Transcript Document

J. de Loecker
Do Exports Generate Higher Productivity?
Evidence from Slovenia
(Journal of Int’l Economics, Sep. 2007)
presented by
Yunrong Li
1

About the paper

Criticism
2
About the Paper
1.
Introduction
2.
Data and preliminary analysis
3.
Export entry and productivity gains
4.
Productivity gains and export
destination
5.
Conclusion
3
1.Introduction






Empirical finding: positive correlation between
the export status of a firm and its productivity
WHY ?
Self-selection hypothesis: only more productive
firms export and are able to compete on
international markets.
Learning-by-exporting hypothesis: exporting
enables firms to learn from their buyers and
competitors through contracts.
Motivation for this paper:
Test for learning-by-exporting hypothesis
Key issue of the test: control for self-selection
process.
4
Why Slovenia ?




Experienced a successful transition from a
socially planned economy into a market
economy in less than a decade.
High growth rate of GDP
Opening up to Western countries; substantial
increase in exports in a very short time.
A natural experiment to test the hypothesis of
learning-by-exporting.
5
2. Data and Preliminary Analysis




Firm level data in the entire manufacturing
sector between 1994 and 2000
The sample consists of 6391 firms and 29,804
observations
In total, 1,872 firms joined the export market
at different points of time.
7% of these entrants are new firms; remaining
1,700 firms are crucial to verify the impact of
“starting to export” on productivity.
6
2. Data and Preliminary Analysis


If exporters have different characteristics
from the non-exporters.
OLS regression model:
xikt     EXPikt   likt   jTime j   k Ind k   ikt
j

k
where x is the characteristics of firm i at
period t in industry k; EXP is a export
dummy; l is the log of number of
employees of firm i.
Interest lies in 
7
2. Data and Preliminary Analysis



Test result: exporters differ significantly
from non-exporters.
Exporters: pay higher wages; sell more;
operate on a larger scale; invest more;
more capital intensive.
These results are in line with Bernard
and Jensen (1995) for the USA, Bernard
and Wagner (1997) for Germany, Isgut
(2001) for Colombia.
8
3. Export Entry and Productivity Gains



3.1
Productivity Dynamics and Export
Status
3.2 Identifying productivity gains arising
from exporting
3.3
Productivity gains upon export for
companies
joining
the
Slovenian
manufacturing industry
9
3.1 Productivity Dynamics and Export Status
The method of estimating productivity is
based on the work of Olley and Pakes (1996).
Two innovations their method:


Control for simultaneity bias without using
instruments available.
Control for potential selection bias. This issue is
important for Slovenia, because selection is
likely to be an intrinsic part of the transition
process.
10
3.1 Productivity Dynamics and Export Status
Innovation of this paper:



Allow for market structure (demand condition,
factor market, etc.) to be different for exporters.
Deflate the “value added” with a Slovenia PPI.
But this is not enough to control for output and
factor price differences for exporters
Hence, the paper estimates the production
function for each 2-digit NACE sector separately
(Assuming firms within the same sector face
the same input prices)
11
3.1 Productivity Dynamics and Export Status

The model for estimating productivity:
ijt  yijt  blj lijt  bkj kijt
y, l, k denote the log of output as measured
by value added, labor and capital.
blj and bkj are OP-EXP estimators for labor
and capital for industry j.

stands for a measure of productivity.
12
3.1 Productivity Dynamics and Export Status



Conventional productivity index: weigh
firms’ productivity by their market share.
Normalize the first year 1994 to 1.
Comparison: by 1999, productivity index
increase by 16% for exporters and only
10% for non-exporters.
13
3.2 Identifying productivity gains arising from
exporting



Key issue: control for self-selection process
while
testing
for
learning-by-exporting
hypothesis.
Solution: create a counterfactual control group
for each starter, “matching technique”.
We don’t observe the counterfactual controls;
therefore, we need to match each starter with a
control group who has similar characteristics as
the starter but don’t export.
14
3.2 Identifying productivity gains arising
from exporting



How to select the control group:
To find a group as close as possible to the
starter (treatment group) in terms of the
predicted probability to start exporting.
We cut off the firms who always export
during the entire periods.
Apply the ‘propensity score matching’
method by Rosenbaum and Rubin
(1983a,1984b and 1984)
15
3.2 Identifying productivity gains arising from
exporting
16
3.2 Identification productivity gain from exporting
17
3.2 Identifying productivity gains arising from
exporting



