The Influence of Exports on the

Labour Productivity of Fiji

EC303
Project

Abstract
The objective of this research is to investigate the impact of exports on the labour productivity
of Fiji. Previous studies have found that exports does did to increase in productivity in many
countries. The methodology used for investigation is econometric analysis where exports in
used as a function in the production equation. To test for the validity of the estimated
regression output, the model has been tested for multicollinearity, structural breaks,
autocorrelation, heteroskedasticity and normality. Apart for a problem of structural break, our
model did pass all the other assumptions of classical linear regression model. The main result of
this study has been that exports does have a statistically significant impact on the level of
labour productivity of Fiji where labour productivity increases by 0.065% due to a 1% increase
in the level of exports on average. The conclusion which we came up from this result has been
that exports create competition for the local producers in the overseas market, thus, they
remedy their production capabilities to par required in the international markets which
eventually leads to increase in the domestic labour productivity.

2
Introduction
The GDP per capita of Fiji has been gradually rising in constant terms of 0.8% per year
since 1980. Compared to other pacific island countries, Fiji has benefitted from relatively higher
growth rate of real GDP. There exists many explanation of changes and deviations in labour
productivity with links to macroeconomic variables. One of the determinants of labour
productivity could be determined to be exports. Thus the objective of this research study is to
determine the impact of Fiji’s export on its labour productivity through the use of econometric
analysis.

Previous studies have shown that export has largely contributed toward the growth of
host countries labour productivity. The main analysis for this relationship has been that export
creates competition in the host country which encourages them to invest in research and
development to increase labour productivity and reduce the product price. In addition, many
studies also reasoned that export leads to spillover effects such as importation of techniques
and advanced technologies which grows the exporting countries labour productivity.

It is of important to understand the determinants of labour productivity in Fiji for policy

implication in the attainment of economic growth. Capital per labour in the production function
is an important determinant for labour productivity to which the government can enhance to
further the rate of economic growth. This research could guide the government to determine if
the level of exports has any impact on the level of labour productivity. If significant effect is
found the government could be recommended to enhance the export level through various
policies in order to increase labour productivity.

In Fiji’s economic context, the aim is to find the relationship between exports and labour
productivity. With the use of classical linear regression modelling techniques, in this study a
regression model will be developed. The regression model will be tested for functional form,
structural breaks, multicollinearity, heteroscedasticity, autocorrelation and normal distribution
of error terms. The final model will be adjusted for the disturbances to determine the
statistically significant impact of exports on labour productivity and spurious results.

The structure of this research paper will be as follows: Section 2 provides a review of
past literature on the topic. Section 3 provides the description of the econometric models being
used for analysis. Section 4 provides information on the distribution of the data. Section 5
provides the full output of the estimation being carried out in this research and lastly section 6
gives the mains conclusion of this research and policy suggestions.

3
Literature review
Review of previous literature and research on our topic has been the foundation to
better understand the relationship between export and labour productivity. Most of the
previous literatures which are reviewed of which Wagner (2001) and Aw, Chung, & Roberts
(2000) has compared the effects of exporting and non exporting firms on labour productivity.
The authors’ have concluded that the resulting effect of exporting firms on labor productivity is
far higher than non- exporting firms.

Wagner (2001) in his paper using the matching approach evaluated the causal effects of
export on the plant performance and its impact on economic activity using the panel data from
Germany. Comparison of export starters and non-exporters results showed that contribution of
export starters to growth in labor productivity was 6% higher than non-exporters. The outcome
of the results showed that increase in exports has a weaker positive effect on labor
productivity. On the hand, increase in exports has very significant positive effects on growth of
employment and wages.

Aw, Chung, & Roberts (2000) also did an analysis on the relationship between export
and productivity using micro-level evidence from Korea and China. The experience and results
from five exporting industries in China on average had higher productivity compared to the
non-exporters in the country. The author summarized this relationship with the evidence of
knowledge and technology spillover because of the active role and engagement in exporting
market. Additionally, the research also made the conclusion that firms were more productive
due to engagement in world export market which is highly competitive. However, with two
country analysis, Korea had much less correlation between export and productivity than China
backed up by several factors.

