Pollution Reduction by Rationalization Hypothesis and Water Pollution in China

China is a major manufacturing and trading nation but has limited freshwater resources (about 6% of the world's freshwater; per-capita availability below world average and one-fifth of the US average). Industrial production is a key source of serious water pollution, creating a water scarcity challenge. While prior literature has focused more on air pollution and aggregate links between trade and pollution, the water pollution–trade nexus at the firm level is less explored. Key questions include: How do firms respond in terms of water pollution to trade openness? Does intra-industry agglomeration exacerbate or mitigate water pollution in the presence of international trade? Why does trade incentivize some firms to pollute more and others less? Which firms gain from trade while undertaking water pollution? Building on heterogeneous-firm trade theory (Melitz, 2003) and the firm-level environmental hypotheses of Cherniwchan et al. (2017)—particularly the Pollution Reduction by Rationalization (PRR) hypothesis—this paper proposes and tests a mechanism linking intra-industry agglomeration, firm productivity, factor reallocation, and water pollution. The core research question is: What is the effect of firm productivity on firm-level water pollution, conditional on export activity? The hypothesis is that trade liberalization and agglomeration increase productivity, leading initially to higher pollution due to scale, but subsequently to lower pollution as resources reallocate from dirty, low-productivity firms to cleaner, high-productivity firms and abatement and techniques improve, yielding an inverted U-shaped relation between productivity and water pollution.

The study relates to extensive trade–environment literature. At the aggregate level, prior work often decomposes trade effects into composition, scale, and technique effects (e.g., Antweiler et al., 2001; Copeland and Taylor, 2013). Evidence on pollution havens and composition effects is mixed and often small compared to scale and technique effects (Cole and Elliott, 2003; Cole, 2004). Trade liberalization’s environmental impacts vary by pollutant and development stage, with some findings that exporters or export plants pollute less (Holladay, 2016; Forslid et al., 2018; Banerjee et al., 2021; He and Wang, 2020; Pei et al., 2021), while other studies show outsourcing of environmental burdens (Wiedmann and Lenzen, 2018) or pollutant-specific outcomes (e.g., reduced SO2 in China after WTO accession: Cui et al., 2020; but increased air pollution in other contexts: Lin, 2017). Cherniwchan (2017) shows NAFTA reduced PM10 and SO2 emissions from U.S. manufacturing. Firm heterogeneity has been increasingly integrated into trade–environment frameworks (Cherniwchan et al., 2017), inspiring hypotheses like PRR, distressed-and-dirty, and pollution offshoring. Studies on agglomeration show links to productivity and emissions (Hu et al., 2015; Cheng, 2016), and that larger or more productive firms may adopt cleaner water technologies (Qi et al., 2021; LaPlue, 2019). Environmental regulation interacts with trade and firm selection (Egger et al., 2021), with mixed evidence on regulation effectiveness and costs (He et al., 2020; Shapiro and Walker, 2018; Duan et al., 2021). In water pollution specifically, national-level regulation may be insufficient, requiring international coordination (Carrascal Incera et al., 2017). The Environmental Kuznets Curve (EKC) for water pollution is context-dependent (Lee et al., 2010), and spatial heterogeneity in economic gains and environmental costs exists (Zhang et al., 2019). This paper differs by using firm-level data on wastewater, explicitly modeling heterogeneous firms, agglomeration, and export intensity to test the PRR mechanism at the micro level.

