3.1.1 Outcome variable – environmental actions

The outcome variable is grouped into three main categories, as follows: (a) the sum of the actions taken by the enterprise in a particular year to address different environmental issues (environmental_action); (b) the sum of the investments taken by the enterprise to purchase equipment to address different environmental issues (environmental_investment); (c) the sum of the value of such equipment, in Vietnamese Dong, that the enterprise has purchased to address different environmental issues (environmental_equipment). These indicators represent enterprises’ intentional actions to address environmental issues within their immediate location. For example, the sum of the actions by the enterprise to address environmental issues includes the sum of all other actions apart from investment in equipment for environmental pollution control. In contrast, the sum of investments in equipment for environmental actions and the cost of such investments are diverse as the latter accounts for the monetary component of the equipment purchase.

The first measure of environmental action, environmental_action, is the sum of the number of actions taken by the enterprise to address environmental issues, including air quality, fire, heat, lighting, noise, waste disposal, water pollution, soil degradation/pollution and other environmental issues. On average, the surveyed enterprises took one environmental action for each of the three survey periods. This is similar to the average (i.e., 1.124) for the entire period's environmental actions (see table 1).

Table 1.

Summary statistics of the variables of interest

Note: SD implies standard deviation, whose values are included in the brackets.

The second measure of environmental treatment, environmental_investment, is also computed as the sum of the number of the enterprise's investments in equipment to address different environmental issues, including air quality, fire, heat, lighting, noise, waste disposal, water pollution, soil degradation/pollution and other environmental issues. The summary statistics of this variable, as reported in table 1, also show a single investment in equipment for environmental actions for the entire period and across the individual survey years.

Finally, the purchase of equipment to treat different environmental issues (i.e., environmental_equipment) is measured in Vietnamese Dong (VND) as the sum of the amount for the acquisition of equipment to treat air quality, fire, heat, lighting, noise, waste disposal, water pollution, soil degradation/pollution and other environmental issues. The summary statistics for the period 2011 show that the average value of equipment purchased by the sampled enterprises was 3.798 million VND, while the average value of such purchases for the 2013 and 2015 periods was 4.291 and 4.184 million VND, respectively. The average equipment purchase for the entire period was 4.093 million VND.

It is important to state at this point that the environmental compliance guideline of Vietnam stipulates these factors (air quality, fire, heat, lighting, noise, waste disposal, water pollution, soil degradation/pollution)Footnote ⁴ as those environmental actions for public wellbeing, and biodiversity preservation (Nguyen, Reference Nguyen2017). It is, therefore, reasonable to consider these dimensions of environmental issues for our analysis.

3.1.2 Participation in the export market

Our primary explanatory variable, participation in the export market, is a continuous variable that measures the extent of an enterprise's foreign market participation. The primary survey data records the percentage of direct exports of the sampled enterprises to other countries outside the domestic market (direct export). The enterprises record an average of 2 per cent direct export to total sales value (see table 1).

Given the extent of participation of the sampled enterprises in the export market, what remains unclear is how increasing export participation affects the sampled enterprises’ environmental actions. Table 2 shows that enterprises that engage in the export market have significantly higher actions to address environmental issues, invest in equipment to address environmental issues, and have higher monetary value for equipment purchases to address different environmental issues.

Table 2.

Difference in mean of outcome variables across different groups of enterprises

Note: *** denotes significance at the 1 per cent level.

3.1.3 Other covariates

Other covariates are included in the estimation to minimize omitted variable bias issues, improve the model's accuracy, and follow the literature on environmental actions and export participation. The covariates included are a dummy for infrastructure availability (i.e., paved road) in the location of the sampled enterprises, and other enterprise-level characteristics, such as asset holding of enterprises, the value of tax payment by the enterprise, total labor force of the enterprise, the age of the enterprise, the squared value of the enterprise's age, the number of customers of the enterprise, extent of technology usage, enterprise-level corruption, stock of equipment, and measures of innovation. The characteristics of the enterprises’ owners are also considered, including educational qualification, gender and age. The summary statistics of these variables are presented in table 1.

3.2.1 Instrumental variable estimation

A key concern in estimating equation (1) is the endogeneity concern with export participation arising from two notable sources – simultaneity and omitted variable bias. First, the environmental actions of enterprises could drive participation in the export market because environmental initiatives by enterprises could signal better operational standards, which could facilitate access to global markets (Jaffee and Masakure, Reference Jaffee and Masakure2005). In turn, higher operational standards could signal to the international community that such a business is compliant with specific requirements for products and production processes to participate in the global market (see Trifkovic, Reference Trifkovic2017).

Second, those unobserved variables might otherwise have a confounding effect on export market participation and environmental actions. There could be other time-invariant variables peculiar to the Vietnamese business environment that could affect both environmental actions and export market participation. For instance, the enterprise laws of 2000 and 2005 and those rules that are connected with Vietnam's membership in the World Trade Organization (WTO) may have a significant impact on industrial development in the country (Perkins and Anh, Reference Perkins and Anh2009), which could also affect the overall framework of incentives for improved business operations of enterprises. Such confounding effects could result in severe misspecification of the relationship of interest.

