Professional Documents
Culture Documents
152163, 2016
0305-750X/ 2015 Elsevier Ltd. All rights reserved.
www.elsevier.com/locate/worlddev
http://dx.doi.org/10.1016/j.worlddev.2015.11.012
1. INTRODUCTION
This study examines the exportnancial constraint relationship, focusing on how small and medium enterprises
(SMEs) dier from other rms. 1 To analyze the relationship,
we consider both the export participation (extensive margin)
and the share of exports in total sales (intensive margin).
Accounting for whether rms reveal that they need loans or
have sucient capital, we evaluate the association between
exports and both the use of bank loans (quantity of nance)
and the availability of overdraft facilities (exibility of
nance). The objective is to understand better the role of
nance on international trade of small rms in developing
and emerging-market economies.
Studying the exportnancial constraint relationship has
both policy and intellectual merits. The Global Financial Crisis inicted a sudden drought of international liquidity, a
sharp drop in international trade, and widespread recession
and unemployment. The severity of the crisis has generated
interest internationally on supportive policies and government
interventions that can help recover rm performance and
jump-start the economy. As a growth strategy, a country
can export its way out of the crisis and recession, but this
course of action requires domestic rms to be highly productive and well equipped with the necessary operational support
to enter and survive international competition. Overcoming
nancial constraints, having sucient capital, managing cash
ow eciently, and readily accessing working capital are
important pre-requisites for rms to be competitive in the global markets. That is the focus of this paper.
Access to credit is challenging for rms that really need it,
most of which are small rms. Credit rationing is known to
be a barrier to a rms growth, especially in developing
* The authors would like to thank the editor and three anonymous
reviewers for constructive comments on this paper. Helpful suggestions
were also provided by Naoyuki Yoshino, Chayodom Subhasri, Amornrat
Kritsophon, Wako Watanabe as well as the participants of several
conferences. These include: the 2014 ADBI-IMF-JFSA International
Conference on Financial System Stability, Regulation, and Financial
Inclusion (Tokyo); the 14th OECD-ADBI Tokyo Roundtable on Capital
Market Reform in Asia (Tokyo); and the 2014 Asia Pacic Economic
Association (Bangkok). Paulo Mutuc and Menaka Arudchelvan provided
able research assistance for earlier versions of this paper. Final revision
accepted: November 23, 2015.
152
153
154
WORLD DEVELOPMENT
Table 1. Denition of small and medium enterprises (SMEs), rm population, and enterprise surveys data
Country
Number of
employees
Chile
Israel
Korea
Mexico
Thailand
Turkey
250
100
300
250
200
250
812,193 (98%)
501,676 (97%)
5,143,696 (99%)
4,671,883 (99%)
2,913,617 (99%)
2,291,904 (99%)
86.1%
82.8%
92.1%
86.5%
70.6%
91.5%
2010
2013
2005
2010
2006
2013
The denition of SMEs by number of employees and alternative criteria, and the national population of SMEs are drawn from Financing SMEs and
Entrepreneurs 2014 report. The Enterprise Surveys sample of SMEs and the corresponding survey year are based on the Surveys 2015 database.
30.1%
73.4%
SME Exports = SME Exports * National Exports
SME Output National Exports National Output
57.1%
SME Output
National Output
38.7%
National
Output
(GDP)
SME Exports
22.1%
SME Output
38.7%
National
Exports
73.4%
155
100.0
90.7
90.0
85.5
80.0
71.9
70.0
65.8
60.0
55.0
53.0
51.7
50.0
49.3
44.3
42.0
40.9
40.0
30.0
20.0
30.9 29.6
29.6
19.1
29.1
27.4
26.2
20.4
17.1
16.7
13.8
9.3
10.0
0.0
Chile
Israel
Korea
Mexico
Thailand
Turkey
Figure 2. Export participation and nancial constraints. This gure provides a summary illustration of export participation (% exporting rms in total rms)
and nancial constraints [based on bank loans (% total nance), loan needs (% of rms), and the availability of an overdraft facility (% of rms)] for all
rms in the sample drawn from the Enterprise Surveys, covering Chile (17.27%), Israel (8.08%), Korea (10.00%), Mexico (24.75%), Thailand (17.44%),
and Turkey (22.47%).