Empirical strategy and models: The study employs a two-stage structural approach aligned with PRR logic. Stage 1 examines the effect of intra-industry agglomeration on firm productivity (competition effect). Stage 2 evaluates how firm productivity (and its square) and export intensity affect firm-level water pollution, controlling for composition, technique, and scale effects, as well as inputs and abatement capacity. Stage 1 (productivity equation): Productivity_it = β0 + β1 Agg_it + β2 Agg_it^2 + β3 Grad_ct + β4 FDI_ct + β5 Export_it + β6 KL_it + β7 Wage_it + β8 Emp_it + ε_it. Productivity is labor productivity (total revenue per average employees). Agglomeration is an intra-industry agglomeration index at the city–industry–year level based on firm employment distribution; both linear and squared terms capture potential nonlinearity. Controls include city graduates (human capital), city FDI, firm export intensity (exports/revenue), capital–labor ratio (KL; total assets per average employees), wage (total payroll per employee), and firm size (employees). Stage 2 (pollution equation): E_it = α0 + α1 Productivity_it + α2 Productivity_it^2 + α3 Export_it + α4 KL_it + α5 rKL_it + α6 KL_it^2 + α7 Wage_it + α8 rWage_it + α9 rWage_it^2 + α10 Revenue_it + α11 Water_it + α12 Purify_it + α13 rWater_it + ε_it. E_it is firm-level polluted water volume. Export is export intensity. KL is capital–labor ratio, with rKL its value relative to national average. Wage is firm average wage, with rWage relative wage and its square. Revenue proxies scale. Water is clean water input volume; rWater is relative water usage to national average. Purify counts water-purification equipment (abatement capacity). All variables are in logs unless noted. Agglomeration measure: Intra-industry agglomeration at city–industry level is constructed using the distribution of employment across firms in the same city–industry following Cheng (2016) and O'Donoghue and Gleave (2004), capturing the extent to which employment in industry j is concentrated among firms in city c and year t. Interactions and marginal effects: To probe mechanisms, fixed-effects models estimate interactions between productivity and export (export effect), productivity and KL (composition effect), and productivity and wage (technique effect): E_it = a + a1 c.Productivity##c.Export + a2 c.Productivity##c.KL + a3 c.Productivity##c.Wage + controls + FE + ε_it. Additional interaction tests examine Employees##Export to assess labor reallocation, and exporting agglomeration (Agg_export)##KL and Agg_export##KL^2 to study resource reallocations under export clustering. Average marginal effects are graphed to identify thresholds where effects change sign. Data: The panel merges China's Annual Survey of Industrial Firms (ASIF; economic indicators, firms with annual income >5 million Yuan), Annual Environmental Survey of Polluting Firms (AESPF; environmental indicators), and national aggregates from China Statistical Yearbook (CSY). Firm-level ASIF and AESPF are merged by firm name and year. The study period is 2000–2012, the most recent firm-level years available to the authors for these datasets. Productivity is primarily measured as labor productivity to maintain full time coverage (TFP available only for 2000–2007). Econometric approach: Given the structural linkage where Stage 1’s dependent variable (productivity) enters Stage 2 as an explanatory variable, the system is estimated by three-stage least squares (3SLS) to address endogeneity. Fixed-effects models are used for interaction analyses in the mechanism and extensive discussion sections. Models include year and sector fixed effects. Sample sizes in main tables are around 20,200–20,930 firm-year observations with 13,184 firm identifiers for FE interaction models. R-squared values in reported models range around 0.58 (productivity equation) and 0.82 (pollution equation).

- Intra-industry agglomeration and productivity: Agglomeration has a statistically significant inverted U-shaped effect on labor productivity. At lower levels, agglomeration raises productivity (competition/knowledge effects), but beyond a threshold, the marginal effect turns negative. - Productivity and water pollution: Firm labor productivity exhibits a statistically significant inverted U-shaped relationship with wastewater emissions. Initially, rising productivity increases pollution (scale), but after a threshold, higher productivity reduces pollution, consistent with PRR. This supports the view that resource reallocation and selection effects lower emissions among high-productivity exporters. - Export intensity: Export intensity is positively associated with wastewater emissions in baseline estimates (higher exports scale up production). Interaction results show Export×Productivity is positive and significant, indicating the export effect on pollution is stronger at higher productivity levels in the lower-productivity range. Average marginal effects reveal a productivity threshold (log productivity ≈ 6.2) at which the marginal effect of export on pollution switches sign: negative below 6.2 and positive above 6.2, evidencing stage-dependent impacts. - Composition effect (capital–labor ratio): KL has a positive effect at lower levels but a negative effect at higher levels (inverted U). The interaction Productivity×KL is negative and significant, suggesting that increasing productivity weakens the pollution-raising effect of capital intensity; firms can become more capital-intensive with less water pollution as productivity increases. - Technique effect (wage): Wage shows an inverted U-shaped relation with pollution (positive then negative), and Productivity×Wage is positive and significant, implying productivity can amplify the technique (wage-related) effect on pollution in certain ranges. - Scale, inputs, and abatement: Revenue (scale) and Water input are positively associated with polluted water volume. Relative water use (rWater) is negatively associated with pollution, suggesting that firms using more water relative to the national average may have higher productivity/efficiency leading to relatively less polluted water. Purification equipment count (Purify) is positively associated with pollution, consistent with firms installing more abatement as they produce more polluted water (lower abatement cost burden encourages higher output/pollution). - Labor reallocation channel: Employees×Export interaction is negative and significant, indicating that additional employees reduce the marginal effect of export intensity on pollution, consistent with trade-induced reallocation of labor from dirtier to cleaner firms and the adoption of labor-saving technologies. - Export agglomeration and resource reallocation: Exporting agglomeration interacted with KL is positive in a simple model, but when including KL^2 interactions, Agg_export×KL becomes negative and Agg_export×KL^2 positive, indicating a nonlinear relationship: export clustering does not always raise pollution and can reduce it beyond certain capital intensity levels due to competition-driven productivity gains—again consistent with PRR. - Regional disparities: Eastern China’s firms are more likely to generate higher wastewater emissions than firms elsewhere; central and western regions show relatively lower pollution, reflecting spatial heterogeneity in exporting activity, industrial structure, and enforcement. - Quantitative notes from reported tables: 3SLS models report n ≈ 20,200 with R² ≈ 0.58 (productivity equation) and R² ≈ 0.82 (pollution equation). Interaction FE models report n ≈ 20,930 and 13,184 firm IDs. Thresholds from marginal effects: log productivity ≈ 6.2; log export intensity ≈ 4.7 for employees’ conditional effects.