To address this endogeneity issue, this study relies on an instrument expected to inform an exogenous variation in the export market participation and not directly affect the outcome variable. In formulating the instrument, this study relied on the gravity model framework's argument that export market participation is strongly correlated with the distance to potential market destination and geographical location. Such an argument is premised on the idea that transportation costs generally rise with the distance between the trading partners, affecting export market participation (see Disdier and Head, Reference Disdier and Head2008).

The main data source has information on the city/province location of the sampled enterprises in Vietnam, and the World Integrated Trade SolutionFootnote ⁷ hosts data for the first 25 countriesFootnote ⁸ where the exports of Vietnam are directed, the value of such exports, and the sector that provides such exports across the sample years (2011, 2013, and 2015). Matching these two datasets, the distance between each enterprise's location and the potential destination of exports across sectors is calculated. Further, each destination country's individual weightsFootnote ⁹ were determined, which were then interacted with the distance variable to compute the instrument. The weights are included as proxy variables for product prices or quality (Wagner, Reference Wagner2016a). As noted in Wagner (Reference Wagner2016b), positive correlations exist between the quality of exported goods and the distance to destination countries.

Therefore, the weights of the destination countries and each export value are different across sectors. Hence, the sector-specific measure of an enterprise's average distance from the city/province where it is located to the export market is computed as the instrument for this study. Similar in spirit, Bratti and Felice (Reference Bratti and Felice2012) examined the relationship between export participation and product innovation of Italian manufacturing firms using an instrument that was computed in a similar fashion.

Formally, the instrument is computed as:

\[\mathop \sum \limits_{J = 1}^{25} \textrm{weigh}{\textrm{t}_{sjt}} \times \textrm{distanc}{\textrm{e}_{pj}},\]

where $\textrm{distanc}{\textrm{e}_{pj}}$ is the distance between the location of the enterprise in city/province p and the capital city of the potential export destination country j, which is taken from the coordinates of the Google Earth distance calculator, and measured in kilometers; and $\textrm{weigh}{\textrm{t}_{sjt}}$ is the weight of the export destination country j on the total exports of sector s over time t, which is mathematically represented as $\textrm{Export}_{sj}^t/\mathop \sum \nolimits_{j = 1}^{E = 25} \textrm{Export}_{sj}^t$. Therefore, based on this computation, two enterprises in the same city/province are likely to have different export distances assuming they operate in different sectors. Likewise, two enterprises in the same sector that operate from a different city/province are also expected to have different export distance because they operate in different geographical locations.

Two conditions need to be met for the instrument to be valid. The first condition can easily be verified from the first-stage regressions presented in the subsequent section. The second condition is more challenging to verify because it cannot be tested. However, it is reasonable to assume that the distance between an enterprise location and the export destination country would not affect its decision to engage in environmental actions. However likely, the instrument could be loosely correlated with foreign knowledge spillovers, especially for countries closer to each other. For example, spillovers occur as the number of enterprises located in countries closer to each other increases, and the distance between such countries could capture such spillovers and supply chains (Piermartini and Rubinova, Reference Piermartini and Rubinova2014). As a result of such spillovers, the extent of innovation in the export-origin country increases, which could matter for environmental treatment (Leoncini et al., Reference Leoncini, Marzucchi, Montresor, Rentocchini and Rizzo2019). This threat is likely to apply when considering a one-to-one measure of the distance between countries. However, the distance between the city/province and 25 destination countries and the weight of export across sectors will systematically diminish the knowledge spillover threat.

Some may still argue that geographic location – i.e., the city/province where the enterprises are located – may determine the extent to which enterprises engage in environmental action. For instance, enterprises located in a city/province that borders other countries may be subject to environmental regulation compared to those in other locations, based on the argument that due to the closeness of the city/province to other countries with some measures of environmental regulations (or not), the enterprises’ environmental actions in such a city/province could be easily influenced. As a result, this study's instrument, which includes elements of the distance between the city/province where the enterprises are located and the potential export destination, will pick up such influences from neighboring trading partners. However, this criticism is not particularly relevant in the case of this study given that the neighboring countries that border the city/province from which the sample was selected are minimal trading partners with Vietnam – i.e., Nghe An borders the Lao People's Democratic Republic (Laos), Quang Nam borders Laos, and Long An borders Cambodia (Vu et al., Reference Vu, Nguyen, Smith and Nghiem2015). For example, Laos records less than 1 per cent export share with Vietnam, and Cambodia records about 2 per cent export share for the entire study period (World Integrated Trade Solution, 2019).

It is essential to further state that the reliance on geographical information to identify enterprises’ locations and then calculate the distance with a potential trading partner could imply that enterprises could choose to site their locations in a particular city/province depending on the extent of environmental regulation. Such strategic choices are more likely to be associated with larger enterprises, which is not the case with this study sample. More so, it is likely that the exclusion restriction could be violated because the export weights are likely to be correlated with other observable characteristics, such as productivity, which could be correlated with environmental actions. We further present in the robustness check the consistency of our estimate by exploiting the time variation component of the enterprise-level data.

