Local Spillovers from Foreign Direct Investment in the Ethiopian Manufacturing Sector
1. Local Spillovers from Foreign Direct
Investment in the Ethiopian Manufacturing
Sector
Presentation of preliminary results from ongoing research. Major Research Grant No. 1386
Project title: Ref 1386 "The Role of Asian Investment in Manufacturing in Catalyzing Economic
Development in Africa"
Friday, July 24, 2015.
EEA-Annual Meeting, Addis Ababa, Ethiopia
Maggie McMillan (Tufts-IFPRI, PI), Girum Abebe (EDRI), Deborah Brautigam (SAIS -
Johns Hopkins), Iñigo Verduzco-Gallo (IFPRI), and Michel Serafinelli (University of
Toronto)
3. Motivation
Contributions (we hope)
• Policy:
– Big push in Ethiopia to industrialize; to large extent, relying on FDI
– Justification is that this will have significant spillover effects on the rest
of economy
– Our paper aims to evaluate these claims and understand effects of FDI
in Ethiopia
• Academic:
– Contribute to agglomeration and FDI literature.
– Little to no credible evidence on magnitude of FDI and agglomeration
spillovers in manufacturing in less-developed countries (particularly in
Africa)
– We have some data on potential mechanisms behind spillovers so we
can look into what might drive the results
4. Significant inflows of FDI since 2000
Source: Own calculations with data from CSA
0
20
40
60
80
100
120
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
Stock of FDI capital in
billions of Birr
FDI stock 1992-2014
Source: EIC 2015
0
5,000
10,000
15,000
20,000
25,000
30,000
35,000
40,000
45,000
0
50,000
100,000
150,000
200,000
250,000
1995_96
1997_98
1998_99
1999_00
2000_01
2001_02
2002_03
2003_04
2004_05
2005_06
2006_07
2007_08
2008_09
2009_10
2010_11
2011_12
2013_14
Total employment in Manufacturing in
Ethiopia
No. workers No. workers (FDI China) No. workers (FDI other)
5. Important role of Asian investors
Largest Operational Manufacturing Projects
Country of Origin Percent
China 27.29
India 9.28
Turkey 5.57
China/Ethiopia 4.91
Sudan 2.95
India/Ethiopia 2.51
Saudi Arabia/Ethiopia 3.28
Saudi Arabia 1.42
Britain 1.42
USA/Ethiopia 1.64
Italy/Ethiopia 2.29
USA 0.98
Source: Bloomberg Businessweek.
http://www.businessweek.com/articles/2014-07-
24/ethiopia-vies-for-chinas-vanishing-factory-jobs#p2’
Source: EIC 2015
6. FDI represents sizeable share of existing
firms in main manuf. Sectors (2013/14)
Top 10 sectors with FDI
ISIC Activity (2-digit) All firms FDI firms FDI firms as
share of all
firms
Manufacture of food products and beverages 464 54 12%
Manufacture of other non-metallic mineral products 399 77 19%
Manufacture of furniture; manufacturing n.e.c. 233 34 15%
Tanning and dressing of leather; manufacture of luggage
handbags saddlery harness and footwear
102 24 24%
Manufacture of fabricated metal products except
machinery and equipment
101 18 18%
Manufacture of rubber and plastics products 85 21 25%
Manufacture of chemicals and chemical products 84 23 27%
Publishing printing and reproduction of recorded media 73 10 14%
Manufacture of wood and of products of wood and cork
except furniture; manufacture of articles of straw and
plaiting materials
54 9 17%
Manufacture of textiles 47 16 34%
Source: Own calculations with data from CSA
7. Data
• EIC data is largely aggregated
• We have been working with CSA to get more
disaggregated information on FDI and linkages
• CSA manufacturing census data from 1995/96 to
2013/14
• The latest round has module on FDI
• Module asks about:
– Locational preferences for both local and foreign firms
– Horizontal linkages
– Vertical linkages
• Backward linkages
• Forward linkages
8. Framework (1)
• Following Aitken and Harrison (1999)
𝑌𝑖𝑗𝑡
= 𝛼 + 𝛾𝑋𝑖𝑗𝑡 + 𝛽1 𝐹𝐷𝐼_𝑃𝑙𝑎𝑛𝑡𝑖𝑗𝑡
+ 𝜷 𝟐 𝐹𝐷𝐼_𝑆𝑒𝑐𝑡𝑜𝑟𝑗𝑡 + 𝛽3 𝐹𝐷𝐼_𝑃𝑙𝑎𝑛𝑡𝑖𝑗𝑡
∗ 𝐹𝐷𝐼𝑆𝑒𝑐𝑡𝑜𝑟 𝑗𝑡
+ 𝛿𝑗𝑡 + 𝜀𝑖𝑗𝑡
• 𝑌𝑖𝑗𝑡 is the log of output measured by sales
revenue for plant 𝑖 operational in sector 𝑗 at
time 𝑡.
