Download Slides 4

CES Lecture
Spacey Parents vs Spacey Hosts in FDI
Harald Badinger
Department of Economics
Outline
I
Introduction
I
I
Modes of Interdependence in FDI
Previous Studies
I
Specifying Spaceyness
I
Empirical Model
I
I
Basic Model
Specifying Spaceyness
I Market-size related spaceyness
I Remainder spaceyness
I
Econometric Issues
I
Estimation Results
I
Conclusions
Outline
2 / 22
Modes of Interdependence in FDI
I
“. . . both export-platform and complex-vertical motivations imply that
FDI decisions are multilateral in nature . . . ”
(Blonigen et al., 2007, p. 1304).
I
“Although [previous] work has focused on outward investment and the
choice among host locations, it is just as important to recognize that
third country effects may be important for inbound FDI as well.”
(Blonigen et al, 2008, p. 183).
Introduction
3 / 22
Modes of Interdependence in FDI
I
Spacey-Hosts-Hypothesis: (Shocks on) a given parent country’s
outward FDI in some host country affect the same parent country’s
FDI in other host countries.
I
Spacey-Parents-Hypothesis: (Shocks on) a given host country’s inward
FDI from some parent country affect inward FDI from other parent
countries in the same host country.
I
Spacey-Third-Countries: in multi-country general equilibrium, bilateral
FDI for a given parent-host country pair will also affect and depend on
(determinants of and shocks on) FDI between other, third parent and
host countries.
Introduction
4 / 22
Previous Studies
I
I
I
Supportive to Spacey-Parents and Spacey-Hosts Hypothesis
Assessed only in isolation from each other
No study has considered the role of spacey third countries
Introduction
5 / 22
This Paper
I
I
I
Considers all three modes of interdependence simultanteously
Distinguishes market-size related from remainder spaceyness
Cross-sectional data on FDI among 22 OECD countries in 2000
Introduction
6 / 22
Specifying Spaceyness
I
I
I
I
Denote (the log of the stock of outward) FDI from parent country
i in host country j by yij .
With I parent countries and J host countries, there are
N = I × (J − 1) observations, which can be collected in the
N × 1 vector y ≡ {yij }.
The structure of interdependence between FDI of two parent-host
country pairs is then reflected in the N × N matrix S ≡ {sij,i 0 j 0 }.
The elements sij,i 0 j 0 depict the interdependence between FDI from
parent i to host j and FDI from parent i 0 to host j 0 .
Specifying Spaceyness
7 / 22
Specifying Spaceyness:
Spacey Hosts
I
I
I
I
The elements sij,i 0 j 0 depict the interdependence between FDI from
parent i to host j and FDI from parent i 0 to host j 0 .
Spacey-Host relationships: {sij,i 0 j 0 } =
6 0 for i = i 0 , j 6= j 0 , whereas
0
0
{sij,i 0 j 0 } = 0 for i 6= i or j = j .
We collect these Spacey-Hosts relationships in the matrix
H
S H ≡ {sij,i
0 j 0 }.
Elements are specified as (inverse) ‘economic distance’ between
two host countries j and j 0 :
H
sij,i
0 j 0 = exp(ln GDPj + ln GDPj 0 − ln DISTjj 0 ).
Specifying Spaceyness
8 / 22
Specifying Spaceyness:
Spacey Parents
I
I
I
I
The elements sij,i 0 j 0 depict the interdependence between FDI from
parent i to host j and FDI from parent i 0 to host j 0 .
Spacey-Parent relationships: {sij,i 0 j 0 } =
6 0 for i 6= i 0 , j = j 0 ,
0
whereas {sij,i 0 j 0 } = 0 for j 6= j or i = i 0 .
We collect these Spacey-Parents relationships in the matrix
P
S P ≡ {sij,i
0 j 0 }.
Elements are specified as (inverse) economic distance between
two parent countries i and i 0 as
P
sij,i
0 j 0 = exp(ln GDPi + ln GDPi 0 − ln DISTii 0 ).
Specifying Spaceyness
9 / 22
Specifying Spaceyness:
Spacey Third Countries
I
I
I
I
The elements sij,i 0 j 0 depict the interdependence between FDI from
parent i to host j and FDI from parent i 0 to host j 0 .
Spacey-Third Countries relationships: {sij,i 0 j 0 } =
6 0 for
i 6= i 0 , j 6= j 0 , whereas {sij,i 0 j 0 } = 0 for i = i 0 or j = j 0 .
We collect these Spacey-Third-Countries relationships in the
T
matrix S T ≡ {sij,i
0 j 0 }.