The first-stage estimation, therefore, is generally defined as:

(2)\begin{equation}\textrm{Expor}{\textrm{t}_{ispt}} = \alpha + \beta \textrm{Instrumen}{\textrm{t}_{spt}} + \delta {X_{ispt}} + {\varsigma _s} + {\rho _p} + {\tau _t} + {\varsigma _s} \times {\tau _t} + {\varepsilon _{ispt}}.\end{equation}

The second-stage regression model, on the other hand, is:<CE: Please check the below equation is correct here>

(3)\begin{equation}E{t_{ispt}} = \alpha + {\widehat {\gamma V}_1} + \delta {X_{ispt}} + {\varsigma _s} + {\rho _p} + {\tau _t} + {\varsigma _s} \times {\tau _t} + {\varepsilon _{ispt}}.\;\end{equation}

Testing the coefficient of $\gamma \;{\rm of}\;{\rm the}\;{\rm estimated}\;{\rm residual}\;\overbrace{v}\limits_{1}$ in equation (3) evaluates whether the export market participation is indeed endogenous.Footnote ¹⁰ The 2SLS estimation yields the average effect of a unit increase in export participation for those groups affected by the instrument. The standard errors are clustered at the city/province- and sector-level because the instrument was constructed at these levels. The estimation also controls for the sector, and city/province fixed effect in the estimations ($\textrm{i}\textrm{.e}\textrm{.},\;{\varsigma _s}\;\textrm{and}\;{\rho _p})$, aiming to adjust for geographic and sectorial time-invariant factors that could affect the relationship of interest. For instance, these fixed effects capture the production technology differences across sectors and sub-national differences across cities/provinces in pollution control. The time fixed effects ${\tau _t}$ are also controlled for, to address those unobserved factors that only vary over the survey periods. To improve the identification, the ${\varsigma _s} \times {\tau _t}$ fixed effect is also adjusted.

3.2.2 Further empirical strategy

A second empirical strategy estimates a difference-in-difference (DID) matching strategy.Footnote ¹¹ Despite the fact that our sample includes more non-exporting enterprises than exporting (see figure A1 in the appendix, an important issue with our identification is that the 2SLS strategy only estimates the local average treatment effects. The 2SLS estimates the effect of participating in the export market for enterprises for which the instrument influences its actual export participation. However, there may be systematic differences between exporting and non-exporting enterprises, and to net out these differences while identifying the effect, a matching procedure is conducted and then the DID is estimated using the matched sample.

The matching procedure is such that a binary variable ‘1’ was created for enterprises with values above 0 per cent sales export to total export in the period 2011, and those with 0 sales export to total export are classified as ‘0’. Mathematically, this transformation is illustrated as ${E_i} = 1$ if enterprise i had an export sales to total sales value that is greater than 0 per cent in the year 2011, and ${E_i} = 0$ if not. Following this transformation, a propensity score matching (i.e., nearest neighbor matching criteria) is then estimated whereby the enterprises that participate in the export market are matched to the non-exporting enterprises with the closest propensity scores. Noting that the propensity score is a conditional probability of export market participation, it was derived from the logistic regression estimation in which the outcome variable is dichotomous, taking the value ‘1’ for export participating enterprises and ‘0’ for non-exporting enterprises (see further explanation in the appendix and the logistic regression used to derive the propensity scores in appendix table A1Footnote ¹²).

Relying on this sample after implementing the propensity score matching on the covariates by imposing common support, a DID estimation on this sample was then carried out. This exercise aims to deal with the observable characteristics that may be present across the sampled enterprises in the 2011 period (see Khandker et al., Reference Khandker, Koolwal and Samad2010). That is, we difference out the permanent confounders of the true effect (as in the DID) while capturing transitory shocks (as in matching techniques). The DID predicts the environmental actions of enterprises by adopting the 2011 survey period as the baseline and the survey periods 2013 and 2015 as the follow-up period.

Therefore the estimated average effect of exporting on environmental actions is:

(4)\begin{align} E{t_{ispt}} & =\alpha + \beta \,\textrm{Year} \times \textrm{Trea}{\textrm{t}_{ispt}} + \alpha \,\textrm{Yea}{\textrm{r}_t} + \sigma \textrm{Trea}{\textrm{t}_{isp}}\notag\\ & \quad + \delta {X_{ispt}} + {\varsigma _s} + {\rho _p} + {\tau _t} + {\varsigma _s} \times {\tau _t} + {\varepsilon _{ispt}},\; \end{align}

where $\textrm{Yea}{\textrm{r}_t}$ is the dummy ‘1’ for the period 2013 and 2015 (and 0 for 2011), $\textrm{Trea}{\textrm{t}_{isp}}$ is a binary variable for the enterprises that export, and the main estimated effect is the interaction of these two variables – i.e., $\textrm{Year} \times \textrm{Trea}{\textrm{t}_{ispt}}$. Therefore, the coefficient of the interaction term between the two groups of enterprises for the periods (baseline and follow-up years) illustrates the expected difference in enterprises’ environmental actions as a result of their export participation.