INTRODUCTION

The analysis of labor productivity and the export dynamic in the manufacturing sector has acquired a certain interest in economic research, due to the rise of the manufacturing sector and its incursion in the evolution of technological industry on the developed countries economy, allowing a greater image of the development level in terms of productivity, industrialization, and export index (World Bank, 2019).

In general terms, the objective of this research is to show the impact generated by exports on the dynamics of labor productivity in the Colombian manufacturing sector, to analyze the behavior of productivity and the process of technology transfer in the manufacturing industry, and to stimulate the economic policy generation focused on the export supply diversification, which allows increasing the manufacturing industry participation in the output of the economy.

Considering the objective of the research, it is necessary to analyze the differentiation in the labor productivity index between exporting and non-exporting firms, by using the hypothesis test of Learning by Exporting and Self-section, which describe the causal relationship between labor productivity and international sales in the manufacturing sector, to encourage the growth of foreign trade of industrial establishments. Besides, the period 2005-2015 is studied due to the external shocks that significantly affect the productivity index and the export dynamic of the Colombian manufacturing sector.

The research added value lies in obtaining significant results that support the studied hypotheses, since the research carried out for the causal relationship analysis between productivity and exports in the Colombian manufacturing sector focuses on macroeconomic aspects, and the geographical locations of industrial activity (Barrientos & Lotero , 2010); on the other hand (Fernandez, 1998) studies the impact of exports on the agglomeration processes of the Colombian industry, as well as analyzing how is the effect of internal logistics costs, and their impact on the international markets entry cost. Finally, the research approach developed by (Giraldo, 2015) toward sunk costs and market entry in the Colombian manufacturing sector.

For the testing process of the learning by exporting and Self-selection hypotheses, certain techniques will be used for data panel analysis, through the fixed effects methodology and the estimator developed by (Arellano & Bond, 1991) for dynamic panel data, where instrumental variables estimator is integrated. The purpose of applying the panel data model is to capture the unobservable heterogeneity between individuals throughout the study time since heterogeneity cannot be detected with time series studies. For this reason, this technique allows for a more dynamic analysis by incorporating the temporal dimension of the data, which generates added value to the study, particularly over long periods (greater than 10 years) (Mayorga & Muñoz, 2000).

The empirical evidence supports the positive effect that the export effect has on firms on the increase in labor productivity indices, where it is observed that the levels of labor productivity increase in Colombian manufacturing firms that decide to export, compared to firms that only focus on the domestic market.

Finally, the document is structured as follows: the first part corresponds to the introduction; the second part analyses the Colombian manufacturing generalities, a brief literature review, and the methodological framework. In the third part, the results of the empirical model are analyzed, and the conclusions are presented.

COLOMBIAN MANUFACTURING SECTOR OVERVIEW

The Colombian manufacturing industry has presented a significant drop in the industrial growth index from the economic opening experienced in the 1990s (Garcia, 2002). In graph 1 it is showed the dynamics of the added value of Colombian industry, where it stands out the year 2009 which registers a considerable fall, caused by the global financial crisis at the end of 2008, the moderate growth of the economy of the United States and China, a decrease in the real average rate of manufacturing industry net production around -3.5%, and a fall in exports index of -3.6%.

Graph 1
Value Added Index (% of GDP) for Colombia
National Accounts World Bank. Open data panel.

The slowdown in the Colombian manufacturing industry was accentuated thanks to external shocks, which drastically affected the dynamics of the national economy through different transfer mechanisms, of which the most relevant was the abrupt fall in international prices of commodities, especially the Brent oil reference price, which stood at less than $50 USD/barrel in 2014. This implies a significant decrease in national income from exports and foreign direct investment that generated a fiscal deficit in Colombian economy (Banco de la República, 2016).

On the other hand, graph 2 shows the industrial groups with the highest share of industrial added value in Colombia on average from 2005-2015 . It can be inferred that the subsectors associated with the transformation of natural resources are those with the best value-added index, the best index of labor productivity, and the highest investment rate in capital stock.

Graph 2
Participation of industrial subsectors in the added value for Colombia (average 2005-2015)
Boletín Técnico. Encuenta Annual Manufacturera (DANE, 2015)

Besides, it is evident that the industrial subsectors with high technological incorporation such as the basic iron and steel industry, the chemical and pharmaceutical subsector, as well as the assembly and manufacture of motor parts, vehicles and engine manufacturing, present a low rate in the industrial added value participation and a significant loss in productivity rates (Clavijo, et al., 2012). This loss of dynamism is product of financial resources migration towards the mining and oil industry experienced by the Colombian economy.