Table 1). Second, there is no panel dimension of the rm characteristics (the statistics of Table 2 summarize cross-section
information). Third, there are minor dierences in the survey
questions across countries and years. The rst and second
issues dictate that the panel estimation methodology is not
applicable for our empirical analysis and thus we are less
assertive on the implications of causality. We have also veried and cross checked the survey questions so that the third
issue has no inuence on the estimation sample, empirical
analysis and the main ndings.
3. EMPIRICAL RESULTS
The issue of endogeneity is the key challenge involved in
analyzing the exportnancial constraint relationship using
rm-level data. Since a rms exports and nancial constraints
are both likely to be inuenced by rm characteristics, estimating an export equation using any nancial variable (a proxy
for nancial constraints) in the regressors, runs the risk of estimation bias due to the correlation between nancial variable
and the error term. Instead, a predicted component of nancial constraints based on additional information from rm
characteristics, or the instrument, should be used in the export
equation to properly estimate the exportnancial constraint
relationship with rm-level data.
We use a two-stage estimation procedure. Denote i for rm
identication and j for sample country, the 2nd-stage equation
of interest is
y ij Y ij b X ij d uij
where y is an export variable [export participation (y d ), share
of exports in total sales (y r )], Y is an endogenous regressor
(bank loan, overdraft facility), and X is a vector of exogenous
regressors.
156
WORLD DEVELOPMENT
Table 2. Summary statistics
Variable
Obs.
Mean
Std. dev.
Min.
Max.
5942
5981
5981
5981
5981
5981
5611
5919
5933
5401
5981
5383
5971
5981
17.20
0.38
0.85
181.91
21.12
0.68
8.55
16.46
0.33
16.95
0.40
0.62
0.64
0.95
31.32
0.49
0.36
747.28
17.45
0.47
25.59
22.89
0.47
26.85
0.49
0.48
0.48
0.50
0
0
0
0
0
0
0
0
0
0
0
0
0
0.52
100
1
1
21,955
210
1
100
100
1
100
1
1
1
1.93
This table provides summary statistics of variables in the estimation. The sample includes rms from Chile (1,033 rms; 17.27%), Israel (483 rms; 8.08%),
Korea (598 rms; 10.00%), Mexico (1,480 rms; 24.75%), Thailand (1,043 rms; 17.44%), and Turkey (1,344 rms; 22.47%). The SME status dummy
variable is based on the national denition of small and medium enterprises. Number of employees is the number of permanent, full-time employees at end
of previous year. Firm age is the year an establishment began operations. Foreign ownership is the % of equity owned by private foreign individuals,
companies or organizations. Employees with university degree is the % of full-time permanent workers with at least a bachelor degree at end of previous
year. ISO certication indicates whether the establishment has an internationally-recognized quality ISO certication. Bank loan is a proportion of a rms
working capital that was nanced by borrowing from banks. No-need-for-a-loan respondent indicates that the establishment did not apply for any loan
because, as responded in the survey, there is sucient capital. Overdraft facility indicates whether or not the establishment has an overdraft facility.
Financial audit indicates whether nancial statements are checked and certied by an external auditor in a previous year. Market-based (vs. bank-based)
nancial structure is a ratio of stock market capitalization to deposit money banks. The data are drawn from the Financial Development and Structure
Dataset (November 2013) and the Enterprise Surveys (March 2015).
Y ij P Z ij mij
where Z is a vector of exogenous instruments; Euij Z ij 0.
The two-stage estimation is a classical instrumental variable
regression, with the auxiliary equation is the projection of Y
on Z, and the sampling assumption requires that
y ij ; Y ij ; Z ij ; X ij are i.i.d.
Table 3 reports the rst-stage results (auxiliary equation),
focusing on the association of bank loans and overdraft facilities (dependent variables) with loan needs and rm characteristics (explanatory variables). The estimation is done for the
pooled sample of all rms, for the subsample of non-SME
rms, and for the subsample of SME rms.