The findings directly address the research question by demonstrating that firm productivity, shaped by intra-industry agglomeration and trade, has a nonlinear effect on water pollution: productivity growth initially raises wastewater due to scale but eventually reduces it as rationalization, selection, and reallocation mechanisms dominate. This pattern supports the Pollution Reduction by Rationalization (PRR) hypothesis in China’s manufacturing sector. Trade liberalization allows high-productivity firms to export and expand, while low-productivity, dirtier firms shrink or exit; labor and other factors reallocate towards cleaner, more productive firms. Interaction analyses show how export effects are conditional on productivity, capital intensity, and wages, clarifying when scale effects dominate and when composition/technique improvements mitigate pollution. Export agglomeration further enhances competition and productivity, contributing to eventual pollution reduction. Despite positive scale effects (revenue) and initial export-driven increases in wastewater, the reallocation and efficiency channels outweigh these, leading to net reductions among more productive exporters. Spatial analysis indicates regional heterogeneity, with eastern regions exhibiting higher wastewater emissions, reflecting concentration of exporting industries and agglomeration patterns. Overall, the results integrate firm heterogeneity into the trade–environment nexus and provide micro-level evidence for PRR in water pollution.

The study provides firm-level evidence from China (2000–2012) that intra-industry agglomeration raises productivity, and that productivity has an inverted U-shaped relationship with water pollution. High-productivity firms that export ultimately emit less polluted water, validating the Pollution Reduction by Rationalization hypothesis for wastewater. Composition and technique effects also matter: with rising productivity, capital intensity can be associated with lower pollution, and wage-related technique effects interact with productivity. Policy implications include prioritizing productivity enhancement and facilitating exports while enforcing water pollution controls and supporting abatement, particularly within sectoral agglomeration strategies. Coordinated industrial and water protection policies linked to export promotion can be effective. Future research avenues include: exploiting detailed firm-level import data to analyze intra-industry trade and pollution bi-directionally; integrating patent data to study green technology innovation and spillovers on wastewater reduction; and testing related hypotheses (distressed-and-dirty industries, pollution offshoring) across pollutants and contexts.

- Measurement of productivity: Labor productivity is used as the main proxy for firm productivity to maintain coverage over 2000–2012; total factor productivity (TFP) is available only for 2000–2007, and labor productivity is acknowledged as an imperfect proxy. - Trade detail limitations: Lack of detailed import and destination-specific export data prevents distinguishing pollution effects by trade partners or by intra-industry trade flows at the firm–country level. The paper notes it cannot resolve whether exported versus imported modes within the same industry produce more or less wastewater due to data constraints. - Data access and scope: ASIF and AESPF microdata are used under license and are not publicly available; the study period ends in 2012, which may limit generalizability to more recent regulatory or technological changes. - Potential endogeneity and measurement error: Although 3SLS addresses structural endogeneity between productivity and pollution equations, unobserved factors or measurement errors in environmental variables (e.g., wastewater volumes, purification equipment) may remain. Regional policy heterogeneity and enforcement intensity are controlled via fixed effects but may not be fully captured.