Naude (2011), investigates the relationship between exports diversity and economic
performance using the annual data of 1662-2000 by using Applied General Equilibrium (AGE)
models. Result indicates that evidence of a U-shape relationship between income per capita
and export specialization and given the results from Granger causality testing are inconclusive
and not robust with regards to export diversification measures.

Feeny (2013), investigates whether the drivers of economic growth are the same as
those genuine progress in South Korea over the period 1970-2005 by using Genuine Progress
Indicator (GPI). According to the result exports, inflation capital and research and development
are all import in determining the growth in GDP per capita but only physical capital is the driver
of genuine progress. Hamed (2016), investigate on economic diversification and economic
growth in developing countries for the period 2000-2009 by using the Generalized Method of

4
Moments (GMM). Results reveals that and increase in export diversification have significantly
positive effect on the rate of economic growth.

Using data on Italian manufacturing firms, Castellani (2000) estimate regressions of the
growth rate in labor productivity on a measure of export behavior. According to Castellani
(2000), a positive effect on productivity when export behavior is measured as the share of
foreign sales. Conversely, when export behavior is measured as a dummy indicating a firm's
participation in the export market it has no impact on the rate of growth of value added per
worker. This means that entering the export market does not produce any learning experience,
while participating in international activities, specific investments and knowledge accumulated
through time and experience of foreign export contexts are needed in order to capture the
benefits from internationalization.

Beckerman (1962) argues that the causality is from export to growth, which underline
the importance of exporting activity that determines a multiplier effects on investment and
output both in the exporting industry as well as households. In addition, exporting industry is
capable of resource reallocation from the non-trade industry to the export industry, which will
boost the overall productivity of the country. Alternatively, Caves (1971) puts forth an
argument claiming that the direction of the causality is from productivity growth to export.
Krugman (1984) supported Caves argument explaining that, due to economic growth
enhancement of skills and knowledge is being articulated, and this gives the opportunity for
international integration, which will determine exports.

Chen (2007) investigates a causal link between exports and growth in Taiwan by testing
for Granger causality with the use of bounds testing methodology. The main findings in relation
to Taiwan revealed that there exists a long run relationship between exports and labour
productivity and the causality flows from real exports to real output. The author recommends
from the result that Taiwan could benefit through export-led growth strategy to increase the
level of domestic output.

5
The Econometric Model
To investigate the influence of exports on labour productivity, we will use the production
function as the theoretical base for analysis. The equation describes the basic relationship
between output, capital stock, labor inputs and other factors.

Y =f (K , L , X) Equation (1)

Where Y is the real Gross Domestic Product, K is the level of physical capital, L is the labour
input and X is the other factors having an impact on output. It is assumed that output is
dependent on the physical capital, labour units and other factors. In the neoclassical production
function it also has technology but for simplicity we will omit technology variable.

If we divide both side of equation (1), we will get output per labour unit as a function of capital
per labour and other factors.
Y K L
=f ( , , X )
L L L

y=f ( k , X ) Equation (2)

Where y is output per labour unit and k is capital per labour unit in equation (2).

Since X represents the other factors which causes labour productivity, for the purpose of this
study, we have included exports as one of the factors influencing the labour productivity.

y=f ( k , E ) Equation (3)

Now taking the logarithms of the equation we will get the following production function in the
log-linear form.

log ( y )=β 0 + β 1 log ( k ) + β 2 log ( E ) + μ Equation (4)

If we take the lagged variables of dependent variable and the independent variables we get the
following equation. During the model specification, moving from general to specific, the lagged
which was significant were the output per labour and capital per labour. On the other hand, the
lag of exports ( Et ) variable was highly statistically insignificant, thus it was excluded from the
model.

log ( y t )=β 0 + β 1 log ( yt −1 ) + β 2 log ( k )+ β3 log ( k t−1 ) + β 4 log ( Et ) +ut Equation (5)

To prevent any spurious results from equation 5, it will be tested for multicollinearity, autocorrelation,
heteroskedasticity, normality of the errors and functional form. If for any assumptions is not satisfied
then the model will be adjusted using the correction procedures assigned to each assumption to give
the significant results.

6
Test for Multicollinearity
The classical linear regression model assumes that there is no multicollinearity for the model to
be BLUE. The table below shows the correlation between the independent variables of k and
Exports.

