # of species within biological genera.
Simon (1955), Simon & Bonini (AER 1958): distribution of …rm sizes.
Gabaix (QJE 1999): distribution of city sizes.
Luttmer (QJE 2007): distribution of …rm sizes.
Chatterejee & Rossi-Hansberg (2009): distribution of …rm sizes across
sectors.

Thomas Chaney (Chicago) ECON 357: Lecture 7 7 / 23


Gibrat’s law

Random growth of …rm size,

Sti +1 = γit Sti

with γit i.i.d. draws from p.d.f. fS (γ)


Gibrat’s law: for "large" …rms, growth rate of size is roughly
independent of …rm size,

fS (γ) = fS 0 (γ) = f (γ) for S, S 0 "large"

Thomas Chaney (Chicago) ECON 357: Lecture 7 8 / 23


Gabaix (QJE 1999)

Continuous time stochastic growth process for population size (Pit ),

dPit /Pit = γdt + σdBit

with γ (σ) average (s.d. of) growth of city size, and B a Brownian
motion.
Normalization,
dSit /Sit = µdt + σdBit
with Sit = Pit / (total population), and µ = γ (S ) γ̄.

Thomas Chaney (Chicago) ECON 357: Lecture 7 9 / 23


Degenerate log-normal limit

This process converges to a log-normal,

St = S0 exp σ2 t/2 + σBt

No steady state: log-normal keeps "fanning out".

Thomas Chaney (Chicago) ECON 357: Lecture 7 10 / 23


Small perturbation generates Zipf

Assume there is a minimal city size, Smin (can be arbitrarily small).


Re‡ected geometric Brownian motion close to the barrier,

dSt /St = max (µdt + σdBit , 0) at St = Smin

Proposition
If S follows a re‡ected geometric Brownian motion, then the distribution
of S converges to a Pareto distribution,
ζ
Pr S̃ > S = (S /Smin )

with ζ = 1/ (1 Smin /S̄ ).

Thomas Chaney (Chicago) ECON 357: Lecture 7 11 / 23


Yule distribution (Yule 1925, Simon & Bonini 1958)

Proposition
Assume Gibrat’s law for existing …rms, and entry of new …rms at Smin .
Then the size distribution converges to,

f (S ) = KB (S, ζ + 1)

with B ( , ) the Beta distribution, which asymptotes to a Pareto


distribution (p.d.f.),
f (S ) ! MS (ζ +1 )
S !+∞

for a constant ζ given by,


1
ζ=
1 g /G
with g the growth from new …rms, and G the growth from existing …rms.

Thomas Chaney (Chicago) ECON 357: Lecture 7 12 / 23


Limitations of Simon’s results

1 To match Zipf’s law, ζ 1, one needs g /G 0.


With g /G 0, the process converges extremely slowly.
2 One needs that growth of # of cities (…rms) > growth of population.
Counterfactual.

Thomas Chaney (Chicago) ECON 357: Lecture 7 13 / 23


Using …rm size distribution to learn about economic forces

Luttmer (QJE 2007) can match:


ratio of (…xed) entry cost to (…xed) operating cost.
ratio of (…xed) entry cost to (…xed) imitation cost.
Chaterjee & Rossi-Hansberg (2009) can match:
relative contribution to growth of new …rms versus existing …rms.
size of "teams" within …rms.

Thomas Chaney (Chicago) ECON 357: Lecture 7 14 / 23


Chatterjee & Rossi-Hansberg (2009): all …rms

Match relative contribution to growth of new …rms versus existing …rms


Thomas Chaney (Chicago) ECON 357: Lecture 7 15 / 23
Chatterjee & Rossi-Hansberg (2009): sectors

Thomas Chaney (Chicago) ECON 357: Lecture 7 16 / 23


Chatterjee & Rossi-Hansberg (2009): sectors (cont’d)

Thomas Chaney (Chicago) ECON 357: Lecture 7 17 / 23


Chatterjee & Rossi-Hansberg (2009)

Model makes prediction about both …rm size distribution and some
fundamental parameters.
Match …rm size distribution and predict:
1 Relative contribution to growth of new versus existing …rms
Aggregate data gets the rigth g /G .
2 Size of "teams" within …rms:
Sectoral data gets the right size of "teams".

Thomas Chaney (Chicago) ECON 357: Lecture 7 18 / 23


Size distribution, innovation and growth

If innovation is a contest (the best wins), then Extreme Value


Distribution Theory (EVT) can be used.
Kortum (ECMA 1994), Eaton & Kortum (IER 1999), Lucas (2009).
Stable asymptotic size distribution of …rm sizes (EK), incomes
(Lucas).

Thomas Chaney (Chicago) ECON 357: Lecture 7 19 / 23


Social networks (Matt Jackson, Stanford)

If …rms meet the "friends of their friends", then stable network


emerges.
Jackson & Rogers (AER 2007) test some predictions of network
theory in di¤erent social settings.

Thomas Chaney (Chicago) ECON 357: Lecture 7 20 / 23


Social networks and trade (Chaney 2010)

Jackson & Rogers (AER 2007) embedded into geographic space:


1 Distribution of the number of exports destinations is assymptotically
Pareto.
2 Distance from exporters increases with number of foreign contacts.
3 The more contacts a …rm has, the more likely it is to enter new
countries.
4 If Japan is close to Korea, an exporter to Japan is more likely to enter
Korea.
5 If Spain trades a lot with Chile, an exporter to Spain is more likely to
enter Chile.

Thomas Chaney (Chicago) ECON 357: Lecture 7 21 / 23


How many …rms export to M markets?

Thomas Chaney (Chicago) ECON 357: Lecture 7 22 / 23


How far does a …rm export to?

Thomas Chaney (Chicago) ECON 357: Lecture 7 23 / 23

You might also like