Transcript PowerPoint

Productivity and Taxes as Drivers of FDI
Heckman Selection model
Rahul Anand
Econ 764 class presentation
Outline

Theoretical framework of the determinants of FDI
flow

Analytical framework with productivity as a driving force



M&A FDI flows
Greenfield FDI flows
Extending the framework to include corporate taxation as
an additional driving force

Econometric Approach: Heckman Selection model

Empirical Evidence
Theoretical Framework

Focus on bilateral FDI flows among members
of OECD

Study of two set of driving forces


Productivity
Taxation
Important feature of this model: fixed set up
costs of new investments (distinguishing FDI
flows from portfolio flows)

Two margins of FDI decision



Intensive margin: determining the magnitude of
flows, based on standard marginal productivity
conditions
Extensive margin: whether to make new
investment at all
Productivity and Taxes may affect the two
margins in different, possibly conflicting ways
Set up costs can be industry-specific, giving
rise to two way rich-rich, as well as rich-poor,
FDI flows
A Stripped-Down Model of Foreign
Direct Investment

Source to host FDI flows typically include
many observations with zero flows, may be
an indication of existence of fixed set up
costs

A stripped down model of FDI with fixed set
up costs
Model






Consider a pair of source and host country
Free capital mobility, fixing world interest rate
at r
Host country: denoted by H
Source country : denoted by S
A representative industry whose product
serves for both consumption and investment
Firms last for two periods







In 1st period: continuum of NH firms that differ
by an idiosyncratic productivity factor, ε
Firm with productivity factor of ε is referred to
as ε –firm
G(.): Cumulative distribution of ε
g(.) : Density function
Number of ε-firms = NH*g(ε)
KH0 = initial net capital stock of each firm
(assumed same for all firms)
If invest I, augmented capital stock K = KH0 +I

Gross output in period 2: AHF(K,L)(1+ε)

AH = country (H) specific aggregate productivity
parameter
Assume, CH = fixed set up cost of investment (same
for all firms, independent of ε)


Fixed cost has two components

CSH = cost borne by FDI investor in his own country
(management time and other expenses at home
headquarters)
Cost incurred in host country: assume it involves labor
input only, LHC
CH  CSH  w L ,
C
H H
where wH : wage in the host country
F : strictly concave, exhibiting deminishin g returns to scales
and diminishin g marginal products of labor and capital
Average cost curve - U shaped so that perfect competitio n can
prevail
Consider an  - firm that invests I  K - K 0H in the first period
to augment its stock of capital to K
Its present va lue becomes V  ( AH , K H0 ,  , wH )  CH
Where
 AHF ( K , L)(1   )  wL  (1   ) K
0 
V ( AH , K H ,  , wH )  max
 ( K  K H )

( K ,L )
1 r


Where
 : physical rate of depreciati on
r : fixed rate of interest

0
Demand for K, K ( AH ,  , wH ) is given by

AHFK(K, L)(1   )  r  
Demand for L, L ( AH ,  , wH ) is given by

AHFL(K, L)(1   )  wH
lower bound on  is - 1
Upper bound is  that is G( )  1
However an  - firm may choose not to invest at all
and avoid lumpy set up cost.
In this case the present value is  A F ( K , L)(1   )  w L  (1   ) K 
V ( A , K ,  , w )  max 

1 r


Labor demand of the firm, L ( A , K ,  , w ) is given by
0

H
0
H
H
H
H
L
-
0
H
A F ( K , L)(1   )  w
0
H
K
H
0
H
H
H
H
H


Firm invests if PV(Investment)>PV(without
investment)
A firm with higher ε (higher productivity firm)
benefits more from investment
Invests if
V ( A , K , , w )  C  V ( A , K , , w )


0
H
H
H
0
H
H
H
H
From envelope theorem
V V
A  F [ K ( A ,  , w ), L ( A ,  , w )]  





 1  r  F [ K , L ( A , K ,  , w )]

We assume, K ( A ,  , w )  K and assume K and L




H
H
0
H

H
H
0
H
H

H
H
0
H
H
H
are complementary in production (F  0). Thus,
KL
L ( A , , w )  L ( A , K , , w )


