Based on the theory of heterogeneous enterprise trade and enterprise micro-data, this paper investigates the influence of direct export and indirect export on enterprise productivity, and explains the knowledge spillovers of the hypothesis of “learning by exporting” from the perspective of the relationship between export and enterprise productivity. Empirical results show that: direct export improves enterprise productivity which increased with the rising of direct export intensity. By contrast, indirect export exerts little influence on enterprise productivity. We conclude that direct export is an important channel for “learning by exporting” which is of great significance to improvement of enterprise productivity.
More enterprises are entering global market for better profits and higher productivity with the increasingly rapid progress of economic globalization. Domestic and foreign scholars have paid high attention to micro-level investigation on relationship between export and enterprise productivity in recent years. Their researches were mainly about two hypotheses on the basis of heterogeneous enterprise trade theory. The first was “self-selection” hypothesis which assumes that only enterprises with high productivity are able to afford sunk costs for participating in international trade in the export market. The second was the hypothesis of “learning by exporting” which is about the positive role of export in promoting enterprise productivity.
Bernard & Jensen (1999) [
There are fewer literatures about “learning by exporting” compared with those supporting “self-selection”. Bigsten et al. (2004) [
Lileeva & Trefler (2010) [
1) Gross output
As the World Bank did not provide gross output values of enterprises in its database. However, since LnY denotes the log value of added outputs in this paper, sample business inventories are basically flat compared to that of last year. Under the situation that sample business inventories are nondecreasing, the added value of the sales revenue includes all enterprise’s added outputs. so gross output index in this paper is replaced by gross annual sales and its logarithm (LnY) is interpreted as variable.
2) Direct export and indirect export
In the database of the World Bank, investigated Chinese enterprises provided data about their gross sales and shares of domestic sale, direct export and indirect export. These data can be used to calculate amount of direct export (DE) and indirect export (IDE). It should be noted that non-exporters in this paper refer to enterprises that make 100% of sales in domestic market; indirect exporters refer to enterprises that have a positive ratio of indirect export and zero ratio of direct export; and direct exporters refer to enterprises that have a positive ratio of direct export and zero ratio of indirect export.
3) Control variables
On one hand, enterprise productivity is also influenced by other knowledge and channels. On the other hand, China is a large economic entity. Enterprises in different industries and regions may noticeably differ from each other in productivity. Therefore, the paper introduces as many control variables as possible in models to control other influencing factors on enterprise productivity and to adapt to heterogeneity of enterprises on the premise of available data. Control variables include: 1) foreign direct investments (FDI): Caves 1974 [
We believe that enterprises in same industries or with similar regional backgrounds may have common in one or more than one features as data from multi-industry and trans-regional enterprises are used in this paper. More obvious homogeneity is found in enterprises in same industries or with similar regional background than multi-industry enterprises or those with different regional backgrounds due to observable and unobservable factors. Region and industry are used as two dummy variables in order to decrease the possibility of estimation error. Major variables are illustrated in
This paper aims to use samples of Chinese manufacturing enterprises to study whether direct and indirect exporters have higher productivity level than enterprises not engaged in exports. To do this, we estimate following production function:
where the left side of the function denotes gross output and the right side denotes various inputs and other technical elements influencing gross output. The model is used to control variables influencing output and investigate the productive difference between enterprises with heterogeneity of export models. According to assumption, influence of explaining variables on output level reflects the influence on enterprise productivity on the condition that other factors are under control. In order to make estimation, assume that the production function can be rewritten as Cobb-Douglas equation:
where i denotes an enterprise, and
Variable type | Symbols | Meanings | Description |
---|---|---|---|
Dependent variables | LnY | Log value of gross output | Gross output replaced by gross annual sales |
Explaining variables | DE | Direct export | Dummy variable: DE = 1, direct export; DE = 0, non-direct export |
IDE | Indirect export | Dummy variable: IDE = 1 indirect export; IDE = 0, non-indirect export | |
Control variables | FDI | Foreign direct investments | Dummy variable: FDI = 1, foreign direct investment available FDI = 0, no foreign direct investment |
RD | Research and development investments | Dummy variable: RD = 1, R&D investments available in the last 3 years RD = 0, no R&D investments available in the last 3 years | |
LnCU | Capital utilization | Ratio of actual outputs in current year in potential maximum outputs | |
LnHU | Human capitals | Average years of education of staff | |
LnAGE | Enterprise age | Existence years of enterprises | |
RG | Region | Dummy variable: RG = 1, provincial capital, non-capital cities | |
ID | Industry | Dummy variable |
DE is a dummy variable denoting direct exporter (DE = 1, direct export; DE = 0, non-direct export); IDE is also a dummy variable denoting indirect exporter (IDE = 1 indirect export; IDE = 0, non-indirect export); and μ represents random error.