9. Framework (1)
• Main interest is 𝜷 𝟐, the coefficient on the variable
𝐹𝐷𝐼_𝑆𝑒𝑐𝑡𝑜𝑟 which captures the proportion of FDI in a given
sector
– The sign and magnitude indicates the nature of FDI spillover effect
towards domestic firms in the same sector as the FDI.
𝐹𝐷𝐼_𝑆𝑒𝑐𝑡𝑜𝑟𝑗𝑡 =
𝑖 𝐹𝐷𝐼_𝑃𝑙𝑎𝑛𝑡𝑖𝑗𝑡 ∗ 𝐸𝑚𝑝𝑡𝑖𝑗𝑡
𝑖 𝐸𝑚𝑝𝑡𝑖𝑗𝑡
10. Results (𝑌𝑖𝑗𝑡 , OLS): Spillovers could be local
National Region Zone Town
Share of FDI in the firm (𝐹𝐷𝐼_𝑃𝑙𝑎𝑛𝑡) 0.01 0.02 0.06 0.04
[0.06] [0.065] [0.071] [0.064]
Share of FDI in the sector (𝐹𝐷𝐼_𝑆𝑒𝑐𝑡𝑜𝑟) 0.25 0.13 0.24 -0.48
[0.16] [0.18] [0.19] [0.353]
Share of FDI in the firm*Share of FDI in the sector(𝐹𝐷𝐼_𝑃𝑙𝑎𝑛𝑡*𝐹𝐷𝐼_𝑆𝑒𝑐𝑡𝑜𝑟) 0.005 0.006 0.006* 0.016***
[0.003] [0.004] [0.003] [0.005]
Share of FDI in the sector at regional level 0.19
[0.13]
Share of FDI in the firm*Share of FDI in the sector at regional level -0.002
[0.003]
Share of FDI in the sector at zone level 0.033
[0.128]
Share of FDI in the firm*Share of FDI in the sector at zone level -0.002
[0.002]
Share of FDI in the sector at town level 0.813**
[0.351]
Share of FDI in the firm*Share of FDI in the sector at town level -0.012***
[0.005]
Number of observations 13055 13055 13055 13055
11. Results (Difference; 𝑌𝑖𝑗𝑡 -𝑌𝑖𝑗(𝑡−𝑛) )
First Diff. Second Diff. Third Diff. Fourth Diff.