Econometric Model:
Re-scale the time periods: a firm starts
exporting at s=0, note that different firms
may start to export in different periods.
is
is the productivity of firm i at period s
STARTi takes on 1 if a firm starts to export,
and 0 for s≠0
18
3.2 Identifying productivity gains arising from
exporting

The probability model of starting to export:
Pr STARTi ,0  1  h(i ,1 , ki ,1 , PRIVATEi ,1 )


Where  is normal cdf; -1 mean before starting to
export; PRIVATE takes on 1 if the firm i is privately
owned; a set of year dummies and industry
dummies is also included to control for aggregate
demand and supply shocks.
Denote the predicted prob. by pi
Select a matching firm j for i based by:
| pi  p j | min( pi  pk )
k{ EXP 0}
19
3.2 Identifying productivity gains arising from
exporting

Difference in difference Method (DID)

N firms start exporting at s=0



a set of controls C(i) for firm i, and Nic denote the
number of controls in C(i).
Every starter is matched with a set of control firms.
The matching is always performed once a firm
starts to export and s={0, 1,…,S}
20
3.2 Identifying productivity gains arising from
exporting

Calculate the effect of learning-by-exporting  LBE
in two ways:
s
 LBE


1
Ns
1
(

 is 
i

jC ( i )
wij cjs )
S
S
1
S
 LBE

( is1    wij cjs )

N S i s 0
s 0 jC ( i )
Where wij =1/ N ic , note that N changes with s,
s
S
so ,   LBE
  LBE
s

We can also calculate
for year-to-year produc.
1
1
growth rate (s  s1 ) and growth rate compared
to pre-export level (s1  11 )
21
3.3 Productivity gains upon export for companies
joining the Slovenian manufacturing industry
22
3.3 Productivity Gains upon Export Entry in
Slovenia Manufacturing
Industry specific result
23
4. Productivity gains and export destination





Eaton and Kortum (2004 and 2005) use destination
to understand the importance of fixed cost in entering
export market.
The first paper to investigate productivity gains from
exporting
by
distinguishing
between
various
destinations.
8 groups: Africa, Asia, North-America, South-America,
Western Europe, Southern Europe, Central and Eastern
Europe, others (Australia and New Zealand)
On average 90 percent of firms export to Western
Europe, Southern Europe and Central and Eastern
Europe. 1/3 firms export to Asia and North America.
The pattern across industries is quite stable, which
implies that the difference across industries can’t be
solely explained by difference in destination.
24
4.Productivity gains and export destination





Divide the destinations into low income and
high income regions.
Split the firm sample into two groups: only
export to high income regions or low income
regions.
Estimate the productivity gains from export
entry separately for these two groups of firms.
Only run the regression on the entire
manufacturing sector.
Estimate the instantaneous productivity gains,
since the destination info is only cross sectional.
25
4.Productivity gains and export destination


Result:
Firms exporting only to high income regions
experience a higher productivity gains than the
overall sample, and so much so do firms exporting
to low income regions.
26
4.Productivity gains and export destination


Evidence for learning-by-exporting hypothesis:
If only more productive firms could export,
then export destination would not matter much.
But we found that destination affects firms
productivity gains.
Comment: This reasoning should be based on
assuming that before exporting, both kinds of
firms (export to high or low) have similar
productivity level, because it might be that
more productive firms can export to high
income regions.
27
5. Conclusion






Estimate total factor productivity based on Olley
and Pakes (1996)
Introduce a matching technique to build up a
counterfactual control group.
On average, exporters become more productive by
8.8% than the control group.
Productivity increases in future years following the
decision to export (can be seen from cumulative
effect).
The magnitude and timing of the learning effects are
quite different across sectors.
Significant higher productivity gains for firms
exporting to high income regions.
28
Criticism



The part I appreciate:
Exporters gain produc. from exporting because
exporters can learn from their importers and
competitors through contracts, but not because
of increasing return to scale.
Van Biesebroeck (2006): exploit increasing
return to scale with access to larger market
demand. But this paper finds out that in
Slovenia, even firms with no significant
increasing return to scale can still gain from
exporting.
29
Criticism



Question left unanswered:
Assume that all differences between exporters
and non-exporters are captured by the
observables including productivity;
But productivity gains from starting to export
can be driven by unobservables which are
highly correlated with exporting status.
However, this can’t be tested in the framework
of this paper.
30
Criticism



Some ideas:
This paper divides exporters into two groups: only
export to high income regions or only to low income
regions. This makes a big drop in the sample size. I
would advise to include those firms who export, say
more than 60%, of their total output to high/low
income regions.
Geography is an important issue. Firms lie in better
regions have higher probability of starting to export.
31