H
0
H
H
H
H
Thus,
F [ K ( A ,  , w ), L ( A ,  , w )]  F [ K , L ( A , K ,  , w )]


H
H
0
H
H
H

0
H
H
H

Therefore a cutoff ε exists(ε0) such that an ε–
investment firm will make a new investment if
and only if ε >ε0

V+(AH,KHO, ε0,wH) –CH= V-(AH,KHO, ε0,wH)

wH is determined in equilibrium by a
clearance in the labor market
N
0
 0 ( AH ,CH , K H
,WH )

H
L ( A , K ,  , w ) g ( ) d

0
H
1
H
H
 N {1  G[ ( A , C K , w )]}L
0
H
N
0

H
H,
C
H
H
H
 L ( A , K ,  , w ) g ( ) d  L
H

0
H
H
0
 0 ( AH ,CH , K H
,WH )
Dividing by N
0
 0 ( AH ,CH , K H
,WH )

H
H
L ( A , K ,  , w ) g ( ) d

0
H
1
0
H
H
H
 {1  G[ ( A , C K , w )]}L
0
0


H
H,
C
H
H
H
 L ( A , K ,  , w ) g ( ) d  L

0
H
0
 0 ( AH ,CH , K H
,WH )
H
0
H
H





Labor abundance is manifested in wage differences
If labor per firm in host > labor per firm in source (i.e,
LHo> LSo )
In addition no. of firms –a measure of abundance of
entrepreneurship- abundance of labor means
scarcity of entrepreneurship
If wages equal then demand per firm is same in both
countries and market clearing condition could not
hold for both countries
Hence, wH< wS (Host employ more worker per firm,
in equilibrium Source firm effectively more
productive)
Mergers and Acquisitions FDI



M & A: acquisitions of existing host firms
Source country entrepreneurs endowed with
some intangible capital, or know how,
stemming from their specialization/expertise
comparative advantage:
Set up cost by source FDI investors in host
country <
Set up cost by host investors in host country

CH*= CSH* + wH LHC < CH

Foreign investor can bid up the direct
investors of the host country in the purchase
of the investing firm in the host country

Each firm (with ε >ε0 ) is purchased at its
market value, V+(AH,KHO, ε0,wH) –CH

New owner also invests: K+(AH,ε,wH) –KHO in
the firm
The amount of FD Investment made by a firm with ε >ε0

FDI ( A , C , K ,  , w )  V ( A , K ,  , w )  C  K ( A ,  , w )  K
*
H

0
SH
0
H
H
H

*
H
H
H
SH
H
0
H
(Acquisition cost is V  C  w L but w L is a part of FDI)

*
C
H
SH
C
H
H
H
Aggregate notional FDI is given by
FDI ( A , C , C , K , w )  N
N
H
*
*
0
H
SH
H
H

 FDI (A , C , K , w  ) g ( ) d
H
H
 0 ( AH ,C H* ,C S* H , wH )
*
0
SH
H
H,
FDI is the actual FDI only if
N
 ( AH , C , C , w )  
0
*
*
H
SH
Eq(12)
H
Otherwise it is zero. Actual FDI is denoted by
FDI ( A , C , C , K , w )  FDI ( A , C , C , K , w ) if eq(12) holds
*
A
H
H
*
SH
0
H
H
N
H
*
*
0
H
SH
H
0
Eq (12) : Selection Condition equation
Eq (11) : Flow equation
H
Otherwise
Eq(11)
Aggregate Productivity Shock: Flow and
Selection


How shock to AH affects FDI flows to the host
country?
Case 1: wH Fixed

Three positive effects of positive shock on the
notiaonal FDI on Flow equation



Raises the marginal productivity of capital, increasing
the amount of investment made by each investing firm
It raises the value of each such firms, increasing the
acquisition price which is a part of notional FDI
Increasing the number of firms purchased by FDI
investors (by lowering ε0 )
Derivation of results:
By total differenti on of FOCs w.r.t to K and L
K 
 FKFLL  FKLFL

0
2
AH AH ( FKKFLL  FKL )
Differenti ating the equation V   CH  V 
V  V 
sign (

)  sign[ F ( K H0 , L )  F ( K  , L )]  0
AH AH

Selection Condition Equation:



