Transcript The effect of human capital on FDI: A meta-regression

The effect of human capital on FDI:
A meta-regression analysis
Artane Rizvanolli, AAB-Riinvest University
Ancona, 21 May 2010
Contents






Introduction: FDI and growth
Rationale for MRA
Sample
MRA model
Empirical results
Conclusion and further research
Introduction: FDI and growth

FDI conventionally considered beneficial




Especially important for transition economies



Technology and know-how transfer (?)
Spillovers (?)
Hence, overall productivity and growth (?)
Need for restructuring and modernisation (at firm and economy
level)
Limited domestic resources
However, are the benefits automatic?
The rationale for meta-regression analysis (MRA)
•
Theory: human capital (HC) attracts FDI
–
•
No consensus in the empirical literature
–
•
Enhancement of productivity, technology adoption and
adaption
Negative, positive and insignificant results found
Potential reasons for the diversity of results?
–
–
Wide range of specifications, HC measures, countries
Lack of “universal” relationship between HC and FDI:
differences in motivation for FDI, sector of economic
activity, etc.
The rationale for meta-regression analysis (2)
MRA as a means of





Quantifying a survey of empirical literature
Analysing the sensitivity of results to different study
characteristics (!)
Identifying and quantifying the “genuine” effect of HC, if present
Identifying publication bias (?)
Informing the specification of further research on the HC-FDI
relationship: which measures?
Sample

Around 30 regression analyses identified



Some excluded




EconLit, SSRN, Google Scholar
References in papers
Measures not convincing/comparable
No results reported
Only interaction/squared terms
Preferred regressions only (?)
Sample (2)

28 studies with a total of 231 regressions

t-stats range -7.8 - 7.7, with a mean of 0.93

Structure:



Developing, transition, mixed, China, developed
Mostly secondary and tertiary education measures
Majority(static and dynamic) panels
Model


Linear regression: weighted to give each study the
same weight, clustered robust (cluster: study),
dependent variables divided by SEpcc
Dependent variable: t-statistic of HC variable
Moderator variable
Description
Constant
Provides an estimate of publication bias (bias across the
whole range of results in the literature)
1/SEpcc
SE of the PCC (standardised measure of association) – a
precision measure; provides an estimate of the “true”
effect in the literature in terms of the PCC
FDIFLOW
Flow measures for FDI used
FDIREL
FDI measured relative to population/GDP
HCFLOW
Flow measures for FDI (enrolment, decision to invest)
Model (2)
Moderator variable
Description
LITERACY , PRIMARY,
TERTIARY, SECTER,
AVGYRED
HC measure: Literacy/illiteracy rate, primary education,
tertiary education, secondary and tertiary combined,
average yrs of education (RC: secondary education)
PANEL, DYNAMIC P.,
QUALITYDV
Static panel, dynamic panel, quality dependent variable model
(RC: cross-section)
DEVELOPED,
TRANSITION, MIXED,
CHINA
Sample according to group of countries (RC: Developing countries)
HCCOST
If model controls for HC cost
HCPROD
If model controls for HC productivity
PUBYR
Year of publication (working paper)
MEDIANYR
Median year of the period covered in the study
NOEXPVAR
Number of explanatory variables in the model (includes FEM
dummies)
ENDOGENEITY
If attempts were made to address endogeneity
Preliminary results
•
Bi-variate MRA
–
•
no publication bias OR “genuine effect”
Dependent
variable
Coefficient
t-statistic
p-value
Con
0.60
0.99
0.33
INVSEEpcc
0.04
1.28
0.31
Multi-variate MRA
–
–
Same result as above
Full model mis-specified
–
–
Ramsey RESET test : F(3, 205) = 94.52 , Prob > F =
Suffers from multicollinearity
0.0000
Preliminary results (2)

Testing down: standard procedure in MRA


Improves functional form
Significantly reduces multicollinearity

Some variables highly correlated with INVSEPCC (PERIOD,
MEDIANYR, LABCOST, TNOEXPVAR?, EDNOGENITY?,
HCSTOCK?)
Preliminary results (3)
Variable
Coefficient
p-value
Constant
-0.014
0.96
INVSEEpcc
-0.002
0.96
CROSS
0.127
0.19
QUALDV
0.228
0.00
MIXED
0.091
0.01
DEVELOPED
0.152
0.05
TRANSITION
0.113
0.10
CHINA
0.143
0.00
AVGYRED
0.081
0.13
TERTIARY
-0.029
0.41
LABPROD
0.043
0.27
PRIMARY
-0.027
0.60
DYNAMIC
0.003
0.89
-0.044
0.25
FDIREL
Conclusion and further research


Heterogeneity in HC-FDI literature can be explained to a
very limited extent (!)
Appears to be no genuine effect in the literature:


Models not specified correctly?
Further research: specify model in accordance with theory

human capital variable: level and stock/flow
Thank you!
Questions & Comments?