K Exports
0.69847
K 1 7
Export 0.69847
s 7 1

The table shows that the correlation between k and Exports is 69.85% which deems that there
is no exact correlation. Thus, the model does not have multicollinearity.
Test for Structural Break: Chow Breakpoint Test
To test for structural break, chow breakpoint test will be used to determine the year at which
the break occurs. The Chow Breakpoint test works by fitting the equation separately for each
sub sample to see if there is a significant difference in the estimated equations. If the
parameters are unstable while fitting the equation then it is deemed that a break has occurred
at that particular point.
Table 1: Output for Chow Breakpoint Test
Chow Breakpoint Test: 1987
Null Hypothesis: No breaks at specified breakpoints
Varying regressors: All equation variables
Equation Sample: 1981 2010

F-statistic 2.758388 Prob. F(5,20) 0.0472

Log likelihood ratio 15.73470 Prob. Chi-Square(5) 0.0076
Wald Statistic 13.79194 Prob. Chi-Square(5) 0.0170

The null hypothesis of no breaks at 1987 has been rejected since all the three p-values are less than 5%.
We only found 1987 to be the breakpoint of our model.

Remedial for Structural Breaks: Application of Dummy Variables

Structural breaks can cause instability in the parameters, which could be only solved by
correcting for structural break. We created a dummy variable DUM87, where for year 1987
value of ‘1’ was placed. ‘DUM87’ was then included in the equation to run the regression of the
model. The following was the result.

7
Dependent Variable: LY
Method: Least Squares
Date: 10/12/16 Time: 19:40
Sample (adjusted): 1981 2010
Included observations: 30 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

LY(-1) 0.308501 0.126519 2.438381 0.0225

LX 0.064763 0.013782 4.699099 0.0001
LK 0.671371 0.179057 3.749489 0.0010
LK(-1) -0.554632 0.183506 -3.022421 0.0059
DUM87 -0.078240 0.029914 -2.615484 0.0152
C 3.817541 0.761990 5.009960 0.0000

R-squared 0.934780    Mean dependent var 9.109625

Adjusted R-squared 0.921192    S.D. dependent var 0.101693
S.E. of regression 0.028548    Akaike info criterion -4.097606
Sum squared resid 0.019560    Schwarz criterion -3.817367
Log likelihood 67.46409    Hannan-Quinn criter. -4.007955
F-statistic 68.79682    Durbin-Watson stat 2.534856
Prob(F-statistic) 0.000000

The dummy variable ‘DUM87’ is significant and it is negatively related to labor productivity
variable. In 1987, Fiji had a political coup which caused a negative impact on the labour
productivity.
log ( y t )=β 0 + β 1 log ( yt −1 ) + β 2 log ( k )+ β3 log ( k t−1 ) + β 4 log ( Et ) + β 5 DUM 87+ut Equation (6)

Test for autocorrelation: Durbin’s h statistic

Autocorrelation violates the assumption of classical linear regression model where the variable
is correlated with itself at different point in times. It is mostly applied where the error term is
correlated with its error term at different point in times.
Since the regression model also includes the lag of independent variable, the Durbin’s h statistic
test will be done to test for autocorrelation.
log ( y t )=3.82+0.31 log ( y t −1 ) + 0.065 log ( X t ) + 0.67 log ( k t )−0.55 log ( k t−1 ) −0.08 DUM 87+ ε t

d = 2.534856 h statistic = -1.015 α = 0.05 hc = -1.96

Since h < hc, we failed to reject the null hypothesis of no autocorrelation. Thus, the
autoregressive model is not auto correlated.

Test for Heteroscedasticity: Breusch-Pagan-Godfrey Test

8
One of the classical assumption of linear model is that the error is supposed to be
homoscedastic. The assumption of homoscedasticity states that the variance of error is
constant for all the observations:
2
var ( ε i )=σ for all i

While the assumption of heteroskedasticity states that the variance of error is not constant or
all individual observations:

var ( ε i )=σ 2i for all i

Using the Breusch-Pagan-Godfrey test, it would be determined whether equation 5 suffers from
the problem of heteroskedasticity.