A rise in AH reduces the likelihood that ε0 exceeds
ε¯ (as profitability of investment increases, the
threshold condition for investment decreases)
So the likelihood of satisfying the selection
condition increases implying the realization of
notional FDI flow
Thus a positive aggregate productivity shock
raises actual FDI through both the flow and
selection condition equation
Case 2: wH not fixed



Wage is determined through market clearing
conditions
Increase in demand of labor raises wH ,
raising the fixed set up cost wH LHC
COUNTERING THE ABOVE THREE
EFFECTS
With unique equilibrium the initial effects are
likely to dominate these subsequent counter
effects, so the notional FDI still rises
(governed by the flow equation)

Effect on Selection Condition Equation


Increase in set up cost reduces the advantage of
carrying out positive FDI flows at all
As wH rises ε0 rises, reducing the likelihood of
satisfying the selection condition

So, the follow up effects of a positive shock
works in the opposite direction and may
dominate it

AS has no effect : as free capital mobility
Greenfield FDI

Establishing a new firm (where KHO =0)

Newcomer doesn’t know in advance ε of the
potential firm and takes G(.) as the CDF.

Assumption- ε is revealed before he decides
whether or not to make new investment
Expected Value of the new firm

V(A, C , w)   max{V ( A ,0,  , w )  C ,0}g ( )d

*
H
nH
H
1
nH
where C is the set up cost of greenfield investment
nH
Suppose, Greenfield entrepreneurship is in limited capacity
Can set up a firm at home or abroad but not in both
Invest in host country if and only if
V(A , C , w )  V(A , C , w )
Eq(15)
*
H
nH
*
H
S
nS
S
Selection Condition Equation for Greenfield Investment
Now A plays a role, positive shock to it increases the likelihood
S
that source country entreprenuers will stay at home, reducing
greenfield FDI flow

Entrepreneur must decide which host country to
invest in and should also outbid competitors from
other source countries
*
Then V(A ' , C ' , w ' ) in the selection condition eq(15) must
H
nH
H
*
be maximum over all V(A , C , w ) for potential other host countries
H
nH
H
V(A , C , w )  arg max V(A ' , C ' , w ' )  V(A , C , w )
*
H
nH
*
H
H 'D
H
nH
*
H
S
S
S
Where D is the set of potential host countries
Each entreprenuer in the source country who decides to make greenfield
FDI invests according to marginal productivity conditions.
Aggregatio n over these provides a flow equation of greenfield FDI.
Effect of Positive Productivity Shocks

Positive Shock to AH


Positive Shock to AS


Positive effects on both the notional FDI flows and
on the likelihood of these flows to actually
materialize
Doesn’t affect the notional flows but reduces the
likelihood of such flows to occur at all
Positive Shock to AH’(productivity of other
potential hosts)

Likelihood of having greenfield FDI is negatively
affected
Source Country and Host Country
Corporate Taxation


Tax rates of the source country and host
country may have different effect on the two
decisions (flow and selection condition
equations)
Case of a parent firm developing a new
product line


Develop it at home and produce it at a subsidiary
abroad
Determined by productivities and tax
considerations
Issue of Double Taxation




Income of a foreign affiliate typically taxed by
the host country
If source country also taxes: double taxation
Double taxation is typically relieved at the
source country level (by granting exemption
or granting tax credits)
In practice, foreign source income is far from
being taxed at the source country rate


Various reduced rates for foreign source income
Taxed only upon repatriation
Thus host country tax rate is more relevant for investment decision
Taxes affect decision by the different treatment of depreciation.
True rate of depriciation  
H
Rate allowed for tax purposes   '
H
For M & A FDI :
V ( A , K , ,  , w ) 

0
H
H
H
H
[ A F ( K , L)(1   )  w L](1   )    ' K  (1   ) K

Max
 ( K  K )
1  (1   )r


(Assuming subsidiary used debt in the host country tofinance new investment)
H
H
H
H
H
0
H
H
( K ,L )
H
Applying Envelope Theorem, V


