Localization vs. Diversity Economies Marshalls Scale Economies By - - PowerPoint PPT Presentation

Localization vs. Diversity Economies

“Marshall’s Scale Economies” By J. Vernon Henderson Journal of Urban Economies (2002) “Agglomeration Economies and the High-Tech Computer” By Nancy Wallace and Donald Walls Fisher Center Working Paper (2004)

Beia Spiller, June 3, 2008

“Marshall’s Scale Economies”: Introduction Localization/Marshall, Arrow Romer (MAR) Economies:

• Externalities arising from other plants in the same

industry Urbanization/Jacobs Economies:

• Externalities arising from scale or diversity of local

economic activity outside of own industry

• Result is important for urban development planning:

industrial concentration vs. diverse metro areas

Introduction Other issues:

• Differing impacts for single or multi-plant firms
• Different levels of externalities due to new or

established firms

• Lagged vs. current externality effects

Industries

• 5 major capital goods or machinery industries
• construction, metal working, special industrial,

general industrial, refrigeration machinery and equipment

• 4 major high-tech industries
• computers, electronic components, aircraft, medical

instruments

Data

• Census of Manufacturers- 1963, and every five years

between 1967-1992.

• Most of this data covers multi-plant firms (corporate)
• Annual Survey of Manufacturers- non-census years
• This data is taken for only non-affiliate plants
• Restrictions:
• Eliminates entries with imputed factors (keeps only

entries with surveyed inputs and outputs)

• Keeps only those firms that appear at least twice

(results in mostly large/established firms in sample)

Data

Measuring Effects General Production Function:

• =

− + + − + =

2

) ( ) ( ) ( ln ) ( ln ) ( ln

s kj kj j s k k

t f t s t E t X t y ε δ β α

k, j: firm, location yk(t): output ln Xk(t): vector of plant inputs ln Ej(t-s):vector of industrial environmental variables in t-s (t): time fixed effects fkj: plant location fixed effects

Estimation

• High tech and machinery industries are pooled:
• ’s and ’s constrained to be same within each

industry.

• (t) are separate for each industry

Diversity Measures

• 1. The degree of MSA specialization in a set of activities:
• −

=

i i j ij j

t E t E t E t E t S

2

) ( ) ( ) ( ) ( ) (

Eij(t): employment in industry i in city j Ej(t): total employment in city j i

ij t

E ) (

Ei(t): national employment in industry i E(t): total national employment i

i t

E ) (

Sj(t): ranges from 0 (perfectly diverse) to 2 (perfectly specialized)

More Diversity Measures

• 2. Overall MSA scale or total employment in the

relevant industries.

• 3. MSA scale by counts of plants.
• 4. County level effects.

Estimation Issues Fixed effects will influence both environmental variables and inputs- OLS is biased, fixed effects estimation valid. Exogeneity Assumptions:

• plant inputs and industrial environmental variables

are strictly exogenous for all t to the kj(t)

• capital stock and labor/materials are state variables,

so are exogenous to the kj(t)

Results: Localization/MAR Economies

(1): OLS, (2): Plant/Location FE

Results: Localization/MAR Economies

(1): OLS, (2): Plant/Location FE

Endogeneity Issues: Some Experiments

• 1. Add MSA-time fixed effects: to control for shocks

that affect RHS variables and error term a. Results are similar except for high-tech non- affiliates: appear to benefit more from externalities than corporate firms. b. Small sample and loss of efficiency

• 2. Instruments to control for endogeneity of Xk and Ej

a. Predetermined values of Xk and Ej b. Suffers from small sample size and weak instruments c. No evidence of correlation between Xk and kj

Other Localization Specifications

• Localization effects only impact high-tech companies

Other Localization Specifications

Dynamic Externalities

• No lagged effects for machinery or high-tech corp.
• For non-affiliate high-tech: only 5 year lagged effects

present.

Jacobs-Urbanization Economies

Jacobs-Urbanization Economies

Jacobs-Urbanization Economies Why does diversity appear to affect corporate machinery firms?

• Unmeasured business service inputs.
• Corporate machinery firms outsource more materials.

Thus: Urbanization economy may be illusionary.

Conclusions Localization externalities:

• mainly for high-tech industries.
• very localized within county.
• non-affiliates benefit more from these externalities.

Diversity externalities:

• no evidence to support static or dynamic Jacobs

economies.

“Agglomeration Economies”: Introduction

• Role of agglomeration on the production decisions of

firms in the high-tech computer cluster.

