University of Southern California Dissertations and Theses

The economy-environment trade-off in China's air pollution policies

(USC Thesis Other)

The economy-environment trade-off in China's air pollution policies

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Copyright 2024 Renzhi Shen
THE ECONOMY-ENVIRONMENT TRADE-OFF IN CHINA’S AIR POLLUTION POLICIES
by
Renzhi Shen
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PUBLIC POLICY AND MANAGEMENT)
December 2024

i
TABLE OF CONTENTS
LIST OF TABLES .........................................................................................................................iii
LIST OF FIGURES ....................................................................................................................... iv
ABSTRACT.................................................................................................................................... v
CHAPTER I. INTRODUCTION.................................................................................................... 1
Overview..................................................................................................................................... 3
CHAPTER II. THE UNINTENDED CONSEQUENCES OF SHORT-TERM BANS ON AIR
QUALITY IN CHINA.................................................................................................................... 7
1. Introduction......................................................................................................................... 7
2. Results............................................................................................................................... 10
3. Discussion......................................................................................................................... 18
4. Data and Methods............................................................................................................. 21
CHAPTER III. THE ROLE OF INDUSTRIAL DIVERSITY IN SUSTAINABLE
DEVELOPMENT: EVIDENCE FROM CHINA......................................................................... 25
1. Introduction....................................................................................................................... 25
2. Background....................................................................................................................... 27
3. Empirical framework ........................................................................................................ 30
4. Data ................................................................................................................................... 37
5. Empirical results ............................................................................................................... 40
6. Conclusion and discussion................................................................................................ 44
CHAPTER IV. THE SECTORAL ECONOMIC COST OF CHINA’S AIR POLLUTION
REGULATION CAMPAIGN ....................................................................................................... 47
1. Introduction....................................................................................................................... 47
2. Background....................................................................................................................... 49
3. Empirical framework ........................................................................................................ 51
4. Data ................................................................................................................................... 54
5. Results............................................................................................................................... 56
6. Robustness checks ............................................................................................................ 59
7. Discussion and conclusion................................................................................................ 64
CHAPTER V. CONCLUSION OF DISSERTATION .................................................................. 67

ii
1. Summary of findings......................................................................................................... 68
2. Implications for policy...................................................................................................... 69
3. Future research directions................................................................................................. 70
REFERENCES ............................................................................................................................. 72
APPENDICES .............................................................................................................................. 80
Appendix A – For Chapter II .................................................................................................... 81
Appendix B – For Chapter III................................................................................................. 136

iii
LIST OF TABLES
Table 2. 1 Summary statistics ....................................................................................................... 40
Table 2. 2 Main result Part 1: Industrial diversity impact on environmental regulation
performance .................................................................................................................................. 42
Table 2. 3 Main result Part 2: Industrial diversity impact on the industry-pollution relationship 43
Table 3. 1 Summary statistics ....................................................................................................... 55
Table 3. 2 Main analysis results.................................................................................................... 57
Table 3. 3 Regression result for 2018 as redefined policy year.................................................... 61
Table 3. 4 Regression result for the primary and tertiary sectors ................................................. 63

iv
LIST OF FIGURES
Figure 1. 1 Controlled residual trends and main analysis results ................................................. 12
Figure 1. 2 Effect of the policy on Continuous Emission Monitoring system firms' emission .... 14
Figure 1. 3 Estimated effect of the policy by hour periods........................................................... 16
Figure 1. 4 Estimated effects of the policy by different plant density quartiles ........................... 17
Figure 2. 1 Trends of pollution ..................................................................................................... 38
Figure 2. 2 2015 status of cities: entropy-AQI ............................................................................. 39
Figure 2. 3 2015 status of cities: entropy-ln(secondary sector product)....................................... 39
Figure 3. 1 Average city-level secondary sector size by treatment and control groups................ 56

v
ABSTRACT
Industrializing, developing countries face strong domestic demands for economic growth
alongside emerging environmental issues. Given limited resources and weak institutions, these
competing demands create a trade-off in the policy agenda, representing a common challenge for
environmental policymaking in these nations. This dissertation seeks to deepen the understanding
of the economy-environment trade-off in developing countries by examining China’s air pollution
policies, which have achieved significant pollution alleviation over the past decade.
This dissertation comprises three essays centered on the economy-environment trade-off.
Chapter II explores the influence of political and economic needs on the design of the 2014 APEC
short-term pollution ban in Beijing. While the policy leveraged the flexibility of command-andcontrol regulation to minimize economic costs, it ultimately led to an unanticipated increase in
pollution, exacerbated by weak enforcement institutions. Using a difference-in-differences (DiD)
approach, the study separately estimated and identified a significant reduction in air pollution
during the APEC event, followed by a larger pollution spike immediately afterward. Overall, the
policy is deemed inefficient due to the disproportionate increase in pollution exposure for the
public. These findings offer important insights into how economic incentives and weak institutions
can undermine the efficiency of environmental policy design, while also constraining the range of
available policy instrument options.
Chapter III investigates the urban economic factor of industrial diversity and its potential
to provide economic resilience against policy shocks, thereby easing this trade-off. Employing
fixed effects models and utilizing data on ambient air pollution and industrial firms, the research
examines the efficacy of industrial diversity in enhancing the outcomes of China's "2+26 Cities"

vi
regional air pollution reduction policy scheme, alongside exploring the mechanism. The findings
indicate that cities characterized by greater industrial diversity witness more substantial reductions
in air pollution levels within the area of the regional policy aimed at pollution abatement.
Furthermore, the study exhibits the buffering effect of industrial diversity, which mitigates the
correlation between industrial sector productivity and air pollution levels. The conclusion of the
paper suggests that developing countries could benefit from fostering industrial diversity
throughout their industrialization endeavors, as a strategic approach to mitigate the economyenvironment trade-off, thereby facilitating a more sustainable development pathway.
Basing on the economy-environment trade-off framework, Chapter IV attempts to examine
the economic impact of the "2+26 Cities" policy. Adopting the difference-in-difference-indifference (DDD) approach, this study investigates the influence of the policy on the secondary
sector within the affected cities. The analysis reveals a significant contraction in the secondary
sector's economic activity, supporting the hypothesis that stringent environmental regulations can
impose short-term economic costs, particularly on heavily industrialized regions. The study also
explores the differential impact across various economic sectors, finding that while the secondary
sector experienced a notable decline, the tertiary sector remained relatively unaffected. These
findings highlight the complex trade-offs between environmental protection and economic growth,
offering valuable insights for policymakers in developing countries who must balance industrial
development with environmental sustainability.

1
CHAPTER I. INTRODUCTION
Through rapid modernization and industrialization, China experienced tremendous
economic growth in the past decades, with improvements in the life and welfare of the enormous
population and the alleviation of poverty for most. Yet, together with the benefits of
industrialization, came with the price of deterioration and pollution to the environment due factors
including the heavy consumption of fossil fuels and the lack of well-executed environmental
regulations. Air pollution became one of the major issues for public health concerns, with the
infamous fine particulate matters (PM 2.5, PM 10 ) as the main pollutant peaking in 2013.
Public concerns for air pollution arose, and numerous policies have since been established
to reduce air pollution levels and address the public health threat. Studies show that China’s policy
agenda has shifted from taking economic performance as the central focus to including
environmental factors for critical consideration. Substantial progress in air pollution reduction
have been made and documented, with the average PM 2.5 concentration cutting half only from
2013 to 2019 (Council on Foreign Relations, 2024). This aligns with the increase of annual “blue
sky days”, an alternative measurement for overall air quality and an environmental governance
goal.
But at what price? Many of the pollution-addressing policies come in the form of industrial
regulations, in hopes of reducing emission at its sources. These pro-environment efforts put
potential industrial economy development into the opportunity costs, and is creating a competition
in the country’s central policy agenda. Even having multiplied, China’s per capita income is still
only a small fraction relative to the developed countries’ figures, and the people’s demand for the

2
bettering of life remains strong. On the other hand, the grown per capita income has also brought
up the marginal benefit of public health improvements: air quality has become more valuable than
ever before to the wealthier Chinese people. Economic growth and environmental improvement
have joined into a trade-off in the country’s allocation of administrative and policy resources.
The arrival at this dilemma can be expected from the very beginning. It is precedented in
the West when the industrialization brought wealth along with polluted air and water to be treated
after, with the most recent example of the 1980s Clean Air Act in the United States. The studies of
the EKC (Environmental Kuznets Curve) have also revealed in the developed world the pattern of
the turning point and recovering of environmental quality as income grows to a certain extent,
indicating the shift of policy agenda from the economy to the environment. However, the challenge
for China may be even more difficult. When the cost of compliance to the environmental policies
rose in the now developed countries, their industries were able to move to the developing countries,
like China, and keep profit by taking advantage of the lower environmental standards and labor
costs. They were therefore able to improve their domestic environment and maintain the income
levels. For present days China, there are far less destination countries for such strategy, and the
existing industrial policies appear to be keeping the manufacturing industries regulated at home.
All of these observations suggest that China is at the turning point where environmental
policymaking has become increasingly important but also tricky. The government faces limited
policy implementation resources to be allocated, institutional weaknesses constraining policy
design, and equally strong public demands for economic development and environmental
improvement. How China’s policies address these challenges provide important reference for the
industrializing, developing countries from the Global South, as they share more common
conditions than with the western countries in the prime industrialization period. China’s experience

3
and lesson may help these countries with their cleaner industrialization and more efficient
environmental preservation in the near future.
This dissertation puts at the heart of its framework the developing countries’ trade-off
between the economic and environmental concerns in the policy agenda. Building on the existing
literature on China’s environmental policy performance and influencing factors, this dissertation
attempts to put the discussion into the context of the economy-environment trade-off to provide
practical insights for industrializing developing countries. More specifically, the dissertation
discusses three widely applicable questions:
1. How might the attempt of minimizing economic costs, combined with institutional
limitations in policy tools, lead to unexpected inefficiency in an industrial regulation?
2. What factors may potentially ease the local economy-environment trade-off to allow
for smoother transition of the policy agenda focus and less impact to the economy?
3. How does the economic impact of a successful regional environmental policy scheme
look like?
Overview
This dissertation contains three studies, surrounding two specific Chinese environmental
policies addressing the air pollution issue: the short-term ban of air pollutant emission around the
2014 Beijing APEC conferences, and the “2+26 Cities” policy with the target of reducing longterm air pollution in the region surrounding the Beijing-Tianjin-Hebei area. Both policies are
regional policies focusing on air pollution alleviation, with largely overlapping policy effective
area, which is also the most heavily polluted area in the country in terms of ambient air quality.

4
The studying of these policies became feasible partly due to the establishment of the nationwide
air pollution monitoring system by China’s Ministry of Ecology and Environment (MEE),
providing accessible and reliable data at the monitor-hour level since 2014.
These environmental policies carry strong “command-and-control” features, common in
developing countries where institutional capabilities may not allow the implementation of marketbased policy instruments. In the 2014 short-term ban for APEC, the policy design for the temporary
regulation was likely motivated by the aim to minimize economic impact to the industries while
achieving political mission of removing the air pollution during the conferences as a “gesture of
hospitality” to the foreign leaders. However, the command-based regulation naturally neglected
private incentives and led to an inefficient “temporal spillover” of air pollution exposure.
In contrast, the “2+26 Cities” policy was established in 2017 as a long-term policy scheme
to address the serious air pollution issue in northern China. Retrospectively, it has successfully
reduced air pollution levels in the policy region in a relatively short period of time. Taking it as a
positive example, details can be learnt about the factors that influence the efficiency of the
regulations, and more importantly for the developing countries, what is the economic cost to
achieve the environmental improvement.
This dissertation aims to analyze the two policies in three separate studies. The studies
draw from the literature of the broad field of environmental economics, as well as from the fields
of China’s environmental policy, political economy, economic resilience, etc. Each study has its
respective empirical framework and theoretical focus, yet together they are constructed organically
into this whole dissertation as a joint discussion of the applied environmental policy design in the
developing world scenario. The dissertation puts much emphasis on providing practical insights to
the environmental policy making in the industrializing, developing countries using China’s cases

5
as valuable precedent examples. While political and governmental differences exist, the common
limitations in resources and institutions offer similarly limited sets policy instruments and
implementational concerns alike.
In the following chapters, the studies are presented as three individual essays. In Chapter
II, the study focuses on the 2014 short-term ban for APEC. While the policy aimed for a temporary
cleaning of the regional air quality to welcome foreign leaders, with minimal costs and disturbance
to the industries, it was documented that soon after the event the air pollution exacerbated
drastically. Understanding the short-term objective of the policy, the study adopts the differencein-difference (DiD) approach and separately measures the impact of the APEC ban on air pollution
during and post the policy effective period. The results revealed significant increase in pollution
in the post policy period, in contrast to the significant decrease during the policy. The findings
further reveals that the post policy pollution increase exceeds the reduction during the policy,
concluding in an unexpected overall increase of public air pollution exposure. Explanation is
drawn from the recouping behavior of private industries to cover their losses, leading to the
discussion of the incomplete regulation design and potential improvements to be made.
In Chapter III, the study looks into the “2+26 Cities” policy and its performance in reducing
air pollution in the area. More importantly, the influence of industrial diversity on the policy
performance is examined at the city level. Recognizing the strengthened industrial regulations as
the central part of the regional environmental policy scheme, the trade-off between local industrial
economy and the environmental agenda arises, generating pressure for local policy implementation.
As industrial diversity provides economic resilience in the industrial sector, the cities with higher
industrial diversity would be expected to face less economic pressure for the environmental
regulation execution. The statistical result supports the hypothesis, providing support for the

6
economy-environment trade-off theory in environmental governance, and suggesting industrial
diversity as a critical concern for developing countries path of industrialization to smoothen their
future transition of the policy agenda.
Chapter IV continues to examine the “2+26 Cities” policy. While most related studies focus
on the effectiveness of air pollution policies in pollution reduction, the economic costs have been
largely neglected. In the developing country scenario, its crucial to be aware of the economic costs
to evaluate the efficiency of the environment improving policies. The study establishes a
framework of assessing the impact of the air pollution regulating policy scheme on the secondary
sector size. The separate DiD, year fixed effect, and difference-in-difference-in-difference (DDD)
models allow for different assumptions for how the economic effect may take form. The analysis
results suggest that the “2+26 Cities” policy reduced the secondary sector size in the policy area
or slowed down the growth. More specifically, the DDD result indicates that the policy may be
executed in a phasing-in fashion, leading to a gradual deviation of the treated cities from the control
group.

7
CHAPTER II. THE UNINTENDED CONSEQUENCES OF SHORT-TERM BANS ON
AIR QUALITY IN CHINA
1. Introduction
China's industrialization has been a significant driver of its remarkable economic growth.
However, this progress relied heavily on the usage of fossil fuels, and has led to severe urban air
pollution, imposing substantial social costs and posing serious public health threats. As a
developing country, China faces strong domestic pressures to sustain economic growth and
improve average income levels. This creates a core policy dilemma: balancing economic
development with environmental protection, as each represents a distinct aspect of social welfare.
It is, therefore, imperative for the government to address environmental issues carefully, ensuring
that the necessary measures do not inflict substantial harm on the industries that underpin the
national economy.
In addition to this economy-environment trade-off, environmental governance and
policymaking in developing countries is met with additional challenges. Among the environmental
policy instrument options, the market-based instruments are often preferred for their advantages
of reducing pollution behavior to an efficient extent (Goulder & Parry, 2008) and offering an
opportunity to generate revenue for further environmental investment (Parry & Bento, 2000). Yet,
the effective implementation of the market-based instruments requires well-established institutions
for robust monitoring and enforcement, which is often not applicable to the developing countries
(Blackman, 2010; Bell & Russell, 2002). As a result, developing nations tend to rely on command-

8
and-control regulations, which, despite their clear compliance mechanisms and immediate impact,
impose higher costs on industries (Cole & Grossman, 2003).
Amidst this backdrop, during the 2014 APEC (Asia-Pacific Economic Cooperation)
conferences in Beijing, China implemented a short-term ban on pollution to prepare for the event.
Similar to strategies used during the Olympic Games and World Expos, such major international
events are viewed as opportunities to enhance the host country’s image and influence (Karamichas,
2012; Kim et al., 2014). For the Chinese government, clearing air pollution and removing urban
smog for APEC was seen as an essential gesture of hospitality. To achieve this goal without causing
heavy compliance costs to the industries, the pollution ban was designed to cover the event period
and a short time before, with the ban lifted immediately after the conferences concluded. This
micromanaging strategy underscores the considerations driving developing countries' preference
for command-and-control measures, as long-term market-based instruments are not feasible due
to institutional limitations, and economic costs are heavily weighted even when pollution reduction
is imperative.
This study examines the impact of Beijing’s short-term pollution ban during the 2014
APEC on ambient air pollution levels during and after the policy's implementation. The policy's
design and execution lacked comprehensive post-event provisions, resulting in incomplete
regulation of polluting behavior. By analyzing the policy's impacts at different stages of
implementation, this study evaluates its effects on air pollution levels throughout the event and
assesses the efficiency and validity of the policy design. The study aims to contribute to the
understanding of command-and-control regulations and environmental governance in developing
countries.

9
The results indicate that the 2014 APEC pollution ban effectively reduced air pollution
levels in the target area during the event. However, it also led to a significant rebound in pollution
immediately after the event, with levels surpassing the initial reductions. This outcome increased
overall pollution exposure and potential health hazards for the public, undermining the policy's
justification and its viability as a standard approach for hosting major events. The policy affected
41 cities, impacting over 200 million people. The findings suggest that the temporary production
ban was far more costly than previously perceived and less desirable as a common approach for
managing pollution during major events.
Specifically, the study employs a Difference-in-Difference (DiD) approach to separately
measure the effects of the pollution ban during and after the event, based on the policy's
implementation timeline. The regression results show an 8.6% reduction in the Air Quality Index
(AQI) during the event and a 16.5% increase post-event compared to baseline levels. The average
AQI increase post-event was 1.92 times the average pollution reduction during the event—a ratio
defined in this study as the “Inefficiency Factor,” indicating the high inefficiency of the pollution
reduction during the event. Further analysis by time of day and by areas with varying levels of
manufacturing plant density confirms the rebound effect and reveals the underlying industrial
behavior: with the regulation's incomplete design and implementation, industries were
incentivized to intensify production immediately after the event, driving a disproportionate
increase in pollution.
This study contributes new perspectives to the existing literature on environmental policy
instruments, particularly in the context of developing countries. First, by acknowledging the
economy-environment trade-off and institutional challenges in policy agendas, the study
introduces an empirical framework centered on the comparison between the targeted effect and the

10
unexpected cost for evaluating environmental policy performance. This framework highlights the
unintended effects of environmental policies, which are more likely to occur in the context of weak
institutions and incomplete regulation design. Second, the inefficiency mechanism revealed by the
empirical evidence introduces an additional challenge to environmental policy implementation in
developing countries: the establishment and shift of industrial norms. By adopting a two-stage
model to assess policy effects, the study directly examines the policy process and identifies
industrial behavior as a neglected factor in incomplete policy design. This approach not only offers
a new method for analyzing campaign-style policies with unsustainable effects—common in China
and other developing countries with similar resource constraints—but also stimulates discussions
on policy design improvements to address identified regulatory gaps.
2. Results
a. Sharp rebound and increased pollution exposure.
As the host of the 2014 APEC event and the target of temporary production restrictions,
the 41 cities surrounding Beijing experienced a significant reduction in pollution levels during and
shortly before the event. However, immediately following the conclusion of APEC, pollution
levels sharply rebounded, exceeding the pre-event normal levels by a margin greater than the initial
reduction. To illustrate this process, we utilized monitor-day-based pollution data and generated
controlled residual trends that excluded the effects of the policy intervention (Fig. 1.1 A and B).
The statistically similar trends between the policy-affected area and the control group before and
after the policy window validate our model assumptions regarding a common national pollution
trend. The distinct divergence in trends during the policy period clearly shows how the pollution

11
levels in the policy area first dropped before the APEC event and then surged sharply afterward,
surpassing the national trend.
To quantify the pollution reduction and subsequent rebound caused by the policy, we
defined the "During" period (October 22 – November 11, 2014) to include the APEC event and
the two preceding weeks to capture all pre-deployment activities. The "Post" period (November
12 – December 3, 2014) was defined as the three weeks following the APEC event. Our main
analysis estimates that the average effects on the AQI levels in the policy area during these periods
were -9.88 points for the "During" period and +18.95 points for the "Post" period1
(Fig. 1.1 C;
Appendix A, Table S10). For context, an 18.95-point increase in AQI during the same period in
the previous year (2013) would have raised the average pollution level in the regulated cities to
exceed the "moderate pollution" threshold of China's ambient air quality standards (Fig. 1.1 D). In
the most populous cities (with populations over 10 million) within the regulated area, where
baseline pollution levels were already concerning, such an increase would bring air quality close
to or even beyond critical thresholds associated with severe health risks.
The point estimates may be influenced by different definitions of the policy periods. To
ensure robustness, we conducted an alternative event study analysis around the policy window,
and the results were consistent with the main analysis (Appendix A, Section 9d, Fig. S1.4). Another
potential issue with the point estimates is that the regulated cities were not randomly selected but
were chosen based on their proximity to Beijing, which could introduce spatial heterogeneity
effects. To address this concern, we performed robustness checks using different bandwidths for
the control group areas (including closer control groups) and employed models with relaxed
1 For reference, the average AQI level for all the untreated observations in our dataset across the country over the period from Jan. 2014 – Dec.
2015 is 114.7.