Taking into account the current situation of the Colombian manufacturing sector, the National Government and Ministerio de Comercio, Industria y Turismo seek to promote public policy strategies focused on promoting a change in the productive structure of the economy and export diversification towards the goods of the manufacturing industry which have higher added value.

To achieve this objective, the Colombian government proposes the development of the manufacturing industry through the implementation of public spending policy reform starting in 2012, where the goal is to stimulate investment in research and development (R&D ) and increase the capital stock through Law 1530 of 2012 (Secretaria del Senado, 2020). This law corresponds to the budget regulation of the Sistema Regional de Regalias, which is the mechanism for investment in physical infrastructure and connectivity, R&D, and investment in education and production process innovation, where the priority of this public policy is to increase the development of scientific knowledge and the implementation of technological development in the national industrial sector.

Regarding the Colombian manufacturing exports, they presented a deficit in their trade balance; in graph 3 it is possible to appreciate the dynamics of Colombian exports.

Graph 3
Export growth rate by industrial sector in Colombia (2006-2015)
Base de datos de exportaciones Departamento Nacional de Estadística de Colombia (DANE) .

According to the export report of the Departamento Nacional de Estadística (DANE, 2015), Colombian manufacturing sector international sales presented a fall of -11% that is equivalent to $ 1,135 million dollars in 2015, as a result of the moderate growth of the Chinese economy and the European Union economy, specifically the German economy which presented a decrease of 2.3%. On the other hand, exports to Latin America showed a loss in their dynamism thanks to the lower growth of the aggregate demand for manufactured and intermediate goods from Brazil, Chile, and Argentina, which had a drop of 8% on average.

In general terms, Colombian foreign sales decreased by 32.5% in 2015, where oil and gas exports registered a 46.8% drop (DANE, 2016), the largest fall in the past decade caused by external shocks, which generated a fiscal deficit of 3.73% of GDP in 2015 that forced the National Government to process a long-term tax reform which reduces the vulnerability of public finances.

In summary, the Colombian manufacturing exports for the period 2010-2015 presented a contraction of -13%, where the international sales of non-metallic minerals fell by 50.3%. Additionally, there was a decrease in exports of chemical products, food and beverages, which represented 30%. This situation caused a significant deficit in the trade balance of the national industry.

BACKGROUND STUDIES

In the economic literature, the causal relationship between productivity and exports is analyzed by taking as a point of reference the research work of (Bernard & Bradford, 1999), where they demonstrate with empirical evidence that exporting companies in the United States manufacturing sector increase their productivity and capital stock levels before exporting, to face the costs related to the insertion of their products in the international market.

The hypotheses that describe this type of behavior are learning by exporting and self-selection. The Learning by exporting hypothesis explains the positive relationship between productivity and exports, caused by the positive impact generated by technology transfer on the productivity of manufacturing companies, carried out by the interaction process that companies have with international market players (suppliers, customers, marketers, and others) (Bernard & Bradford, 1999).

The self-selection hypothesis on its side indicates that only firms, which show the best productivity and quality index in their final or intermediate good, can carry out international sales to markets that present social and economic characteristics similar to the country of origin (Bernard & Bradford, 1999).

Regarding the hypotheses mentioned, three research lines are identified that analyze the causal relationship between exports and labor productivity in the manufacturing sector. The first research line focuses on analyzing the Learning by exporting and self-selection hypotheses. The second research line focuses on the “entry cost” of the international trade of manufactured goods. The third research line studies the geographic characteristics associated with the industrial agglomeration of exporting firms in the manufacturing sector.

From the first research line regarding the analysis of the Learning by exporting and self-selection hypotheses, the work of (Bernard & Bradford, 2004) for the manufacturing industry in the United States stands out, where the empirical evidence is overwhelming in arguing that only the most productive firms are capable for entering the international market. On the other hand, the empirical evidence of the positive effect of technology transfer indicates that there is an increase in the productivity index in the short term, but this effect tends to disappear in the long term.

To carry out the estimates they propose a panel data analysis with random effects that allow predicting the causal relationship between the variables. Besides, they quantify the decision to export through a dummy variable, which establishes the condition of having a higher income than the international market entry cost. Finally, the transition period that exists when a firm sells abroad and decides to leave the international market is taken into consideration.

The conclusion of (Bernard & Bradford, 2004) is that exporting firms in the North American manufacturing sector increased their productivity rates in an average range of 12% -19%, compared to non-exporting firms for the period 1984- 1992. Surprisingly, the evidence that argues the effect of Learning by Exporting tends to be not statistically significant, due to the difficulty of modeling the decision to export made by firms in the manufacturing sector.