The dummy variable No-need-for-a-loan respondent,
used as a proxy of loan needs, is negative and statistically signicant across the pooled sample and subsamples of SME and
non-SME rms, in the use of bank loan equation (columns I
III) and in the availability of overdraft facility equation (columns IVVI). When rms responded in the survey that they
have sucient capital, this additional information is associated
negatively with the observed use of bank loans and the availability of overdraft facilities in the data, all else being equal.
In the pooled sample of all rms, we also nd that SME status is associated with lower use of bank loans and availability
of overdraft facilities. Other rm characteristics are also statistically correlated with the use of bank loans and the availability
of overdraft facilities as follows. Overdraft facilities are more
available to older rms. The percentage of foreign ownership
is negatively associated with the use of bank loans and the
availability of overdraft facilities. Firms in the manufacturing
sector are more likely to use more bank loans. Financial audits
are positively associated with higher use of bank loans and
availability of overdraft facilities, specically for SMEs. We
also note that the market-based nancial structure measure,
as well as the period before and the period after the global
nancial crisis, are associated with higher use of bank loans
and availability of overdraft facilities in the sample countries.
control vars.
No-need-for-a-loan respondent (dummy var.)
SME Status (dummy var.)
Firm age (years)
Foreign Ownership (%)
Manufacturing sector (dummy var.)
Financial audit (dummy var.)
Market-based nancial structure measure
Before the global nancial crisis (200507) obs.
After the global nancial crisis (201113) obs.
Constant term
Obs.
R-sq.
Log likelihood
Non-SM coe.
(std. err.)E
(II)
18.363
(2.147)***
17.015
(.733)***
.486
(.450)
.134
(.027)***
4.169
(2.141)*
.391
(3.730)
6.790
(3.223)**
23.266
(3.633)***
4.722
(3.356)
15.391
(6.382)**
729
.239
3399.373
.050
(.232)
.052
(.018)***
2.730
(.772)***
5.319
(.730)***
6.754
(1.183)***
14.584
(1.357)***
11.209
(1.242)***
5.510
(1.940)***
4608
.150
21202.679
Pooled Sample
coe. (std. err.)
(IV)
Non-SME coe.
(std. err.)
(V)
.310
(.041)***
.242
(.062)***
.051
(.013)***
.003
(.001)***
.006
(.045)
.328
(.045)***
1.928
(.068)***
2.125
(.085)***
2.209
(.069)***
2.641
(.124)***
5013
.264
(.114)**
2606.674
.093
(.028)***
.003
(.001)**
.145
(.117)
.219
(.212)
1.741
(.201)***
2.046
(.206)***
1.911
(.190)***
2.501
(.366)***
784
380.894
SME coe.
(std. err.)
(VI)
.318
(.044)***
.039
(.014)***
.002
(.001)**
.019
(.049)
.331
(.047)***
1.972
(.073)***
2.144
(.097)***
2.257
(.074)***
2.902
(.117)***
4229
2220.993
This table reports 1st-stage results of the instrumental variable estimation. The dependent variables are a rms use of bank loans as measured by a proportion of working capital that was nanced by
borrowing from banks (%) and rms availability of an overdraft facility (0/1 dummy variable). No-need-for-a-loan respondent indicates that the establishment did not apply for any loan because, as
responded in the survey, there is sucient capital. The sample is based on rm-level observations from the Enterprise Surveys, covering Chile (17.27%), Israel (8.08%), Korea (10.00%), Mexico (24.75%),
Thailand (17.44%), and Turkey (22.47%). Standard errors are in parenthesis; *** (**, *) denotes statistical signicant at 1 (5, 10) percent level.
Table 3. Loan needs, bank loan, overdraft facility, and rm characteristics 1st stage estimation (auxiliary equation)
1st-Stage of instrumental variable estimation
(Auxiliary equation)
157
158
Table 4. Exports and nancial constraint 2nd stage estimation (equation of interest)
2nd-Stage of instrumental variable estimation (Equation of interest) control vars.
.183
(.045)***
Non SME
coe.
(std. err.) (II)
.104
(.139)
SME coe.