12
assumptions regarding parallel trends to better accommodate local trends (Appendix A, section 9a,
Table S1.13, section 9c, Table S1.15). These checks consistently produced results in line with our
main analysis. All these findings corroborate our initial observation: while pollution levels dropped
during the APEC event, the subsequent increase was significantly larger, leading to greater overall
pollution exposure and revealing the inefficiency of the policy.
Figure 1. 1 Controlled residual trends and main analysis results

13
b. Revealing the industrial mechanism of pollution rebound
The pollution ban policy was implemented across multiple sectors to reduce emissions,
making it essential to understand what specifically contributed to the inefficient outcome and the
underlying mechanism. The hypothesis tested for this mechanism is that the two-staged effects
resulted from the temporary ban on polluting industries, which reduced their emissions before and
during the event, but then significantly ramped up production afterward. This surge in production
would exceed the normal rate, leading to inefficiency, particularly in terms of emission intensity
(Appendix A, section 4). To test this hypothesis, we utilized an additional dataset of emission data
collected from manufacturing plants under the Continuous Emission Monitoring System (CEMS).
If the manufacturers' recoup behavior was responsible for the inefficient pollution rebound, the
emission patterns should align with our observed changes in ambient air pollution levels.
However, using the same model and settings as the main analysis but with different
dependent variables, the regression on the emission data produced divergent results. As is exhibited
in Fig. 1.2, the emissions of total suspended particles, SO₂, and NOx from the monitored factories
all decreased during the "During" period and did not rebound in the "Post" period. These findings
align with the policymakers' intentions but do not correspond with our observations of ambient
pollution levels. This discrepancy suggests that the inefficiency observed in ambient pollution
levels may not be directly attributable to the recoup behavior of manufacturing plants, indicating
that other factors may be contributing to the pollution rebound.

14
Figure 1. 2 Effect of the policy on Continuous Emission Monitoring system firms' emission

This result is likely due to the fact that the companies monitored by the CEMS are a
selected group that is closely monitored and subject to stricter enforcement of emission standards.
These companies tend to be larger, more essential, and already compliant with regulations. While
CEMS-monitored companies may account for the majority of industrial pollutant emissions, they
do not necessarily represent the most egregious polluters, which are often smaller companies that
are more adept at evading inspections. This suggests that the policy's industry-reaction effects and
their distribution are influenced by the heterogeneity in government enforcement stringency across
different companies, leading to varying local effects.
To test this theory, we conducted two mechanism-validating regressions to examine the
policy's impact by different times of day and varying levels of manufacturing plant density—two
dimensions where government enforcement of pollution standards may differ. Manufacturing
plants might shift their operations to nighttime, when emissions are less observable and less likely

15
to be detected. Spatially, areas with higher plant density could pose greater challenges and costs
for government inspectors, making enforcement less stringent.
In the first regression aimed at examining the industrial mechanism behind the pollution
rebound, we analyzed the hour-period-specific effects of the production ban policy. We divided
the day into four six-hour periods (P1-P4, starting at 6 am) and estimated the specific effects of
the policy for each period. The results, presented in Fig. 1.3, indicate that the night periods (P3 and
P4, covering 6 pm to 6 am) experienced a significantly higher rebound effect than the daytime
periods. This finding supports the hypothesis that industries engaged in recoup behavior by adding
night shifts, which amplified the rebound effect. This was not only due to increased production
intensity but also because the effect was measured against the normal operational hours, during
which there would typically be no shifts or emissions at night. These results support our theory
that there is a negative relationship between government enforcement and emission behavior.
The model controls for fixed effects across the hour periods, assuming a common daily
cycle pattern for urban pollution. This assumption was relaxed in a robustness check that allowed
for city-specific daily pollution cycle patterns as fixed effects around the hour periods. The
robustness check produced results consistent with the original hour-period regression findings
(Appendix A section 9f, Table S1.22).

16
Figure 1. 3 Estimated effect of the policy by hour periods
Numbers in brackets indicate the difference between the effects of the respective periods and that of P1 (6am12pm) in the same policy period, and the stars indicates the statistical significance of the difference (***, **,
and * represent significance at 1%, 5% and 10%, respectively). Redline emphasizes the result for Post period
effect for P1, as the Post period effects for the other hour periods are to be compared to it.
In the second mechanism test, we categorized the pollution monitors into quartiles based
on the aggregated counts of industrial plants within a 20-mile radius, which serves as a proxy for
plant density in each area (Q1-Q4, with Q1 representing the lowest plant density). The regression
then estimates the effects of the policy on pollution levels for each quartile, controlling for quartilespecific baseline fixed effects. The results indicate that the lowest quartile (Q1) experiences the
smallest rebound effect, while the other quartiles exhibit significantly larger rebounds and smaller
initial pollution reductions (Fig. 1.4). This finding suggests a spatial correlation between the
density of industrial plants and local pollution levels, supporting our hypothesis that the recouping
of industrial production is the primary driver of the pollution rebound.
Additionally, we ruled out the possibility that the total production in an area is the key
factor influencing the rebound magnitude by conducting a robustness check. This check

17
incorporated monitor quartiles based on the total sales of industrial plants within a 20-mile radius
(Appendix A, section 9g, Fig. S1.8). The results of this analysis further corroborate the conclusion
that plant density, rather than total production, is the critical factor in the observed pollution
rebound.
Figure 1. 4 Estimated effects of the policy by different plant density quartiles
Numbers in brackets indicate the difference between the effects of the respective periods and that of Q1 (lowest
plant density), and the stars indicates the statistical significance of the difference (***, **, and * represent
significance at 1%, 5% and 10%, respectively). Redline emphasizes the result for post ban Q1, as the Post period
effect for the other quartiles are to be compared to it.

18
3. Discussion
The main analysis of this study, which estimates the “During” and “Post” policy effects of
the 2014 APEC pollution ban, revealed an average reduction of 9.88 points in the AQI of the
policy-covered cities during the event, accompanied by an unexpected pollution rebound of 18.95
points above the counterfactual baseline. The ratio of these effects, 1.92, can be interpreted as
follows: for each unit of AQI reduction achieved during the APEC event, there was a consequent
1.92-unit increase in AQI after the event, representing an unexpected cost. We define this ratio as
the “Inefficiency Factor” (Fig. 1.2 C). A ratio greater than one indicates that the policy was
inefficient, ultimately leading to an overall increase in pollution exposure for the public.
While the back-of-the-envelope calculation here focuses strictly on air pollution exposure,
the calculation of the “Inefficiency Factor” could be expanded to include the economic costs to the
public and industries, as well as the actual health impacts resulting from exposure to extreme air
pollution levels due to the rebound. The adoption of the “Inefficiency Factor” as a policy
evaluation tool highlights the unintended consequences of the policy, which are more likely to
occur in developing countries where resources and institutional capacities are limited.
Furthermore, the industrial mechanism analyses (Section 2.2) demonstrate that even with
major firms and primary sources of air pollution under strict regulation through the CEMS program,
inefficient pollution rebound can still occur in times and locations with weaker oversight. On the
one hand, the CEMS emission analysis results (Fig. 1.2) suggest that institutions are effective in
enforcing regulations, preventing emission rebounds in the Post period among monitored large
firms. On the other hand, the results from the hour-period and plant density quartile-specific
analyses (Fig. 1.3, Fig. 1.4) suggest that smaller, unmonitored firms, which contribute a smaller

19
portion of total production and baseline emissions, can exploit enforcement loopholes,
undermining the environmental policy’s effectiveness without proper institutional oversight.
The significant gap between the “During” and “Post” policy effects cannot be entirely
attributed to underdeveloped institutions. It is also a consequence of the incomplete design of the
temporary short-term ban. Enforcement details are critical: industries responded to the regulations
by exploiting weaknesses in the government’s enforcement. A solution could involve incorporating
industries’ potential reactions into the policy design. The 2014 APEC pollution ban’s simplistic
focus on reducing air pollution during the event, without considering industries’ incentives to
recoup, is one of the core reasons for the subsequent pollution rebound. For this specific case,
improvements could include adopting a policy patchwork to anticipate and mitigate pollution
leakage. For example, within the APEC pollution ban framework, the government could require
regulated industrial firms to submit production plans covering a period longer than the immediate
post-event period, helping to flatten production intensity over time and avoid extreme pollution
spikes. Additionally, announcing the ban well in advance would allow regulated firms to plan their
production more efficiently, minimizing both compliance costs and emissions, and reducing the
need for a pre-event production surge.
These approaches align with discussions in the literature on policy combinations. Fullerton
& Wolverton (2005) discusses the use of a two-part instrument as a second-best option for
pollution mitigation, while Lehmann (2010) and Rogge & Reichardt (2016) explore the potential
of policy mixes more broadly. Although much of this literature remains theoretical, Rivera (2021)
found that Santiago’s pollution-warning-triggered temporary driving restrictions were effective,
particularly when complemented by increased public transport ridership, highlighting the
importance of having efficient substitutes available. Li et al. (2019) further emphasize the benefits

20
of comprehensive policy design, showing that Hangzhou’s pollution mitigation efforts for the 2016
G20 event led to both short-term and long-term pollution reductions, thanks to the focused
allocation of resources.
The findings of this paper also provide a detailed examination of the role of institutional
development in environmental governance and how it limits environmental policy implementation
in the context of developing countries. They indicate that for environmental policies to be effective
and efficient, institutions need to be sufficiently developed. The literature has long acknowledged
that institutional limitations in developing countries can hinder the effective implementation of
market-based instruments, leading to a reliance on command-and-control regulations, which are
less efficient but easier to enforce (Blackman & Harrington, 2000; Blackman, 2010; Goulder &
Parry, 2008). Building on this literature, our findings suggest that, even when regulatory policies
are used, the performance of these policies is also contingent on the level of institutional
development. In other words, institutional capacity not only determines the feasibility of more
efficient environmental policy instruments but also affects the efficiency of policy implementation
for any given instrument.
This emphasis on the relationship between the efficiency of environmental policies and
institutional development brings the discussion back to the economy-environment trade-off that is
central to the agendas of industrializing, developing countries. As industrialization progresses,
environmental degradation worsens, impacting productivity, while rising incomes increase
demand for a better environment, heightening the urgency to address environmental issues
(Greenstone & Jack, 2015). On the other hand, the demand for further income growth competes
with the demand for environmental improvement, and economic development is also linked to
institutional development. As we argue, this institutional development is correlated with the

21
government’s ability to design and implement environmental policies more efficiently. These
perspectives illustrate the complex interconnections between economic and environmental factors,
making it particularly challenging for developing countries to balance these priorities in their
agendas.
4. Data and Methods
a. Data
The hourly concentrations of PM10, PM2.5, SO₂, and NO₂, along with the AQI
automatically calculated from these pollutant concentrations, are reported by China’s Ministry of
Ecology and Environment (MEE) through a network of ambient air pollution monitors across the
country (Appendix A, Section 6). For this study, we computed the daily mean values of AQI and
pollutant concentrations for the period from January 2013 to July 2015. The unit of measurement
for AQI is in points, while pollutant concentrations are measured in micrograms per cubic meter
(µg/m³). The treatment group is identified by a dummy variable, set to 1 if the pollution monitor
is located in one of the 41 cities affected by the air pollution regulation (Appendix A, Section 1).
Meteorological variables, including wind direction, wind speed, precipitation, and
temperature, are incorporated into each monitor-day observation by merging data from NOAA’s
Integrated Surface Database, using the date and the closest weather station to the pollution monitor.
Industrial characteristics are also included, aggregated from the China’s Annual Survey of
Industrial Enterprises 2008 data to the monitor level, within a 20-mile radius of each monitor.
These aggregated variables include the count of industrial plants, total sales, total labor size, total
assets, and the percentage distribution of various industrial categories within the radius.

22
Additionally, we use a dummy variable to indicate whether the observation date falls within the
winter heating period for the city where the pollution monitor is located, as winter central heating
can impact local air pollution levels due to increased fossil fuel consumption. This information
was obtained from local government announcements.
b. Preliminary residual trend analysis
To provide an initial observation of the impact of the 2014 APEC pollution ban and to
gather the necessary information for defining the "During" and "Post" periods for the two-stage
main analysis based on the Difference-in-Differences (DiD) model, we conduct a preliminary
analysis using the residual trend of pollution. First, we employ an ordinary least squares (OLS)
regression to estimate the impact of control variables on the pollution indicator dependent variables.
The regression equation is as follows:
𝑃𝑜𝑙𝑙𝑢𝑡𝑖𝑜𝑛𝑖𝑡 = 𝛾𝑋𝑐𝑡 + 𝛼𝑖 + 𝜃𝑡 + 𝜖𝑖𝑡 [1]
Where 𝑖 represents a pollution monitor station located in city 𝑐, observed on the date 𝑡.
𝑃𝑜𝑙𝑙𝑢𝑡𝑖𝑜𝑛𝑖𝑡 represents daily average reading of the AQI or pollutant concentration. 𝛼𝑖
represents
a set of monitor fixed effects, which absorbs pollution baselines for the location of the monitors
and prevent the heterogeneity of the treated area against untreated. 𝜃𝑡
represents a set of date
fixed effects, which absorbs the national trend of pollution by assumption. 𝑋𝑐𝑡 represents the
time-varying control variables associated to city 𝑐, including precipitation, temperature, wind
direction, their respective quadratic terms, and wind direction. The dummy variable for the
winter heating period is also included. Province-month fixed effects are included to allow
different general regional trends.

23
By obtaining the initial estimates for the control variables, we can calculate the fitted values
and residuals for each observation, which are then aggregated by taking the daily residual mean
for both the policy-treated group and the control group. The resulting "conditional residual" trends
represent, for each group, the extent to which the actual pollution levels deviate on average from
the counterfactual fitted values, assuming no policy intervention had occurred. As observed in Fig.
1.2 A and B, the control group’s residual trends do not deviate from the fitted values (residual =
0), which supports the validity of the control variable estimates from the initial OLS regression.
Using these conditional residual trends, we define the "During" and "Post" periods of
policy implementation. Considering the start date of the APEC event on November 5, 2014, we
include the two weeks preceding this date in the "During" period to capture all potential mitigation
efforts deployed in preparation for the event. Thus, October 22 is designated as the first day of the
"During" period, and November 11, the final day of the APEC conferences, is set as the last day,
making the "During" period span 21 days. As shown in Fig. 1.2 A and B, the lack of significant
deviation between the two trends before any potential policy intervention (before October 22, Day
-14 relative to the APEC start date) further validates our control variables and the underlying
assumption of parallel trends between the two groups.
The "Post" period is defined to begin on the first day after the APEC conference, November
12, 2014. As shown in Fig. 1.2 A, the AQI level for the treated group increased rapidly following
the conclusion of the APEC event and then gradually returned to the normal level (at residual = 0).
On Day 28, December 3, the treated group’s trend aligned with both the normal level and the
control group’s trend. Therefore, we designate this day as the last day of the "Post" period, making
the "Post" period span 22 days. Observations with dates falling within these two periods, as defined
above, are marked with the respective dummy variables.

24
c. Main analysis
With the "During" and "Post" policy periods established, we proceed to estimate the
treatment effects for these two periods using the Difference-in-Differences (DiD) regression
approach, with standard errors clustered at the monitor and year-quarter levels. The regression
function is as follows:
𝑃𝑜𝑙𝑙𝑢𝑡𝑖𝑜𝑛𝑖𝑡 = 𝛽1(𝐷𝑢𝑟𝑡
∙ 𝑇𝑟𝑒𝑎𝑡𝑐) + 𝛽2(𝑃𝑜𝑠𝑡𝑡
∙ 𝑇𝑟𝑒𝑎𝑡𝑐) + 𝛾𝑋𝑐𝑡 + 𝛼𝑖 + 𝜃𝑡 + 𝜖𝑖𝑡 [2]
Beside the control variables and fixed effects from function [1], 𝐷𝑢𝑟𝑡 and 𝑃𝑜𝑠𝑡𝑡
represent
the “During” period and “Post” period of the policy as defined previously. 𝑇𝑟𝑒𝑎𝑡𝑐
represents
whether city 𝑐 falls in the treatment group: the 41 policy-regulated cities. 𝛽1 and 𝛽2 represents
respectively the treatment effect of the APEC policy on the pollution levels in the policy area
over the “During” and “Post” policy periods.

25
CHAPTER III. THE ROLE OF INDUSTRIAL DIVERSITY IN SUSTAINABLE
DEVELOPMENT: EVIDENCE FROM CHINA
1. Introduction
Environmental issues are particularly severe in industrializing developing countries. As
these nations pursue economic development to enhance the welfare of their citizens, they often
face the unintended consequence of environmental deterioration, a byproduct of industrial activity
that poses a significant threat to public health. Conversely, imposing strict environmental
regulations can adversely impact industries, creating a challenging trade-off between economic
growth and environmental sustainability, especially given the limited resources available for
policy-making and implementation.
This paper aims to explore the potential of industrial diversity as an urban economic factor
that can buffer this economy-environment trade-off and improve the efficacy of environmental
regulation policies. Understanding the impact of industrial diversity on this trade-off, and
subsequently on the performance of environmental regulations, is crucial for policy decisions in
developing countries as they chart their developmental trajectories. While the typical path of
development involves initial environmental degradation followed by recovery, the opportunity to
alleviate the economic-environment trade-off could enable healthier, more sustainable
development and significant welfare improvements for current and future generations.
To this end, the paper examines the effects of industrial diversity on ambient air pollution
levels within the context of implementing environmental policies. It leverages China’s “2+26
Cities” environmental policy scheme to analyze the impact of industrial diversity at the city level

26
on policy performance in terms of pollution reduction. Furthermore, the study tests the influence
of industrial diversity on the relationship between local air pollution and the size of the secondary
sector's output. Given that the correlation between local industrial output and air pollution mirrors
the economy-environment relationship, the impact of industrial diversity on this correlation serves
as a conditional proof of its buffering effect. The research utilizes data from industrial firms, socioeconomic yearbooks at the city level, and ambient air quality monitors across China, covering the
period from 2015 to 2019. Employing fixed effects models, the study finds that cities with higher
industrial diversity experience additional pollution reduction under the “2+26 Cities” policy, and
that these cities exhibit a weaker negative correlation between industrial output and air pollution
levels.
The motivation and theoretical foundation of this paper build upon the work of Greenstone
and Jack (2015), who highlighted the challenges of poor environmental quality in developing
countries. Acknowledging the economy-environment trade-off as a central issue, this study
introduces industrial diversity as a potential solution. While much of the existing literature on
industrial diversity focuses on its impact on local economic performance, particularly in terms of
employment and productivity, this paper extends the concept to the broader context of the
economy-environment trade-off. The hypothesis is that the resilience provided by industrial
diversity could ease the tension between economic resource constraints and the pressing need for
environmental quality improvement in developing countries, ultimately leading to better
environmental policy outcomes. By exploring this hypothesis within the Chinese context, this
study offers insights that could be valuable for environmental policy-making in other developing
nations.

27
The subsequent sections of the paper present the background, empirical framework, and
data used in the study. The results section demonstrates that industrial diversity is associated with
improved environmental policy performance in pollution reduction and is capable of mitigating
the negative correlation between pollution and secondary sector size, thereby illustrating its
buffering effect. Finally, the discussion and conclusion section reviews the study's implications
and limitations and outlines potential directions for future research.
2. Background
2.1. Developing Country Environmental Policy
Greenstone & Jack (2015) have highlighted the complexity of environmental challenges in
developing countries, primarily stemming from low income levels among citizens and constrained
governmental budgets. These economic limitations result in a relatively lower marginal
willingness to pay (MWTP) for environmental improvements when weighed against the immediate
needs for consumption and economic development. Empirical studies have consistently shown a
significantly lower MWTP in developing countries for environmental goods such as clean air
(Hammit & Robinson, 2011; Ito & Zhang, 2020), electric vehicle infrastructure (Tan & Lin, 2020),
and clean water (Null et al., 2012; Genius et al., 2008; Wang et al. 2010). When MWTP for
environmental improvements is low, resources tend to be prioritized for consumption or
investment in economic growth. Another factor contributing to the environmental dilemma in these
countries is the high marginal cost of environmental improvement, which translates to relatively
low efficiency in environmental investments and policy efforts. This inefficiency is often due to
limitations in institutional capacities related to policy design and implementation (Greenstone &

28
Jack, 2015; Oliva, 2015; Duflo et al., 2013; Duflo et al., 2018), coupled with lower technological
capability and resource constraints.
The overarching question of “why is environmental quality so poor in developing countries”
(Greenstone & Jack, 2015) encapsulates a central theme in recent environmental economics and
policy studies concerning these nations. This question is closely linked to discussions surrounding
the environmental Kuznets curve (EKC), which describes the relationship between economic
development and environmental degradation (Gross & Krueger, 1995). The EKC is often depicted
as an inverted U-shape, with countries initially experiencing environmental degradation during
industrialization, followed by improvements as income levels rise. While the validity and
universality of this model are subjects of ongoing debate (McConnell, 2001; Stern, 2004; Dinda,
2004), the literature generally agrees that a critical factor for the EKC's turning point is the shift in
preferences toward environmental quality as incomes increase. Thus, the dilemma for developing
countries lies in achieving sustainable national welfare improvements amid the stark trade-off
between economic growth and environmental protection.
China, in particular, has been the focus of numerous studies due to its status as a major
global manufacturer and the environmental challenges it encountered during its rapid
industrialization (Liu & Raven, 2010; He et al., 2012). Despite the tension between its economic
and environmental goals, China has made significant progress in reducing pollution and restoring
the environment (Tie & Cao, 2009; Jalil & Feridun, 2011; Jin et al., 2016; Xu, 2021). Notably,
China reached its EKC turning point earlier and at a lower income level compared to historically
industrialized nations, suggesting that it managed to control environmental degradation more
effectively and within a shorter timeframe (Xu, 2021; Mahmood et al., 2023). In the long run, the
economy-environment contradiction in China appears to have been mitigated by technological

29
advancements, innovation, and regional cooperation (Peng et al., 2020). As such, China’s
experience offers valuable lessons for other developing and industrializing countries in navigating
the economy-environment trade-off through a variety of environmental and industrial policies.
2.2. Industrial Diversity
The concept of industrial diversity, despite being extensively studied, has elicited mixed
viewpoints due to its multifaceted nature. Previous research has predominantly focused on its
impact on regional economic performance, particularly in terms of employment and productivity.
For instance, Attaran (1986) employed entropy as a measure of economic diversity but found
inconclusive results regarding its correlation with growth, unemployment stability, or per capita
income. Kemeny & Storper (2015), using the Herfindahl-Hirschman Index (HHI) to measure
economic specialization (the inverse of diversity), observed a positive correlation between
specialization and income. Simon (1988) also used the HHI to measure industrial diversity and
suggested that it reduces regional frictional unemployment rates. However, Mizuno et al. (2006)
pointed out the neglect of structural factors in these analyses. They incorporated the “location
quotient,” which is similar to Kemeny & Storper’s concept of relative specialization, along with
the HHI, and found that structural elements significantly influenced employment but not industrial
diversity. Contradictory findings by Diamond & Simon (1990) and Simon & Nardinelli (1992)
further illustrate the complexity of industrial diversity’s role in urban and industrial economies.
These divergent studies, coupled with varying measurement approaches, underscore the
complexity of industrial diversity and its multifaceted impact on urban and industrial economies.
Frenken et al. (2007) proposed a framework that distinguishes between related and unrelated
varieties of economic activity. Related variety refers to the diversity among firms or industrial
branches that share externalities, such as direct interdependencies or shared managerial knowledge