On the other hand, (De Loecker, 2007)presents empirical evidence that supports the Learning by Exporting hypothesis for the Slovenian manufacturing sector for the period 1994-2000. Where they use labor productivity as an approximate measure of the total factor productivity index of the manufacturing sector, and they conclude that exporting companies are on average 29.59% more productive than non-exporting companies.

To obtain statistically significant results they propose the creation of control groups, which quantify the decision to export, in terms of the expected probability of starting their sales abroad. For this reason, they use Propensity Score Matching (PSM), where a probit model is estimated. The results show that on average the companies that start to export are 8.8% more productive due to firms increasing their participation in the international market, which indicates a positive effect of the learning-by-exporting hypothesis.

In the case of Latin America, (Alvaréz & García, 2010) analyze the causal relationship and the impact that exports and technological innovation have on the productivity index of the Chilean manufacturing industry, where the empirical evidence favors the hypothesis that only the most productive firms are capable of exporting. The econometric model implemented is based on the one developed by (De Loecker, 2007).

The methodological approach of (Alvaréz & García, 2010) focuses on the Matching technique in the context of differences in differences, which allows obtaining more precise estimators because it is possible to control for unobservable heterogeneity between the signature groups. The results of the estimates are consistent in arguing an increase in the productivity index of 10% on average of firms that begin to export, which indicates that only firms that previously improved their productivity and installed capacity can enter the international market.

The second research line regarding "entry costs" refers to the transactions carried out by firms to enter the international market. For the case of Colombia, the research of (Giraldo, 2015) estimates the export determinants of the Colombian manufacturing sector through a macroeconomic model of simultaneous equations, where the supply, and demand of exports and the costs of entry to the international market (freight, international physical distribution, tariffs, among others) are taken into account in order to estimate the price and income elasticity of the Colombian manufacturing sector. The results show that exports from the Colombian manufacturing sector are focused on food products, textiles, industrial chemicals, basic iron and steel metals, and non-ferrous metals; being food products the exports mainline with an average participation of 25% of total exports for the period 1998-2013.

Finally, the third research line focuses on the analysis of variables associated with economic geography and industrial agglomeration. The research of (Greenaway & Kneller, 2008) studies the geographic agglomeration effect of the United Kingdom manufacturing industry, and its relationship with the export potential, as well as examining the empirical evidence of the Spillover¹ between the exporting and non-exporting firms. For this reason, the analysis focuses on demonstrating the impact of the size of the firm, productivity, and geographic agglomeration.

Taking these antecedents into consideration, the goal of this article is aimed at carrying out an analysis of the causal relationship between productivity and exports in the Colombian manufacturing sector. The main objective of this research is to establishing inputs for the generation of economic policy that allows improving the productive processes of the national manufacturing sector, to increase competitiveness and international sales of manufactured goods, which allows stimulating the growth of this sector in the share of GDP, which has lost dynamism in the last 30 years.

The added value of this research its analyzing the process of technological transfer existing in foreign trade operations, which is measured through the knowledge acquired in the international market and foreign direct investment, in order to promote investment in R&D projects that allow increasing the implementation of new technologies in the productive system of the Colombian manufacturing industry.

Currently, Colombia has one of the lowest rates in R&D investment, as well as investment in scientific activities in the OECD member countries. Colombian investment in these types of activities is less than 0.5% of the GDP, compared to developed economies such as Japan, China, the United States, Germany, and France, which have percentages of investment in R&D activities more than 3% of its GDP on average (Observatorio Colombiano de Ciencia y Tecnología, 2016).

For this reason, the Colombian Government approves law No. 2190 of 2016 corresponding to the budget regulation of Sistema Regional de Regalías, whose purpose is the creation of equity conditions in the income distribution from the exploitation of non-renewable natural resources, to promote the development of productivity and competitiveness of the different departments, in the form of decentralized budgets for investment in physical infrastructure and connectivity, investments in R&D, education and in research groups; that increase the capital stock of the national industry and improve the export offer diversification.

METHODOLOGY

To carry out the estimates, the micro-data available from the Encuesta Anual Manufacturera (EAM) performed by the Departamento Administrativo Nacional de Estadística (DANE) of Colombia is used, which has a census of more than 3,000 industrial establishments and disaggregated information regarding the industrial subsector, production levels, investment in assets, employment level, among other economic variables. Additionally, it is used the micro-data from the Encuesta de Desarrollo e Innovación Tecnológica (EDIT) carried out by DANE at the same census of industrial establishments, where it appeared investment in scientific activities, investments in the technology transfer process, and investments in research and development.

Another fundamental aspect is that the DANE micro-data count with complete series for 2005-2015, which allows us to analyze the impact generated on economic growth and the industrial production index in Colombia by the external shocks such as the world financial crisis in 2008, the moderate growth of the Chinese economy, and the drop in international commodity prices.