Data used in this paper are 2011 survey data about Chinese enterprises published by the World Bank in 2013 after conducting face-to-face interviews with enterprises. Simple random sampling or stratified random sampling was used in the survey to ensure the randomness of samples. Certain sampling methods were designed to select enterprises that were representative in industry, scale ownership, export and region.
In order to ensure validity, data are processed as follows: data lacking export information and from enterprises in non-manufacturing and unknown industries are deleted to ensure that all variables including output, fixed assets and number of wage laborers are positive. The valid samples are consisted of 1578 private manufacturers from 11 manufacturing industries covering food, textile, clothing and chemicals in 25 prefecture-level cities. These samples are decomposed into domestic market enterprises, direct exporters and indirect exporters according to their export models.
There are 1060 or 67.3% domestic market enterprises, 331 or 16.5% direct exporters and 187 or 16.2% indirect exporters among the total 1578 valid samples.
Before estimating the function (1), we firstly conduct comparative analysis on the mean value of domestic market enterprises, direct exporters and indirect exporters and set domestic market enterprises as baseline group. The mean value t of direct exporters and indirect exporters is inspected respectively, and the results are shown in
Comparative analysis on inspection results in
DE are 0.019 (4.466 - 4.447) higher than non-export enterprises in capital utilization (log values), and the
Variables | Baseline group | DE | IDE | Complete sample | Sample size |
---|---|---|---|---|---|
LnY | 16.659 | 17.841*** | 16.907** | 16.936 | 1579 |
LnL | 4.249 | 5.143*** | 4.545** | 4.471 | 1579 |
LnK | 15.391 | 16.252*** | 15.790** | 15.619 | 1148 |
LnM | 15.370 | 16.598*** | 15.878*** | 15.687 | 1446 |
LnEN | 13.301 | 13.994*** | 13.445*** | 13.462 | 1366 |
LnAGE | 2.430 | 2.451 | 2.460 | 2.438 | 1541 |
LnCU | 4.447 | 4.466** | 4.455 | 4.452 | 1548 |
LnHU | 2.299 | 2.323*** | 2.274 | 2.301 | 1546 |
RD | 0.368 | 0.612** | 0.530 | 0.438 | 1565 |
FDI | 0.040 | 0.188*** | 0.080*** | 0.076 | 1574 |
Notes: ***, **, * indicate noticeable mean difference under significance level of 1%, 5% and 10%, respectively.
difference is significant in statistics; and the former is much higher than the latter in terms of average years of education of staff (log values).
Further analysis discovers that log value (LnY) of output level average 17.841 and 16.907 in DE and IED respectively, both higher than the 16.659 in non-export enterprises. However, comparison finds that average values of characteristic variables of IDE are largely between those of non-export enterprises and those of DE. Overall, difference between DE and non-export enterprises is more noticeable than that between IDE and non-export enterprises.
Least square regression is used in the function (2) in order to estimate the contributions of different export models (DE and IDE) to productivity. Results are shown in
As many independent variables are used in models, variance inflation factors (VIF) are used to check multicollinearity before employing OLS regression in the function (1) in order to avoid possible influence caused by
Variables | Coefficient and significance level | |
---|---|---|
LnL | 0.337*** | 0.342*** |
(0.023) | (0.023) | |
LnK | 0.041*** | 0.040*** |
(0.015) | (0.015) | |
LnM | 0.498*** | 0.499*** |
(0.018) | (0.018) | |
LnEN | 0.161*** | 0.157*** |
(0.017) | (0.018) | |
DE | 0.099* | 0.098* |
(0.054) | (0.055) | |
IDE | 0.019 | 0.034 |
(0.065) | (0.067) | |
LnAGE | 0.067* | 0.067* |
(0.040) | (0.040) | |
LnCU | 0.117 | 0.116 |
(0.111) | (0.112) | |
LnHU | 0.560*** | 0.510*** |
(0.100) | (0.102) | |
RD | 0.019 | 0.004 |
(0.044) | (0.045) | |
FDI | 0.011 | 0.007 |
(0.074) | (0.074) | |
ID | No | Yes |
RG | No | Yes |
R-squared | 0.8560 | 0.8583 |
Adj R-squared | 0.8544 | 0.8550 |
Notes: ***, **, * indicate noticeable mean difference under significance level of 1%, 5% and 10%, respectively.
