and Industry Aggregation Karolien Lenaerts and Bruno Merlevede - - PowerPoint PPT Presentation

Horizontal or Backward? FDI Spillovers, Input-Output Tables and Industry Aggregation

Karolien Lenaerts and Bruno Merlevede

Outline

2

• Overview of the Literature
• Research Topic in the Paper
• Empirical Approach & Data
• Estimation Results
• Conclusions

Overview of the Literature

Foreign Direct Investment

4

• Governments all over the world develop

policies to attract multinationals (MNEs)

• Benefit from both direct and indirect

effects of multinational activity:

• direct effects: employment, infrastructure
• indirect effects: FDI spillovers

FDI Spillovers

5

• Markusen (1995): when investing abroad,

MNEs bring proprietary technology with them to compete with local firms

MNE invests & brings technology Technology leaks

• r is transferred

intentionally The technology is adopted by domestic firms, raising their productivity level

FDI Spillovers

FDI Spillovers

horizontal and vertical FDI spillovers

raw materials final goods goods spillovers

Upstream Supplier Foreign Subsidiary Downstream Customer Local Competitor

backward spillover forward spillover horizontal spillover

6

Supply Chain

Research Topic in the Paper

Research topic in the paper

8

• Link with the literature:
• mixed empirical evidence on FDI spillovers
• many different explanations
• Our contribution: importance of the level of

industry aggregation in input-output tables

Research topic in the paper

9

• Focus: level of industry aggregation in the

input-output (IO) tables

• Why?
• spillovers are constructed from IO-tables

→ technical coefficients of vertical spillovers

• level of aggregation in the IO-tables determines

classification in horizontal or vertical spillovers

FDI Spillovers

10

FDI spillovers: computation

Proxy for the share of industry j’s

• utput produced by foreign firms

Proxy for the foreign presence in industries supplied by industry j

(linkages between MNEs and suppliers)

Proxy for the foreign presence in industries that supply industry j

(linkages between MNEs and clients)

Horizontal or vertical?

11

a b a b a b a b a b a b a b a b a b a b a b a b 3 1 2 3 3 Intermediate Consumption Industry Intermediate Consumption 1 2 1 2 3 Industry 1 2

IO-tables at aggregated and detailed level of industry aggregation

Importance of the diagonal

12

Consider specific sectors:

X = off-diagonal elements within the same 2-digit industry

X / within NACE 2-digit intermediate supply X / total intermediate supply

FOOD 26 % 12.1% CHEMICALS 47% 6.5% MINERAL PRODUCTS 37% 4.3%

Research topic in the paper

13

Additional research question:

• Inclusion of within-industry intermediate supply and

use of goods (include the diagonal of the IO-table) → BK as supplier-customer relationship → potential solution when tables are aggregated?

Empirical Strategy & Data

Empirical Strategy

15

• Havranek & Irsova (JIE, 2011): best practice

→ Javorcik (AER, 2004)

• FDI spillover analysis: two-step estimation procedure

in a production function framework

• Two-step estimation procedure
• first step: estimate total factory productivity (TFP)
• second step: relate the estimated TFP to FDI

spillover variables, control variables, time, industry and region dummies

Empirical Strategy

First step: estimate total factor productivity Issue: (potential) endogeneity between input choices and productivity

• OLS estimates will be biased
• Alternative methods: OP, LP, ACF, DPD
• Alternative specification: translog

16

Empirical Strategy

Second step: relate the estimated TFP to FDI spillover variables, control variables and time, region and industry dummies

Equation to estimate: TFPijrt = αi + ψ1 f(FDIjt-1) + ψ2 Zi(j)t-1 + ξijrt ∆TFPijrt = ψ’1 ∆f(FDIjt-1) + ψ’2 ∆Zi(j)t-1 + αt + αj + αr + εijrt

17

Data

18

• Romanian manufacturing firms with at least

five employees on average (1996-2005)

• Data sources:
• firm-level data: Amadeus database Bureau Van Dijk
• input-output tables: Romanian Statistical Office

→ detailed input-output table (NACE 3) → collapse to more aggregated level (NACE 2)

Data

Why Romania?

• Excellent coverage in the Amadeus database
• Characteristics of FDI in Romania:
• entry in the late 1990s
• concentrated in manufacturing industries

Note: stylized facts confirmed: foreign firms are larger (labour, capital, output) and more productive

19

Estimation Results

Estimation results

21

ACF OP DPD TL Agg. Det. Agg. Det. Agg. Det. Agg. Det.

HOR 1.908** 1.205*** 0.578** 0.344* 0.543** 0.327* 0.579** 0.346**

[0.734] [0.463] [0.240] [0.175] [0.239] [0.177] [0.242] [0.176]

BK 2.553 2.146** 1.426* 1.059*** 1.367* 1.071*** 1.287* 1.027***

[1.746] [0.958] [0.752] [0.330] [0.731] [0.325] [0.745] [0.328]

# obs. 73,255 73,255 96,681 96,681 96,728 96,728 96,728 96,728 R² 0.079 0.074 0.063 0.062 0.059 0.059 0.068 0.067

***/**/* denotes significance at 1/5/10 percent

Aggregated versus detailed input-output tables: zero-diagonal definition

Estimation results

22

The level of industry aggregation matters!