(std. err.) (III)
1.599
(.541)***
.137
(.040)***
9.213
(2.605)***
22.088
(2.772)***
.001
(.053)
2.391
(4.097)
26.629
(6.009)***
17.450
(4.121)***
.017
(6.346)
725
.574
.440
.257
67.36
36.57
19.087
(7.253)***
8.729
(1.432)***
.855
(.280)***
.190
(.020)***
6.694
(1.113)***
10.258
(.976)***
.027
(.024)
17.277
(4.988)***
3.989
(6.229)
1.608
(5.637)
21.696
(4.065)***
4960
.325
.004
.001
100.85
51.35
21.750
(28.182)
SME coe.
(std. err.) (VI)
.305
(.267)
.196
(.021)***
6.708
(.985)***
9.093
(.934)***
.001
(.019)
7.987
(1.421)***
1.187
(1.831)
12.743
(1.528)***
6.811
(2.175)***
4542
.304
.975
.000
518.99
292.26
1.005
(.837)
.102
(.044)**
7.501
(2.876)***
22.781
(2.930)***
.160
(.067)**
15.947
(15.431)
43.932
(19.788)**
33.248
(17.102)*
12.118
(9.153)
765
.561
.546
.476
8.18
4.07
25.290
(7.697)***
.591
(.315)*
.208
(.025)***
5.757
(1.238)***
8.488
(1.074)***
.006
(.027)
22.959
(5.475)***
14.613
(6.797)**
4.295
(6.181)
19.497
(4.807)***
4195
.206
.004
.000
90.64
46.20
This table reports 2nd-stage results of the instrumental variable estimation. The dependent variable is a rms share of exports in total sales (%). The endogenous regressors are bank loans (% of total
nance) and overdraft facilities (0/1 dummy var.); their instrumented estimates from the 1st-stage estimation are used. The sample is based on rm-level observations from the Enterprise Surveys,
covering Chile (17.27%), Israel (8.08%), Korea (10.00%), Mexico (24.75%), Thailand (17.44%), and Turkey (22.47%). Standard errors are in parenthesis; *** (**, *) denotes statistical signicant at 1 (5, 10)
percent level.
WORLD DEVELOPMENT
9.999
(1.256)***
.568
(.237)**
.199
(.018)***
7.655
(.921)***
10.931
(.886)***
.010
(.018)
7.300
(1.340)***
2.677
(1.757)
13.238
(1.433)***
14.923
(2.371)***
5267
.366
.691
.000
596.78
335.75
Non SME
coe.
(std. err.) (V)
.206
(.048)***
Pooled Sample
coe.
(std. err.) (IV)
159
Table 5. Marginal Eects of Control Variables: Loan-Needs SMEs vs. Other Firms
Dependent var. is export participation
(0/1 dummy var.)
Logit Estimation Control vars.
Export participation
coe./std. err.
Loan needs,
non SME
(I)
Export participation
coe./std. err.
Sucient capital,
non SME
(II)
0.0008
0.0148
0.7731
0.5987
0.0102
0.0103
0.0013
0.0012
0.1257
0.0464
0.2548
0.0468
0.0010
0.0009
0.4737
0.3012
565
.244
0.0038
0.0237
1.7139
1.1831
0.0006
0.0217
0.0045
0.0015
0.0903
0.0696
0.2376
0.0665
0.0001
0.0017
1.3087
0.6285
259
.281
Export participation
coe./std. err.
Loan needs,
SME
(III)
Export participation
coe./std. err.
Sucient capital,
SME
(IV)
0.0084
0.0070
1.7412
0.3792
0.0059
0.0074
0.0041
0.0005
0.1267
0.0202
0.1973
0.0237
0.0004
0.0005
1.1273
0.2030
2715
.170
0.0043
0.0134
0.9240
0.7051
0.0061
0.0142
0.0040
0.0005
0.1346
0.0235
0.2399
0.0362
0.0006
0.0009
0.6863
0.3896
1981
.195
0.03
(0.9850)
4.86
(0.0882)
1.60
(0.4491)
22.32
(0.0000)
This table reports average marginal eects based on logit regressions of export participation (0/1 dummy variable) as a dependent variable. The
endogenous regressors are bank loans (% of total nance) and overdraft facilities (0/1 dummy var.); their instrumented estimates from the 1st-stage
estimation are used. Constant term and before/after global nancial crisis dummy variables included but not reported. The sample is based on rm-level
observations from the Enterprise Surveys, covering Chile (17.27%), Israel (8.08%), Korea (10.00%), Mexico (24.75%), Thailand (17.44%), and Turkey
(22.47%). Standard errors are calculated by the delta method.