30
(Frenken et al., 2007; Jacobs, 1969). In contrast, unrelated variety, grounded in portfolio theory,
depends on the absence of relationships between different economic activities and serves as a
mechanism for risk spreading. This distinction has guided numerous empirical studies, including
those by Fritsch & Kublina (2018) and Aarstad et al. (2016) on enterprise management, as well as
by van Oort et al. (2016) and Cortinovis & van Oort (2015) on regional economic growth in Europe.
Particularly, the risk-spreading feature of industrial diversity, especially in terms of
unrelated variety, is perceived to contribute to economic resilience. Following Frenken et al.’s
(2007) framework, Brown & Greenbaum (2017) found that counties with high industrial
concentration performed better in employment during economic upturns, while more diverse
industries were more resilient against employment shocks. Similarly, Tan et al. (2020) examined
how industrial structure and agency factors influenced regional economic resilience in Chinese
resource-based cities. Feng et al. (2023) discussed industrial structure in the context of regional
integration, noting its significant role in influencing economic resilience. These ongoing
discussions and case studies highlight the evolving understanding of industrial diversity’s role in
strengthening economic resilience.
3. Empirical framework
In this study, the central hypothesis is that industrial diversity can mitigate the economyenvironment trade-off in developing countries. When environmental regulations are introduced,
economies with higher resilience—fostered by industrial diversity—are expected to experience
less immediate production loss and recover more quickly from the regulatory shock (Hallegatte,
2014). This is particularly relevant in developing nations, where environmental regulations can

31
impose significant costs and disrupt essential industrial productivity. In such cases, cities with high
industrial diversity and, consequently, higher economic resilience may encounter less economic
and environmental pressure. With a reduced economy-environment trade-off, environmental
regulations are likely to perform better, both in economic terms and in achieving environmental
improvement goals. This paper examines the hypothesis that industrial diversity enhances local
economic resilience and alleviates the economy-environment trade-off by improving the
effectiveness of environmental regulations.
3.1.Part 1: Effect of industrial diversity
This study investigates the relationship between industrial diversity and ambient air quality
in cities affected by an air pollution reduction policy scheme in China. Air pollution issues,
particularly urban smog, became increasingly prominent during the country's industrialization,
with pollution levels peaking around 2013. The problem is especially acute in northern China,
where many heavy and mining industries are concentrated, and where geographical and climatic
conditions hinder the dispersion of air pollutants. Notably, Beijing, the national capital, is at the
center of this region. As a key political and international hub with a population exceeding 20
million, the urgent need for the government to address air pollution and urban smog in Beijing and
the broader northern China region became clear.
The regional environmental policy scheme known as the “2+26 Cities Atmospheric
Pollution Treatment Action Plan” (referred to in this paper as the “2+26 Cities” policy) targeted an
area centered around Beijing, including Tianjin and 26 prefecture-level cities2
. These cities are
2 Beijing and Tianjin are among the four direct-administered municipalities (cities). They are administratively at the
provincial level, the Level-1. Prefecture-level cities are the Level-2 cities governed by the provinces. In terms of
size, Beijing and Tianjin are between provinces and prefecture-level cities, closer to the latter.

32
located within a common geographical pathway where air pollutants are likely to travel, resulting
in highly correlated ambient air pollution levels across the region. Announced in 2017, the policy
aimed to reduce air pollution levels in the covered area and increase the number of "blue sky days."
It provided broad guidelines for policy approaches, such as industrial regulations and urban traffic
controls, and set local goals for reducing air pollutant concentrations. Local governments were
then responsible for developing more detailed guidelines and implementing specific policies to
achieve these goals. Following the initial announcements, the “2+26 Cities” policy was carried
over the years with new and updating local policies. Recordsindicate that air pollution in the policy
area was effectively reduced during this period.
The hypothesis to be tested posits that industrial diversity contributes to pollution reduction
in response to the environmental policy shock, with cities exhibiting higher industrial diversity
achieving greater reductions in pollution. This study employs a fixed effects model to estimate the
effect of the industrial diversity index on local air pollution, considering the pollution reduction
policy's impact while controlling for the effect independent of industrial diversity. The regression
model is as follows:
𝑃𝑖𝑡 = 𝛼𝑖 + 𝛽1𝑅𝑒𝑔𝑖𝐷𝑡 + 𝛽2𝐷𝑡𝑇𝑖 + 𝛽3𝑅𝑒𝑔𝑖𝐷𝑡𝑇𝑖 + 𝛿𝑋𝑖𝑡 + 𝜃𝑡 + 𝜀𝑖𝑡 [3]
Where 𝑃𝑖𝑡 represents the average air quality level of city 𝑖 in year 𝑡. 𝑅𝑒𝑔𝑖
is the dummy variable
for the treatment of the environmental policy scheme, which indicates whether the city 𝑖 of the
observation belongs to the “2+26 Cities” policy area. 𝐷𝑡
indicate the timing of the observation, in
terms of whether the “2+26 Cities” policy is in effect. 𝑇𝑖
represents the treatment of the local
industrial diversity, in the form of a cross-sectional industrial diversity index calculated for city 𝑖,
based on the 2015 industrial firm data. 𝛼𝑖
represents the city fixed effects, which absorbs the
baseline differences in air pollution due to city characteristics, including the direct effect of the

33
industrial diversity index differences between the cities on pollution. 𝑋𝑖𝑡 is the set of control
variables potentially influencing air pollution, including local GDP per capita as a proxy for
general level of urban development, and weather variables of temperature, windspeed, and
precipitation, which directly influence air pollution level. 𝜃𝑡
represents the year fixed effects,
which indicates the assumed common national trend of pollution reflected by the control group
cities outside of the “2+26 cities”, absorbing the influences of the general impact of national
environmental policy agendas.
In Equation [3], the model's main estimator of interest, 𝛽3, estimates the average effect of
the industrial diversity index on air pollution in the "2+26 Cities" area over the course of the policy.
A negative sign for 𝛽3 would indicate that cities with higher industrial diversity achieve greater
pollution reduction during the policy, thereby supporting the theory that industrial diversity
enhances environmental policy performance. 𝛽1 estimates the industrial-diversity-neutral impact
of the "2+26 Cities" environmental regulations on air pollution. A negative value for 𝛽1 would
confirm the general effectiveness of the policy in reducing air pollution in the regulated area,
compared to changes in air pollution levels in the rest of the country. 𝛽2 estimates the effect of
industrial diversity on air pollution in the control group during the "2+26 Cities" policy period. It
is important to note that China’s broader agenda and efforts for environmental improvement and
air pollution mitigation extend beyond the "2+26 Cities," potentially affecting the control group
cities as well. Consequently, industrial diversity may also contribute to pollution reduction in these
areas. As the key estimator, 𝛽3 should be interpreted in conjunction with 𝛽1 and 𝛽2, representing
the effect of industrial diversity in addition to the direct impact of the "2+26 Cities" policy and the
baseline effect of industrial diversity.

34
The measurement of the industrial diversity index in this study is based on the entropy
approach. The focus of this paper is on the shock-buffering, risk-absorbing, and resilience-building
characteristics of industrial diversity. According to Frenken et al. (2007), these features primarily
stem from the unrelated variety within an industry, while related variety concerns the synergy or
competition arising from the agglomeration of firms. As demonstrated by Attaran (1986), entropy
calculation naturally allows for the disaggregation of entropy into between-set and within-set
components. By selecting an appropriate level of industry categorization for the entropy
calculation, this study aims to achieve a valid separation of unrelated and related variety.
Following China’s national standard for industrial classification, entropy is calculated at
the "Industrial Branch Level I" (referred to as "branch" throughout the text). The analysis is
restricted to the officially documented secondary sector of the economy, which includes the
industrial sector—encompassing mining, water and power, and manufacturing industries. The
formula for entropy by branch employment (referred to as "employment entropy") is as follows:
𝐸𝑛𝑡2015𝑖 = − ∑ 𝑆𝑖𝑙 ln 𝑆𝑖𝑙
𝐿
𝑙=1
[4]
Where 𝐸𝑛𝑡2015𝑖
, the measurement for 𝑇𝑖
in the regression Equation [3], stands for the entropy by
branch employment of city 𝑖 in year 2015. 𝑆𝑖𝑙 stands for the employment share of the industrial
branch 𝑙 in city 𝑖, among a total of 𝐿 existing branches at the "Industrial Branch Level I" for the
entirety of the secondary sector. That is, industrial diversity defined in this study is determined
entirely by the industrial branches existing and their relative shares, regardless of the number or
sizes of firms in each branch or the overall size of the secondary sector of economy. The theoretical
minimum value of entropy is 0, reachable only when there is only one industrial branch taking
100% of the secondary sector employment.

35
3.2. Part 2: Examining the mechanism
In addition to evaluating the effect of industrial diversity on enhancing the performance of
environmental regulations, this study also aims to investigate whether this effect is achieved by
easing the economy-environment trade-off, thereby acting as a buffering mechanism. If the
presumed trade-off is indeed influential in the context of China's environmental policy
implementation, one would expect to observe a negative correlation between pollution reduction
progress and the growth of the industrial sector, all else being equal. This correlation, referred to
as the “industry-pollution relationship,” would serve as an indicator of the trade-off at work.
Furthermore, if industrial diversity functions as a buffer, cities with higher levels of industrial
diversity should display a weaker negative correlation between these two variables.
This nuanced analysis aims to validate two key hypotheses: first, that a trade-off between
economic growth and environmental quality exists and is evident in the context of China's
environmental policy implementation; and second, that industrial diversity can buffer this tradeoff, enabling the simultaneous pursuit of economic development and environmental sustainability.
Validating these hypotheses would provide strong support for the theory that industrial diversity
fosters economic resilience, thereby helping developing countries better navigate the challenges
of improving environmental quality amidst economic resource constraints.
To test the above hypotheses, the regression model for the mechanism analysis estimates
both the impact of the city secondary sector size on ambient air pollution levels, and the combined
impact of the city secondary sector size and industrial diversity index upon the air pollution. The
regression model is as follows:

36
𝑃𝑖𝑡 = 𝛼𝑖 + 𝛽1𝐿𝑜𝑔𝑆𝑒𝑐𝑖𝑡 + 𝛽2𝑅𝑒𝑔𝑖𝐿𝑜𝑔𝑆𝑒𝑐𝑖𝑡 + 𝛽3𝐿𝑜𝑔𝑆𝑒𝑐𝑖𝑡𝑇𝑖
+𝛽4𝑅𝑒𝑔𝑖𝐿𝑜𝑔𝑆𝑒𝑐𝑖𝑡𝑇𝑖 + 𝛿𝑋𝑖𝑡 + 𝜃𝑡 + 𝜀𝑖𝑡 [5]
In addition to the same variables from Equation [3], 𝐿𝑜𝑔𝑆𝑒𝑐𝑖𝑡 represents the natural logarithm of
the gross product of the secondary sector in city 𝑖 in year 𝑡. 𝛽1 captures the baseline correlation
between the local secondary sector size magnitude and the pollution level – the industry-pollution
relationship – when industrial diversity index is 0, which is only a theoretical situation with one
industrial branch taking up the whole of the secondary sector employment. 𝛽2 estimates the effect
specific to the “2+26 Cities” region. 𝛽3 is the estimator of this analysis’ main interest, as it
measures by how much each unit of industrial diversity would alter the industry-pollution
relationship. If the sign of 𝛽3 coefficient is opposite from that of 𝛽1, especially if 𝛽1 is positive, it
is considered as a sign of the buffering effect of industrial diversity on the industry-pollution
relationship, and thus on the environment-economy trade-off. 𝛽4 estimates this buffering effect
specific to the “2+26 Cities” region.
Unlike in the first step analysis, where the effect of the industrial diversity is associated
with the performance of the pollution reduction policy, the mechanism of the industry-pollution
relationship and the buffering effect of the industrial diversity is theoretically universal. Therefore
𝛽1 and 𝛽3 outcomes are what is sought after and are sufficient for the examination of the buffering
mechanism. Meanwhile, if the effect is only found in 𝛽2 and 𝛽4 for the “2+26 Cities”, questions
need to be raised for the validity of the theory, and whether the detected environmental policy
enhancing effect of industrial diversity is due to heterogenous qualities of the “2+26 Cities” region.

37
4. Data
This study utilizes data from China’s ambient air quality monitoring system to measure the
dependent variable of air pollution levels. The Air Quality Index (AQI) data is provided by the
National Environmental Monitoring Headquarters Platform, which operates a network of 1,497
monitors across the country. These monitors collect hourly concentration readings of multiple
pollutants. Established in 2014, the system achieved better monitor coverage and data
completeness by 2015. For the purposes of this study, the data is averaged at the annual-city level
to reflect yearly trends and is limited to the period from 2015 to 2019 to ensure data quality and
alignment with the study’s scope. Figure 2.1 presents the average AQI trends for the "2+26 Cities"
and the control group over these years, illustrating the general changes in air pollution levels.
The industrial data used in this study comes from China’s Annual Survey of Industrial
Enterprises (ASIE) for 2015. Firm-level employment data is aggregated by industrial branches
(Industrial Branch Level I) and by cities, allowing for the calculation of city-level entropy by
branch employment (employment entropy) using Equation [4]. Since the ASIE data is only
available annually up to 2015, this study employs a time-invariant 2015 entropy value as the best
available proxy for city-level industrial diversity. Figure 2.2 shows the distribution of city AQI
levels in relation to employment entropy as of 2015. Notably, both the treatment and control groups
include cities with a wide range of industrial diversity levels, though the "2+26 Cities" generally
have higher air pollution levels.
Provincial statistical yearbooks provide data on the annual gross product of the secondary
sector at the city level from 2015 to 2019, published by China’s National Bureau of Statistics and
provincial statistics departments. Figure 2.3 illustrates the distribution of city secondary sector size
in relation to employment entropy as of 2015. The yearbooks also provide GDP per capita, used

38
as a control variable to approximate the overall levels of economic development and government
resources in each city. Additional control variables include daily weather data from the NOAA
Integrated Surface Database, which is aggregated to the annual city level. These weather
variables—temperature, wind speed, and precipitation—control for climate-related influences on
pollution trends. All the variables used in the main analysis are summarized in Table 2.1.
Figure 2. 1 Trends of pollution

39
Figure 2. 2 2015 status of cities: entropy-AQI
Figure 2. 3 2015 status of cities: entropy-ln(secondary sector product)

40
Table 2. 1 Summary statistics
Policy Scheme Regions
Regression variables 2+26 Cities No policy
AQI 104.113 67.473
(1.247) (0.492)
Entropy by branch employment 2.731 2.666
(0.039) (0.012)
Natural logarithm of second sector product 16.459 15.754
(0.068) (0.032)
GDP per capita 58966.386 56497.187
(2270.386) (969.133)
Precipitation 2.402 3.895
(0.403) (0.269)
Wind speed annual average 10.425 10.835
(0.952) (0.541)
Temperature annual average 57.911 59.398
(0.206) (0.287)
Observations 140 1249
5. Empirical results
Regression results for Equation [3] are exhibited in Table 2.2, presenting estimations for
the impact of industrial diversity on the air pollution reduction performance of the “2+26 Cities”
environmental regulations. In Models 1 and 2, the timing of the policy (𝐷𝑡 of Equation [3]) is
factored as a single dummy variable indicating 2017, the starting year of the policy scheme, or
after. For Models 3 and 4, more flexibility is added to the policy timing term by using a set of year
dummies for years 2016-2019, considering the notable pollution reduction efforts prior to the
establishment of the “2+26 Cities”, and the gradual process of the central government guidelines
turning into local execution after the official announcement.

41
Models 1 and 3 estimates the overall effect of the “2+26 Cities” policy on pollution levels.
Model 1 shows a significant pollution reduction of 9.9 unit in AQI in the “2+26 Cities” region
post-policy, comparing to the control group trend. Model 3 aligns with this result and indicates
specifically that this difference appeared since 2017. The results of Models 1 and 3 support the
presumption that the “2+26 Cities” region experienced additional pollution reduction during the
course of the policy scheme compared to the rest of the country.
Models 2 and 4 examines the influence of industrial diversity during the pollution reduction
in the policy region. Model 2 shows that on average, control group cities experienced 3.4 units of
additional AQI reduction post-policy for each unit of additional local industrial entropy by branch
employment. For the “2+26 Cities”, they experience 8.7 units more AQI reduction for each unit of
higher industrial entropy. The year-specific estimates in Model 4 supports the average effect
estimates in Model 2, that higher employment entropy is significantly correlated with more
pollution reduction in the “2+26 Cities”, in addition to the moderate signs of this effect in the
control group cities. For the coefficients of the direct policy effect, they become positive given
they are the estimates of the policy effect conditional on the employment entropy being 0. As Table
2.1 indicates that the average employment entropy being around 2.7 in the “2+26 Cities” region,
with the standard deviation being 0.039, the total effects of the policy region add up to the average
effect estimates in Models 1 and 3.

42
Table 2. 2 Main result Part 1: Industrial diversity impact on environmental regulation
performance
Model Specification
Key estimators (1) (2) (3) (4)
Effect of policy scheme
"2+26 Cities" × Post -9.914*** 13.604
(1.992) (9.913)
"2+26 Cities" × 2016 -3.102 26.490***
(1.931) (4.961)
"2+26 Cities" × 2017 -8.463*** 27.002***
(2.586) (9.235)
"2+26 Cities" × 2018 -12.570*** 24.185**
(3.102) (11.079)
"2+26 Cities" × 2019 -13.902*** 29.939**
(2.931) (14.129)
Effect of industrial diversity
Post × Employment entropy -3.370***
(0.891)
2016 × Employment entropy -0.820
(0.932)
2017 × Employment entropy -3.812***
(1.240)
2018 × Employment entropy -5.862***
(1.393)
2019 × Employment entropy -1.774
(1.231)
"2+26 Cities" × Post × Employment entropy -8.734**
(3.521)
"2+26 Cities" × 2016 × Employment entropy -10.868***
(2.014)
"2+26 Cities" × 2017 × Employment entropy -13.071***
(3.380)
"2+26 Cities" × 2018 × Employment entropy -13.641***
(4.014)
"2+26 Cities" × 2019 × Employment entropy -16.311***
(5.053)
Year FE Yes Yes Yes Yes
City FE Yes Yes Yes Yes
Observations 1,186 1,186 1,180 1,180
Adjusted R2 0.936 0.937 0.939 0.942
Adjusted R2 within 0.087 0.100 0.142 0.181
Note: Standard errors are clustered at the city level. ***, **, and * represent significance at 1%, 5% and 10%, respectively.

43
Table 2. 3 Main result Part 2: Industrial diversity impact on the industry-pollution
relationship
Note: “LogSec” stands for the natural logarithm of the city secondary sector gross product, same as in Equation 3. Standard errors
are clustered at the city level. ***, **, and * represent significance at 1%, 5% and 10%, respectively.
Table 2.3 presents the results for the Equation [5] regression analysis, the test for whether
industrial diversity buffers or weakens the negative correlation between local air pollution and city
secondary sector product size (the industry-pollution relationship). Model 1 confirms the existence
of the industry-pollution relationship, as the natural logarithm of the secondary sector gross
product is positively and significantly correlated with local AQI. Within each of the control group
cities, 1 unit increase of “LogSec” increase is estimated to be on associated with 7.4 units of AQI
increase on average. The interaction term of "2+26 Cities” and “LogSec” is not significant,
indicating the pattern in these cities is not statistically different from the control group. In Model
2, the coefficient of interest shows that for each unit of employment entropy difference, the
industry-pollution relationship is to reduce by 4.4 units on average. Although the result is at a
modest statistical significance level of 10%, it aligns with the theory of industrial diversity’s
Model Specifications
Key estimators (1) (2)
LogSec 7.389*** 18.749***
(2.100) (6.264)
"2+26" × LogSec -15.278 74.685
(12.813) (56.363)
LogSec × Employment entropy -4.421*
(2.249)
"2+26" × LogSec × Employment entropy -34.168
(22.456)
Year FE Yes Yes
City FE Yes Yes
Observations 994 988
Adjusted R2 0.925 0.925
Adjusted R2 within 0.029 0.042

44
buffering effect. This effect is applied not specifically to the “2+26 Cities” but universally to all of
the cities, supporting the assumptions of the buffering effect mechanism.
6. Conclusion and discussion
In this paper, I propose the potential of industrial diversity as a mechanism to alleviate the
economy-environment trade-off in developing, industrializing countries. This theory is examined
through a two-step analysis of the impact of industrial diversity on the performance of China’s
regional environmental policy scheme, the “2+26 Cities” policy in terms of ambient air pollution
reduction, and then its impact on the industry-pollution relationship. The results from regression
analyses support the theory. In the first-step analysis, higher local entropy by industrial branch
employment in the “2+26 Cities” region is associated with a significant additional reduction in the
Air Quality Index (AQI) since the establishment of the policy. Moreover, the second-step analysis
result affirm the theory of industrial diversity's buffering effect, which reduces the tension between
economic growth and environmental preservation when environmental regulation shocks occur.
For cities across the country, the result shows a significant correlation between AQI and the cities’
secondary sector gross product, which diminishes with increased employment entropy. This
suggests that cities with greater industrial diversity have less stress between their economic and
environmental policy agendas, likely due to the economic resilience brought by the industrial
diversity, and can address environmental issues more effectively.
The results are consistent with previous studies on the resilience-providing feature of
industrial diversity (Brown and Greenbaum, 2017; Izraeli and Murphy, 2003; Watson and Deller,
2017). Furthermore, by linking local air pollution with the secondary sector's gross product, this

45
study directly measures the influence of industrial diversity on this relationship, highlighting its
buffering effect. While previous studies focus on the properties of industrial diversity in terms of
its contribution to employment and productivity, this paper proves the applicability of the
resilience industrial diversity in environmental policy and governance. The treatment of the
environmental regulation as an economic shock suggests broader context of industrial diversity
discussion.
To the literature on environmental economics in developing countries, this study
contributes by introducing industrial diversity as a potential solution to their environmental
challenges. It highlights industrial diversity's capacity to reconcile resource competition, a
fundamental source of many institutional failures in addressing environmental issues. The
alleviation of economy-environment trade-off tensions provides an opportunity for developing
countries to smooth their development trajectory, potentially leading to an earlier and lower turning
point on the Environmental Kuznets Curve (EKC), thereby improving welfare for current and
future generations. Though the study utilizes a specific environmental policy in China, the policy
itself only provides the context of a regulation shock to the economy to facilitate the pollution
reduction. Also, as the second step analysis result delivers, not restricted to the policy region but
universal to all cities in China, and most likely in other developing, industrializing countries as
well.
On the other hand, the study recognizes certain limitations and areas that remain
unexplored. It is important to note that the measure of industrial diversity used in this study,
entropy by industrial branch employment, primarily captures unrelated variety as per Frenken et
al. (2007). While the entropy method at the industrial branch level attempts to capture unrelated
variety in the local secondary sectors, there is no certainty that it completely isolates any inter-

46
branch connections. The observed reduced correlation between pollution and industrial sector size
could be attributed to risk spreading among industrial branches, but synergistic effects among
different branches cannot be ruled out with the data used in this study. Similarly, the heterogenous
factors among the cities can be related to the local industrial diversity, and cannot be completely
ruled out for their connection to the dependent variable due to the complexity of industrial diversity.
The study’s conclusions on industrial diversity, even with the finding of the buffering effect
and its contribution to pollution reduction, only provide modest policy advice for practice. For
developing countries facing the economy-environment trade-off in the present or future, the results
of this study suggest policies promoting industrial diversity could benefit these countries to create
a smoother, healthier development trajectory. Yet it needs to be acknowledged that industrial
diversity results from a combination of resource endowments and a series of policy decisions over
extended periods. It often entails an opportunity cost of foregone productivity due to specialization.
This study, therefore, focuses more on understanding the properties of industrial diversity rather
than advocating for its improvement as a policy agenda. While it provides evidence of industrial
diversity's buffering effect during environmental regulatory shocks, it does not delve into the
specific affected industrial branches or workforce reemployment dynamics, crucial aspects for
understanding effective industrial diversity conditions. Thus, much remains to be explored about
industrial diversity, particularly its role in environmental policy and its contribution to economic
resilience against shocks.

