border effects in public procurement

Introduction Data The role of governments: a simple decomposition Plant-level evidence Model Quantification Conclusions

Border effects in Public ProcurementThe Aggregate Effects of Governments’ Home Bias

Manuel Garcıa-Santana Marta Santamarıa

UPF, CREi (visiting Princeton) University of Warwick

STEG - Plenary Workshop, Sep. 2021

IntroductionGovernment Procurement

Government purchases represent a big fraction of World’s GDP

- 11-20% depending on country and year

- Governments are the largest buyers in many industries

- Potentially powerful industrial policy tool

Development/macroeconomic implications not well understood

Low import penetration rates

Firms Households Governments

France 17.02 13.09 2.24

Spain 17.22 13.48 2.90

Notes: Imports/total expenditure. Source: WIOD input-output tables.

Garcıa-Santana, Santamarıa Border effects in Public Procurement 1/18

France 17.02 13.09 2.24

Spain 17.22 13.48 2.90

France 17.02 13.09 2.24

Spain 17.22 13.48 2.90

IntroductionLocal share of procurement across French and Spanish regions

Share of procurement value

awarded to plants located

within the same region

Around 53% on average

Compares to 32% for total trade

Not driven by composition

– Machinery and elec. equip.: 50% vs. 37%

– Transport equipment: 41% vs. 19%

– Food: 78% vs. 27%

– Furniture: 51% vs. 18%

IntroductionLocal share of procurement across French and Spanish regions

Share of procurement value

awarded to plants located

within the same region

Around 53% on average

Compares to 32% for total trade

Not driven by composition

– Machinery and elec. equip.: 50% vs. 37%

– Transport equipment: 41% vs. 19%

– Food: 78% vs. 27%

– Furniture: 51% vs. 18%

IntroductionRelevant questions

Are governments responsible in explaining these patterns?

* Challenge: potentially driven by two types of origin-destination factors:

1. Natural frictions: τod abruptly increases when o ≠ d

- ex: geography, information frictions, path dependence, etc.

2. Governments’ home-bias

- governments intentionally discriminate against non-local firms

- Joe Biden: “we will not purchase anything that is not made in America”, Sep. 2020

- increasing cross-border participation is one of the EU Commission’s main goals

What are the consequences for aggregate efficiency and welfare? (not today)

* It benefits local firms

* Inefficient provision of public goods

IntroductionThis paper

1. Dataset of procurement contracts awarded in Spain and France

- more than 1 million contracts in 38 regions

- info about the contract: value, product, etc.

- info about the seller: location, firm, etc.

- info about the buyer: location, agency (around 10,000 only in Spain!), etc.

→ key: classify buyers into different gov. types

2. New plant-level evidence on inter-regional and international procurement flows

- participation increases systematically with market size

- sales distributions are similar across markets

- big “border effects” both at the region and country level

3. Strategy to identify the role of governments in explaining these border effects

– Sub-national vs. national governments have offices in the same location

- Ex: both the Federal Gov. and the Gov. of California have offices in San Francisco

- Ex: both the central Spanish Gov. and the Catalan Gov. have offices in Barcelona

– Use variation within a plant-origin-destination across government-types

- Allows us to control for τod!

– Hypothesis: geographical scope of governments main driver of home bias

- Regional governments discriminate against firms from other regions

- National governments discriminate against firms from other countries

4. Quantitative exercise

– Multi-region trade model with heterogeneous firms applied to procurement

- Chaney (2008), Breinlich, Cunat (2014)

– Use the model to quantify the aggregate effects of governments’ home bias

Procurement DataOur sample (all EU countries potentially)

We worked with France and Spain so far

Classify “buyers” into different gov. types

1. LOCAL: 18.56 %2. PROVINCIAL: 16.13 %3. REGIONAL: 23.58 %4. NATIONAL: 41.71 %

Work with 2 government types: national and sub-national

Big overlap in purchases across governments

- ex, motor vehicles: sub-national spend e5M and national spend e2M

- ex, medical consumables: sub-national spend e14M and national spend e5M

Years: 2011-2019

Plants: 192,155

The role of governments: a simple decomposition

The role of governmentsA simple decomposition

Share of procurement value in region r awarded to firms from r

πrr =XrrXr

=Xnatrr +Xsub

Xnatr +Xsub

Simple manipulations allow to decompose it into 4 different components

0.53«πrr =

0.13³¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xnatr

Xnatr +Xsub

0.29³¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xnatrr

Xnatr)

´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶πnatrr

0.87³¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xsubr

Xsubr +Xsub

0.56³¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xsubrr

´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶πsubrr

Naive counterfactual: set πsubrr = πnatrr and re-compute πrr

πrr =XrrXr

=Xnatrr +Xsub

Xnatr +Xsub

0.53«πrr =

0.13³¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xnatr

Xnatr +Xsub

0.29³¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xnatrr

Xnatr)

´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶πnatrr

0.87³¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xsubr

Xsubr +Xsub

0.56³¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xsubrr

´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶πsubrr

πrr =XrrXr

=Xnatrr +Xsub

Xnatr +Xsub

0.53«πrr =

0.13³¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xnatr

Xnatr +Xsub

0.29³¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xnatrr

Xnatr)

´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶πnatrr

0.87³¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xsubr

Xsubr +Xsub

0.56³¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xsubrr

´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶πsubrr

πrr =XrrXr

=Xnatrr +Xsub

Xnatr +Xsub

0.53«πrr =

0.13³¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xnatr

Xnatr +Xsub

0.29³¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xnatrr

Xnatr)