Under these precedents and the micro-data characteristics, a model based on the panel data technique is developed, which allows capturing the unobservable heterogeneity between individuals throughout the study time, since this heterogeneity cannot be detected with time-series studies or with cross-sectional studies.

For this reason, this technique allows a more dynamic analysis by incorporating the temporal dimension of the data, which generates added value to the study, particularly in long term (t ≥ 10) (Mayorga & Muñoz, 2000).The general specification of the model with panel data regression analyzed by (Baltagi , 2005) is the following:

Panel data models are interpreted through the error component term (Parra, 2016), from equation (1) the error term $u_{it}$ is decomposed as follows:

Where the term $u_i$ represents the unobservable effects that differ between the study units, which do not vary over time; ${\lambda }_t$ represents non-quantifiable effects that vary over time, but not between study units; finally, $e_{it}$ is the purely random error.

The estimation of ${\alpha }_i$ and $\beta$ is carried out by ordinary least square, considering that this model presents a significance loss on its degree of freedom. The F test² is a great help in this type of model to check if ${\alpha }_i=\alpha$ for any $i$. On the other hand, it should be noted when you want to include a constant term you only have to enter N-1 dummy variables. The estimator of the fixed effects model has the following representation:

Within the analysis of the causal relationship between productivity and exports, it is common to find endogeneity between the error term and the explanatory variables of the model, because there is evidence of a common tendency relationship and self-selection between productivity and exports corresponding to a selection bias in the empirical model.

In addition, a double causality relationship is observed and indicates that there is a relationship between the unobservable characteristics of productivity such as managerial skills, externalities, and opportunities in the international market, competitive advantages, among other factors that affect the behavior of export companies of Colombian manufacturing sector.

Consequently, this endogeneity condition triggers a bias problem in the estimation of the parameters, for this reason the estimator $\stackrel{⏜}{{\beta }_{\mathit{\text{M}}\mathit{\text{R}}\mathit{\text{E}}}}$ by generalized least squares and $\stackrel{⏜}{{\beta }_{\mathit{\text{M}}\mathit{\text{F}}\mathit{\text{E}}}}$ by ordinary least squares are no longer consistent, unbiased, and less variance.

To solve the endogeneity problem, the methodology developed by (Arellano & Bond, 1991) is used; the difference generalized method of moments used the first differences to control the unobservable heterogeneity and internal instruments, specifically the lag of the dependent variable and another group of variables such as capital and labor input, and lags regarding export status, which helps to solve the problem.

The estimator developed by Arellano & Bond is the product of a modification to the random effects model where the matrix representation refers to a dynamic panel data model represented in equation (4), and whose assumptions are:

However, the presence of lags of the dependent variable in the model shows a correlation between $y_{i,t-1}$ and $(c_i+{\epsilon }_{it})$ of the form $\mathit{\text{C}}\mathit{\text{o}}v[y_{i,t-1}/(c_i+{\epsilon }_{it}\left)\right]={\sigma }_c^2+\delta \mathit{\text{C}}ov[y_{i,t-2}/(c_i+{\epsilon }_{it}\left)\right]$ . Therefore, the estimator of OLS and CGM remain inconsistent and biased. To face this correlation (Arellano & Bond, 1991) propose an approach based on implementing the first differences.

Additionally, the instrumental variable estimator is used under the panel data structure, which is represented by equation (7) and (8):

From equation (7) and (8), it is possible to execute the least squares method in two stages, taking as a reference the characteristics of the matrices ${\text{Z}}_i^{\prime }$, ${\widetilde{\mathit{\text{X}}}}_i$ and ${\tilde{y}}_i$ and the estimator of instrumental variables is obtained $\stackrel{⏜}{{\theta }_{\mathit{\text{I}}\mathit{\text{V}}}}$ in equation 8, which is unbiased and of less variance, and allows correcting the endogeneity problem in the initial model, and thus developing the estimates that describe the impact of exports on the productivity growth in the Colombian manufacturing sector.

EMPIRICAL MODEL

Thanks to the structure of the micro-data from the EAM and EDIT industrial census, it is possible to create panel data for the period 2005-2015, where 96.214 observations were preliminarily obtained. However, after imputing the missing data to obtain a balanced panel, 35.230 observations are obtained corresponding to 3,523 firms in the Colombian manufacturing sector.

The dependent variable of the model corresponds to Labor Productivity (LP) as an approximate measure of Total Factor Productivity (TFP), Because the use of TFP as a dependent variable of the model generates autocorrelation bias, since obtaining the TFP index, it is necessary to perform a linear regression for panel data with fixed effects and generates a severe problem of autocorrelation and endogeneity between the error term and the explanatory variables of the model, which seriously affect the results of the estimates for the Colombian manufacturing sector (Camacho , 2011).