Elements are specified as (inverse) economic distance between
two third countries i and j, i.e.,
T
sij,i
0 j 0 = exp(ln GDPi + ln GDPj − ln DISTij ).
Specifying Spaceyness
10 / 22
Specifying Spaceyness:
Summary
I
Summing up, we have defined three N × N (weights) matrices reflecting one
mode of interdependence each, corresponding to the Spacey-Parents
hypothesis (S P ), the Spacey-Hosts hypothesis (S H ), and the
Spacey-Third-Countries hypothesis (S T ):
(
0 for j 6= j 0 or i = i 0 ,
P
S P ≡ {sij,i
0 j0 } ≡
exp(ln RGDPi + ln RGDPi 0 − ln DISTii 0 ) otherwise,
(
0 for i 6= i 0 or j = j 0 ,
H
S H ≡ {sij,i 0 j 0 } ≡
exp(ln RGDPj + ln RGDPj 0 − ln DISTjj 0 ) otherwise,
(
0 for i = i 0 or j = j 0 ,
T
S T ≡ {sij,i
0 j0 } ≡
exp(ln RGDPi + ln RGDPj − ln DISTij ) otherwise.
To rule out self-influence the main diagonal elements of the weights matrices
P
H
T
0
0
are set to zero, i.e., sij,i
0 j 0 = sij,i 0 j 0 = sij,i 0 j 0 = 0 for i = i , j = j . Moreover,
each weights matrix is row-normalized to ensure well-behaved asymptotics.
Specifying Spaceyness
11 / 22
The Basic Empirical Model
Starting point is the following parsimonious specification:
yij = β0 + β1 ln GDPi + β2 ln GDPj + β3 ln GDPPCi
(1)
+β4 ln GDPPCj + β5 ln DISTij + β6 CBij + q ij δ + uij .
I
Dependent variable yij : (natural log of the) stock of nominal outward FDI
from parent i to host j.
I
RHS variables: GDP, GDP per capita (GDPPC ) of the parent and host
country, geographical distance between the parent and the host country
(DISTij ), common border dummy (CBij ), a set of controls related to parent
and host countries’ institutional quality (q).
I
The cross-sectional sample refers to the year 2000 and comprises
i = 1, . . . , I = 22 parent and j = 1, . . . , J = 21 host countries, making a
total of N = 462 observations.
I
Data: Bilateral FDI stocks from the UNCTAD and OECD, GDP from the
World Bank’s WDI, geographical variables from CEPII, institutional variables
from ICRG.
Empirical Model
12 / 22
Specifying Spaceyness:
Market-Size Related Interdependence
I
I
I
Measuring (economic) country size in terms of GDP, the
market-size-related Spacey-Parents hypothesis is incorporated
into Equation (1) by including an (economic) distance weighted
average of all parent countries’ GDP.
We collect these distance weighted averages of GDP in the N × 1
vector s̄P ≡ {s̄ijP }.
For observation ij, we have
s̄ijP =
I X
J
X
P
sij,i
0 j 0 ln GDPi 0 .
i 0 =1 j 0 =1
Empirical Model
13 / 22
Specifying Spaceyness:
Market-Size Related Interdependence
I
Accordingly, the market-size-related Spacey-Hosts hypothesis is
associated with the explanatory variable s̄H ≡ {s̄ijH }, given by
s̄ijH =
I X
J
X
H
sij,i
0 j 0 ln GDPj 0
i 0 =1 j 0 =1
and the market-size-related Spacey-Third-Countries hypothesis is
associated with explanatory variable s̄T ≡ {s̄ijT }, given by
s̄ijT =
I X
J
X
T
T
sij,i
0 j 0 ln GDPi 0 j 0 ,
i 0 =1 j 0 =1
where GDPijT is GDP of the rest of the world (ROW), i.e.,
GDPiT0 j 0 ≡ GDPW − GDPi − GDPj .
Empirical Model
14 / 22
Specifying Spaceyness:
Remainder Interdependence
I
While market-size-related spillovers are possibly the predominant
source of interdependence, we do not a priori expect them to capture
spaceyness in FDI completely.
I
We summarize channels of interdependence unrelated to market size
as remainder interdependence.
I
Since reminder interdependence in FDI is not included as an
explanatory variable in Equation (1), it will be captured by the error
term uij and can thus be modeled and estimated by using a spatial
regressive disturbance process:
u = ρP S P + ρH S H + ρT S T + ε,
(2)
where ε ≡ {εij } is an idiosyncratic error term, which is assumed to be
independently though not necessarily identically distributed, i.e.,
εij ∼ i.d. (0, σij2 ).