ACF OP DPD TL Agg. Det. Agg. Det. Agg. Det. Agg. Det.

HOR 1.908** 1.205*** 0.578** 0.344* 0.543** 0.327* 0.579** 0.346**

[0.734] [0.463] [0.240] [0.175] [0.239] [0.177] [0.242] [0.176]

BK 2.553 2.146** 1.426* 1.059*** 1.367* 1.071*** 1.287* 1.027***

[1.746] [0.958] [0.752] [0.330] [0.731] [0.325] [0.745] [0.328]

# obs. 73,255 73,255 96,681 96,681 96,728 96,728 96,728 96,728 R² 0.079 0.074 0.063 0.062 0.059 0.059 0.068 0.067

***/**/* denotes significance at 1/5/10 percent

Estimation results

23

The level of industry aggregation matters!

ACF OP DPD TL Agg. Det. Agg. Det. Agg. Det. Agg. Det.

HOR 1.908** 1.205*** 0.578** 0.344* 0.543** 0.327* 0.579** 0.346**

[0.734] [0.463] [0.240] [0.175] [0.239] [0.177] [0.242] [0.176]

BK 2.553 2.146** 1.426* 1.059*** 1.367* 1.071*** 1.287* 1.027***

[1.746] [0.958] [0.752] [0.330] [0.731] [0.325] [0.745] [0.328]

# obs. 73,255 73,255 96,681 96,681 96,728 96,728 96,728 96,728 R² 0.079 0.074 0.063 0.062 0.059 0.059 0.068 0.067

***/**/* denotes significance at 1/5/10 percent

Estimation results

24

The level of industry aggregation matters!

ACF OP DPD TL Agg. Det. Agg. Det. Agg. Det. Agg. Det.

HOR 1.908** 1.205*** 0.578** 0.344* 0.543** 0.327* 0.579** 0.346**

[0.734] [0.463] [0.240] [0.175] [0.239] [0.177] [0.242] [0.176]

BK 2.553 2.146** 1.426* 1.059*** 1.367* 1.071*** 1.287* 1.027***

[1.746] [0.958] [0.752] [0.330] [0.731] [0.325] [0.745] [0.328]

# obs. 73,255 73,255 96,681 96,681 96,728 96,728 96,728 96,728 R² 0.079 0.074 0.063 0.062 0.059 0.059 0.068 0.067

***/**/* denotes significance at 1/5/10 percent

Estimation results

25

The level of industry aggregation matters!

• Aggregated table: horizontal

↔ Detailed table: horizontal and backward

• Upward bias of horizontal spillover coefficient
• Bias against finding significant backward spillovers
• Results hold for FE and LP

Estimation results

26

Zero-diagonal or non-zero-diagonal?

***/**/* denotes significance at 1/5/10 percent

ACF OP

Agg. Det. Agg. Det. zero non-zero zero non-zero zero non-zero zero non-zero

HOR 1.908** 2.020** 1.205*** 0.712 0.578** 0.700* 0.344* 0.216

[0.734] [1.002] [0.463] [0.497] [0.240] [0.382] [0.175] [0.190]

BK 2.553 2.251 2.146** 2.344** 1.426* 0.918 1.059*** 0.923***

[1.746] [1.964] [0.958] [1.004] [0.752] [0.710] [0.330] [0.323]

# obs. 73,255 73,255 73,255 73,255 96,681 96,681 96,681 96,681 R² 0.079 0.076 0.074 0.074 0.063 0.060 0.062 0.061

Estimation results

27

Zero-diagonal or non-zero-diagonal?

ACF OP

Agg. Det. Agg. Det. zero non-zero zero non-zero zero non-zero zero non-zero

HOR 1.908** 2.020** 1.205*** 0.712 0.578** 0.700* 0.344* 0.216

[0.734] [1.002] [0.463] [0.497] [0.240] [0.382] [0.175] [0.190]

BK 2.553 2.251 2.146** 2.344** 1.426* 0.918 1.059*** 0.923***

[1.746] [1.964] [0.958] [1.004] [0.752] [0.710] [0.330] [0.323]

# obs. 73,255 73,255 73,255 73,255 96,681 96,681 96,681 96,681 R² 0.079 0.076 0.074 0.074 0.063 0.060 0.062 0.061

***/**/* denotes significance at 1/5/10 percent

Estimation results

28

Zero-diagonal or non-zero-diagonal?

• Aggregated table: no impact

↔ Detailed table: horizontal effect disappears

• No solution for the biases

Estimation results

29

Contribution to TFP level– sector 15

Estimation results

30

Contribution to TFP level– sector 24

Conclusions

Conclusions

32

• Literature: mixed evidence of FDI spillovers

→ channels, determinants, measurement

• In this paper:
• level of industry aggregation in IO-tables matters
• zero-diagonal versus non-zero-diagonal definition
• In the analysis of FDI spillover effects, use IO-tables

with a sufficiently detailed industry classification!