On the other hand, within the explanatory variables of the model are the gross investment as an approximate measure of the capital stock of the firms, the cost and expenses of the employed personnel as an approximate measure of the salary, the amount invested in internal and external R&D, the investment in technology transfer, and finally the variable related to exports. The labor productivity Index is expressed as follows:

Where $\mathit{\text{P}}{\mathit{\text{L}}}_{\mathit{i}\mathit{t}}$, is the labor productivity of the firms, ${\mathit{\text{Y}}}_{\mathit{i}\mathit{t}}$ refers to the added value of the industry, and ${\mathit{\text{L}}}_{\mathit{i}\mathit{t}}$ corresponds to the number of workers employed in the industry. In this way, the index assesses the performance of a company and, at a macro level, the performance of the industry within the study period.

In terms of methodology, there is the reference framework developed by (Bernard & Bradford, 1999), (Newman, et al., 2017) and (Isgut, 2001). Additionally, those firms that have exported in all periods are eliminated from the data panel because this behavior generates a bias in the results of the estimates. For this reason, the number of observations went from 35,230 to 31,588, which represents the elimination of around 400 companies that have this characteristic.

To evaluate the self-selection and learning by exporting hypotheses, equation (10) is used, and the exporting company³ category is conditioned to capture the effect generated by investment in technology transfer that allows increasing productivity and competitiveness of the Colombian manufacturing sector, which presents a significant loss in its growth.

The empirical results of equation (10) are summarized in table 1, where there are variables created that capture the effect of technology transfer and the learning by exporting effect derived from the entry of firms to the international market, as well as variables that describe wages, capital stock, and investment in R&D.

On the other hand, the variables that capture the effect of learning by exporting are: LagExport, which is a dummy variable that indicates whether the firm made exports in the previous period, this variable is related to the effect of the hypothesis since it allows to measure, which is the impact of exports on the productivity index. Likewise, the ExportExp variable refers to the years of experience that companies have in the international market, measured as the number of consecutive years that the company exports. Finally, ExportStop is a dummy variable that indicates a company exit from the international market.

The expected results for the estimators associated with these created variables must be positive and significant to argue the presence of learning by exporting in the Colombian manufacturing sector. A positive and significant value of the coefficient associated with the LagExport variable shows the presence of this effect in the firms, through the relationship between productivity and the export status of the firm, measured by consecutive years of exporting.

Table 1
Empirical results. Panel Data model under Fixed Effects methodology.

Encuesta Anaual Manufacturera (2005-2015)Note: They have a balanced data panel. This table only shows estimators with statistically significant results. Columns (1), (2) and (3) include control variables associated with the CIIU Rev. 3 a.c⁴ industrial subsector, geographic region and period. The F test of significance is specified. Robust standard errors are in parentheses. Significance levels *** p <0.01, ** p <0.05, * p <0.10.

Dependent Variable: Labor Productivity (Log LP)	Ordinary Least Squares Fixed Effects (1)	Ordinary Least Squares Fixed Effects for Self-Selection (2)	Ordinary Least Squares Fixed Effects for Learning By exporting (3)
Export		0,155***	0,155***
		(0,0011)	(0,0011)
Lag Export	-0,021	-0,0431***	-0,0429**
	(0,010)	(0,010)	(0,010)
Lag LP	0,276***	0,274***	0,274***
	(0,0055)	(0,0055)	(0,0055)
ExportExp			0,00162
			(0,0028)
ExportStop			0,00675
			(0,025)
ExportExp / ExportStop			-0,0058
			(0,0066)
Log Lab	-0,186***	-0,224***	-0,224***
	(0,009)	(0,0010)	(0,0010)
Log Kap	0,233***	0,213***	0,212***
	(0,0055)	(0,0056)	(0,0056)
Technicians Salary		-0,0000031***	-0,0000035***
		(0,00000041)	(0,00000072)
Workforce Salary		0,0000041***	-0,0000042***
		(0,00000068)	(0,00000068)
Administrative staff salary		0,0000044***	0,0000044***
		(0,00000059)	(0,00000059)
Import		0,0212*	0,0212*
		(0,0088)	(0,0088)
Internal R&D investment		-0,0000077*	-0,0000077*
		(0,00000035)	(0,00000035)
External R&D Investment
Technology Transfer		-0,0000069*	-0,0000068*
		(0,00000033)	(0,00000033)
R2	0,1462	0,1578	0,1578
F Test of Global Significance	409, 42	1215,79	332,69
Number of Companies	3170	3170	3170
Number of Observations	31588	31588	31588

The empirical evidence of this research agrees with the results of (Bernard & Bradford, 1999) for the United States manufacturing industry, (Alvaréz & García, 2010) for Chilean manufacturing industry, and (Camacho , 2011) for the Mexican manufacturing industry; that the effect of learning by exporting does not generate significant results, since the increase in productivity generated by technology transfer process is not constant or observable in the long term.