Share of FDI in the firm (𝐹𝐷𝐼_𝑃𝑙𝑎𝑛𝑡) 0.023 -0.055 -0.288*** -0.353***
[0.102] [0.113] [0.103] [0.122]
Share of FDI in the sector
(𝐹𝐷𝐼_𝑆𝑒𝑐𝑡𝑜𝑟) 0.189 0.490* 0.806** 1.02***
[0.217] [0.276] [0.333] [0.385]
Share of FDI in the firm*Share of FDI
in the sector
(𝐹𝐷𝐼_𝑃𝑙𝑎𝑛𝑡*𝐹𝐷𝐼_𝑆𝑒𝑐𝑡𝑜𝑟)
0.008 0.004 0.005 0.007
[0.005] [0.005] [0.007] [0.009]
Number of observations 7481 4866 3486 2556
Differencing helps remove fixed effects at the plant level; helps us
observe spillover effects over longer time horizon
12. Results: Interpretation
• Evidence of positive spillover from FDI to
domestic firms in sectors with FDI presence
– First FDI seems to benefit nearby domestic firms
• Point estimate imply that increasing foreign equity by one
percentage points is associated with a 0.8 % increase in
productivity of domestic firms located in the same town and
sector as the FDI firm
– Second, the differenced model shows that the
spillover effects tend to become stronger over time
• The effects seems to be felt beyond the locality where FDI is
present
13. Result: Possible channels of spillover
• We are working to parametrically identify
mechanisms but already from the data we have
evidence on
– Vertical linkages ( Local suppliers) :
• 10% of firms report selling to FDI firms
• Of firms receiving technology transfers about 25% did
so from FDI firms
– Most of these transfers (~60%) from FDI firms
aimed at improving production processes and
product design
– Horizontal linkages (Competition) :
• In 15% of firms, competitive pressure from FDI has lead to Δ in
prod. Techniques
– Of these, 57% did so by observing/copying FDI firms
– And most of them (76%) have not perceived FDI firms tried to
prevent this
14. Result: Possible channels of spillover
– Horizontal linkages (labor channel) :
– 9% of firms have hired employees who have worked
in FDI firms before
– These workers do seem to make positive
contributions to productivity in local firms:
» 62% report former FDI workers contributing to
better production technologies
» 13% to bringing export knowledge to the firm
» 11% help improve managerial practices
15. Framework (2)
• Likely that if spillover effects exist, they will have a local
effect (as indicated in the previous results based on
Framework (1)).
• FDI firms don’t choose location at random, rather they
choose based on characteristics of locality (potentially
unobservable)
• If unobservables are correlated with productivity then
simple productivity differences between incumbent
firms in non-FDI localities v.s. FDI localities would be
biased.
• Better identification strategy following Greenstone,
Hornbeck and Moretti (2010):
17. Framework (2)
• Failed attempts
– Use localities (towns) where FDI firms “almost”
located but, for exogenous reasons (administrative,
budget) did not
– These “Losing” towns are potentially a good
counterfactual for “Winning” towns
– We can further control for pre-existing trends, firm
and industry idiosyncrasies as well as other
differences.
• Govt. designation
– If we know decision rule by govt. we can control for
this
18. Framework (2)
• Pair-wise comparison is methodologically superior to both
the regression formulation and DID-matching based on
observables.
• In our case, we have information about other localities
(towns) FDI firms considered before opening the current
plant
• We can use that information to link “treated” towns(i.e.
had a FDI firm opening in towns) with “control”
towns(those not selected by FDI firm)
• We also observe location and year when FDI firm started
operations
20. Econometric specification
• Parameters of interest are:
– 𝜃1 which represents a mean shift in productivity
of incumbent plants in a “treated” towns
– 𝜃2 which measures the trend break in productivity
for incumbent plants in a “treated” towns
• Specification also controls for a number of
pre- and post-treatment levels, allows for
trend breaks and controls for firm, industry,
case and industry-year fixed effects.
21. Road ahead
• We are now trying to complement some
information on “losing” locations in order to
be able to follow our main identification
strategy
• Cleaning the location variables to reduce
measurement error
• Will expect results from the main model in
two to three months time.
Presentation of preliminary results from ongoing research. Major Research Grant No. 1386
Project title: Ref 1386 "The Role of Asian Investment in Manufacturing in Catalyzing Economic Development in Africa"
Name of the Principal Investigator(s): Margaret McMillan