correlation between independent variables. Specifically, larger value of VIF indicates more serious multicollinearity. There is no multicollinearity when VIF is between 0 and 10, strong multicollinearity when VIF is between 10 and 100 and serious multicollinearity when VIF is larger than 100. VIFs of all dependent variables in this paper are smaller than 10 and the tolerance (1/VIF) is larger than 0.1. Thus there is little multicollinearity in the models specified by this paper. Empirical results in
As sample enterprises are from different regions and more than one industry, the paper employs region control variable (RG) and industry control variable (ID) to do re-estimation. It’s important to note that all enterprises among samples are from major Chinese cities and divided into two types based on their location in administrative centers or not (ID = 1, enterprises located in provincial capitals or direct-controlled municipalities; ID = 0, enterprises located in cities except provincial capitals or direct-controlled municipalities). Dummy variables of region and industry are not listed in
Empirical results in model (1) have proved that DE has higher productivity level than non-export enterprises while it is not significant for IDE. Robustness test is further empirical research on DE and IDE. Firstly, DE and non-export enterprises are selected from samples as new research objects. The percentage of direct export in sales volume of enterprises (namely, direct export intensity) is decomposed into five levels (0% - 20%, 20% - 40%, 40% - 60%, 60% - 80%, 80% - 100%) to compare productivities of enterprises with different direct export intensity. Non-export enterprises constitute a reference group.
Take
where Control is a control variable reflecting heterogeneity of enterprises and specific variables are same with those in model (1). The results of OLS regression are shown in
Variables | Coefficient and significance level | |
---|---|---|
DE1 | 0.165** | 0.137* |
(0.082) | (0.081) | |
DE2 | 0.017 | 0.022 |
(0.088) | (0.088) | |
DE3 | 0.246** | 0.254** |
(0.126) | (0.127) | |
DE4 | 0.354** | 0.348** |
(0.153) | (0.154) | |
DE5 | −0.032 | −0.036 |
(0.114) | (0.119) | |
ID | No | Yes |
RG | No | Yes |
R-squared | 0.855 | 0.858 |
Adj R-squared | 0.853 | 0.854 |
Notes: ***, **, * indicate noticeable mean difference under significance level of 1%, 5% and 10%, respectively.
OLS regression method is used to analyze data of Chinese manufacturing enterprises and several statistical methods are used to test the robustness of analysis results, proving the relationship between direct export and productivity. Just as the prior assumption, we find that DE enterprises have higher productivity level than non-export enterprises but IDE enterprises little differ from non-export enterprises in this term. We attribute this difference to following reasons: first, direct exporters have more access to learning from overseas market and they can update production technologies and improve product quality in an accurate and timely manner via direct contact with international market, cooperators, competitors and consumers. Second, continuous contact with overseas market enhances direct exporters’ learning ability. Interaction with technology and market system of many other countries enables direct exports to deeply understand demands of foreign customers and the latest technical information, then signing contracts or starting equity cooperation with leading enterprises in the international market. This paper verifies and testifies the feasibility of this theoretical analysis.
In term of policy, export enterprises usually choose direct export considering healthy and sustainable development and the limitation of indirect export in utilizing foreign market experience, technologies and knowledge to improve productivity. Enterprises still performing indirect export should start transition to direct export in order to improve productivity.
Although empirical analysis in this paper controls many factors influencing enterprise productivity and the robustness of results is tested, there are limitations in samples as the research objects are multi-industry enterprises and there only cross-section data in 2011. If more data are available in following researches, time series can be added to the analysis of panel data and choropleth tools can be used to further study the relationship at higher level. In addition, results of separate researches on different industries are more convincing on the premise of adequate available data.
XiaolongZhang, (2016) Export Mode Heterogenicity and Enterprise Productivity—A Test about Learning by Exporting. Modern Economy,07,84-91. doi: 10.4236/me.2016.71010