In table 1 the results of column (3) stand out, where the coefficient of the variable Export is statistically significant and indicates a positive impact on the levels of labor productivity of exporting firms. The results argue that firms with at least two consecutive years of presence in the international market have an increase in their labor productivity index of 15.5% on average.

On the other hand, the coefficient associated with the variable of investment in R&D and investment in technology transfer present in column (3) shows a negative impact on the labor productivity of firms in the manufacturing sector, which indicates a deficiency in investments in scientific activities, investments in R&D and implementation of new technology in the productive process of companies. Originating a limited implementation of economies of scale that allow increasing the competitiveness levels in the Colombian manufacturing industry.

In the same way, the coefficient associated with the LagLP⁵ variable presents a positive and statistically significant behavior, which indicates the presence of an increase in the labor productivity of the firms before exporting, this implies that companies experience an increase on average of 27.4% in their labor productivity before entering to the international market, which supports the self-selection hypothesis in the Colombian manufacturing sector.

From another point of view, the results obtained by (Newman, et al., 2017) for the Vietnam manufacturing industry supports the positive and significant effect of the learning by exporting hypothesis in the manufacturing sector. For the estimations, they use the Arellano-Bond Difference Generalized Method of Moments, and the empirical evidence shows that there is an improvement in labor productivity through an increase in investment in R&D and technology transfer.

In table 2 column (2) it is showing the results of the estimations under the difference generalized method of moments, where the value of the coefficient associated with the LagExport surprisingly is negative and statistically significant, the similar result obtained through the fixed effects method, which is evidence that Colombian manufacturing industry is not generating an increase in its labor productivity index from the experience in the international market, instead, there is a loss of -4.6% on average.

This negative behavior is generated by the low optimization of the production factors, the low index of qualified labor, and the absence of economies of scale that directly affect the competitiveness of the Colombian manufacturing sector.

The empirical evidence agrees with the results obtained by (Alvaréz & García, 2010) for the Chilean manufacturing industry, where firms classified as exporters show negative and not significant productivity index, indicating that the effect of learning and technology transfer is not generating any impact on the firm productivity that decide to export for the first time in the study period.

On the other hand, when applying the dynamic panel estimator of (Arellano & Bond, 1991), it is observed that the coefficient of the variables ExportExp and ExportStop show negative and non-significant values. This leads to the conclusion that the learning effect and productivity growth originated by international trade operations tend to disappear when firms leave the international market.

Table 2
Empirical Results. Panel Data Model with Difference Generalized Method of Moments (Arellano-Bond)

Encuesta Anaual Manufacturera (2005-2015)Note: There is a balanced data panel. The present table only shows the estimators with statistically significant results. Columns (1) and (2) include control variables associated with the ISIC rev industrial subsector. 3 BC, geographic region and period. Chi-Square test is specified. Robust standard errors are in parentheses. Significance levels *** p <0.01, ** p <0.05, * p <0.10.

Dependent variable: Labor Productivity (Log LP)	Difference Generalized Method of Moments (Arellano-Bond) (1)	Difference Generalized Method of Moments (Arellano-Bond (2)
Export		0,135***
		(0,0177)
Lag Export	-0,0846***	-0,0460***
	(0,016)	(0,01823)
Lag PL	0,269***	0,264***
	(0,0254)	(0,026)
ExportExp		-0,00140
		(0,0049)
ExportStop		-0,03814
		(0,2629)
ExportExp / ExportStop		0,0040
		(0,0052)
LogLab	-0,291***	-0,321***
	(0,0288)	(0,295)
LogKap	0,325***	0,3029***
	(0,0195)	(0,209)
Technician Salary		-0,0000037 ***
		(0,00000010)
Workforce Salary		-0,0000034***
		(0,00000093)
Administrative staff salary		0,0000039***
		(0,00000081)
Import		0,0608***
		(0,0136)
Internal R&D investment		-0,00000039**
		(0,00000012)
External R&D Investment
Technology Transfer		-0,0000069*
		(0,00000033)
R2	N/A	N/A
Wald Chi Square Test	525,87	1035,68
Number of Companies	3167	3169
Number of Observations	25200	28388

In contrast, the impact of the capital stock LogKap on productivity is positive and significant, which is an indicator that firms are investing in fixed assets that allow increasing their productive capacity, and therefore there is an increase in the labor productivity index of 30.2% on average. A result that agrees with the exogenous growth theory developed by (Solow, 1956), where he concludes that the productivity index depends on the level of capital per worker, which means that an increase in machinery and equipment investment generates increases in capacity and labor productivity.