47
CHAPTER IV. THE SECTORAL ECONOMIC COST OF CHINA’S AIR POLLUTION
REGULATION CAMPAIGN
1. Introduction
Over the past few decades, China has experienced rapid economic growth driven by its
aggressive industrialization process. However, this growth has come at a significant environmental
cost, with pollution issues becoming increasingly prominent since the early 2000s. By the early
2010s, environmental degradation had reached a level of public concern that could no longer be
ignored. In response, the Chinese central government began integrating environmental targets into
its core policy agenda, which had previously focused primarily on economic development (Zheng
et al., 2021). A pivotal moment occurred in 2014 when air pollution, specifically urban smog
caused by particulate matter, was identified as a critical issue.
Since then, a variety of environmental policies have been implemented at different
governmental levels, leading to substantial progress in pollution alleviation. Urban ambient air
pollutant concentrations have significantly decreased, and improvements have been observed in
other areas such as water pollution reduction, biodiversity and ecosystem protection, and counterdesertification efforts3
. The speed at which China has pursued its environmental agenda, alongside
its prior rapid industrialization and economic growth, has drawn significant attention. China's
recent environmental remediation efforts, particularly in air pollution reduction, have been notably
faster than those of countries that underwent similar processes in the past, such as the United States
following the enactment of the 1970 Clean Air Act (Greenstone et al., 2021).
3 Report on the State of the Ecology and Environment in China 2019. Ministry of Ecology and Environment of the
People’s Republic of China, 2020.

48
This rapid progression is particularly significant in the context of the global climate change
threat, where the atmospheric capacity for carbon emissions is increasingly limited. Developing
countries, which are still striving for industrialization and improved welfare for their citizens, face
stringent constraints. As a leading developing economy, China’s experience in balancing economic
growth with environmental protection offers valuable lessons for other emerging industrializing
countries. China's achievements and the lessons learned during this process could provide crucial
examples and references for other nations seeking sustainable development paths4
.
This paper places the economy-environment trade-off at its core, examining the impact of
China's “2+26 Cities” policy, established in 2017. This regional environmental policy scheme aims
to reduce ambient air pollution in the area surrounding Beijing. While the policy has been effective
in reducing air pollution, this study focuses on its economic impact, particularly the industrial
regulations associated with it. Using difference-in-difference models and their variations, the study
finds a reduction in city-level secondary sector size due to the “2+26 Cities” policy, supporting the
validity of the economy-environment trade-off theory.
In the following sections, this paper will first review the literature on environmental policy
theories and empirical studies in the context of developing countries. It will then describe the
empirical framework, detailing the policy and the empirical strategy used for estimation. The
subsequent sections will present the data used, followed by the statistical results, their
interpretation, and robustness checks. Finally, the paper concludes with insights into the economyenvironment trade-off and the implications for policymaking.
4 World Development Report 2021: Data for Better Lives. World Bank, 2021

49
2. Background
Environmental policy in developing countries has gained significant attention over the past
decades, especially in the context of sustainable development. Developing countries face unique
challenges in formulating and implementing environmental policies (Greenstone & Jack, 2015).
These challenges include limited financial resources, lack of technical expertise, and competing
development priorities. Nonetheless, several countries have made strides in adopting policies
aimed at reducing pollution and conserving natural resources. For instance, China has
implemented policies like the Air Pollution Prevention and Control Action Plan in 2013 and the
Three-Year Action Plan for Winning the Blue Sky Defense Battle in 2018 (Zheng et al., 2021).
These policies have led to significant air quality improvements, particularly in major urban centers.
Similarly, India introduced the National Clean Air Programme (NCAP) aimed at reducing
particulate matter pollution by 20-30% by 2024 (Ministry of Environment, Forest and Climate
Change, 2019). However, the effectiveness of such policies often depends on enforcement
mechanisms and the capacity of local governments.
The relationship between economic growth and environmental degradation in developing
countries is complex. While environmental policies aim to preserve natural resources and protect
public health, they also have significant implications for the economy and industries. The costs
associated with environmental policies include direct compliance costs (Jaffe et al., 1995), indirect
costs due to changes in market dynamics (Aldy & Pizer, 2015), and opportunity costs (Hanley et
al., 2015). Studies have examined the economic impact of environmental policies, focusing on
effects on economic growth, employment, and productivity. Some studies suggest that
environmental regulations can slow down economic growth in the short term due to increased
compliance costs (Kozluk & Zipperer, 2015; Tian et al., 2020; Dechezleprêtre & Sato, 2017).

50
Environmental policies often require significant investments in new technologies and
infrastructure, straining capital resources and affecting economic performance (Ambec et al., 2013;
Gray & Shadbegian, 2003). Stringent environmental policies can affect the competitiveness of
domestic industries, leading to potential negative impacts on trade and economic performance
(Ederington & Minier, 2003; Levinson & Taylor, 2008). Environmental policies can have mixed
effects on employment, with some studies highlighting potential job losses in certain sectors
(Walker, 2011; Greenstone, 2002). However, other studies argue that these policies can lead to
sustainable long-term growth by promoting green technologies and reducing environmental
degradation (Borghesi, 2000; Stern, 2004). The specific impacts on employment and productivity
are also debatable. For example, Berman & Bui (2001) found that stricter air quality regulations
in the United States did not significantly affect employment in the oil refinery sector, suggesting
that the overall impact on jobs may be neutral or even positive. Lanoie et al. (2008) found that
Canadian manufacturing plants subject to stringent environmental regulations experienced
productivity gains due to increased innovation and efficiency improvements.
For industrializing developing countries, the core dilemma lies in balancing economic
growth and welfare improvement with maintaining environmental quality to support public health.
This balance is particularly challenging given limited resources and institutional capacities. The
economy-environment trade-off thus becomes central to government agenda setting and resource
allocation. Intense industrialization can drive environmental degradation, while stringent
environmental policies may constrain industrial activities, reduce incomes, and limit future
productivity growth (Grossman & Krueger, 1995). Understanding the mechanisms of this tradeoff, including the performance and efficiency of different policy tools, is crucial for informed
environmental policymaking in developing countries (Cole et al., 2006).

51
3. Empirical framework
China's rapid economic growth through industrialization over the past few decades has led
to significant environmental challenges, particularly since the early 2000s. The central government
recognized air pollution, specifically urban smog caused by particulate matter, as a critical issue
to address. This shift was notably marked in 2014 when environmental targets were integrated into
the national policy agenda, which had previously focused predominantly on economic growth
(Zheng et al., 2021).
China's experience offers valuable lessons and references for emerging industrializing
countries, highlighting both successes and challenges in balancing economic growth with
environmental protection. To examine the economic impact of environmental policy, this study
focuses on the “2+26 Cities” policy, a major regional environmental policy campaign in China.
The policy aims to reduce pollution in a cluster of industrialized northern cities surrounding
Beijing. Environmental standards and goals were set for these cities and carried out by local
governments to control industrial emissions and urban pollution. The policy was first announced
in 2017 and gradually implemented by local governments in the following years. During this
period, significant reductions in air pollution were observed nationwide, particularly in the policy
area. The environmental campaign and local regulations were expected to impact and add
compliance costs to more polluting industries. This study examines this process to reveal the
overall impact on the economy and sectoral structure.
The impact of the “2+26” policy is examined in a quasi-experimental setting to understand
its influence on city-level economic productivity. The policy's influence on the secondary sector,
tertiary sector, and total economy is evaluated and compared for a comprehensive understanding
of its complex impact. The policy treatment is defined as starting from 2017, the year of the policy

52
announcement, with the policy area defined as the “2+26 Cities.” Observations from 2017 and
beyond are considered treated temporally, and those located in the “2+26 Cities” are treated
spatially.
The difference-in-difference (DiD) approach is adopted for the estimation model, with
fixed effects employed to allow for city-specific baselines. The regression model is specified as
follows:
𝑉𝑖𝑡 = 𝛼𝑖 + 𝛽1𝐷𝑖 + 𝛽2𝑃𝑜𝑠𝑡𝑡 + 𝛽3𝐷𝑖𝑃𝑜𝑠𝑡𝑡 + 𝛾𝑌𝑒𝑎𝑟𝑡 + 𝛿𝑋𝑖𝑡 + 𝜀𝑖𝑡 [6]
Where as 𝑉𝑖𝑡 represents the economic sector indicator of city i in year t. 𝛼𝑖
is the set of city fixed
effect dummies. 𝐷𝑖
is the dummy variable for the spatial treatment indicating whether city i is
included in the “2+26” cities. 𝑃𝑜𝑠𝑡𝑡
is the dummy variable for the temporal treatment, indicating
whether year t is 2017 or after. 𝑌𝑒𝑎𝑟𝑡
is the linear variable for year t. 𝑋𝑖𝑡 serves as the control
variable to proxy the city development level, measured by per capita GDP.
The estimators 𝛽1 and 𝛽2 respectively represent the difference of the dependent variable
between the policy area and the control group, and the difference of the control group before and
after the policy. While 𝛽1𝐷𝑖
is in effect absorbed by 𝛼𝑖
, 𝛽2 assumes a common trend in the
dependent variable for the treatment and control group cities in the absence of the policy, and is to
pick up any effect of the policy timing on both groups. 𝛽3 is the key estimator of interest in this
study for the policy impact, for the interaction effect of the spatial and temporal treatment, defining
the cities covered and treated by the policy after the announcement of the “2+26 Cities” policy
scheme. 𝛾 assumes and estimates a linear annual trend correlation, and 𝛿 estimates the assumed
common linear effect of the per capita GDP.

53
The model is specified alternatively with year fixed effect added to the baseline model
(two-way fixed effect), as in Equation [7], and with separate linear trends pre- and post-policy
respectively for the treatment and control groups (Difference-in-Difference-in-Difference, DDD),
as in Equation [8]. The former specification allows for non-linear common annual trend in the
absence of the policy, and draws the focus of the estimation to the coefficient of the interaction
term of post-policy and policy area – the policy effect of interest – as the potential effect of postpolicy is absorbed. The latter specification relaxes the DiD model assumption that the effect of the
policy is immediate, and takes into consideration that the policy effect may instead result in
changing the dependent variable trend in the policy area. The equations are as follows:
𝑉𝑖𝑡 = 𝛼𝑖 + 𝛽1𝐷𝑖 + 𝛽2𝑃𝑜𝑠𝑡𝑡 + 𝛽3𝐷𝑖𝑃𝑜𝑠𝑡𝑡 + 𝜃𝑡 + 𝛿𝑋𝑖𝑡 + 𝜀𝑖𝑡 [7]
𝑉𝑖𝑡 = 𝛼𝑖 + 𝛽1𝐷𝑖 + 𝛽2𝑃𝑜𝑠𝑡𝑡 + 𝛽3𝐷𝑖𝑃𝑜𝑠𝑡𝑡 + 𝛽4𝐷𝑖𝑌𝑒𝑎𝑟𝑡 + 𝛽5𝑃𝑜𝑠𝑡𝑡𝑌𝑒𝑎𝑟𝑡
+𝛽6𝐷𝑖𝑃𝑜𝑠𝑡𝑡𝑌𝑒𝑎𝑟𝑡 + 𝛾𝑌𝑒𝑎𝑟𝑡 + 𝛿𝑋𝑖𝑡 + 𝜀𝑖𝑡 [8]
In Equation [7], the term of post-policy (𝑃𝑜𝑠𝑡𝑡
) is to be absorbed by the year fixed effect 𝜃𝑡
, and
only 𝛽3 would be estimated in the regression. In Equation [8], 𝛽4 estimates the difference in the
pre-policy dependent variable trends between the treatment and control groups. 𝛽5 estimates the
difference in the control group trends pre- and post-policy. 𝛽6 is the estimator for the difference in
the pre- and post-policy difference in the treatment and control group trends.
As is mentioned, the regression models are used to estimate the policy effects respectively
on the secondary sector, the tertiary sector, and the total economy at the city level. While the policy
is expected to have a direct negative impact on the secondary sector given the regulations and the
strengthened production standards, the reallocation of the production factors of labor force and

54
capital may potentially overflow into the tertiary sector and lead to unforeseeable influences. The
primary sector is not considered as a subject of such mechanism due to China’s system of urban
and rural resident identification, which reduces the likelihood of workforce spillover from the
secondary sectors which are mostly located in the urban area to the rural area where the agricultural
sectors are located. The city-level total economy would provide information on the overall
economic impact of the environmental policy, with various mechanisms of policy impact
adjustment in effect.
4. Data
In this study, given the quasi-experiment setting of the “2+26 Cities” policy, the core
independent variables of treatment consist of dummy variables defining the effective timing and
area of the policy. This information is provided by the official document of the “Action Plan for
the Atmospheric Pollution Treatment in the Beijing-Tianjin-Hebei and Surrounding Region,”
published in 2017 by the Ministry of Ecology and Environment. The region centers Beijing and
Tianjin, with 4 provincial capital cities and 22 other prefecture-cities included.
The dependent variables for the policy impact assessment are the local economy statistics
by sectors. This data is aggregated from provincial statistical yearbooks of the 31 provinces (and
province level municipalities) of mainland China over the years 2010-2019. The yearbooks are
produced and published by the respective statistics department of province level governments. The
data provides for the dependent variables of secondary sector added value , tertiary sector added
value, and total GDP at the city-year level. In the regression analysis, the dependent variables are
taken for their natural logarithm to resemble the relative magnitude of the sector or economy size,
in consideration of the vast difference across cities.

55
Table 3. 1 Summary statistics
Sample Groups
Regression variables 2+26 Cities Control group
ln(secondary sector added value) 16.373 15.645
(0.042) (0.020)
ln(tertiary sector added value) 16.173 15.495
(0.062) (0.021)
ln(GDP) 17.077 16.425
(0.050) (0.018)
Entropy by employment 2.731 2.648
(0.028) (0.009)
Secondary sector share in GDP 51.293 47.880
(0.942) (0.374)
Tertiary sector share in GDP 42.267 40.962
(0.799) (0.333)
Observations 280 2553
Note: Standard deviations of variables are in bracket. Entropy by employment, Secondary sector share in GDP, and
Tertiary sector share in GDP are aggregated to the city level from the 2015 ASIE, therefore are time-invariant.
For the follow-up mechanism analyses and robustness checks to be discussed in the
sections to come, the share of secondary sector added value in GDP and the share of tertiary sector
are also collected from the same data source of the provincial statistical yearbooks. These variables
are used to explore how the sectoral structure of local economy may influence the policy impact
on total product. Similarly, industrial diversity and structure of the industrial sector economy are
also examined for their influence on the policy. These variables are calculated from the firm level
industrial data provided by China’s Annual Survey of Industrial Enterprises (ASIE) of 2015. The
details of the variable calculation are to be covered in the respective sections of the analyses. The
summary statistics of the mentioned variables are listed in Table 3.1, with the trends of the
secondary sector size (logarithm of secondary sector added value) exhibited in Fig. 3.1.

56
Figure 3. 1 Average city-level secondary sector size by treatment and control groups
5. Results
Three model specifications (DiD, Year FE, and DDD) are used for the regression analysis,
providing flexibility in assumptions for how the control group cities resemble the treatment group
of the “2+26 Cities” area and how the policy effect may take form, and in the assumptions for how
the policy effect may take form. The three models are run for the secondary sector, the tertiary
sector, and total GDP respectively to compare the policy effects across sectors and the total
economy. The regression results are summarized in Table 3.2.

57
Table 3. 2 Main analysis results
Note: Standard errors are clustered at the city level. ***, **, and * represent significance at 1%, 5% and 10%,
respectively. The year variable is in relative terms, with the reference year 2016 as year 0.
Starting with the secondary sector regressions, with the logarithm of the city level
secondary sector added value (LogSec) as the dependent variables, the DiD model (Column 1) is
based on the strong assumption of a parallel constant national linear trend of secondary sector
growth between the treated “2+26 Cities” and the rest of the country. The model returns the results
of a significant positive nationwide annual growth rate, and a significant average decrease in city
level secondary sectors after the policy year of 2017. Most importantly, the model estimates a -
0.063 impact of the policy, at the 5% significance level, on the log secondary sector added value
of the “2+26 Cities”, in addition to the drop on all of the cities. The year FE model (Column 2)
supports this estimation with the same estimated coefficient for the policy effect, albeit only at the
Model Specifications
DiD Year FE DDD
Independent variables (1) (2) (3)
Post -0.067*** -0.036***
(0.011) (0.009)
2+26 × Post -0.063** -0.063* 0.077***
(0.032) (0.032) (0.017)
Year 0.062*** 0.065***
(0.003) (0.004)
2+26 × Year -0.013*
(0.007)
Post × Year -0.023***
(0.007)
2+26 × Post × Year -0.037*
(0.019)
Year FE No Yes No
City FE Yes Yes Yes
Adj. R-squared 0.967 0.970 0.967
Adj. R-squared within 0.406 0.002 0.409
Observations 2,976 2,976 2,976

58
10% significance level, with the common linear growth assumption relaxed. This model controls
for the year fixed effects and assumes parallel nonlinear trends between the treatment and control
group cities. The DDD model of Column 3 allows for different pre-policy linear trends for the two
groups of cities, and difference in their respective changes of the trends post-policy. The regression
result confirms the positive growing trend of the secondary sector and the significant shrink of the
sector size after 2017 in the control group cities. It also finds the pre-policy growing trend of the
secondary sector in the “2+26 Cities” is slower (10% significance level) than the control group,
and that the growing trend of the control group slowed down post-policy (1% significance level).
In addition, as the featured estimator of this model, the three-way interaction term of “2+26 × Post
× Year” coefficient indicates that the “2+26 Cities” policy slowed down the treatment group
growing trend of the secondary sector even more than that of the control group post-policy, by
0.037 unit per year in terms of log sector added value at the 10% significance level over the period
of 2017-2019. Although the estimated immediate policy effect, the interaction term of the “2+26
Cities” and Post, is overturned to be positive, it is due to the kinked trend of the policy region and
the selection of the reference year for the regression, and is not to be interpreted as a definitive
immediate impact of the policy.
In general, the regression results of the three model specifications show negative impact of
the “2+26 Cities” policy on the city level secondary sectors of economy. The results suggest the
economic cost, especially to the secondary industry, introduced by the enforcement of
environmental policies in the form of industrial regulations. This provides evidence, combined
with the reduction of air pollution in the same region over the same period of time, to support the
theory of the economy-environment trade-off. In other words, the environmental policy scheme
may be considered as improving the environmental at the cost of industrial development.

59
However, it is important to acknowledge the limitations of this study using the economic
sector data at the year-city granularity: it is difficult to disentangle the impact of the “2+26 Cities”
environmental policy scheme and that of many other policies and incidents taking place in the
same time period. In the following section, robustness check analyses are adopted in attempt to
reveal the process and mechanism of the “2+26 Cities” policy in a rather well-rounded manner,
and in turn enhance the understanding of the EKC turning point in the developing country scenario
and provide insight in in related policy making.
6. Robustness checks
6.1. Alternative anchor for policy year
Although it has been viewed as a clear-cut, effective-immediately policy in the main
analysis for simplicity, the reality for the “2+26 Cities” policy is not at all carried out in such
fashion. As the post-policy treatment group trend in the DDD result suggests, the policy was
implemented in a gradual process. In March 2017, China’s central government issued the Action
Plan document and announced the “2+26 Cities” to be included in the policy. The Action Plan
featured the missions and specific goals to be accomplished, directions of effort, and how the local
governments should establish their respective policies for action and how they should cooperate
with local companies to effectively implement. Accordingly, the provincial governments issued
their versions of province specific policy documents to identify local pollution issues and measures
to be taken, as well as production and emission standards. Later in October 2017, the central
government announced the air pollution regulation measures specific to the 2017-18 winter heating
season, restricting industrial production until March 2018, to make space for the necessary
emission from public heating.