0
H
FOC w.r.t K,
   '
Flow eq, since  '   flow declines with
1
Source country parent firm will undertake the project if and only if
w L (1   )  C (1   )  V ( A , K , ,  , w )
Eq(18)
A F ( K , L)(1   )  r 
H
H
H
H
K
H
H
C*
H
H

*
H
SH
S
0
H
H
H
H
H
H
Effect of tax rates

Tax rate in source country ςs affects positively
the decision by a parent firm to invest in host

ςH has a negative effect on this decision

ςs is irrelevant for the determination of the
magnitude of FDI flows which are negatively
affected by ςH
There is a cutoff level,  ( A , C , L , C , K , , , w )
H
0
H
C*
*
0
H
SH
H
H
S
H
Firms above  invest and acquired by FDI, while those below make no new investment
0
Cutoff level,  is implicity defined by eq(18)
0
An increase in  reduces  as it reduces the after tax source country set up cost
S
0
V declines in  but increase in  reduces the after tax host country set up cost

H
H
However, if the first effect dominates, increase in  raises 
H
0
Aggregate Notional FDI
FDI ( A , w L , C , K , , , w ) 
H
N
H
C*
*
0
H
SH
H
H
S
H

 0 ( AH ,CH
, LCH*

FDI( A , w L , C , K , , ,  , w )g ( )d
H
,C S* H
, K H0
, H , S , wH )
H
C*
*
0
H
SH
H
H
S
H
Where
FDI( A , w L , C , K , , ,  , w )  V ( A , K , ,  , w )  C (1   )  K ( A , ,  , w )  K
H
H
C*
*
0
H
SH
H

H
S
0
H
H

*
H
H
H
SH
S
H
H
H
0
H
Actual FDI is equal to notional FDI only when  is below 
0
 ( A , C , L , C , K , , , w )  
0
H
H
C*
*
0
H
SH
H
H
S
Eq(21), it is the selection condition equation
H
The actual flow of FDI is
FDI ( A , w L , C , K , , , w )  FDI ( A , w L , C , K , , , w )
A
H
H
C*
*
0
H
SH
H
H
S
H
N
H
H
C*
*
0
H
SH
H
H
0
S
H
if condition (21) holds
Otherwise
Increase in  reduces actual FDI flow and the likelihood that such flow will occur
H
Increase in  increases the likelihood that FDI flows will occur
S
Econometric Approach

FDI flow two fold decisions:
 Whether to invest: “threshold” selection eq.
 How much to invest: “flow/gravity” eq.

There may be zero actual flows

Hence, selection of actual (s,h) countries
endogenous (as opposed to exogenous in
traditional gravity models)

So, Heckman Selection Model used
Heckman Selection Model

Jointly estimate the likelihood of surpassing
the threshold and the magnitude of the FDI
flow (provided threshold is surpassed)
Flow Equation
Y Y *  X β u
if π *  0
Y 0
if π  0
Y * : latent var iable denoting flow of notional
FDI from source country i to host country j
in period t
X : explanator y variables
 : coefficien t vector
u : error term
Y : actual flow of FDI
ijt
ijt
ijt
ijt
ijt
ijt
ijt
ijt
ijt
ijt
ijt



Y*ijt can be positive or negative
Yijt is zero not only when Y*ijt is negative but
also when Y*ijt is positive but below the
threshold profit
π*ijt = π*’ijt / σ π*’ = (Wijtρ– Cijt) / σ π*’





π*’ijt : indicates if FDI will be made or not
(depending on positive or negative)
Wijt : explanatory variables
Cijt : fixed cost of setting up new investment
ρ : vector of coefficients
σπ*’ : standard deviation of π*’
Set up cost

C*ijt = Aijt δ + vijt
 Aijt : explanatory variables
 δ : vector of coefficients
 vijt : error term

Substituting for C*ijt in the previous equation

π*ijt = Zijtθ + εijt
Where,



Zijt = (Wijt, Aijt)
θ = (ρ/σ π*’ ,- δ/σ π*’ )
εijt = - vijt/σ π*’

Assuming and uijt and vijt are normally distributed
with zero means. It follows that εijt is N(0,1). The
error terms εijt and uijt are bivariate normal:
u    0  
    N  , 
    0   
  