Regarding the estimates of the labor factor LogLab in the Colombian manufacturing sector present in Table 2 column (3), it is inferred that the wages of workers and technical personnel are generating a negative effect on labor productivity of 32, 1% on average. This result implies that factors such as low-skilled labor, the economic slowdown, and the inflationary process, affected the dynamics of wages in the Colombian economy and the labor productivity index (FEDESARROLLO, 2018).

According to the research of (Gori & Sakamoto, 2018), there is a relationship between wages and labor productivity for the different types of workers. The results show that there is a positive relationship between salary and the degree of economic development, this implies that external shock can influence the salary rate and productivity index. In the case of the Brazilian economy, the earnings grow initially much faster than productivity and it caused a reduction in the quality of employment and socio-economic development.

The empirical evidence from Colombia is similar to the Brazilian economy because there is a gap between the technical wage and the professional wage, which harms the labor productivity and the technological development of the economy. This situation affects the participation of the manufacturing sector in the GDP, due to the lack of new technologies and innovation implementation in the economy productive structure.

In conclusion, the solution of the endogeneity problem presented between the variables of labor productivity and exports does not change the level of significance, signs of the coefficients, and the general effect on the estimates of the learning by exporting hypothesis. This means that the effect of technology transfers existing in the international trade and interaction with international market agents are not generating a considerable increase in the labor productivity of the Colombian industrial sector.

CONCLUDING REMARKS

The result indicates the existence of empirical evidence that supports the self-selection hypothesis, where it is observed that the levels of labor productivity increase in firms that decide to export by 27% on average, compared to firms that only focus on the domestic market. Contrarily, the costs associated with employed personnel show a negative behavior due to factors such as inflation, the lack of qualified labor, and the unemployment rate in the industrial sector, which affect the growth of labor productivity.

On the other hand, the capital stock reflects a positive and significant impact on labor productivity of 30% on average; a result that is acceptable because of the installed capacity, and capital goods are not showing the expected effect of economies of scale that allow maximizing production levels. This empirical evidence is relatively low compared to other similar economies in South America such as Chile, Brazil and Uruguay, which have percentages greater than 40% according to the Global Competitiveness Index published by the World Economic Forum (World Economic Forum, 2015).

Likewise, the results in terms of salary show a negative effect on labor productivity, which implies the level of wage and the job skills in the industrial sector are not generating the expected effect. For this reason, the private and public sectors must formulate strategies to improve the employment generation and investment in education, and training in areas of technological development that are the most value-added subsector to diversify the export offer in Colombia and promote the national manufacturing sector growth.

Based on the empirical evidence obtained, it is possible to connect the negative result in terms of technology transfer and learning by exporting hypothesis with the deindustrialization process that the Colombian industry is experiencing, where there is a migration of resources and capital to the mining and services sector, which distort the GDP growth due to the volatility of the income derived from the commodities exports (Polline & Lourenço, 2013).

The recommendations in terms of economic policy that improve the productivity of the Colombian manufacturing sector are linked to the theories of increasing public spending as a factor that enhances economic development. The most relevant theory that argues the hypothesis of an increase in public spending is the macroeconomic theory developed by Keynes (Beneti, 2000), which argues that public spending has a positive effect on aggregate demand and the output of the economy. Additionally, he argues that public spending can be used as a countercyclical tool to reduce the gap in real production.

In consequence, the economic policy research in Colombia should turn towards an analysis substantial increase in public and private investment in R&D; because this type of investment does not exceed 0.5% of GDP, well below developed economies (Yazgan & Yalçinkaya, 2018). Additionally, it is proposed the creation of a public policy that promotes foreign direct investment in the Colombian industrial sector, focuses on the long term.

Finally, this type of initiative will stimulate the development of the manufacturing sector through the implementation of technological development and innovation processes in the industry productive structure, tin order to increase the participation of the Colombian manufacturing industry in the GDP. Lastly, this is the best strategy for stopping the dependence on commodities exports of low value-added, which have caused a significant drop in the Colombian industry and the trade balance.

BIBLIOGRAPHY

Alvaréz, R., & García, A. (2010). Productividad, Innovación y Exportaciones en la Industria Manufacturera Chilena. El Trimestre Económico. No. 77, 155-184.

Arellano, M., & Bond, S. (1991). Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. Review of Economics Studies. 58, 277-297.