• Trade-off between labor market pooling and labor

market poaching

Functional Definition of the Computer Cluster New definition:

• Services SICs are included
• High-tech firms may include establishments not

normally considered “high-tech” Computer Cluster includes:

• Manufacturing (hardware)
• Services (software)
• All establishments that meet this criteria or report to

a headquarter that does. These are found from the SIC and NAICS codes

Data Information at firm level:

• NETS database: 1989-2002.
• Identifies geographic location of each firm: both

its headquarters and its establishments. 401 8-digit SICs from several computer-related manufacturing and service industries

• any firm that ever had this SIC or reported to one that

did = 539,942 establishments tracked over the period.

Issues with Data

• 1. Several firms operate exclusively within high-tech

manufacturing

• Not representative of high-tech firms.
• 2. Most capital, R&D capital, and outsourcing

decisions are made at headquarters

• Can’t treat individual establishments in multi-

establishment firms as individual decision makers.

• 3. Lack of controls for relationship between plants and

HQ- can lead to underestimates of std.err.

Geography of high-tech Computer Cluster

Geography of high-tech Computer Cluster

Measures of Geographic Network of Each Firm

• 1. Inner Network Interactivity:

2

circ e e w

i j j i ij

× × × =

• π

ei = employment in firm i

1 for HQ and outlying establishments

wij = .5 between separate establishments

0 for single establishment firms

circ = 60 mile radius of influence

Measures of Geographic Network of Each Firm

• 2. Outer Network Interactivity:

2 ij i j j i ij

dist e e w

• ×

× = ei = employment in firm i wij = same as above, for establishments outside of 60 miles radius of influence. distij

2 = distance between firms i and j

Measures of Geographic Network of Each Firm

• 3. Firm’s total MSA-level labor market exposure to

workers in computer manufacturing SICs.

• 4. Firm’s total MSA-level labor market exposure to

workers in computer services SICs. Purpose of these measures:

• Decisions made at HQ will likely affect dispersed

production units

• Location of establishments will likely be coordinated

with proximity to other employees of similar type

Firm Structure and Estimation of Production Function

• Cobb-Douglas production function, 4 inputs (Xi): labor,

purchased inputs, R&D capital, physical capital.

• +

=

+

i

i i

X Q ε α α ln ln ln

• Translog functional form:
• +

+ + =

i i j j i ij i i

X X X Q ε α α α ) )(ln (ln ) 5 (. ln ln ln

• Inputs and outputs are endogenous choices determined

by exogenous market state variables.

Firm Production Data COMPUSTAT database: 177 publicly-traded firms operating from 1989-2002 Output: total sales Labor inputs: total employees Purchased inputs: “Cost of Goods Sold” Capital Stock: “Net Purchased Plant and Equipment” R&D Capital: research expenditures

Production Function Results Assumption: geographic dispersion and labor market structure are fixed ex ante to production decisions.

• Implication: use lagged exogenous network structure

as explicative variables.

• +
• +

=

− +

k i

t k i i i i

Z X Q ε γ α α

1 ,

ln ln ln ln

Z: natural log of all four geographic network measures.

Accounting for Heterogeneity in Firm-Level Technology Method follows Mundlak and Hellinghausen (1982) Coefficients: functions of variables reflecting inter-firm heterogeneity. New Cobb-Douglas Specification:

j j Vj j

y ε β α + + = X where

) , ( ~

2 nj j

N

j

I 0 σ ε

→

Accounting for Heterogeneity in Firm-Level Technology Coefficients on variable factors are functions of exogenous state variables:

j t j j

ω δ β + =

−1 ,

Z

and

) , ( ~

• →

N

j

ω

This implies: j are random coefficients, distributed normally, with mean

δ

1 , − t j

Z

.

Accounting for Heterogeneity in Firm-Level Technology Zj,t-1: K x L matrix of ex-ante decisions concerning the spatial configuration of the firm’s establishments

• =

→ → → → → → − jK j t j

' ~ ' ~

1 1 ,

z z Z

• Where

jk

' ~ z

is an lk-element vector of variables that determine network technology effects on the firm’s variable input choices.

Accounting for Heterogeneity in Firm-Level Technology Combining the two equations gives:

) (

1 , j j j t j j j

y ω ε δ X Z X + + =

−

• Can estimate firm-level elasticities and standard errors.

Can test for location effects on production decisions.

• OLS is biased because of correlation between error term

and independent variables. Use restricted maximum likelihood

• Assumption: diagonal elements of Zj,t-1 are same within

each firm.

Results

Results

Results

Results

Results

Results

• Considerable heterogeneity in firm-level technology
• Smaller, more concentrated firms have a productivity advantage

(reflects Henderson’s results that non-affiliate plants benefit more from localization externalities)

• Localization effects of employment are significant
• Employment appears to be substitutable with other purchased

inputs

• Overall production function has constant returns to scale
• Factor elasticities sum to 1.21

Conclusions

• Ability of firms to be near a large pool of computer

service employees allows for more flexible production

• Large network interactions diminish productivity
• Appear to overcome benefits from accessibility to

employee pool.