60
Understanding how the policy is phased in and carried out, it is arguable that in the fiscal
year of 2017 the execution and influences of the policy was not yet fully in place, especially when
this study is considering the effects of the policy in terms of the cities’ annual economic
performance. Motivated by this perspective, the following analysis alternatively redefines the year
2018 as the temporal treatment of the policy with the same regression models run. This one-yearafter analysis is to provide comparison with the results from the main analysis, potentially finding
a clearer timing for the policy effects to take place, and potentially allowing a more straightforward
interpretation of the regression results.
The regression result of the one-year-after analysis is exhibited in Table 3.3. The DiD and
year FE model results in Column 1 and 2 resembles that for the original main analysis, with the
interaction term of the “2+26 Cities” and the post-policy period significantly negative at -0.094
(5% significance level), indicating a substantial additional reduction in the secondary sector size
in the policy covered cities. The DDD model, in contrast with the secondary sector counterpart,
show no significant effect for the “2+26 × Post” or the “2+26 × Post × Year” terms, suggesting no
policy impact detected when the policy year is defined as 2018 to make the pre-post comparison.

61
Table 3. 3 Regression result for 2018 as redefined policy year
Note: Standard errors are clustered at the city level. ***, **, and * represent significance at 1%, 5% and 10%,
respectively. The year variable is in relative terms, with the reference year 2017 as year 0. Post redefined as 2018 and
after.
6.2. Policy effects in the primary and tertiary sectors
While the effect of the “2+26 Cities” policy in the secondary sector is examined in the main
analysis, the models are not exclusive to the influence of other policies. This is increasingly
important to address given the estimated effects of the policy, or the temporal treatment of the
policy as “post 2017”, shows significant outcomes outside of the “2+26 Cities” policy region. It is
therefore helpful to check whether, and potentially how, the spatial and temporal treatments of the
“2+26 Cities” policy extend into the other sectors of economy. If the estimated results for the
policy effects in the other sectors closely resemble that of the secondary sector, the argument that
Model Specifications
DiD Year FE DDD
Independent variables (1) (2) (3)
Post -0.045*** 0.015
(0.012) (0.011)
2+26 × Post -0.094** -0.094** -0.044
(0.037) (0.038) (0.038)
Year 0.058*** 0.060***
(0.003) (0.004)
2+26 × Year -0.009
(0.007)
Post × Year -0.045***
(0.010)
2+26 × Post × Year -0.003
(0.021)
Year FE No Yes No
City FE Yes Yes Yes
Adj. R-squared 0.966 0.970 0.967
Adj. R-squared within 0.404 0.003 0.405
Observations 2,976 2,976 2,976

62
the recession in the secondary sector is caused by the environmental policy and industrial
regulations would be compromised. To address such threat for identifying the source of the effects,
the same model specifications are applied in regressions using the log sector added values of the
primary and tertiary sector as dependent variables. The regression results are exhibited in Table
3.4.
From the result of DiD models of both sectors, significant average growing trends of the
sector added value sizes can be observed, same as the reduction of the sector sizes for control
group cities across the country. These effects are found also in the main analysis for the secondary
sector. The key estimator “2+26 × Post” show different results. For the primary sector, the policy
effect is estimated to be a significant reduction of 0.142 in the log of primary sector added value
(LogPri) as an average effect to the “2+26 Cities” after 2017, aligning with the previously found
negative effect of the policy on the secondary sector. In contrast, the tertiary sector does not show
significant effect in the policy region post the establishment of the policy. This comparison is
repeated in the year FE model, where the parallel growing trend and control group post policy
effects are absorbed, with the estimator of policy effect in the policy region exactly the same as in
the DiD specification. In the DDD models for the two sectors, while the post policy reduction in
the control group cities remain for the primary sector, this effect in the tertiary sector is relocated
to the estimator of “Post × Year”, indicating this reduction took form as a gradual process in the
tertiary sector. The primary sector also lost the significant coefficient for the “2+26 × Post”
estimator, yet recording a significant gradual reduction of LogPri in the “2+26 Cities” area of 0.030
per year starting from pre-policy. Neither sector finds significant reducing trend specific to the
“2+26 Cities” region post-policy, as is found for the secondary sector.

63
Table 3. 4 Regression result for the primary and tertiary sectors
Note: Standard errors are clustered at the city level. ***, **, and * represent significance at 1%, 5% and 10%,
respectively. The year variable is in relative terms, with the reference year 2016 as year 0.
In summary, the tertiary sector, roughly equivalent to the service sector, does not show any
impact from the establishment of the “2+26 Cities” region. While there is significant reduction in
the growth rate since the year of 2017, it is unlikely to be directly caused by the environmental
policy scheme as it is applied to all cities. Similar effects found for the secondary sector may also
be due to the slowing down of general economy. The primary sector, roughly equivalent to the
agricultural sector, shows deviation between the “2+26 Cities” and the control group cities.
However, the DDD model indicates that this deviation may not be associated with the policy of
this study’s interest, but a gradual process started long before 2017 the establishment of the
environmental regulations. In comparison, although all three sectors have experienced reduction
LogPri LogTer
DiD Year FE DDD DiD Year FE DDD
Independent variables (1) (2) (3) (4) (5) (6)
Post -0.103*** -0.109*** -0.043*** -0.002
(0.009) (0.010) (0.006) (0.006)
2+26 * Post -0.142*** -0.142*** 0.018 -0.012 -0.012 -0.001
(0.021) (0.021) (0.019) (0.027) (0.027) (0.012)
Year 0.069*** 0.073*** 0.121*** 0.123***
(0.002) (0.002) (0.002) (0.002)
2+26 * Year -0.030*** -0.003
(0.005) (0.006)
Post * Year -0.005 -0.025***
(0.004) (0.004)
2+26 * Post * Year -0.005 0.001
(0.010) (0.010)
Year FE No Yes No No Yes No
City FE Yes Yes Yes Yes Yes Yes
Adj. R-squared 0.982 0.984 0.983 0.990 0.990 0.990
Adj. R-squared within 0.675 0.030 0.681 0.909 0.000 0.910
Observations 2,976 2,976 2,976 2,976 2,976 2,976

64
or slowing of growth in various forms, the difference in the regression results suggests that the
reduction of the secondary sector size in the “2+26 Cities” is at least partially specifically
associated with the regional environmental policy scheme.
7. Discussion and conclusion
The empirical results of this study provide substantial evidence of the economyenvironment trade-off in the context of China's "2+26 Cities" policy. The regression analyses
demonstrate a significant negative impact of the policy on the secondary sector, confirming the
hypothesis that stringent environmental regulations can lead to a contraction in industrial activities.
This finding aligns with previous studies that highlight the short-term economic costs of
environmental policies, particularly in heavily industrialized regions (Tian et al., 2020;
Dechezleprêtre & Sato, 2017; Kozluk & Zipperer, 2015).
The robustness checks further support the validity of the primary findings. By redefining
the policy year to 2018, the analysis revealed a clearer temporal impact of the policy, emphasizing
the gradual implementation process and its delayed effects on the secondary sector. The
examination of the primary and tertiary sectors provided additional insights, indicating that while
the primary sector experienced some reduction, the tertiary sector remained largely unaffected by
the policy. This sectoral differentiation suggests that the secondary sector bears the brunt of
environmental regulations, which is consistent with the theory that industrial activities are more
directly impacted by pollution control measures (Gray & Shadbegian, 2003).
The significant reduction in the secondary sector's growth rate post-implementation of the
"2+26 Cities" policy underscores the need for careful consideration of the economic implications

65
of environmental policies. While the policy achieved its goal of reducing air pollution, as
evidenced by improved air quality metrics, it also imposed economic costs on the affected cities.
This dual outcome highlights the inherent challenges in balancing environmental protection with
economic growth, particularly in developing countries with limited resources and institutional
capacities (Grossman & Krueger, 1995).
This study contributes to the understanding of the economy-environment trade-off by
empirically analyzing the impact of China's "2+26 Cities" policy. The findings confirm that
stringent environmental regulations can lead to a contraction in industrial activities, supporting the
theory that environmental improvements often come at the cost of economic growth in the short
term. However, the study also illustrates that the impact varies across different sectors, with the
secondary sector being most affected.
The policy implications of these findings are significant for developing countries facing
similar dilemmas. Policymakers must strike a balance between enforcing environmental
regulations and supporting economic growth. Strategies to mitigate the negative economic impacts
of such policies could include:
• Phased Implementation: Gradual implementation of environmental policies can help
industries adjust and adopt cleaner technologies without sudden economic disruptions.
• Innovation Incentives: Providing financial and technical support for innovation in cleaner
technologies can help industries comply with regulations while maintaining
competitiveness (Porter & van der Linde, 1995).
• Sectoral Support: Tailored support for sectors most affected by environmental policies,
such as the secondary sector, can help mitigate economic losses and facilitate a smoother
transition to sustainable practices.

66
• Flexible Policy Instruments: Market-based instruments like cap-and-trade systems and
environmental taxes can provide flexibility for industries to reduce emissions costeffectively (Goulder & Parry, 2008).
China's experience with the "2+26 Cities" policy offers valuable lessons for other
developing countries. The rapid improvements in air quality achieved through stringent regulations
demonstrate that significant environmental gains are possible. However, the associated economic
costs highlight the importance of designing policies that minimize negative impacts on economic
growth. Further research is needed to explore the long-term effects of environmental policies and
to develop strategies that can effectively balance environmental protection with economic
development.

67
CHAPTER V. CONCLUSION OF DISSERTATION
This three-essay dissertation has explored the intricate relationship between economic
growth and environmental protection in China, focusing on the trade-offs and policy implications
that arise from efforts to address air pollution over the past decade. The theories presented in this
dissertation are built upon the acknowledgment of constrained resources in a developing country
context and the inherent competition between economic and environmental demands within
government agendas. This work conceptualizes the economy-environment trade-off and examines
the policy implications associated with this challenge.
Two key environmental policies are studied: the 2014 APEC pollution ban and the "2+26
Cities" regional environmental policy scheme. The former, examined in the first essay, functioned
as a short-term pollution ban aimed at temporary air pollution reduction for political gains at
minimal economic cost. The latter, established as a long-term environmental policy scheme for
regional pollution reduction, is analyzed in the second and third essays. Although these policies
cover similar, overlapping areas, their distinctive designs and outcomes reflect various aspects of
how the economy-environment trade-off influences China’s environmental governance. Through
the analysis of these two policies across three separate essays, this dissertation provides valuable
insights into the challenges and opportunities faced by developing countries as they navigate the
complex terrain of sustainable development.

68
1. Summary of findings
Chapter II, which examines the 2014 APEC pollution ban, revealed significant unintended
consequences of the policy. While the short-term measures effectively reduced air pollution during
the event, the sharp rebound in pollution levels immediately afterward highlighted the
inefficiencies of such temporary interventions. Although the policy was not intended for a
sustainable effect but rather to temporarily reduce pollution at minimal economic cost, the
industrial recoup that followed was not accounted for in the policy design, leading to losses that
exceeded expectations. This phenomenon underscores the limitations of command-and-control
policy designs that fail to account for industrial behavior, as well as the weak institutions that
constrain the government’s capacity for rigorous policy enforcement. These findings highlight the
challenges that developing countries face in their environmental governance, particularly when
navigating the economy-environment trade-off.
In contrast, the "2+26 Cities" policy, which aimed to reduce long-term air pollution in the
Beijing-Tianjin-Hebei region, demonstrated more sustainable results. Chapter III confirmed the
policy's success in reducing pollution levels and showed that industrial diversity, as a key attribute
of local industrialization, enhances the performance of environmental policies in reducing air
pollution. Industrial diversity, measured as entropy by industrial branch product, provided
economic resilience and mitigated the adverse impacts of strict environmental regulations on
industries. The study emphasizes the importance of considering local economic structures when
designing environmental policies, especially in developing countries where the economyenvironment trade-off is more pronounced.
Chapter IV built upon the recognition of the "2+26 Cities" policy’s achievements in
reducing air pollution. In light of the economy-environment trade-off theory, the study sought and

69
found preliminary evidence of the policy’s impact on industries, particularly in terms of reducing
the secondary sector size in policy-covered cities. The analysis also suggests that the economic
costs associated with environmental policies may manifest in different temporal patterns, such as
a combination of immediate impact and gradual influence. While the observations for the "2+26
Cities" policy do not offer further implications regarding the factors influencing economic costs,
the findings support concerns about the economic costs of environmental policies and point to the
need for further research on strategies for developing countries to address this challenge.
2. Implications for policy
The findings of this dissertation have important implications for policymakers in China
and other developing countries. First, the study underscores the critical role of institutional
capacity in the success of environmental policies. Institutional strength determines the feasibility
of environmental policy options and the effectiveness of policy implementation and enforcement.
Developing countries must invest in building institutional capacities to ensure that environmental
policies are not only effective but also manageable and sustainable.
Second, the research highlights the interconnections between industrial and environmental
policies. As environmental policies are implemented, the demand for economic development can
create pressure that hinders their effectiveness, especially at the local level. While policy design
and institutional development are crucial, the impact of environmental policies is also influenced
by the underlying economic structure. The dissertation’s finding that industrial diversity enhances
the effectiveness of air pollution reduction policies suggests that the efficiency of environmental
policies is partly determined by the health of the industrial structure. However, building industrial

70
diversity is a complex process, shaped by local resource endowments and a series of decisions
over the course of industrialization rather than by a few standalone policies. A balanced
industrialization path, planned and followed over the long term, offers a better opportunity to
address environmental issues more smoothly and efficiently.
Finally, this dissertation provides direct evidence of the economic impact on the industrial
sector resulting from the implementation of environmental policy schemes. Centered around the
economy-environment trade-off, these findings support previous arguments that emphasize the
importance of considering the economy and industries in environmental policy-making. For
developing countries, the economic costs associated with efforts to mitigate environmental
degradation are inevitable. Considering institutional development and other potential influencing
factors, it is up to policymakers to make comprehensive decisions regarding the extent, timing,
and form of balancing and shifting priorities on the economy-environment spectrum within their
central policy agendas.
3. Future research directions
While this dissertation has provided valuable insights into the economy-environment tradeoff in China, several areas warrant further investigation. First, data limitations posed a significant
challenge to conducting more detailed analyses of the policies examined in this dissertation.
Although the establishment of the national ambient air pollution monitoring system enabled this
series of studies, the industrial and local socioeconomic data lacked granularity and detail.
Moreover, the COVID-19 pandemic halted global development, including in China, disrupting
economic growth and industrial production, which prevented continuous observation of the long-

71
term impacts of these policies. Future research would benefit from the accumulation of new data
and from new policy efforts in China and other developing countries.
Second, while this dissertation focuses on air pollution—the most commonly recognized
form of environmental degradation associated with industrial processes—environmental issues
linked to economic development can take various forms. Water pollution, deforestation,
desertification, and biodiversity loss can also be studied within the economy-environment tradeoff framework to test the validity of the theory and examine the consequences of human production
behavior. Contextualizing these different environmental issues within the broader scope of social
welfare is both a necessary and valuable endeavor.
Finally, as this dissertation suggests using China as a reference for environmental
governance in developing countries, it is important to recognize the diverse realities of these
nations. Comparative studies between China and other developing countries could shed light on
the broader applicability of the findings and offer lessons for countries at different stages of
industrialization and environmental governance. Additionally, further research is needed to
understand the social dimensions of environmental policies, including their effects on public health,
income inequality, and social welfare.

72
REFERENCES
Aarstad, J., Kvitastein, O. A., & Jakobsen, S.-E. (2016). Related and unrelated variety as regional
drivers of enterprise productivity and innovation: A multilevel study. Research Policy, 45(4),
844–856. https://doi.org/10.1016/j.respol.2016.01.013
Antweiler, W., Copeland, B. R., & Taylor, M. S. (2001). Is Free Trade Good for the
Environment? American Economic Review, 91(4), 877–908. https://doi.org/10.1257/aer.91.4.877
Attaran, M. (1986). Industrial diversity and economic performance in U.S. areas. The Annals of
Regional Science, 20(2), 44–54. https://doi.org/10.1007/BF01287240
Attaran, M., & Zwick, M. (1987). Entropy and Other Measures of Industrial Diversification.
Quarterly Journal of Business and Economics, 26(4), 17–34.
Auffhammer, M., & Kellogg, R. (2011). Clearing the Air? The Effects of Gasoline Content
Regulation on Air Quality. American Economic Review, 101(6), 2687–2722.
https://doi.org/10.1257/aer.101.6.2687
Balland, P.-A., Broekel, T., Diodato, D., Giuliani, E., Hausmann, R., O’Clery, N., & Rigby, D.
(2022). The new paradigm of economic complexity. Research Policy, 51(3), 104450.
https://doi.org/10.1016/j.respol.2021.104450
Bell, R. G., & Russell, C. (2002). Environmental Policy for Developing Countries. Issues in
Science and Technology, 18(3), 63–70.
Berman, E., & Bui, L. T. M. (2001). Environmental regulation and labor demand: Evidence from
the South Coast Air Basin. Journal of Public Economics, 79(2), 265–295.
https://doi.org/10.1016/S0047-2727(99)00101-2
Blackman, A. (2010). Alternative Pollution Control Policies in
Developing Countries. Review of Environmental Economics and Policy, 4(2), 234–
253. https://doi.org/10.1093/reep/req005
Blackman, A., & Harrington, W. (2003). The Use of Economic Incentives in Developing
Countries: Lessons From International Experience With Industrial Air Pollution. In The Theory
and Practice of Command and Control in Environmental Policy. Routledge.
Blackman, A., Li, Z., & Liu, A. A. (2018). Efficacy of Command-and-Control and Market-Based
Environmental Regulation in Developing Countries. Annual Review of Resource Economics,
10(1), 381–404. https://doi.org/10.1146/annurev-resource-100517-023144

73
Borghesi, S. (1999). The Environmental Kuznets Curve: A Survey of the Literature (SSRN
Scholarly Paper 200556). https://doi.org/10.2139/ssrn.200556
Brajer, V., & Mead, R. W. (2003). Blue Skies in Beijing? Looking at the Olympic Effect. The
Journal of Environment & Development, 12(2), 239–263.
https://doi.org/10.1177/1070496503012002005
Brown, L., & Greenbaum, R. T. (2017). The role of industrial diversity in economic resilience:
An empirical examination across 35 years. Urban Studies, 54(6), 1347–1366.
https://doi.org/10.1177/0042098015624870
Chen, J., Li, X., & Zhu, Y. (2023). Shock absorber and shock diffuser: The multiple roles of
industrial diversity in shaping regional economic resilience after the Great Recession. The Annals
of Regional Science. https://doi.org/10.1007/s00168-023-01233-2
Chen, Y., Jin, G. Z., Kumar, N., & Shi, G. (2013). The promise of Beijing: Evaluating the impact
of the 2008 Olympic Games on air quality. Journal of Environmental Economics and
Management, 66(3), 424–443. https://doi.org/10.1016/j.jeem.2013.06.005
Cohen, A., & Keiser, D. A. (2017). The effectiveness of incomplete and overlapping pollution
regulation: Evidence from bans on phosphate in automatic dishwasher detergent. Journal of
Public Economics, 150, 53–74. https://doi.org/10.1016/j.jpubeco.2017.03.005
Cole, D. H., & Grossman, P. Z. (2003). When is Command-and-Control Efficient? Institutions,
Technology, and the Comparative Efficiency of Alternative Regulatory Regimes For
Environmental Protection. In The Theory and Practice of Command and Control in
Environmental Policy. Routledge.
Cole, M. A., Elliott, R. J. R., & Wu, S. (2008). Industrial activity and the environment in China:
An industry-level analysis. China Economic Review, 19(3), 393–408.
https://doi.org/10.1016/j.chieco.2007.10.003
Conroy, M. E. (1975). The Concept and Measurement of Regional Industrial Diversification.
Southern Economic Journal, 41(3), 492–505. https://doi.org/10.2307/1056160
Content, J., & Frenken, K. (2016). Related variety and economic development: A literature
review. European Planning Studies, 24(12), 2097–2112.
https://doi.org/10.1080/09654313.2016.1246517
Dasgupta, S., Wang, H., Wang, H., & Mamingi, N. (2002). Incomplete Enforcement of Pollution
Regulation: Bargaining Power of Chinese Factories. The World Bank.
https://doi.org/10.1596/1813-9450-2756

74
Davis, L. W. (2008). The Effect of Driving Restrictions on Air Quality in Mexico City. Journal
of Political Economy, 116(1), 38–81. https://doi.org/10.1086/529398
Dechezleprêtre, A., & Sato, M. (2017). The Impacts of Environmental Regulations
on Competitiveness. Review of Environmental Economics and Policy, 11(2), 183–
206. https://doi.org/10.1093/reep/rex013
Diamond, C. A., & Simon, C. J. (1990). Industrial Specialization and the Returns to Labor.
Journal of Labor Economics, 8(2), 175–201. https://doi.org/10.1086/298219
Dinda, S. (2004). Environmental Kuznets Curve Hypothesis: A Survey. Ecological Economics,
49(4), 431–455. https://doi.org/10.1016/j.ecolecon.2004.02.011
Duflo, E., Greenstone, M., Pande, R., & Ryan, N. (2013). Truth-telling by Third-party Auditors
and the Response of Polluting Firms: Experimental Evidence from India*. The Quarterly
Journal of Economics, 128(4), 1499–1545. https://doi.org/10.1093/qje/qjt024
Duflo, E., Greenstone, M., Pande, R., & Ryan, N. (2018). The Value of Regulatory Discretion:
Estimates From Environmental Inspections in India. Econometrica, 86(6), 2123–2160.
https://doi.org/10.3982/ECTA12876
Engel, U., & Pötschke, M. (1998). Willingness to pay for the environment: Social structure,
value orientations and environmental behaviour in a multilevel perspective. Innovation: The
European Journal of Social Science Research, 11(3), 315–332.
https://doi.org/10.1080/13511610.1998.9968571
Fang, D., Chen, B., Hubacek, K., Ni, R., Chen, L., Feng, K., & Lin, J. (2019). Clean air for
some: Unintended spillover effects of regional air pollution policies. Science Advances.
https://doi.org/10.1126/sciadv.aav4707
Feng, T., Du, H., Lin, Z., & Zuo, J. (2020). Spatial spillover effects of environmental regulations
on air pollution: Evidence from urban agglomerations in China. Journal of Environmental
Management, 272, 110998. https://doi.org/10.1016/j.jenvman.2020.110998
Feng, Y., Lee, C.-C., & Peng, D. (2023). Does regional integration improve economic resilience?
Evidence from urban agglomerations in China. Sustainable Cities and Society, 88, 104273.
https://doi.org/10.1016/j.scs.2022.104273
Fisher-Vanden, K., & Jefferson, G. H. (2008). Technology diversity and development: Evidence
from China’s industrial enterprises. Journal of Comparative Economics, 36(4), 658–672.
https://doi.org/10.1016/j.jce.2008.07.003