2
ijt
ijt
Y * *
Y*
Y * *
Y*
1
Y*
 

Defining the Indicator function
if if   0
1
D 
0
*
ijt
ijt
otherwise
 itself is a latent variable which is not observed.
We observe D, that is, we do observe whether  is positive or not
E(Y / D  1)  X   E (u / D  1)  X    
*
*
ijt
ijt
ijt
Where
  

Y * *
ijt
ijt

ijt
ijt
Y*
and
 (Z  )
 
is the inverse Mills ratio
( Z  )
ijt
ijt
ijt
 ,  are the density and cummulative unit normal distribution functions
Because Prob(D  1)  Pr ob(  0)  Pr ob(   Z  )  Pr ob(  Z  )
*
ijt
ijt
Pr ob(D  1)   ( Z  )
ijt
ijt
ijt
ijt
ijt
ijt

Maximum Likelihood method is employed to
estimate the flow coefficient vector β and the
selection coefficient vector θ

λijt depends on Xijt . So from equation 7.8.
OLS estimates of β confined to positive
observations of Yijt is biased because such
estimates include also the effect of Xijt on Yijt
through the term βλ λijt
Yijt
Y*=βXX
B
B’
Y=bXOLSX
R
M’
A’
M
A
T’
T
S
XL
XH
Xijt
The Tobit Model

Actual FDI flow may be zero even when
notional FDI flows are not. A significant
portion of a typical sample is zero, but is
roughly continuously distributed for positive
values. In such cases Tobit model is often
employed.
Y  X   
Y  
0
if Y  0
*
ijt
ijt
*
ijt
ijt
if Y  0
ijt
*
ijt
Where
  N (0,  )
2
ijt
Y
It can be shown
E(Y  / D  1)  X     
ijt
ijt
where
 

ijt

ijt
*
Y
and
 (X  /  )
 
 (X  /  )
*
ijt
ijt
Y
*
ijt
Y
Comparing with Heckman equation suggests that
Tobit is a special case of Heckman model when
Y and  are fully positively correlated, that is
*
*
ijt
ijt

Y* *
1
If the estimated value of 
Y* *
in Heckman model is
significan tly below 1, then the tobit estimates are biased.
Empirical Analysis
Productivity and Tax Rate as
Determinants of FDI flows
Data and Descriptive Statistics

Standard Mass Variables


Distance Variables



Source and host population sizes
Physical distance
Whether two countries share a common language
Economic Variables



Source and host GDP per capita
Difference in years of schooling
Financial risk rating

Control for country and time fixed effects

Dependent Variable in all flow equations:
Log of FDI flow

The main variables are grouped as follows:

Standard Country Characteristics




Real GDP per capita
Source and host GDP per capita
Difference in years of schooling
Financial risk rating

Source and Host Characteristics





Physical distance
Whether two countries share a common language
Productivity
Corporate tax rates
Productivity


Approximated by output per worker- measured by
purchasing power parity-adjusted real GDP per
worker
At times instrumented by capital to labor ratio and
years of schooling

Corporate Tax Rates





Statutory rates or by the effective average rates
compiled by Devereux, Griffith, and Klemm
At times instrumented by the statutory corporate
tax rates and GDP per capita
No smoothing of data done: to investigate the
effects of the explanatory variables over the
business cycle
Data on FDI flows from International Direct
Investment (IDI): bilateral flows among 18
OECD countries during 1987-2003
Deflated by US CPI (Urban Consumers)
Empirical Evidence



Since labor productivity and FDI flows both
affected by other variables not controlled for
(such as business cycle variables, interest
rate and unemployment rate), alternatives in
results presented
First regression: only labor productivity
Second regression: labor productivity is
instrumented by the capital-to-labor ratio,
years of schooling and country fixed effects

Tax Variables

First, statutory tax rates

Another alternative is the effective tax rates
compiled by Devereux, Griffith, and Klemm
(difference between the cost of capital in the
corporate sector and the tax-free interest)

Also use the statutory corporate tax rates, GDP
per capita, and country fixed effects as
instruments to generate the fitted values for the
effective tax rates
Predicted Effects
M&A
Flow
Greenfield
Selection
Flow
Selection
Productivity increase, fixed
wages
Host
+
+
+
+
Source
0
0
0
-
+
Amb
+
+
Host
-
-
Source
0
+
Productivity increase, flexible
wages
Host
Source
Tax Increase
Results