´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶πnatrr

0.87³¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xsubr

Xsubr +Xsub

0.56³¹¹¹¹¹¹¹¹¹¹¹¹¹¹·¹¹¹¹¹¹¹¹¹¹¹¹¹¹µ

(Xsubrr

´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶πsubrr

Data Counterfactual

XnatrXnr +Xsubr

πnatrrXsubr

Xsubr +Xsubrπsubrr πrr πrr

Madrid (ESP) 0.72 0.68 0.28 0.80 0.71 0.68 ↓ 5%

Aragon (ESP) 0.02 0.35 0.98 0.78 0.77 0.35 ↓ 55%

Ile de France (FR) 0.68 0.68 0.36 0.88 0.78 0.68 ↓ 13%

Pays de la Loire (FR) 0.05 0.31 0.95 0.65 0.63 0.31 ↓ 51%

mean 0.13 0.29 0.87 0.56 0.53 0.29 ↓ 45%

Plant-level evidence

Plant-level evidenceThree stylized facts about procurement

1 Firms’ participation in procurement increases with market size, BUT

1.a Firms participate disproportionally more in their own country

1.b Within their country, firms participate disproportionally more in their own region

1.c This last pattern particularly strong for sub-national governments

2 The firms’ sales distribution within an origin o is similar across destinations

2.a Well approximated by a Pareto except for the very low end of the distribution.

Firm level evidenceParticipation in procurement by firms located in Catalonia

CAN CYM

EXTGALIBA

RIONAV

ASTMUR

RHABOU

BRECVL

LARMIP

r of f

1 10 100Market size (in awarded value, in billions of eur)

ANDARA

CANCYM

MADVAL

EXTGALIBA

RIONAV

ASTMUR

AUVBREALSCHA

IDFHAN

AQULAR

r of f

A) Sub-national governments

ANDARA

CANCYM

EXT GAL

CVLLIM

r of f

B) National governments

ANDARA

CANCYM

MADVAL

EXTGALIBA

RIONAV

ASTMUR

AUVBREALSCHA

IDFHAN

AQULAR

r of f

A) Sub-national governments

ANDARA

CANCYM

EXT GAL

CVLLIM

r of f

B) National governments

A model of trade in procurementMain ingredients

Standard:

There are R regions indexed by o, r

Shipping goods across regions is costly (τor)

Each region is populated by Lr consumers who supply one unit of labor

Heterogeneous firms produce tradable differentiated intermediate goods

In each region r, two gov-types combine ygor(j) to produce final public goods

Y gr = (R

∑o=1∫

(αgor)1σ ygor(j)

σ−1σ dj)

σσ−1

Governments’ home bias manifests in two forms:

– Intensive margin through a preference parameter (αgor)

– Extensive margin through entry costs (Egor)

Standard:

Y gr = (R

∑o=1∫

(αgor)1σ ygor(j)

σ−1σ dj)

σσ−1

Standard:

Y gr = (R

∑o=1∫

(αgor)1σ ygor(j)

σ−1σ dj)

σσ−1

Identification of αgor (similar intuitions apply for Egor)

Sales to gov-type g in region r by firm j (from region o) given by:

Xgor(j) =X

σ − 1

σ)σ−1

(P grτor

σ−1

Assume:

αgor = 1 ∀g if o = r,

αgor = αg if o ≠ r

We use a dummy to identify the relative home-bias

ln Xgor(j) = β × 1(o ≠ r)

´¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¶non-local

× 1(g = s)´¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¶

Sub-national gov

+µj + δgr + φor + εjrg

⇒αs

αn= exp(β)

Xgor(j) =X

σ − 1

σ)σ−1

(P grτor

σ−1

Assume:

´¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¶non-local

× 1(g = s)´¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¶

Sub-national gov

⇒αs

αn= exp(β)

Xgor(j) =X

σ − 1

σ)σ−1

(P grτor

σ−1

Assume:

´¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¶non-local

× 1(g = s)´¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¶

Sub-national gov

⇒αs

αn= exp(β)

Xgor(j) =X

σ − 1

σ)σ−1

(P grτor

σ−1

Assume:

´¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¶non-local

× 1(g = s)´¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¶

Sub-national gov

⇒αs

αn= exp(β)

Main estimation results

when selling non-locally, plants sell 33% less to sub-national gov.

αn = exp(−0.38) = 0.67

similar identification strategy + data on firms’ participation implies that

En = 1.05

set some other important parameters to standard values

- σ = 5.00 (Elasticity of Substitution)

- θ = 8.00 (Pareto shape)

calibrate the elasticity of τor w.r.t distance and the level of E to match

- overall trade flows

- average participation rates

αn = exp(−0.38) = 0.67

En = 1.05

αn = exp(−0.38) = 0.67

En = 1.05

αn = exp(−0.38) = 0.67

En = 1.05

Quantification

Results: come back to the original question

How much of the low penetration rates in procurement is explained bygovernments’ home bias?

(1) (2) (3)

Agg. local share (πrr) Nat. gov. local share (πnrr) Subnat. gov. local share (πsrr)

Data 0.53 0.29 0.56

Baseline 0.50 0.35 0.52

αs = αn 0.38 (24.39% ↓) 0.35 (0.00% ) 0.38 (26.29% ↓)

Es = En 0.46 (8.90% ↓) 0.35 (0.00% ) 0.47 (9.35% ↓)