Baltagi , B. (2005). Econometric Analysis of Panel Data. West Sussex. England: John Wiley & Sons Ltd.

Banco de la República. (2016). Informe de Gestión 2005-2016. Informe de Transparencia y rendición de Cuentas.

Barrientos, J., & Lotero , J. (2010). Evolución y Determinantes de las Exportaciones Industriales Regionales: Evidencia Empírica para Colombia (1977-2002). Cuadernos de Economía. No. 30 (54).

Beneti, C. (2000). La Estuctura Lógica de la Teoría General de Keynes . Cuadernos de Economía No. 33.

Bernard, A., & Bradford , J. (2004). Why Some Firms Export. The Review of Economics and Statistics, 561-569.

Bernard, A., & Bradford, J. (1999). Exceotional Exporter Performance: Cause, Effect or Both? Journal of International Economics. No. 47, 1-25.

Camacho , D. (2011). Relación Empírica entre la Productividad y las Exportaciones Manufactureras Mexicas, 2000-2008. Análisis Económico. No. 61, vol XXVI, 70-88.

Clavijo, S., Vera, A., & Fandiño, A. (2012). La Desindustrialización en Colombia: Análisis Cuantitativo de sus Determinantes. Publicacions ANIF.

DANE. (2015). Boleín Técnico de Exportaciones. Boletin de Prensa del DANE. Exportaciones 2015.

DANE. (2016). Boletín Técnico de Exportaciones . Boletín de Prensa Exportaciones Departemento Administrativo Nacional de Estadística.

De Loecker, J. (2007). Do exports generate higher productivity? Evidence from Slovenia. Journal of International Economics. No. 73, 69-98.

FEDESARROLLO. (2018). Informe Mensual del Mercado Laboral . ACRIP-FEDESARROLLO. Centro de Investigaciòn Econòmica y Social.

Fernandez, C. (1998). Agglomeration and Trade: The Case of Colombia. Ensayos Sobre Política Económica. No. 33, 85-123.

Garcia, J. (2002). Liberalización, cambio estructural y crecimiento económico en Colombia. Cuadernos de Economía. 21(36), 189-244.

Giraldo, I. (2015). Determinantes de las Exportaciones Manufactureras de Colombia: Un Estudio a Partir de un Modelo de Ecuaciones Simultáneas. Coyuntura Económica: Investigación Económica y Social. No. XLV, 55-99.

Gori, A., & Sakamoto, A. (2018). Does wage reflect labor productivity? a comparison between Brazil and the United States. Brazilian Journal of Political Economy. Vol. 38 No. 4, 629-649.

Greenaway, D. & Kneller, R. (2008). Exporting, Productivity and Agglomeration. European Economic Review. No. 52, 919-939.

Isgut, A. (2001). What´s Different about Exporters? Evidence from Colombian Manufacturing. The Journal of Development Studies. Vol. 37 No. 5, 57-81.

Mayorga, M., & Muñoz, S. (2000). La Técnica de Datos de Panel, una técnica para su uso e interpretación. Departamento de Investigaciones Económicas. Banco Central de Costa Rica.

Newman, C., Rand, J., Tarp, F., & Tue Anh, N. (2017). Exporting and Productivity: Learning from Vietnam. Journl of African Economies. No. 26, 67-92.

Observatorio Colombiano de Ciencia y Tecnología. (2016). Indicadores de Ciencia y Tecnología - Colombia 2016. Observatorio Colombiano de Ciencia y Tecnología.

Parra, F. (2016). Curso de Estadística con R. Instituto Cántabro de Estadística. Documento No. 2.

Polline, M. & Lourenço , C. (2013). Exchange rate, exports and growth: an investigation on the hypothesis of Dutch Disease in Brazil. Brazilian Jounal of Political Economy, Vol 33 No. 1, 82-101.

Secretaria del Senado. (31 de enero de 2020). Secretaria del Senado de la República de Colombia. http://www.secretariasenado.gov.co/senado/basedoc/ley_1530_2012.html

Solow , R. (1956). A Contribution to the Theory of Economy Growth. Quaterly Journal of Economicas. Vol 70 No. 1, 65-94.

World Bank. (2019). Global Economic Prospects. Darkening Skies. Washington D.C.: International Bank for Reconstruction and Developmente / The World Bank.

World Economic Forum. (2015). The Global Competitiveness Report 2015-2016. Insight Report. World Economic Forum.

Yazgan, Ş., & Yalçinkaya, Ö. (2018). The Effects of Research and Development (R&D) Investment on Sustainable Economic Growth: Evidence From OECD Countires (1996-2015). Review of Economics Perspective. Vol 18. Issue 1, 3-23.

Notes

Author notes

julianberrio15@gmail.com