160
WORLD DEVELOPMENT
Table 6. Benchmark Results - Generalized Linear Model (GLM) Estimation
.032
(.078)
.006
(.003)**
.555
(.222)**
1.366
(.224)***
.001
(.004)
1.100
(1.318)
2.355
(1.504)
2.184
(1.551)
3.346
(.827)***
.55
.51
766.13
372.07
812
.055
(.034)
.017
(.002)***
.352
(.112)***
.876
(.112)***
.003
(.002)
2.643
(.826)***
1.981
(.904)**
1.045
(.938)
1.105
(.723)
.50
.62
3093.95
1535.98
4679
This table reports benchmark estimation results based on the Generalized Linear Model (GLM), using maximum likelihood with Bernoulli/binomial
distribution of dependent variable (share of exports in total sales) and logit link function. The endogenous regressors are bank loans (%) and overdraft
facilities (0/1 dummy var.); their instrumented estimates from the 1st-stage estimation are used. The sample is based on rm-level observations from the
Enterprise Surveys, covering Chile (17.27%), Israel (8.08%), Korea (10.00%), Mexico (24.75%), Thailand (17.44%), and Turkey (22.47%). Standard errors
are in parenthesis; *** (**, *) denotes statistical signicant at 1 (5, 10) percent level.
extensive margin (export participation) and the intensive margin (share of exports in total sales):
Ey rij jY ij ; X ij Ey rij jY ij ; X ij ; y dij 1 Pry dij 1jY ij ; X ij
The share of exports in total sales equation is estimated with
a Generalized Linear Model (GLM), using maximum
likelihood with Bernoulli/binomial distribution and logit link
function. 15 The estimation results are reported in Table 6.
As shown, the availability of an overdraft facility variable is
positively associated with the share of exports in total sales in
the pooled sample and in the SME subsample. The use of a
bank loan variable remains insignicant in the equation of
the share of exports in total sales. These results further support
our ndings in the export participation equation that the exibility of nance is an important link in the nancial constraintexport relationship, particularly so for SME rms.
Based on the benchmark estimates of Table 6, we also nd that
the control variables have statistically signicant coecient estimates, both in the intensive margin of exports and in the extensive margin of exports. These variables are SME status (),
percentage of foreign ownership (+), ISO certication (+), manufacturing sector dummy variable (+), and market-based (vs.
bank-based) nancial structure measure ().
As a precautionary exercise and verication, we conduct the
GLM estimation without using the instrumented endogenous
regressors (i.e., skipping the 1st-stage estimation of the nancial
1.00
0.45
0.36
0.20
0.05
-0.05
-0.11
-1.50
161
-0.19
-0.95
-1.00
-0.50
0.00
0.50
1.00
1.50
Figure 3. Economic signicance of control variables on the share of exports in total sales. This gure reports the economic signicance of each control variable,
calculated from a product of variables coecient estimate (Table 6, column I) and its sample standard deviation (Table 2) from the pooled sample of rms.
The data are drawn from the Enterprise Surveys and the Financial Structure Database.
dummy variable, the percentage of foreign ownership, ISO certication, and SME status. In sum, the economic signicance
of the control variables found in our analysis suggests that
the pre-conditions for exporting, both in the extensive margin
and in the intensive margin, collected in the literature (i.e., rm
size, product quality, foreign ownership) remain important factors for rms in our sample countries. Most importantly, we
nd that nancial constraints, specically the exibility of
nance available to rms (as measured by the availability of
an overdraft facility variable) have a large economic signicance on the intensive margin and the extensive margin of
exports, particularly so for the SMEs with loan needs.