Figure 1: Heteroskedasticity Test: Breusch-Pagan-Godfrey

Heteroskedasticity Test: Breusch-Pagan-Godfrey

F-statistic 1.750746    Prob. F(5,24) 0.1614

Obs*R-squared 8.017771    Prob. Chi-Square(5) 0.1553
Scaled explained
SS 4.408612    Prob. Chi-Square(5) 0.4922

2
Test statistic: ε t =a0 + a1 log ( y t−1 ) +a2 log ( X t ) +a 3 log ( k t ) + a3 log ( k t −1 ) + a4 DUM 87+ut

Null hypothesis of no heteroscedasticity H0: a 1=a2=a3=a3=a 4=0

Alternative hypothesis of heteroscedasticity HA: at least one is not equal to zero

Since all the test statistics have p-values of more than 5%, the null hypothesis of no
heteroscedasticity is not rejected, thus the errors are homoscedastic.

Test for Normality of Errors: Jarque-Bera test

The non-normality of the error is bound to have a major impact during hypothesis testing.
When errors in the model is normally distributed, it leads to the assumption that the Ordinary
Least Squares is also normally distributed which derives accurate t statistic for hypothesis
testing. To test for normality, Jarque-Bera test will be used as shown in Figure 1 below.
The critical value of Jarque-Bera test is always 5.99 and it is not affected by either sample size of
the number of parameters.

9
Figure 2: Eviews output for histogram-normality test
8
Series: Residuals
7 Sample 1981 2010
Observations 30
6
Mean 2.44e-15
5 Median -0.003874
Maximum 0.051506
4 Minimum -0.055919
Std. Dev. 0.025971
3 Skewness 0.168363
Kurtosis 2.718297
2
Jarque-Bera 0.240926
1 Probability 0.886510

0
-0.06 -0.04 -0.02 0.00 0.02 0.04 0.06

Since the Jarque-Bera statistic of 0.24 is less than the critical value of 5.99, we fail to reject the
null hypothesis of normality of errors. Thus, the regression model has the normal distribution of
the error term.
Therefore, except for structural break which had been remedied, the model had no case of
multicollinearity, autocorrelation, heteroskedasticity and non-normality of errors

Data
The research covers analysis with sample data from period 1980 to 2010 with total annual
observations of 31. The data for real GDP, employment level and capital stock has been
obtained from Penn World Table. Exports data was obtained from the World Bank’s database
under the development indicators in current values which was converted into constant values
using Fiji’s CPI with the base year of 2010.

Table 2: Descriptive Statistics

MEAN MEDIAN MAXIMU MINIMU STD. SKEWNES KURTOSIS
M M DEV. S
OUTPUT PER 9067.98 8984.12 10808.23 7569.35 919.19 0.33171 2.38589
LABOUR
CAPITAL PER 24657.19 23714.12 31057.48 22109.9 2702.85 1.447678 3.75447
LABOUR
EXPORTS (IN 685.00 634.00 1890.00 149.00 497.00 0.890498 2.92430
MILLIONS
USD)

10
Table 2 shows the descriptive statistics for labour productivity, capital per labour and exports.
The first row shows the averages of output per labour is 9067.98 USD from the period 1980-
2010. This figure is higher than Fiji’s GDP per capita of 4375.41 USD. The maximum output per
labour that was ever attained during the period was 10808.23 USD while the minimum was
7569.35 USD. The differences between the maximum and minimum is close to 3000 USD which
implies that growth of output per labour has been relatively stable.

The second row of Table 2 shows that the average of capital per labour is 24657.19 USD. The
maximum capital per labour is 31057.48 USD while the minimum is 22109.9 USD. The last row
shows the distribution of exports of goods and services data. The average exports during the
period 1980 to 2010 has been 685m USD, the highest export was 1890m USD while the lowest
exports in real values was 149m USD. The skewness for three data set shows that it is positively
skewed implying that most of the data lies on the left part of the normal distribution.

Figure 3: Output per labour and Capital per labour from 1980 to 2010

Output per labour and Capital per labour from 1980 to

2010
35000

30000

25000

20000
USD

15000

10000

5000

0
80 82 84 86 88 90 92 94 96 98 00 02 04 06 08 10
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20

Output per labour Capital per labour

Figure 3 shows the general trend in output per labour and capital per labour from period 1980
to 2010. The trend for output per labour unit is a slight increase over the years. In addition, the
trend for capital per labour unit shows that from 1980 to 1998 it has remained stable while has
increased only from 1999 to 2010 at greater pace. It implies that in the past decade the
government of Fiji has increased the level of capital stock due to is implications on the growth
rate of labour productivity.