Empirical Model
15 / 22
Specifying Spaceyness:
Remainder Interdependence
I
The spatial regressive specification implies that a unit shock to the
error term ε, denoted as e, has a magnified impact on FDI through
spillover effects, given by (I − ρP S P − ρH S H − ρT S T )−1 e.
I
With row normalized weights matrices the multiplier effect is equal
across observations and amounts to 1/(1 − ρP − ρH − ρT ).
I
In contrast to the specification of market-size-related interdependence,
the spatial regressive structure does not allow a linearly additive
decomposition of the effects of the three modes of interdependence.
I
However, the specification still allows to test for the significance of
each single mode of interdependence, the direction of their effects, and
the respective coefficients are indicative of their relative magnitude.
Empirical Model
16 / 22
LS and 2SLS Estimation of
Market-Size Related Interdependence
I
For estimation of the the main equation, a two-stage least squares (2SLS)
approach is used, instrumenting spatially weighted parent GDP (s̄P ), host
GDP (s̄H ), and ROW GDP (s̄T ) by spatial weights of (the log of)
population.
I
To address potential endogeneity concerns related to the interdependence
matrices S, their elements defined in Equation (11) are constructed using
GDP in 1995 (whereas the variables refer to 2000).
I
For conservativeness, we also pursue an alternative approach and construct
weights matrices P ≡ {pij,i 0 j 0 } that are based on a purely population and
geography related gravity model (Frankel and Romer, 1999). Their elements
are defined as pij,i 0 j 0 = exp(ln POPi + ln POPj − ln DISTij ).
I
Summing up, the instruments that will be used for s̄P , s̄H , and s̄T in the
estimation of the main equation are given by p̄ P ≡ P P p P , p̄ H ≡ P H p H ,
and p̄ T ≡ P T p T , where p P , p H , and p T are N × 1 vectors of (the log of)
parent, host, and ROW population respectively.
Econometric Issues
17 / 22
Generalized Moments (GM) Estimation
of Remainder Interdependence
I
Having obtained consistent estimates of the parameters of the main equation,
the residuals û can be used for estimation of the error process.
I
Our estimation procedure builds on the GM estimator by Kelejian and
Prucha (2010) and its extension to higher order spatial regressive models by
Badinger and Egger (2011). The moment conditions are:
n
h
i
oi
h
0
2
N −1 E(ε0 S 0m S m 0 ε) − Tr S m 0 diagN
= 0,
n=1 E(εn ) S m
N −1 E(ε0 S m ε) = 0,
where m 0 = {P, H , T } for m = {P}, m 0 = {H , T } for m = {H }, and
m 0 = {T } for m = {T }, i.e., there is a total of 6 moment conditions. (Tr
indicates the trace operator.)
I
Again, to avoid endogeneity concerns with respect to the weights matrices S,
we will use both the preferred measure of economic distance, as well as
population-based weights matrices P.
Econometric Issues
18 / 22
Spatial Feasible Generalized
Two-Stage Least Squares Estimation
I
Having estimated the disturbance process, the main equation can be
estimated more efficiently by FGLS.
I
The spatial feasible generalized two-stage least squares (FG2SLS)
transformed version of the main equation is given by
T̂ y
= T̂ Xβ + T̂ Qδ + γP T̂ s̄P + γH T̂ s̄H + γT T̂ s̄T + T̂ u,
where T̂ = (I − ρ̂P S P − ρ̂H S H − ρ̂T S T ) is the estimated spatial GLS
transformation matrix, T̂ û = ε̂ is a consistent estimate of ε, and the
transformed instruments are given by T̂ p̄ P , T̂ p̄ H , and T̂ p̄ T .
Econometric Issues
19 / 22
LS and 2SLS Estimates
Estimation Results
20 / 22
Spatial FG2SLS Estimates
Estimation Results
21 / 22
Conclusions
I
Spaceyness matters both in terms of market-size-related and
remainder interdependence.
I
I
Larger and close-by parents investing in a particular host increase the
magnitude of a given parent’s FDI there relative to the FDI of smaller and
more distant parent countries.
Larger hosts in the neighborhood of a particular host increase a given
parent’s FDI there relative to its FDI in smaller and more distant host
countries (export-platform FDI, information spillovers).
I
Support of the Spacey-Parents and the Spacey-Hosts hypotheses with
regard to remainder (unobservable) determinants of FDI.
I
Some but less support of the Spacey-Third-Countries hypothesis.
I
Suggests relative dominance of interdependence through
learning/horizontal motives (export platforms) and vertical integration
motives (intermediate goods trade) over general equilibrium effects
(resource constraints, multilateral factor and output price effects).
Conclusions
22 / 22