Table A-2: Instrumented productivity and tax
equation


Coefficients on capital-to-labor, years of schooling
are significant and positive
Statutory tax rate and GDP per capita are positive
and significant
Productivity as a Driver

Table 4: column 1-uninstrument
productivities, column 2- instrumented
productivities



Source GDP per capita has a positive and
significant effect on flow equations in both
columns
Host GDP per capita has a positive and significant
effect on flow equation in column 2 only
Neither significant in selection equation


Existence of previous FDI (a dummy) may be
indicative of low setup costs. Used as an
exclusion restrictions in the selection
equation. It is significant and positive
Column 1 of Table 4
Productivity host: positive in both flow and
selection but significant in only flow
 Productivity source: negative and significant in
selection
Both results consistent with the analytical framework


Column 2, Table 4:


Productivity host: not significant
Productivity source: negative and significant in
flow and selection

Results consistent with the model

Figure 2 and Figure 3
Tax Variables

First 3 columns of Table 5:






1st column – statutory tax rate
2nd column – effective tax rates
3rd column – fitted effective tax rates
Host tax rate – negative and significant effect on
flow of FDI in flow equations
Source tax rate - positive and significant effect on
flow of FDI in flow equations
Source tax has a positive and significant effect on
the selection mechanism (as predicted by theory)
but only in column 1 (it intensifies in column 4, with
larger set of countries).

Figure 4 and 5

When both set of drivers are used, a problem
of multicolinearity arises. Estimates shown in
A-3. Results don’t change in sign but there
statistical significance weakens.
Conclusion



Examined the role of productivity and
corporate taxation as driving forces
Important feature – fixed set up costs
FDI flows:



M&A
Greenfield
Differ in that alternate investment opportunities in
host countries do not affect M&A


Effect of productivity shock on M&A
considered.
Under fixed wages it has three positive
effects on notional flow of FDI




Raises the marginal productivity of capital, increasing
the amount of investment made by each investing firm
It raises the value of each such firms, increasing the
acquisition price which is a part of notional FDI
Increasing the number of firms purchased by FDI
investors (by lowering ε0 )
Selection Equation

Shock increases profitability hence notional FDI is
realized

When wage rate flexible




The three effects are countered by raising wage in
the host country
In selection condition the rising host country
component of set up cost reduce the likelihood of
positive FDI flows to occur.
Source country shocks does not affect M&A
Greenfield FDI


Positive shock has positive effect on notional
flows and likelihood of flow
Positive source country shock reduces the
likelihood of FDI flows




Empirical findings support the claims of the
model
Host output per worker has a positive effect
in both the flow and selection equation, but
significant only in flow equation
Source country output per worker has a
negative and significant effect on the
selection mechanism
Results are robust
Tax as driver





Host tax rate has negative and significant
effect on flow of FDI in flow equation
Source tax rate has positive and significant
effect in flow equation
Magnitude increasing (source country tax has
a depressing effect on their investment
abroad)
Results fairly robust
Source tax – positive and significant effect on
the selection mechanism




Effects intensifies for larger set of countries
Simulations suggest marked differences in
the sensitivity of flows from the US to OECD
countries
Sensitivity to flow in UK is positive and higher
than EU and Japan.
The sensitivity of these flows to taxes in UK is
negative and high relative to other countries
Empirical Analysis
Existence of Fixed Costs
Data and Variables

OECD Countries (from OECD reports)




24 countries
1981-1998
FDI export from 17 OECD source countries to 24
OECD countries
Three year average (to smooth the
variables): six periods
Explanatory Variables

Country Characteristics






GDP or GDP per capita
Population size
Educational attainment (average years of schooling)
Language
Financial sound rating (inverse of financial risk rating)
Source-Host Characteristics




Geographical distance
Common language (zero-one variable)
Flows of goods
Bilateral telephone traffic per capita (proxy for
international distance)
Estimation


GDP per capita : good predictor of direction of
the direction of flows
Frequency of flows