11
Figure 4: Export of goods and services

Exports of goods and services

9.4
9.2
9
8.8
8.6
logarithms

8.4
8.2
8
7.8
7.6
80 82 84 86 88 90 92 94 96 98 00 02 04 06 08 10
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20

Figure 4 show the line graph of exports of goods and services from period 1980 to 2010. The
graph depicts that exports of goods and services have continued to increase over the period in
constant values despite a slight reduction from 1997 to 2002. Fiji has been able to continually
grow the export sector through various policies such as tourism activities, attracting foreign
direct investment and provision of subsidies to industries engaged in exports.

12
Estimation Output
Estimation technique
The econometric model (Equation 6) has satisfied the assumptions of classical linear regression
model of homoscedasticity, no autocorrelation, and normality of errors, multicollinearity,
structural break and functional form. It is assumed that upon the satisfaction of the
assumptions of CLRM, ordinary least squares can be used for estimation of parameters since it
is both unbiased and consistent.
Regression output for equation 6

Estimated Equation

log ( y t )=β 0 + β 1 log ( yt −1 ) + β 2 log ( E t ) + β 3 log ( k t ) + β 4 log ( k t −1) + β5 DUM 87+ut

log ( y t )=3.82+0.31 log ( y t −1 ) + 0.065 log ( E t ) + 0.67 log ( k t )−0.55 log ( k t−1 )−0.08 DUM 87+ ε t

Std. Error (0.126519) (0.013782) (0.179057) (0.183506) (0.029914)

T-statistic (2.438381) (4.699099) (3.749489) (-3.022421) (-2.615484)
P-value (0.0225) (0.0001) (0.0010) (0.0059) (0.0152)

R-Squared = 0.9348 F-statistic = 68.79682

Adjusted R-squared = 0.9212 Prob(F-statistic) = 0.000000

The results from equation 6 shows a goodness of fit of 92.12% (Adjusted R-squared) which is
relatively good as it shows that much of the variations in labour productivity is explained by
variations in capital per labour, exports and lag of labour productivity. The probability of F-
statistic is within the significance level of 5%, implying that the null hypothesis of no correlation
of determination is rejected in favor of the alternative hypothesis. Thus, lagged labour
productivity, capital per labour and exports does have a statistically significant impact on the
level of labour productivity.

Interpretation of the coefficients

Constant
The constant of 3.82 implies that labor productivity (y) is 3.82% assuming that all other
variables is held constant.
log ( y t −1 )
The lag labour productivity has the coefficient of 0.31. This implies that on average labour
productivity increases by 0.31% with every increase of 1% in lag labour productivity while
holding other things constant. The p-value of lagged labour productivity is 0.0225 which is

13
within the significance level of 0.05, thus we can conclude that lagged labour productivity has a
statistically significant impact in determining the level of labour productivity.

log ( E t )
The variable of exports has the coefficient value of 0.065. This implies that on average labour
productivity increases by 0.065% with every 1% increase in exports while holding other things
constant. The p-value of the variable exports is 0.0001 which is within the significance level of
0.05, thus we can conclude that exports has a statistically significant impact in determining
labour productivity.

log ( k t )
The variable, capital per labour unit has the coefficient value of 0.67. This implies that on
average labour productivity increases by 0.67% with every 1% increase in capital per labour unit
while holding other factors constant. The p-value of the variable capital per labour is 0.0010
which is within the significance level of 0.05, thus we can conclude that capital per labour unit
has statistically significant impact in determining the level of labour productivity.

log ( k t−1 )
The variable, lagged one period capital per labour unit has the coefficient of -0.55. This implies
that on average labour productivity decreases by 0.55% with every 1% increase in lagged one
period capital per labour unit while holding other factors constant. The p-value of lag capital
per labour is 0.0059 which is within the significance level of 0.05, thus we can conclude that lag
capital per labour variable has statistically significance impact in determining the labour
productivity.

DUM 87
The variable, DUM87 has the coefficient of -0.08.This implies that on average the structural
break in 1987 decreases labour productivity by 8% while holding other factors constant. The p-
value of DUM87 is 0.0152 which is within the significant level of 0.05, thus we can conclude
that DUM87 has statistically significant impact in determining the labour productivity.