4. CONCLUSION
Using rm-level data drawn from the World Banks Enterprise
Surveys in Chile, Israel, Korea, Mexico, Thailand, and Turkey,
we analyze factors associated with the exportnancial constraint relationship, focusing on dierences between SMEs and
other rms. The empirical analysis suggests the importance of
(i) accounting for whether a rm actually needs a loan to address
demand conditions and supply constraints for loanable funds,
(ii) dierentiating the quantity of nance (i.e., use of bank loans)
and the exibility of nance (i.e., availability of an overdraft
NOTES
1. We follow the national denition of SMEs based on the number of
full-time employees: Chile250 employees, Israel100 employees,
Korea300 employees, Mexico250 employees, Thailand200 employees, and Turkey250 employees. Section 2 provides a more detail
description of the alternative SME denitions and statistics.
2. Related studies on nance and trade are numerous. For recent ones,
see Paravisini, Rappoport, Schnabl, and Wolfenzon (2015) for evidence
162
WORLD DEVELOPMENT
10. The evidence (e.g., Beck, Demirguc-Kunt, & Levine, 2005) is less
conclusive on whether SMEs causally impact economic growth.
16. See van der Schans (2012) for the importance of bank overdrafts in
SME nance in the UK. Armstrong, David, Liadze, and Rienzo (2013)
provide evidence from the UK that the tight credit supply conditions
continue beyond the height of the Global Financial Crisis and that the
rejection rates of applications for bank nancing have increased
(particularly for low and average risk SMEs previously viewed as safe
for banks) even compared to the period of the Crisis.
REFERENCES
ADB-OECD (2014). ADB-OECD study on enhancing nancial accessibility
for SMEs: Lessons from recent crises.
ADB (2014). Asia SME Finance Monitor 2013.
Antra`s, P., & Foley, C. Fritz. (2015). Poultry in motion: A study of
international trade nance practices. Journal of Political Economy, 123
(4), 809852.
Armstrong, A., David, E. P., Liadze, I., & Rienzo, C. (2013). An
assessment of bank lending to UK SMEs in the wake of the crisis.
National Institute Economic Review, 225, R3951.
Ayyagari, A., Beck, T., Demirguc-Kunt, A. (2007). Small and Medium
Enterprises Across the Globe, 29, 4, 415434.
Beck, T. (2013). Banking nancing for SMEs Lessons from the
literature. National Institute Economic Review, 225, R2338.
Beck, T., & Demirguc-Kunt, A. (2006). Small and medium-size enterprises: Access to nance as a growth constraint. Journal of Banking and
Finance, 30, 29312943.
Beck, T., Demirguc-Kunt, A., & Levine, R. (2005). SMEs, growth, and poverty:
Cross-country evidence. Journal of Economic Growth, 10(3), 199229.
Berger, A. N., & Udell, G. F. (2006). A more complete conceptual
framework for SME nance. Journal of Banking and Finance, 30(11),
29452966.
Bernard, A., & Jensen, B. (2004). Why some rms export. Review of
Economics and Statistics, 86(2), 561569.
163
APPENDIX. A
.094
(.048)**
.008
(.003)***
.554
(.229)**
1.469
(.261)***
.010
(.005)**
.364
(.365)
1.857
(.449)***
1.296
(.382)***
3.526
(.610)***
.52
.50
611.43
294.72
679
.017
(.031)
.015
(.002)***
.562
(.098)***
.746
(.115)***
.003
(.002)
.487
(.210)**
.903
(.260)***
1.430
(.203)***
2.747
(.311)***
.51
.62
2769.31
1373.66
4060
This table reports benchmark estimation results based on the Generalized Linear Model (GLM), using maximum likelihood with Bernoulli/binomial
distribution of dependent variable (share of exports in total sales) and logit link function. The endogenous regressors are bank loans (%) and overdraft
facilities (0/1 dummy var.); both are not instrumented. The sample is based on rm-level observations from the Enterprise Surveys, covering Chile
(17.27%), Israel (8.08%), Korea (10.00%), Mexico (24.75%), Thailand (17.44%), and Turkey (22.47%). Standard errors are in parenthesis; *** (**, *) denotes
statistical signicant at 1 (5, 10) percent level.
ScienceDirect