Close to one among rich countries
Very low or close to zero among poorer countries
Source-Host differences in GDP per capita are
not correlated with the volume of FDI flows
(among the subset of country pairs with positive
flows)

Japan got 1.26% of its GDP from US, whereas Spain
received 6.54% of its GDP from US
Estimation of the Determinants of
bilateral FDI flows

Standard Mass Variables


Distance Variables



Source and host population sizes
Physical distance
Whether two countries share a common language
Economic Variables



Source and host GDP per capita
Difference in years of schooling
Financial risk rating

Control for country and time fixed effects

Dependent Variable in all flow equations:
Log of FDI flow deflated by the unit value of
manufactured goods exports
Econometric Procedure Adopted
1.
As a benchmark, ignoring the selection equation
and estimating the gravity equation twice

By treating all FDI flows in (s,h) pairs with no recorded FDI
flows as “zeros” (OLS-zero)

By excluding country pairs with no FDI flows (OLS-D)

Assign a negligible value as a common low value for the
value of the FDI flows for the zero-flow (s,h) pairs

Here the lowest observed flow between any (s,h) country pair in
the sample is chosen

Rationale of including “zeros” in OLS-zero
case:

Observed zero flows could be because:



Two countries don’t wish to have such flows even in the
absence of fixed costs
Set up costs are prohibitive
Measurement error
So under the assumption of no set up costs and
measurement errors, (s,h) pairs with zero FDI
flows truly indicate zero flows
Tobit

Assumptions




No fixed costs
All FDI flows that are below a certain low
threshold level (“censor”) are due to measurement
errors
Tobit estimator
Three “censor” levels considered



Lowest
0.0
3.00
Heckman Selection Model



To highlight the role of fixed setup costs
against the two benchmarks
Jointly estimate the maximum likelihood of
the flow equation and the selection equation
It accommodates both the measurement
errors and a possible existence of setup
costs

Binary Variable Dijt



Di,j,t = 1 if country i exports positive FDI to country j
at time t
Di,j,t = 0 , otherwise
Assuming that the setup costs are lower if
country i has invested in country j in the past

Di,j,t-k could serve as an instrument in the selection
equation (exclusion restriction)
Results

All results confirm volume of FDI flows is not
affected by deviations from long run averages
of GDP per capita in the source and host
countries

Host country education level, relative to
source-country counterpart:


Tobit: significant effect on flow of FDI
Heckman: manifests through selection and has no
significant effect on the flow of FDI

Nonlinearity in FDI flows: parameters of
interest in the OLS method estimated for
different range of FDI flows

OLS-zero has different coefficients from those of
OLS-D regression

Common language dummy: positive and
significant

Distance: negative and significant in all
formulations

Host country financial sound rating:


Significant (and positive) in only Heckman flow
equation and OLS-D case
Source country financial sound rating


Significant and negative effect on FDI in the tobit
cases and one of the two OLS cases (OLS-zero)
Heckman method suggests that it works through
the selection process rather than having a direct
effect on FDI flows

Existence of previous FDI flows:


Significant and positive effect in the selection
effect indicating that the existence of FDI flows in
the past reduces the fixed cost of setting a new
FDI
Correlation between the error terms in the
flow and the selection equations is negative
and significant – additional evidence for the
relevance of fixed set up costs




Few cases of negative flows in the sample,
indicating liquidations of previous FDI
A dummy variable for negative FDI flows as
an instrument (reasonable as past FDI
liquidations are correlated positively with
past FDI flows, but not a priori correlated with
current FDI flows)
Tobit: positive role to this dummy for flows
(Table 8.6)
Heckman: positive effect comes through the
selection mechanism
Evidence of Fixed Costs

Significant correlation ρ between the error
terms in the flow and selection equations
indicates that the formation of an (s,h) pair of
positive FDI, and the size of the FDI flows
between this pair of countries are not
independent process

ρ negative is consistent with the setup cost
hypothesis
Conclusion

Some evidence of fixed costs

In such case the OLS and Tobit estimates of
the determinants of FDI flows is biased

Heckman method suggests some of the
determinants of FDI flows in OLS and Tobit in
fact influence FDI through selection
mechanism rather than directly through the
flows of FDI
Thanks