75
Freeman, R., Liang, W., Song, R., & Timmins, C. (2019). Willingness to pay for clean air in
China. Journal of Environmental Economics and Management, 94, 188–216.
https://doi.org/10.1016/j.jeem.2019.01.005
Frenken, K., Van Oort, F., & Verburg, T. (2007). Related Variety, Unrelated Variety and Regional
Economic Growth. Regional Studies, 41(5), 685–697.
https://doi.org/10.1080/00343400601120296
Fritsch, M., & Kublina, S. (2018). Related variety, unrelated variety and regional growth: The
role of absorptive capacity and entrepreneurship. Regional Studies, 52(10), 1360–1371.
https://doi.org/10.1080/00343404.2017.1388914
Fullerton, D., & Wolverton, A. (2005). The two-part instrument in a second-best world. Journal
of Public Economics, 89(9), 1961–1975. https://doi.org/10.1016/j.jpubeco.2004.06.011
Genius, M., Hatzaki, E., Kouromichelaki, E. M., Kouvakis, G., Nikiforaki, S., & Tsagarakis, K.
P. (2008). Evaluating Consumers’ Willingness to Pay for Improved Potable Water Quality and
Quantity. Water Resources Management, 22(12), 1825–1834. https://doi.org/10.1007/s11269-
008-9255-7
Goulder, L. H., & Parry, I. W. H. (2008). Instrument Choice in Environmental Policy. Review of
Environmental Economics and Policy, 2(2), 152–174. https://doi.org/10.1093/reep/ren005
Greenstone, M., He, G., Li, S., & Zou, E. Y. (2021). China’s War on Pollution: Evidence from
the First 5 Years. Review of Environmental Economics and Policy, 15(2), 281–299.
https://doi.org/10.1086/715550
Greenstone, M., & Jack, B. K. (2015). Envirodevonomics: A Research Agenda for an Emerging
Field. Journal of Economic Literature, 53(1), 5–42. https://doi.org/10.1257/jel.53.1.5
Grossman, G. M., & Krueger, A. B. (1995). Economic Growth and the Environment*. The
Quarterly Journal of Economics, 110(2), 353–377. https://doi.org/10.2307/2118443
Hallegatte, S. (2014). Economic Resilience: Definition and Measurement (SSRN Scholarly Paper
2432352). https://papers.ssrn.com/abstract=2432352
Hammitt, J. K., & Robinson, L. A. (2011). The Income Elasticity of the Value per Statistical Life:
Transferring Estimates between High and Low Income Populations. Journal of Benefit-Cost
Analysis, 2(1), 1–29. https://doi.org/10.2202/2152-2812.1009
Haq, G., Bailey, P. D., Chadwick, M. J., Forrester, J., Kuylenstierna, J., Leach, G., Villagrasa, D.,
Fergusson, M., Skinner, I., & Oberthur, S. (2001). Determining the costs to industry of

76
environmental regulation. European Environment, 11(3), 125–139.
https://doi.org/10.1002/eet.257
He, G., Lu, Y., Mol, A. P. J., & Beckers, T. (2012). Changes and challenges: China’s
environmental management in transition. Environmental Development, 3, 25–38.
https://doi.org/10.1016/j.envdev.2012.05.005
He, G., Wang, S., & Zhang, B. (2020). Watering Down Environmental Regulation in China*. The
Quarterly Journal of Economics, 135(4), 2135–2185. https://doi.org/10.1093/qje/qjaa024
Israel, D., & Levinson, A. (2004). Willingness to Pay for Environmental Quality: Testable
Empirical Implications of the Growth and Environment Literature. Contributions in Economic
Analysis & Policy, 3(1), 1–29. https://doi.org/10.2202/1538-0645.1254
Ito, K., & Zhang, S. (2020). Willingness to Pay for Clean Air: Evidence from Air Purifier
Markets in China. Journal of Political Economy, 128(5), 1627–1672.
https://doi.org/10.1086/705554
Izraeli, O., & Murphy, K. J. (2003). The effect of industrial diversity on state unemployment rate
and per capita income. The Annals of Regional Science, 37(1), 1–14.
https://doi.org/10.1007/s001680200100
Jacobs, J. (1969). The Economy of Cities. Random House.
Jalil, A., & Feridun, M. (2011). The impact of growth, energy and financial development on the
environment in China: A cointegration analysis. Energy Economics, 33(2), 284–291.
https://doi.org/10.1016/j.eneco.2010.10.003
Jin, Y., Andersson, H., & Zhang, S. (2016). Air Pollution Control Policies in China: A
Retrospective and Prospects. International Journal of Environmental Research and Public
Health, 13(12), Article 12. https://doi.org/10.3390/ijerph13121219
Karamichas, J. (2012). The Olympics and the Environment. In H. J. Lenskyj & S. Wagg (Eds.),
The Palgrave Handbook of Olympic Studies (pp. 381–393). Palgrave Macmillan UK.
https://doi.org/10.1057/9780230367463_24
Kemeny, T., & Storper, M. (2015). Is Specialization Good for Regional Economic Development?
Regional Studies, 49(6), 1003–1018. https://doi.org/10.1080/00343404.2014.899691
Kim, H., Hwang, S.-J., & Yoon, W. (2023). Industry cluster, organizational diversity, and
innovation. International Journal of Innovation Studies, 7(3), 187–195.
https://doi.org/10.1016/j.ijis.2023.03.002

77
Kim, J., Kang, J. H., & Kim, Y.-K. (2014). Impact of Mega Sport Events on Destination Image
and Country Image. Sport Marketing Quarterly, 23(3), 161–175.
Lanoie, P., Patry, M., & Lajeunesse, R. (2008). Environmental Regulation and Productivity:
Testing the Porter Hypothesis. Journal of Productivity Analysis, 30, 121–128.
https://doi.org/10.1007/s11123-008-0108-4
Lehmann, P. (2010). Combining emissions trading and emissions taxes in a multi-objective world
(Working Paper 4/2010). UFZ Discussion Paper. https://www.econstor.eu/handle/10419/44751
Levinson, A., & Taylor, M. S. (2008). Unmasking the Pollution Haven Effect. International
Economic Review, 49(1), 223–254.
Li, B., Wang, F., Yin, H., & Li, X. (2019). Mega events and urban air quality improvement: A
temporary show? Journal of Cleaner Production, 217, 116–126.
https://doi.org/10.1016/j.jclepro.2019.01.116
Li, J., & Huang, J. (2020). The expansion of China’s solar energy: Challenges and policy
options. Renewable and Sustainable Energy Reviews, 132, 110002.
https://doi.org/10.1016/j.rser.2020.110002
Li, P., Lu, Y., & Wang, J. (2020). The effects of fuel standards on air pollution: Evidence from
China. Journal of Development Economics, 146, 102488.
https://doi.org/10.1016/j.jdeveco.2020.102488
Liu, J., & Raven, P. H. (2010). China’s Environmental Challenges and Implications for the
World. Critical Reviews in Environmental Science and Technology, 40(9–10), 823–851.
https://doi.org/10.1080/10643389.2010.502645
Mahmood, H., Furqan, M., Hassan, M. S., & Rej, S. (2023). The Environmental Kuznets Curve
(EKC) Hypothesis in China: A Review. Sustainability, 15(7), Article 7.
https://doi.org/10.3390/su15076110
McConnell, K. E. (1997). Income and the demand for environmental quality. Environment and
Development Economics, 2(4), 383–399. https://doi.org/10.1017/S1355770X9700020X
Mizuno, K., Mizutani, F., & Nakayama, N. (2006). Industrial diversity and metropolitan
unemployment rate. The Annals of Regional Science, 40(1), 157–172.
https://doi.org/10.1007/s00168-005-0046-2
Null, C., Kremer, M., Miguel, E., Garcia Hombrados, J., Meeks, R., & Peterson Zwane, A.
(2012). Willingness to pay for cleaner water in less developed countries: Systematic review of

78
experimental evidence (2012th ed.). International Initiative for Impact Evaluation (3ie).
https://doi.org/10.23846/SR1003
Oliva, P. (2015). Environmental Regulations and Corruption: Automobile Emissions in Mexico
City. Journal of Political Economy.
Parry, I. W. H., & Bento, A. M. (2000). Tax Deductions, Environmental Policy, and the “Double
Dividend” Hypothesis. Journal of Environmental Economics and Management, 39(1), 67–96.
https://doi.org/10.1006/jeem.1999.1093
Peng, B., Sheng, X., & Wei, G. (2020). Does environmental protection promote economic
development? From the perspective of coupling coordination between environmental protection
and economic development. Environmental Science and Pollution Research, 27(31), 39135–
39148. https://doi.org/10.1007/s11356-020-09871-1
Rivera, N. M. (2021). Air quality warnings and temporary driving bans: Evidence from air
pollution, car trips, and mass-transit ridership in Santiago. Journal of Environmental Economics
and Management, 108, 102454. https://doi.org/10.1016/j.jeem.2021.102454
Rogge, K. S., & Reichardt, K. (2016). Policy mixes for sustainability transitions: An extended
concept and framework for analysis. Research Policy, 45(8), 1620–1635.
https://doi.org/10.1016/j.respol.2016.04.004
Ross, W. J., & Leopkey, B. (2017). The adoption and evolution of environmental practices in the
Olympic Games. Managing Sport and Leisure, 22(1), 1–18.
https://doi.org/10.1080/23750472.2017.1326291
Simon, C. J. (1988). Frictional Unemployment and the Role of Industrial Diversity*. The
Quarterly Journal of Economics, 103(4), 715–728. https://doi.org/10.2307/1886071
Simon, C. J., & Nardinelli, C. (1992). Does Industrial Diversity Always Reduce Unemployment?
Evidence from the Great Depression and After. Economic Inquiry, 30(2).
https://www.proquest.com/docview/1297323756/citation/E552FE6C9E5C4938PQ/1
Stern, D. I. (2004). The Rise and Fall of the Environmental Kuznets Curve. World Development,
32(8), 1419–1439. https://doi.org/10.1016/j.worlddev.2004.03.004
Tan, J., Hu, X., Hassink, R., & Ni, J. (2020). Industrial structure or agency: What affects regional
economic resilience? Evidence from resource-based cities in China. Cities, 106, 102906.
https://doi.org/10.1016/j.cities.2020.102906
Tan, R., & Lin, B. (2020). Are people willing to support the construction of charging facilities in
China? Energy Policy, 143, 111604. https://doi.org/10.1016/j.enpol.2020.111604

79
Teets, J. C., & Hasmath, R. (2020). The evolution of policy experimentation in China. Journal of
Asian Public Policy, 13(1), 49–59. https://doi.org/10.1080/17516234.2020.1711491
Tian, Z., Tian, Y., Chen, Y., & Shao, S. (2020). The economic consequences of environmental
regulation in China: From a perspective of the environmental protection admonishing talk policy.
Business Strategy and the Environment, 29(4), 1723–1733. https://doi.org/10.1002/bse.2464
Tie, X., & Cao, J. (2009). Aerosol pollution in China: Present and future impact on environment.
Particuology, 7(6), 426–431. https://doi.org/10.1016/j.partic.2009.09.003
van Oort, F., & Dogaru, S. de G. and T. (2016). Related Variety and Regional Economic Growth
in a Cross-Section of European Urban Regions. In Second Rank Cities in Europe. Routledge.
Walker, W. R. (2011). Environmental Regulation and Labor Reallocation: Evidence from the
Clean Air Act. American Economic Review, 101(3), 442–447.
https://doi.org/10.1257/aer.101.3.442
Wang, H., Xie, J., & Li, H. (2010). Water pricing with household surveys: A study of
acceptability and willingness to pay in Chongqing, China. China Economic Review, 21(1), 136–
149. https://doi.org/10.1016/j.chieco.2009.12.001
Watson, P., & Deller, S. (2017). Economic diversity, unemployment and the Great Recession.
The Quarterly Review of Economics and Finance, 64, 1–11.
https://doi.org/10.1016/j.qref.2016.12.003
Xu, Y. (2021). Environmental Policy and Air Pollution in China; Governance and Strategy.
Routledge.
Zheng, S., Kahn, M. E., Sun, W., & Luo, D. (2014). Incentives for China’s urban mayors to
mitigate pollution externalities: The role of the central government and public environmentalism.
Regional Science and Urban Economics, 47, 61–71.
https://doi.org/10.1016/j.regsciurbeco.2013.09.003

80
APPENDICES

81
Appendix A – For Chapter II
Contents:
1. Policy coverage area and baseline pollution
Table S1 List of provinces and cities affected by the policy
Figure S1 Overview of policy regulated area and baseline air quality level
2. Industrial contribution to pollution
Table S2 Share of industrial contribution in pollution inventories
3. Welfare calculation
Table S3 Estimation of welfare impacts of the policy
4. Industry inefficiency mechanism
5. Policy implementation
a. Time frame
b. Local implementation details
6. Data
a. Data sources
Table S4 List of datasets and sources
Table S5 Number of reporting stations, cities and observations across sample period
Figure S2 Distribution of pollution monitoring stations
b. Data compilation
Table S6 Description of variables
Table S7 Summary statistics for various pollutants in the Before, During and After periods
of the policy

82
Table S8 Summary statistics for manufacturing plants characteristic aggregation at
monitoring stations
7. AQI calculation
Table S9 Reference breaks for pollutant-specific concentration levels
8. Primary results in tabular format
Table S10 Main analysis: effects of the policy
Table S11 Supporting analysis: effects of the policy across different hours periods
Table S12 Supporting analysis: effects of the policy by different monitor quartiles of plant
density
9. Robustness checks
a. Alternative control group band for the monitors
Table S13 Effects of the policy with different control group bands
Figure S3 Areas & Monitoring Stations within 200km from APEC Pollution Regulation
Boundaries
b. One-year-before placebo test
Table S14 One-year-before placebo for the main analysis
c. Relaxed assumptions for trend and local heterogeneity
Table S15 Trend and local heterogeneity assumptions relaxation
d. Event study analysis as alternative approach
Figure S4 Event study result over the policy period for the effect of the policy
e. Local specific results
Figure S5 Estimated effects of the policy in different affected provinces
Figure S6 Estimated effects of the policy in different affected cities
f. Max daily reading as dependent variable instead of daily mean
Table S16 Effects of the policy on daily max pollution readings
g. Visibility Index and fire points as alternative dependent variable
Table S17 Effects of policy on Visibility Index
Table S18 Effects of policy on straw burning

83
h. CEM emission data as alternative policy result indicators
Table S19 Effects of the policy on emissions from firms under Continuous Emission
Monitoring (CEM) system
Figure S7 Effect of the policy on Continuous emission monitoring (CEM) firms’ emission
Table S20 Effects of policy on the emission of State-Owned and non-State-Owned CEM
firms.
Table S21 Effects of policy on the emission of CEM firms by different industry categories
f. Allowing city-specific daily pollution patterns for the hour-period analysis
Table S22 Effects of the policy by 6-hour periods, allowing city-specific daily patterns of
pollution
g. Monitor quartiles based on area total sales instead of plant density
Figure S8 Effects of policy by different manufacturing total sales levels within 20 miles
from monitors
h. Internal consistency of the hour-period and density quartile analyses
Table S23 Effects of policy by combinations of different 6-hour-periods and different plant
density quartiles
Figure S9 Estimated effect of the policy by plant density quartile and hour period
combinations

84
1. Policy coverage area and baseline pollution
The ban centers Beijing, the venue for the 2014 APEC event. Because of the regional
channels of air pollutant transmission, Beijing, Tianjin, and 39 other cities in 5 provinces are
affected by the policy to different degrees. Table S1.1 lists the covered provinces and cities by the
policy.
The regional air pollutant transmission is a form of spillover effect. The pollution levels of
these cities are found to be highly correlated with that in Beijing. The specific criteria for the
selection process are not publicly announced. In general, the selection of cities is based on
proximity to Beijing, their size, and their respective industrial sectors. Province capitals of the five
provinces are all included. They are generally the biggest cities in respective provinces in terms of
population and economy. They are also naturally most responsive to central government orders.
Many of these cities are originally in relatively high levels of pollution. Some of them are
at brinks of critical levels (AQI 100 for “moderate” and 150 for “heavy pollution”) by air quality
standard. Figure S1.1 shows the covered area by the ban, and the baseline pollution levels in the
respective cities.

85
Table S1 List of provinces and cities affected by the policy
Provinces Cities Provinces Cities
Beijing 北京 Beijing 北京 Shanxi 山西 Taiyuan 太原
Tianjin 天津 Tianjin 天津 Datong 大同
Hebei 河北 Shijiazhuang 石家庄 Shuozhou 朔州
Chengde 承德 Xinzhou 忻州
Baoding 保定 Yangquan 阳泉
Langfang 廊坊 Jinzhong 晋中
Zhangjiakou 张家口 Lvliang 吕梁
Xingtai 邢台 Changzhi 长治
Handan 邯郸 Jincheng 晋城
Dingzhou 定州 Yuncheng 运城
Xinji 辛集 Linfen 临汾
Hengshui 衡水 Inner Mongolia 内蒙古 Hohhot 呼和浩特
Cangzhou 沧州 Baotou 包头
Tangshan 唐山 Chifeng 赤峰
Shandong 山东 Jinan 济南 Ulanqab 乌兰察布
Zibo 淄博 Xilingol 锡林郭勒
Dongying 东营 Henan 河南 Zhengzhou 郑州
Dezhou 德州 Xinxiang 新乡
Liaocheng 聊城 Jiaozuo 焦作
Binzhou 滨州 Hebi 鹤壁
Anyang 安阳

86
Figure S1.1 Overview of policy regulated area and baseline air quality level
(A) Location of regulated area within China. (B) Average AQI levels of prefecture-level municipalities
within regulated area over 2013-2018. (C) Average AQI of selected cities over October and November of
2013. The yellow and red dashed lines respectively indicate “medium pollution” and “heavy pollution”
defined by China‘s ambient air quality standards.
From Panel C we can see that the baseline levels of pollution in the regulated cities were close to the critical
thresholds to begin with. Any increase in pollution level at this point would have more significant public
health consequences than if it was upon a lower baseline level.

87
2. Industrial contribution to pollution
The industrials sector is a major contributor to the pollution levels in China, as well as the
regulated area (located in northern China). At the national level, Zheng et al. (2018) estimated in
a bottom-up approach that 57.4% of anthropogenic PM 10 and 50.5% of PM 2.5 emission
inventories are attributed to the industrial sector (not including power generation). In 2014, the
Municipal Government of Beijing announced a study of the sources of PM 2.5 pollution in
Beijing.5 Out of the local sources, 18.1% of PM 2.5 pollution come from industrial production (not
including production related transportation and coal burning). This ratio can be expected to be
even higher from the part of PM 2.5 from peripheral transmission, given the larger sizes of
industrial sectors and looser regulation. In a more specific scope, Duan et al. (2018) estimated that
the iron and steel industry, by itself, on annual average contributes 12.7% of total PM 2.5
concentration in the Beijing-Tianjin-Hebei region, the core region of the regulated area. Table S1.2
summarizes the above observations.
These observations at different levels all point to the significance of the industrial sector to
air pollution in the ban-covered area. Given a major part of the ban is cast on the industrial sector
especially the relatively more polluting industries, the large share of the industrial sector’s
contribution to the pollution level in the regulated area validates our theories for the mechanism
of the policy’s impacts.
5 Beijing Government, available at: http://www.gov.cn/xinwen/2014-04/16/content_2660844.htm. (Archived Feb
2022)

88
Table S1.2 Share of industrial contribution in pollution inventories
3. Welfare calculation
As is mentioned previously in this appendix, the public health welfare calculation for the
impact of this policy is based on the cumulative effects of the policy as added pollution exposure.
𝑁𝑒𝑡 𝑐ℎ𝑎𝑛𝑔𝑒 𝑖𝑛 𝑝𝑜𝑙𝑙𝑢𝑡𝑖𝑜𝑛 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒
= 𝐸𝑓𝑓𝑒𝑐𝑡 𝑖𝑛 ∆ 𝑎𝑖𝑟 𝑝𝑜𝑙𝑙𝑢𝑡𝑖𝑜𝑛 × 𝑇𝑖𝑚𝑒 𝑙𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑒𝑓𝑓𝑒𝑐𝑡
Under this setting, we are able to calculate the cumulative effect of PM 10 changes by
adding up the During policy period impact and the Post period impact, and derive 400.32, with the
unit (ug/m3 * day). With this cumulative effect in net change of pollution exposure, as well as the
local population statistics, we can conduct two back-of-envelop calculation for the welfare loss.
The calculation is shown in Table S1.3.
We use Ito and Zhang’s (2020) estimate of the household annual willingness to pay for PM
10 reduction, $1.34 per household per ug/m3 of PM 10. By assuming this number is evenly downscalable to the time scope of our policy period, we are able to obtain the daily average willingness
to pay. Total population in the 41 regulated cities is 203.9 million.6 The average household size is
2.97 in China in 2014.7 We therefore assume 68.7 million households in the regulated area were
6 From https://www.citypopulation.de/en/china/admin/.
7 From https://www.statista.com/statistics/278697/average-size-of-households-in-china/.
Inventory
National PM10 National PM 2.5 Beijing PM2.5 BTH PM2.5
Pollution source (1) (2) (3) (4)
Industrial production 57.4% (8.1 Tg) 50.5% (5.2 Tg) 18.1%
Iron and steel 12.70%

89
affected by the policy through changes in air pollution levels. We multiply the cumulative effect
of the policy on PM 10 exposure, by the household daily willingness to pay for PM 10 reduction,
and then by the total number of households in the regulated area, to derive $100.90 million.
It is to be noted that Ito and Zhang (2020) findings are only for a regional and single market
of air purifier filters, also they only examine PM10 without controlling for the other pollutants or
considering the correlation. We then use the He et al (2016) finding on the mortality impact of PM
10 improvement, 8.36% improvement on all-cause monthly mortality rate by 10 ug/m3 PM 10
reduction over a month. With the baseline average monthly standardized mortality rate 4.12 per
10,000 people (He et al., 2016), the cumulative additional PM 10 exposure, and the total population
in the regulated area, we derive 9,069 premature deaths could be potentially caused by the policy.
We acknowledge these calculations provide only ballpark figures that offer an image of the
overall welfare impact of the policy through increased pollution exposure overall.