14
Results

The capital per labour and exports coefficient and labor productivity lagged one year
used in the regression shows a significant impact on the level of labour productivity and this can
be determined in several ways. That is, an increase capital per worker and exports improves
standard of living, increase in labor skills, technological advance resulting from innovation
activities and contribute to a higher capital stock for example machinery, equipment and
building construction.
The results shows that exports leads to higher productivity levels and exporter remains
more productive upon export entry which is confirmed by the positive estimated productivity
growth. On average a 1% increase in exports lead to 0.065% increase in the level of labour
productivity of Fiji while holding other factors constant. The coefficient of exports also shows
the importance of exports to international markets and new business development in
technology improvement that results labor productivity increase. Therefore, encouraging firms
to expand exports and facilitating industrial exports can increase labor productivity.
The study show that capital per labour plays an important role in determining
productivity. For example, an increase of 1% capital per labor increases productivity by 0.67%.
This indicate that an increase in capital per labour should be increase through utilization of
more advance technology and physical capital, however needs a larger number of labour force
with higher education. Thus, human capital is important as indirect means to promote the
growth of the economy. The variable capital per labour lagged one year, has a negative
statistically significance impact in determining the labour productivity. Conforming the issue
that past period’s capital per labour has negative impact on the productivity as there it leads to
the problems of depreciation of capital stocks and diminishing skill levels. Overall this research
has emphasized the importance of exports to Fiji in which the local producers engage in
overseas to learn better production techniques which eventually leads to betterment of Fiji’s
productivity.

15
Conclusion
The research project supports our hypothesize theory that exports is a key variable in the
determination of labor productivity within the Fiji context. Inflation is a major concern for Fiji,
as it will provide negative externalities to its consumers. Hence export can be a positive remedy
in terms of increasing labor productivity and in return reduces the prices of commodities.

Classical linear regression models techniques have been used to test the relationship between
exports and labor productivity, and the results show a positive impact of exporting on labor
productivity. Increase in human capital activities will have a significant contribution to the
advancement of productivity within the Fiji export markets. Domestic Exporters will tend to
minimize their production cost to attract foreign markets in buying goods from them. While
testing our Model for structural break, it postulates that there is a structural break in the year
1987. This is due to political instability resulting in a coup, in which the Fiji economy faces a
decline in GDP, which causes a major drawback for domestic markets to compete in the foreign
markets. Political stability is extremely important in terms of trade and economic welfare, as the
life of our industries heavily depends on the decisions that our leaders agreed on.

Policy Implication
To expand Fijis participation in the exporting market both domestic and international, the
government should formulate and implement development strategy that consist of objectives and
targets for the country to achieve in terms of trade in locally produced goods and services for the
domestic market and possibly for export, and possible steps that the government and other
stakeholders should entertain to achieve these objectives.

16
Bibliography
Aw, B. Y., Chung, S., & Roberts, M. J. (2000). Productivity and Turnover in the Export Market:
Micro-level Evidence from the Republic of Korea and Taiwan (China). The World Bank
Economic Review, 40(1), 65-90.
Bernard, A. B. (1996). Exporters and Trade Liberalization in Mexico: Production Struc-ture and
Performance. Department of Economics Working Paper. MIT, Cambridge, Mass.

Castellani, D. (2002). Firm's Technological Trajectories and the Creation of Foreign Subsidiaries.
International Review of Applied Economics 16 (3): 359-371.

Caves, R. E. (1971). Export-Led Growth and the New Economic History, Trade, Balance of Payments, and
Growth, Amsterdam: North-Holland.

Chen, S. (2007). Exactly what is the link between export and growth in Taiwan? New evidence from the
Granger causality test. Economics Bulletin

Clerides, S., S. Lach, and J. Tybout (1998). Is: 'Learning-By-Exporting' Important? Mi- cro-Dynamic
Evidence from Colombia, Mexico and Morocco. Quarterly Journal of Economics 113 (3): 903-947.

Feeny, S. (2013). The Determinants of Economic Growth. Soc Indic Res, 1055-1074.

Hamed, K. (2016). Export diversification and economic growth in some selected developing countries.
African Journal of Business Management, 700-704.

Naude, W. (2011). Export diversification and economic performance. Econ Change Restuct, 99-134.

Wagner, J. (2001). The causal effects of exports on firm size and labor productivity: First
evidence from matching approach. HWWA Discussion Paper.

17