90
Table S1.3 Estimation of welfare impacts of the policy
Welfare calculation
WTP pollution reduction
All-cause
premature death
Policy effect in pollution
Average impact on PM10 in During -9.116 ug/m3
Length of During 21 days
Average impact on PM10 in Post 26.898 ug/m3
Length of Post 22 days
Cumulated net effect 400.32 (ug/m3 * day)
Unit pollution welfare impact
Household annual WTP for PM10 reduction $1.34/(ug/m3)
Household daily WTP for PM10 reduction $0.003671/(ug/m3)
All-cause monthly mortality rate baseline 4.12/10000
Mort. redc. by 10 ug/m3 PM10 reduction 8.36%
Unit impact over policy period $1.47 0.00445% lives
Total population in the regulated area 68.7 M households 203.9 M
Welfare impact $100.90 million 9069 lives

91
4. Industry inefficiency mechanism
As is noted in the main text, the larger pollution rebound than the initial pollution reduction
points to more emission associated with unit production in the Post period production recoup. The
intensification of industrial production and ramping-up of the production rate can undermine the
efficiency in terms of energy intensity and emission intensity. Which means the energy
consumption of unit production may increase as the production rate increase, and similarly, the
emission associated with unit production is likely to increase as the production rate increase. In
general, this is because the original production rate tends to be efficient in terms of energy and
emission intensity by nature of production optimization.
To provide a specific example of potential passage to decrease emission efficiency, we
investigate the details of the iron and steel industry, which we have identified as a major pollution
contributor.8
In the most polluting processes of sintering and ironmaking, dedusting exhaust hoods
are required to be installed. The pollution control of these processes is highly dependent on such
dedusting device.9 The efficiency of these hoods (and other types of dedusting filters in-effect) is
related to the temperature of the emission gas flow. These hoods are electrically powered and work
at fixed level of air flow10, and cannot adjust to increase the air flow or efficiency to cope with the
increased emission. The processes also rely on cooling devices that maintain the working
8 We refer to the MEP published “Emission standard of air pollutants for iron smelt industry”, available at:
https://www.mee.gov.cn/ywgz/fgbz/bz/bzwb/dqhjbh/dqgdwrywrwpfbz/201207/t20120731_234141.shtml. (Archived
Feb 2022)
9
See MEP published “Opinions on moving forward with implementing ultra-low emission for iron and steel
industry”, available at: https://www.mee.gov.cn/xxgk2018/xxgk/xxgk03/201904/t20190429_701463.html.
(Archived Feb 2022)
10 See http://www.wxvcee.com/cn/case2.html. (Archived Feb 2022)

92
conditions. With increased production intensity, heat can build up faster, and the cooling devices
may not catch up. The changed working conditions, especially at the ironmaking blast furnace
outlets, will not be optimal as they should be normally for the dedusting devices, and the
performance of reducing emitted pollutants will be compromised. Relatedly, dedusting and
filtering devices are installed in many other stages of production where materials are exposed to
open air or in transport. These devices normally operate at or under their respective optimal rate,
but their functionality of pollutant removal rate can be undermined by intensified production and
increased material flows.
This applies also to the urban constructions which often use water spraying as a simple but
effective PM mitigation device. The water spraying, automatic or manual, often take regular time
intervals by protocol, but as the construction speed ramps up the regular water spraying program
would not be as effective. Industries and production processes that uses boilers can also be affected
by production intensification in a similar way, as boilers typically have labeled optimal rate of
power. When the power passes the optimal rate, the efficiency is undermined in terms of energy
intensity. As less time is taken for unit production, more energy input, thus emission, is required
for unit production.

93
5. Policy implementation details
a. Policy time frame
The 2014 APEC event starts on Nov. 5, 2014.11 To prepare for the event, part of
the mitigation efforts started shortly before the event. To include the mitigation impact of
these efforts, we include two weeks prior to the event into the During period of the policy
for our analysis. The event lasted until Nov. 11, 2014, the day after which most of the ban
policy is alleviated. In the practice of policy implementation, the local governments applied
different starting dates and spans for the ban. For example, in Hebei, the cities of
Shijiazhuang, Handan, Xingtai, Baoding, and Langfang, started the ban on Nov. 3, while
Tangshan, Cangzhou, and Hengshui started on Nov. 6.12
We do not separately consider these differences because, as is mentioned
previously, the selection of the regulated cities is based on area air pollution transmission.
Even when Shandong hasn’t started mitigation, Beijing’s mitigation which has already
started would still have an effect on Shandong’s air quality. Also, we try to estimate the
average effect of the policy over the entire policy span for the whole policy area. Therefore,
as long as we include the entire time span during which the policy drives the pollution level
to deviate from the normal level, and that we clearly define and distinguish the During and
Post policy periods, the policy time frame and the model for estimation would fit our
purpose. This also backs up our approach defining the During and Post policy periods
basing on the controlled residual trends.
11 Available at: https://www.fmprc.gov.cn/ce/cohk/chn/xwdt/zt/apec/t1196267.htm. (Archived Feb 2022)
12 Available at: http://www.gov.cn/xinwen/2014-11/04/content_2774841.htm. (Archived Feb 2022)

94
b. Local implementation details
The ban policy of our interest is a comprehensive mitigation policy that covers a
large area with mitigation efforts in various directions. The comprehensive mitigation
policy includes many local policies that bans the operation of the most polluting industries
and all urban construction sites in major cities, restricts the operation of power plants and
some essential industries, and restricts driving by license plate numbers. In practice, the
policy implementation in different cities is very different. This includes the different time
spans of local mitigation mentioned above. Moreover, the levels of strictness, and
mitigation approaches vary.
We define the different local policies comprehensively as a ban because of the
strictness of government enforcement as a whole, and the crisp, significant impact in
industry operation, which is associated with the intensive recouping after the policy. At the
local level, if we define local policy implementation with longer time span, stricter
enforcement, and more means of mitigation as higher level of implementation, in general,
Beijing, Tianjin, and the provincial capitals implemented the policy at higher levels than
the rest, and cities closer to Beijing implemented at higher levels. For example, in Hebei,
the province that surrounds Beijing, all air-polluting companies within the 100 km buffer
from Beijing were banned from production, whereas some air-polluting companies in the
100-200 km distance ring reduced production but was not entirely banned. 13 For the
regulated cities in Shandong Province and Inner Mongolia, which are farther from Beijing,
construction halts and driving restriction were not implemented at all. This also applies to
the different local time frame of the ban. The farther, the shorter.
13 Available at: http://news.sohu.com/20141031/n405637662.shtml. (Archived Feb 2022)

95
We do not treat the variation of local implementation levels differently, as we are
more interested in the overall impacts of the comprehensive ban. We acknowledge the
distribution of population may hinder the accurateness of our welfare calculation,
especially as we take the average effects approach. However, to specify all the local
implementation details is not feasible in this project especially as many details are not
available as public information.
As is mentioned, the comprehensive ban also includes sectors and factor that do not
contribute to the Post ban period pollution rebound as we have suggested, such as the
driving restrictions. In the industrial sector, too, while the policy temporarily banned a
significant number of polluting facilities, many factories were closed permanently as a part
of the policy and may not recoup after the ban. These government efforts contribute only
or mainly in the pollution reduction but not in the rebound. This means that the mechanism
we describe for the ban’s impact on the industrial sector does not cover and explain the
whole of our observation in the ban’s impact on pollution levels. But the existence of these
factors also suggests that the pollution reduction impact of the temporary ban on the
industries may be smaller than the total observed reduction, while the pollution-increasing
rebound fully reflect the inefficient recouping of the industries, which suggests even greater
ratio (Inefficiency Factor) for the actual rebound effect and worse inefficiency for the ban’s
impact on industrial production.

96
6. Data
a. Data sources
Table S1.4 shows all the datasets, their sources, and public accessibility. This list
also includes data used only for the robustness checks. The monitor level hourly
pollution data provides for the pollution indicator dependent variables for most of the
analyses. It offers hourly monitor readings of AQI, and concentrations of PM 10, PM
2.5, SO2, NO2, O3, CO. It also offers 8-hour and 24-hour averages of some of these
pollutants that are included in the calculation of AQI.
A total of 1497 air pollution monitoring stations deployed by China’s Ministry of
Environmental Protection (MEP, now Ministry of Ecology and Environment) are
spread across the country, taking hourly readings since 2013. The 1497 monitors are
not installed at once, but in multiple phases through 2013-2015. The monitor
availability and coverage are reflected in Table S1.5. Figure S1.2 shows their
geographic distribution. The recordings from the monitoring stations are published
routinely through China National Environmental Monitoring Headquarters website14
.
Other datasets provide control variables for the main analysis. The daily level
weather data provides the weather control variables, including temperature,
precipitation, wind direction, and wind speed. This data come from National Oceanic
and Atmospheric Administration (NOAA, of the US) Integrated Surface Database
(ISD)15. The database integrates station level hourly reading from various sources over
14 Available at: https://air.cnemc.cn:18007/. The historical data is not officially provided, but is publicly shared at:
https://quotsoft.net/air/. (Retrieved Oct 2019)
15 Available at: https://www.ncei.noaa.gov/products/land-based-station/integrated-surface-database.

97
the globe. It also provides, and we use, the daily averages of all the weather elements.
A total of 407 weather stations are spread across China. Note that for the robustness
check with visibility index as the alternative dependent variable, the visibility index
also come from the weather dataset.
To control for the impact of the centralized winter heating policy, we collect from
popular media, local government documents, and local information websites for the
policy dates. This information is not readily available in a summarized form, and in
many cases even the start and end dates can be announced separately. As is noted under
Table S1.4, the local information platform website16 provides us a useful source for
these scattered pieces of information.
To compile the datasets, we use the spatial shapefile data from United Nations
Office for the Coordination of Humanitarian Affairs (UN OCHA) Humanitarian Data
Exchange (HDX) platform that provides the first (provincial) and second (prefecture level cities)
level administrative boundaries17
. The main analysis datasets are used for most of our
supporting analyses and robustness checks. There are also unique datasets used
exclusively in the supporting analyses and robustness checks.
China’s Annual Survey of Industrial Enterprises ( ASIE) 2008 provides us the
locations of manufacturing plants, their industry category, sales, assets, and labor size.
This data is used mainly in the plant density quartiles supporting analysis. This data
comes from a nationwide annually census enforced by the National Bureau of Statistics
(NBS) of China. It includes firms in mining, manufacturing, and utility industries with
16 See http://www.bendibao.com/index.htm
17 Available at: https://data.humdata.org/dataset/cod-ab-chn.

98
annual sales of 5 million RMB or above. The firms included in this census accounts for
about 95% of total industrial production and 98% of industrial exports of China. NBS
publishes the aggregated data in the official China Statistical Yearbook18
. We were able
to obtain the dataset from the 2008 industrial census. This dataset is available by
purchase through Beijing Fu’ao HuaMei Information Consulting19
.
We also use the dataset from China’s Continuous Emission Monitoring systems
(CEMS), which provides the hourly stack gas emission concentration data at the sensor
level. It includes attributes of TSP (total suspended particles, roughly equals to
particulate matter), SO2, and NOx. In the corresponding robustness check, we use these
attributes as alternative dependent variables. This data is collected from the individual
firms by MEP through the Environment Supervision Center platform20. The dataset
includes 3214 firms and 6080 sensors (monitored smokestacks) with observations from
2014-2017. Since sensors are installed in smokestacks, multiple sensors may be
associated with a single firm.
Lastly, we use Moderate Resolution Imaging Spectroradiometer (MODIS) data of
National Aeronautics and Space Administration (NASA, of US) for active fire points21
and their location and time for the corresponding robustness check. This is a raster type
spatial data that reflect abnormal heat to detect active fire.
18 Available at: http://www.stats.gov.cn/tjsj/ndsj/2008/indexch.htm.
19 See http://www.allmyinfo.com/English/#sect1.
20 See http://www.envsc.cn/.
21 Available at: https://modis-fire.umd.edu/

99
Table S1.4 List of datasets and sources
Table S1.5 Number of reporting stations, cities and observations across sample period
Year No. of stations No. of cities No. of observations
2013 802 158 172597
2014 1361 279 331346
2015 1497 330 544908
2016 1497 330 547902
2017 1497 330 546405
Dataset Source platform Accessibility

Main analysis datasets
Monitor level hourly pollution MEP National Environmental Monitoring
Headquarters platform
Available publicly, historical data
publicly shared by private individual
Station level daily weather NOAA Integrated Surface Database Available publicly
Winter heating plans Multiple online sources Available publicly
Province and cities boundaries UN OCHA Humanitarian Data Exchange
platform Available publicly
Supporting analysis and robustness
check datasets
Factory level manufacturing stats Census of Manufacturing 2008 Unavailable publicly
Factory level emission China CEM systems Unavailable publicly
Fire point NASA MODIS active fire Publicly available

100
Figure S1.2 Distribution of pollution monitoring stations
We made each point of the pollution monitoring stations and weather stations opaque, so the relatively
darker locations are where the monitors or stations are more clustered. While the weather stations are
relatively evenly spreaded, the pollution monitors are not evenly or randomly distributed across the country,
but more clustered in populous areas. These areas tend to be located in urban areas. This supports our scope
and granularity of study, suggesting our findings in terms of policy effects in pollution, though not evenly
applied spatially, reflect close to what is experienced by the general population.

101
b. Data preparation
In preparation for the analyses, we trim the MEP air pollution data to January 2013
- July 2015 as our sample period. This period is relatively closer to the policy period, and
provide sufficient observations for time-invariant monitor fixed effect estimates, which is
a key element of our control strategy. We aggregate the hourly data to daily level by the
mean readings (by daily max for the corresponding robustness check) of AQI, PM 10, PM
2.5, SO2, and NO2 for the dependent variables of analyses22. We also aggregated the
hourly data to the granularity of 6-hour-periods for the hour period supporting analysis.
We drop observations of specific pollutant from monitors with over 25% missing rate out
of total observations for this pollutant.
With the coordinates of the monitors, together with the HDX administrative
boundary spatial data, we use ArcGIS software to determine the location of the monitors
in terms of city and province, and thus whether they belong to the policy area. This
information is indicated as dummies in our dataset. With the dates of the observations, we
also assign them to outside of policy period, or Before (day -28 to -15), During (day -14 to
+6), and Post (day +7 to +28) periods relative to the 2014 APEC start date (Nov 5, 2014,
day 0). These periods are also indicated as dummies in our dataset. The Before period
provides reference for dependent variable levels at the unaffected state, but is not used in
the analysis. Table S1.6 shows the summary statistics of the dependent variables in
different policy periods.
22 We discuss this selection of pollutants in Section 7 AQI calculation.

102
To merge with the weather data, each weather station is matched with the closest
pollution monitoring station based on their coordinates. This is also performed through
ArcGIS. We use temperature, windspeed, and wind direction as control variables. We also
add the squared values of temperature and windspeed. We also merge this data with the
aggregated summary of the respective start and end dates of the heating season in all cities
with central heating in this sample period. Whether the observations fall in the heating
season is indicated by a dummy. Table S1.7 lists the dependent variables (different
pollution measures) and the control variables. 23 The quadratic terms are omitted for
simplicity. The treatment variables are dummy variables that simply reflect dates relative
to the policy periods, and are well explained in the Methodology section in the main text,
thus are not included in this table.
For the plant density quartile analysis, we use the AEIS data and aggregate the data
of the manufacturing plants within 20 miles radius of each MEP pollution monitor. We do
so by first identifying the plants within each of the pollution monitors by the coordinates
on ArcGIS, and then aggregate to each pollution monitors the counts of the plants within
their radius, the total sales of these plants, the total assets, and the sum of their labor size.
We also calculate the percentages of industrial categories for the plants within the radius
of each monitor. We break the monitors into quartiles by their associated counts of plants
within radius, which is equivalent to the density given unified area. The summary stats of
the aggregated attributes are presented in Table S1.8 by different plant density quartiles.
23 The winter-heating period is an annual policy that significantly affect the air quality in northern China. Although
it does not overlap with the APEC policy periods, including the dummy helps us build the counterfactual.

103
For the fire point robustness check, we aggregate the MODIS active fire point raster
data to the counts of fire points by the dates and weather stations, in tabular format through
ArcGIS. We trim the analysis span to 2013 – 2016.

104
Table S1.6 Summary statistics for various pollutants in the Before, During and After periods
of the policy
Treated Control
All Bef Dur Post All Bef Dur Post
Panel A. All stations1
AQI 115.44 136.7 120.02 145.85 82.35 95.75 92.92 94.57
(71.23) (83.61) (67.78) (89.81) (52.26) (47.88) (55.76) (50.44)
NO2 45.24 53.24 54.03 60.53 33.53 38.68 41.09 42.34
(25.94) (26.56) (25.45) (32.32) (21.30) (21.55) (21.85) (24.11)
PM2.5 78.33 95.55 82.37 104.33 55.00 66.10 65.78 65.57
(64.46) (75.49) (58.05) (84.08) (45.53) (42.13) (55.87) (43.78)
PM10 139.57 168.48 147.65 175.98 91.77 109.93 105.59 105.61
(92.54) (100.24) (86.66) (110.56) (69.53) (60.24) (74.37) (61.39)
SO2 50.81 40.54 48.73 73.58 26.81 25.42 29.26 37.15
(50.21) (33.05) (43.32) (58.32) (30.01) (19.76) (27.11) (36.66)
No. of Stations 180 1317
Panel B. Less than 200km from regulation boundaries2
AQI 115.44 136.7 120.02 145.85 100.75 99.59 107.29 119.54
(71.23) (83.61) (67.78) (89.81) (60.45) (55.77) (54.24) (66.18)
NO2 45.24 53.24 54.03 60.53 37.92 41.25 46.85 49.82
(25.94) (26.56) (25.45) (32.32) (21.88) (20.33) (20.70) (25.70)
PM2.5 78.33 95.55 82.37 104.33 68.00 66.74 73.46 85.26
(64.46) (75.49) (58.05) (84.08) (55.50) (51.24) (47.36) (59.93)
PM10 139.57 168.48 147.65 175.98 118.87 118.97 127.16 138.28
(92.54) (100.24) (86.66) (110.56) (79.31) (72.94) (68.80) (80.98)
SO2 50.81 40.54 48.73 73.58 42.67 33.58 42.62 63.00
(50.21) (33.05) (43.32) (58.32) (41.25) (27.88) (30.57) (46.80)
No. of Stations 180 200
Notes:
Standard errors are reported in the parenthesis.
1. In Panel A, we examine all the monitoring stations in our sample.
2. In Panel B, we restrict the analysis to stations less than 200km from the regulation boundaries, which is associated with the
robustness check of alternative control group range.

105
Table S1.7 Description of variables
Variable Description Source
Pollutants
Monitoring Station
AQI Avg daily hourly measure of Air Quality Index in ug/m3 at monitoring
station i on day t
MEP
PM10 Avg daily hourly measure of particulate matter 10 in ug/m3 at monitoring
station i on day t
MEP
PM2.5 Avg daily hourly measure of particulate matter 2.5 in ug/m3 at
monitoring station i on day t
MEP
SO2 Avg daily hourly measure of sulphur dioxide in ug/m3 at monitoring
station i on day t
MEP
NO2 Avg daily hourly measure of nitrogen dioxide in ug/m3 at monitoring
station i on day t
MEP
Weather
temperature Avg daily temperature from nearest NOAA station NOAA ISD
rainfall Avg precipitation from nearest NOAA station NOAA ISD
wind speed Avg wind speed from nearest NOAA station NOAA ISD
wind direction Binary variable equals to 1 to denote wind direction for every 45-degree
angle NOAA ISD
winter-heating Binary variable equals to 1 if day t falls in the winter heating period Multiple sources

106
Table S1.8 Summary statistics for manufacturing plants characteristic aggregation at
monitoring stations
All Q1 Q2 Q3 Q4
No. of Plants 233.50 28.06 121.15 228.05 600.65
(242.40) (22.27) (30.77) (49.94) (302.11)
Total Sales 71528.03 27338.47 89010.28 82891.62 59734.97
(106650.34) (83168.69) (140878.13) (85234.78) (64465.82)
Total Assets 78075.32 38421.85 89519.61 91811.43 69556.88
(104852.84) (126555.55) (114688.06) (93199.04) (71838.21)
Labour Size 112.15 58.66 139.19 127.52 85.87
(117.26) (61.66) (157.30) (94.63) (73.07)
% Coal Power Plant 0.31 0.39 0.38 0.26 0.22
(0.66) (1.01) (0.76) (0.41) (0.35)
% Cement/Lime/Glass 2.30 0.96 2.78 2.13 2.77
(3.52) (2.54) (2.81) (1.48) (6.05)
% Oil/Petrol 0.48 0.34 0.68 0.36 0.39
(0.84) (1.00) (0.97) (0.45) (0.84)
% Chemical 7.68 5.53 8.39 7.56 8.27
(6.68) (7.99) (4.49) (5.06) (9.64)
% Iron/Steel 10.91 22.45 9.59 9.17 6.90
(12.27) (22.21) (6.62) (9.28) (6.14)
% Others 78.31 70.32 78.18 80.52 81.44
(15.91) (24.46) (10.34) (12.22) (17.83)
Note:
Standard errors are reported in the parenthesis.
Q1 indicates the monitor quartile with the lowest number of plants within 20 miles. Q4 indicates the quartile with the highest.

107
7. AQI calculation
As is mentioned in Section 6, the raw data contains hourly records of AQI, PM 10, PM 2.5,
SO2, NO2, O3 and CO. It also includes 8-hour and 24-hour means of some of the pollutants.
We take AQI as our key dependent variable, as it offers a comprehensive reflection of the general
air quality level. Its calculation picks up the most critical type of pollutant at the time, based on an
established scale of standards that is set and provided by the MEP24
. Table S1.9 is the concentration
normalization table for pollutants in AQI. Values in the cells are breaks for normalization when
calculating individual air quality index (IAQI) for each of the monitored air quality indicators.
This table reflects the relative scale for each pollutant.
For the calculation of AQI, IAQI for each of the monitored air quality indicators is first
calculated. The calculation uses the breaks in Table S1.9 and the below formula.
𝐼𝐴𝑄𝐼𝑝 =
𝐼𝐴𝑄𝐼ℎ𝑖 − 𝐼𝐴𝑄𝐼𝑙𝑜𝑤
𝐵𝑃ℎ𝑖 − 𝐵𝑃𝑙𝑜𝑤
∗ (𝐶𝑝 − 𝐵𝑃𝑙𝑜𝑤) + 𝐼𝐴𝑄𝐼𝑙𝑜𝑤
where 𝐶𝑝 measures the hourly concentration of pollutant 𝑝. 𝐵𝑃ℎ𝑖 and 𝐵𝑃𝑙𝑜𝑤 are the higher and
lower breaks of pollutant concentration that 𝐶𝑝 falls between in the respective column in Table
S1.9. 𝐼𝐴𝑄𝐼ℎ𝑖 and 𝐼𝐴𝑄𝐼𝑙𝑜𝑤 are the corresponding IAQI values for 𝐵𝑃ℎ𝑖 and 𝐵𝑃𝑙𝑜𝑤 in the table. The
final AQI is the highest of the IAQI values.
𝐴𝑄𝐼 = 𝑚𝑎𝑥 {𝐼𝐴𝑄𝐼1,𝐼𝐴𝑄𝐼2, … ,𝐼𝐴𝑄𝐼𝑝}
In the raw data, the AQI is automatically calculated. The calculation may not be exactly
the same as we show in the table, given not all recorded pollutant variables match the categories
in the table, but the framework should be the same.
24 Available at: https://www.mee.gov.cn/ywgz/fgbz/bz/bzwb/jcffbz/201203/t20120302_224166.shtml. (Archived
Feb 2022)

108
Note that Table S1.9 includes ozone and CO. We have the hourly readings of them in the
raw data, but we do not include them in our analysis. This is because the photochemical process
and the formation pattern of ozone is very different from the formation of PMs, SO2, and NO2,
and CO is at too low levels to be considered. Neither pollutant was targeted by the policy
implementation. The urban smog was the main pollution issue in Beijing in 2014, and ban policy
was implemented precisely to prevent smog and create blue skies during the APEC event. PMs are
considered the main contributor to smog in Beijing, and they were also the biggest contributor to
AQI around 2014. Hence the policy mainly targets PMs for mitigation. Therefore, in our policy
impact interpretation and result illustrations, we focus on AQI, PM 10, and PM 2.5. We also
include SO2 and NO2 in our analyses, and present the results in tabular format, but they are much
less important in the context of this policy and in our findings.

109
Table S1.9 Reference breaks for pollutant-specific concentration levels
IAQI SO2 24-
hour avg
SO2
hourly
avg (1)
NO2 24-
hour avg
NO2
hourly
avg (1)
PM10
24-hour
avg
CO 24-
hour avg
(mg/m3)
CO
hourly
avg
(mg/m3)
(1)
O3
hourly
avg
O3 8-
hour avg
PM2.5
24-hour
avg
0 0 0 0 0 0 0 0 0 0 0
50 50 150 40 100 50 2 5 160 100 35
100 150 500 80 200 150 4 10 200 160 75
150 475 650 180 700 250 14 35 300 215 115
200 800 800 280 1200 350 24 60 400 265 150
300 1600 (2) 565 2340 420 36 90 800 800 250
400 2100 (2) 750 3090 500 48 120 1000 (3) 350
500 2620 (2) 940 3840 600 60 150 1200 (3) 500
Note:
(1) The 24-hour avg values of SO2, NO2, and CO only apply to the calculation of AQI daily value. For calculation of realtime(hourly) AQI value, hourly avg values of SO2, NO2 and CO are applied.
(2) When the hourly avg of SO2 exceeds 800ug/m3, this indicator is not applicable and will not participate in the calculation of
AQI.
(3) When the 8-hour avg of O3 exceeds 800ug/m3, this indicator is not applicable and will not participate in the calculation of
AQI.
(4) Translated from MEP document.

110
8. Primary results in tabular format
We include here the results of the main analysis and the supporting analyses in tabular
format to offer more specific observation. Table S1.10 shows the complete results for the main
analysis. SO2 and NO2 results are added to this table. We can see that they generally align with
our previous conclusion, but less significant statistically or in terms of magnitude. This is
expectable due to the relatively low baseline of these two pollutants, and them not being the main
target of the policy or the pollution concern in Beijing around 2014. The cumulated net effect
indicates the overall pollution exposure impacts of the policy. This provides the foundation of our
welfare calculation.
Table S1.11 shows the complete results for the hour-period supporting analysis. It is
coupled with Figure 1.2 in the main text. Table S1.12 shows the complete results for the plant
density supporting analysis. It is coupled with Figure 1.3 in the main text

111
Table S1.10 Main analysis: effects of the policy
Pollutants
AQI PM10 PM2.5 SO2 NO2
(1) (2) (3) (4) (5)
Panel A. Effects of ban
During1
-9.875*** -9.116*** -11.049*** -1.808 -1.258**
(1.061) (1.519) (0.948) (1.674) (0.478)
Post1 18.947*** 26.898*** 13.885*** 8.396*** 5.882***
(1.592) (2.306) (1.339) (1.959) (0.596)
Observations 987,492 914,286 981,756 978,824 971,460
Panel B. Percentage and cumulated effects of ban
Baseline Concentration Levels2 114.681 140.311 77.571 50.358 44.737
During Ban (%)3
-8.611 -6.497 -14.244 -3.590 -2.811
Post Ban (%)3 16.522 19.171 17.900 16.673 13.149
Cumulative Effect4 209.453*** 400.330*** 73.446 146.754* 102.999***
Panel C. Avg. number of days in the post ban period
in violation of pollution standards5
Post Ban (22 days) 16.781 37.654 24.286 6.096 31.679
Same period in 2013 12.931 34.059 20.094 10.362 27.630
Note:
Standard errors are clustered at the monitor and year-quarter level. ***, **, and * represent significance at 1%, 5% and 10%,
respectively.
1. “During Ban” and “Post Ban” as the key variables of the policy effect, are in fact abbreviations for “During (or Post) Ban *
treated area”, which is more precise with respect to our identification strategy. This applies to the rest of the tables.
2. Mean of the untreated observations. This is not a statistically robust counterfactual, but to provide a rough reference.
3. Main coefficients divided by “baseline concentration levels”
4. The cumulative effects are calculated through: “During Ban coefficient * 21days + Post Ban coefficient *22days”. 21 days are
22 days are the respective span of the During and Post periods of the policy. This value indicates the overall pollution exposure
impact regardless of the how the two periods are specified as long as all the effects of the policy are included. Significant positive
values suggest that the policy in total increases pollution exposure to the citizens.
5. Panel C reports the average number of days in the post ban period for all monitors that the AQI crosses the “medium level
pollution” line of 150 defined by China’s MEP, and that pollutant levels violates the line WHO’s Interim Target 1 standards (75
for PM10, 75 for PM2.5, 125 for SO2, and 40 for NO2, all in ug/m3). This value is a typical attribute of performance evaluation
for China’s local officials.

112
Table S1.11 Supporting analysis: effects of the policy across different hour periods
Pollutants
AQI PM10 PM2.5 SO2 NO2
(1) (2) (3) (4) (5)
During Ban (6am to 12pm) -10.19*** -7.63*** -11.53*** -3.37 -2.92***
(0.76) (1.18) (0.63) (1.95) (0.43)
During Ban (12pm to 6pm) -10.33*** -11.27*** -10.01*** 0.24 -0.58
(1.28) (2.02) (1.14) (1.58) (0.86)
During Ban (6am to 12pm) -8.78*** -8.01*** -11.64*** -2.5 0.29
(1.31) (1.81) (1.22) (2.05) (0.72)
During Ban (12am to 6am) -9.22*** -9.04*** -10.58*** -2.13 -1.74***
(1.00) (1.29) (0.96) (1.48) (0.33)
Post Ban (6am to 12pm) 16.85*** 25.08*** 12.50*** 11.53*** 4.33***
(1.32) (1.91) (1.11) (2.15) (0.54)
Post Ban (12pm to 6pm) 16.71*** 21.70*** 12.49*** 7.98*** 8.23***
(1.74) (2.59) (1.54) (1.85) (0.94)
Post Ban (6pm to 12am) 24.56*** 32.42*** 17.66*** 8.30*** 6.56***
(1.74) (2.41) (1.56) (2.24) (0.81)
Post Ban (12am to 6am) 22.42*** 29.39*** 15.85*** 5.08** 3.96***
(1.43) (1.92) (1.25) (1.67) (0.47)
Observations 3,924,189 3,577,964 3,890,458 3,877,828 3,844,817
R2 0.46 0.42 0.40 0.46 0.49
Notes:
Standard errors are clustered at the monitor and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.

113
Table S1.12 Supporting analysis: effects of the policy by different monitor quartiles of
plant density
Pollutants
AQI PM10 PM2.5 SO2 NO2
(1) (2) (3) (4) (5)
During Ban (Q1) -10.34*** -13.55*** -9.70*** -2.99 -1.53
(1.14) (1.40) (0.81) (1.89) (0.95)
Post Ban (Q1) 8.55*** 15.18*** 5.05*** 1.99 5.40***
(1.33) (1.80) (1.06) (2.16) (1.06)
During Ban (Q2) -10.21*** -10.86*** -10.23*** 3.35* 0.45
(1.28) (1.70) (1.26) (1.82) (0.70)
Post Ban (Q2) 21.86*** 30.81*** 15.48*** 10.58*** 5.94***
(1.73) (2.40) (1.57) (2.09) (0.77)
During Ban (Q3) -8.95*** -4.97 -11.08*** -2.50 -2.04**
(1.98) (2.87) (1.64) (2.03) (0.93)
Post Ban (Q3) 19.57*** 23.65*** 15.92*** 8.73*** 4.89***
(2.68) (3.76) (2.29) (2.57) (1.04)
During Ban (Q4) -8.51*** -6.02*** -12.06*** -7.86*** -1.88***
(1.03) (1.77) (0.93) (2.40) (0.59)
Post Ban (Q4) 21.06*** 32.60*** 14.09*** 8.75*** 6.99***
(1.71) (2.85) (1.42) (2.72) (0.75)
Observations 987,492 914,286 981,756 978,824 971,460
R2 0.53 0.51 0.50 0.56 0.57
Notes:
Standard errors are clustered at the monitor and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.

114
9. Robustness checks
We conduct a series of robustness checks to validate our primary results and provide more
support for our argument.
a. Alternative control group band for the monitors
Given the assignment of the policy area is based on proximity to Beijing, this
treatment group can be essentially different with the control group. To examine the
influence, we restricted the control group monitoring stations to those within 200, 300,
and 400 km from the boundary of the policy area, and run robustness checks separately
for the three settings. Panels A1-A3 of Table S1.13 shows the results. While the
coefficients are deviate from the main analysis, the pattern of “reduction then rebound”
does not change, therefore they do not undermine the validity of our conclusion. Figure
S1.3 illustrates the setting of the 200 km control group band and Panel B of Table S1.7
summarizes the pollution level statistics under this setting.
b. One-year-before placebo test
The potential existence of unobservable temporal factors may falsely create the
results in our may analysis. To address this concern, we add a placebo test one year
ahead of the policy, setting up the analysis as if the policy took place in the same dates
and periods in 2013. Table S1.14 shows the results. No significance for the false policy
periods is indicated, supporting the validity of the main analysis.
c. Relaxed assumptions for trend and local heterogeneity

115
Our baseline assumption assumes common national trend and monthly provincial
trends. We run a series of regressions specifying different assumptions for the local
(station-specific) trends, allowing for station specific linear year-month trends, station
specific year-quarter fixed effects, and interaction terms of polynomials of distance to
policy area boundary and year-month dummies, respectively. Table S1.15 shows the
results. These results align with and support the validity of the main analysis
d. Event study analysis as alternative approach
Our main regression estimates the average effects over the policy periods. As an
alternative analysis approach, we adopt the event study method to estimate the effect
of each day over the policy window relative to the counterfactual. Figure S1.4 shows
the results, which aligns with our finds with the main analysis. The cumulative effects
shows that the policy increases overall pollution exposure
e. Local specific results
Our main regression estimates the average effects over the policy treated area. To
examine the local differences in the policy impact, we separately run analyses to
estimate the province-specific and city-specific pollution impacts of the policy. Figure
S1.5 and S1.6 show the results. We find these results generally align with the main
analysis findings of the rebound, with variations among provinces and cities.
At the province level, Tianjin, Hebei and Inner Mongolia’s initial drops are
insignificant, although they implemented relatively strict local policies. This can be a
result of variation in policy implementation timing, upfront leakage and incomplete

116
enforcement, but it doesn’t hurt the conclusion of larger rebound and overall
inefficiency. City specific effects match the observation for the province level results,
while providing more details.

117
Table S1.13 Effects of the policy with different control group bands
Pollutants
AQI PM10 PM2.5 SO2 NO2
(1) (2) (3) (4) (5)
Panel A. Effect of Ban with Control Group Restricted to Area within 200km from Regulation Boundary
During Ban -7.434*** -5.651** -7.271*** 1.258 -1.205**
(1.406) (2.044) (1.222) (1.888) (0.523)
Post Ban 15.124*** 22.420*** 10.417*** 8.107*** 5.149***
(1.550) (2.255) (1.327) (2.191) (0.625)
Observations 279,106 257,600 277,413 276,296 272,461
Panel B. Control Group within 300km from Regulation Boundary
During Ban -12.029*** -12.320*** -12.574*** -0.07 -1.174**
(1.370) (1.909) (1.245) (1.866) (0.510)
Post Ban 16.880*** 24.259*** 13.124*** 9.149*** 6.294***
(1.619) (2.237) (1.406) (2.213) (0.639)
Observations 351,207 326,544 349,403 347,897 344,041
Panel C. Control Group within 400km from Regulation Boundary
During Ban -12.029*** -12.320*** -12.574*** -0.07 -1.174**
(1.370) (1.909) (1.245) (1.866) (0.510)
Post Ban 16.880*** 24.259*** 13.124*** 9.149*** 6.294***
(1.619) (2.237) (1.406) (2.213) (0.639)
Observations 351,207 326,544 349,403 347,897 344,041
Notes: Standard errors are clustered at the station and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.

118
Figure S1.3 Areas & Monitoring Stations within 200km from APEC Pollution Regulation
Boundaries
Table S1.14 One-year-before placebo for the main analysis
Pollutants
AQI PM10 PM2.5 SO2 NO2
(1) (2) (3) (4) (5)
During Ban 3.124* 6.391** 4.111*** 6.367 7.954***
(1.373) (2.467) (0.964) (3.324) (0.958)
Post Ban 0.648 -2.241 0.978 13.388** -1.951
(1.856) (2.585) (1.293) (3.844) (1.083)
Observation 311,919 290,026 308,217 306,363 302,856
Notes: Standard errors are clustered at the station and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.

119
Table S1.15 Trend and local heterogeneity assumptions relaxation
Pollutants
AQI PM10 PM2.5 SO2 NO2
(1) (2) (3) (4) (5)
Panel A. Station Specific Linear Year-Month Trends
During Ban -9.725*** -9.206*** -10.984*** -2.127 -1.187**
(0.845) (1.316) (0.746) (1.409) (0.482)
Post Ban 19.011*** 26.572*** 13.897*** 8.105*** 6.022***
(1.425) (2.095) (1.131) (1.679) (0.618)
Observations 987,492 914,286 981,756 978,824 971,460
Panel B. Station*Year-Quarter Dummies
During Ban -5.079*** -8.008*** -5.483*** -8.542*** 1.283***
(1.156) (1.199) (0.981) (0.372) (0.368)
Post Ban 17.415*** 22.791*** 13.775*** 2.978*** 7.271***
(1.375) (1.486) (1.201) (0.573) (0.270)
Observations 987,352 914,158 981,622 978,684 971,325
Panel C. Distance to Regulation Boundary Polynomials*Year-Month Dummies
During Ban -10.033*** -9.457*** -11.017*** -2.015 -1.482**
(1.041) (1.506) (0.936) (1.605) (0.487)
Post Ban 18.854*** 26.526*** 14.027*** 8.072*** 5.545***
(1.575) (2.282) (1.328) (1.879) (0.613)
Observations 987,492 914,286 981,756 978,824 971,460
Notes:
Standard errors are clustered at the station and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.
Panel A, we include the interaction of monitoring station dummies with linear year-month trends to allow pollution trends to vary
linearly for every station.
In Panel B, we flexibly control for pollution trends over time by allowing trends to vary non-linearly by including station
dummies interacted with year-quarter dummies.
In Panel C, we allow pollution trends to vary non-linearly with distance from the regulation boundary by interacting distance
from boundary first and second polynomials with year-month dummies.

120
Figure S1.4 Event study result over the policy period for the effect of the policy
The top and bottom caps indicate the 95% confidence interval for each day estimate, and the standard errors are clustered at the
station level.
The cumulative effect is the sum of the prior daily effect coefficients.
The estimated effect of each day over the policy period on AQI for the regulated and unregulated areas. The resulting trend generally
aligns with the main result, and the dashed line indicating the cumulative effect suggests eventual increase of total pollution.

121
Figure S1.5 Estimated effects of the policy in different affected provinces
Note: Standard errors are clustered at the station level.

122
Figure S1.6 Estimated effects of the policy in different affected cities
Note: Standard errors are clustered at the station level.

123
f. Max daily reading as dependent variable instead of daily mean
Exposure to pollution at an exceptional level may cause substantially more serious
health impacts than at relatively lower levels even in a short period, meanwhile our
usage of daily mean of pollution levels as dependent variable fails to address this
situation. To address this, we reconstruct the dependent variables by adopting the daily
maximum from the hourly records of the raw data. If the max daily readings are not
significantly impacted by the policy as the daily means are, it is arguable that the public
health implication of our results are not as strong. Table S1.16 shows the results of this
analysis. The impact of the policy result in even larger numbers in daily max pollution.
It does not deviate much from our main analysis and does not undermine our
conclusion.
g. Visibility Index and fire points as alternative dependent variable
We use Visibility Index data from NOAA monitors and active fire counts data from
NASA’s Moderate Resolution Imaging Spectroradiometer as alternative dependent
variables. The logarithms of these two variables are also included in the analyses as
dependent variables. The poorer the air quality, the lower the Visibility Index should
be. We use this analysis to address the concerns for potential pollution data
manipulation by government officials. The fire points proxy for the agricultural activity
of straw burning, which provides a competition to the industrial mechanism theory.
Table S1.17 and Table S1.18 shows the results.
The logarithm of the Visibility Index shows significant drop in the Post period of
the ban, which aligns with and supports our finding of the rebound effect. The fire

124
points significantly dropped in the During period, which aligns with the policy
implementation. But it does not increase in the Post period, suggesting that it may not
contribute to the rebound. These results do not undermine, if not support, our previous
findings and theories.
h. CEM emission data as alternative policy result indicators
We use the Continuous Emission Monitoring system data from MEP to investigate
the effect of the policy on emission. This data is from a different source and takes a
different angle compared to the ambient air pollution data, and adopts a bottom-up
approach directly tackling the emission behavior. Given the CEM monitors are built in
factory pipes and emission outlets, they are less affected by noises such as weather.
The analysis setting is the same as the main analysis. Table S1.19 and Figure S1.7 show
the results of the CEM analysis. The results do not align with our main analysis, as
there is no significant rebound.
We trace the reason of not aligning to the fact that the CEM system monitors only
a selective list of manufacturing firms. These firms generally tend to be the major firms
under close watch of the local environmental protection agencies. The CEM system is
built specifically to ensure the all-time compliance of these firms to the pollution
standards. Given their status, the firms under CEM system are reasonably different
from the ones that would utilize chances to get around inspection to emit more, as is
suggested by our supporting analyses. Table S1.20 provides the results for our separate
analyses for non-state-owned and state-owned enterprises under CEM system. The
results are consistent with the main CEM emission analysis. Table S1.21 provides more

125
detailed results for our estimate of the policy effect on the emission of firms in different
industry categories. We notice that except for the Oil/Petrol industry showed a
significant increase in TSP emission During the policy, all other estimated effects are
either negative or insignificant, not showing any sign of the rebound effect and recoup
behavior.

126
Table S1.16 Effects of the policy on daily max pollution readings
Pollutants
AQI PM10 PM2.5 SO2 NO2
(1) (2) (3) (4) (5)
Panel A. Effects of Ban on Pollutants Concentration
During Ban -13.690*** -17.971*** -19.135*** -5.056 -3.906***
(1.733) (2.739) (1.889) (3.176) (0.573)
Post Ban 35.173*** 47.817*** 22.153*** 16.106*** 3.950***
(2.419) (3.837) (2.847) (3.634) (0.753)
Observations 988,440 914,420 981,883 986,042 978,599
Notes:
Standard errors are clustered at the station and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.
Table S1.17 Effects of policy on Visibility Index
Panel A: Visibility Index
Visibility ln(Visibilty)
(1) (2)
During Ban -0.164 -0.120**
(0.395) (0.048)
Post Ban -0.428 -0.179***
(0.363) (0.046)
Observations 652,651 651,217
Notes:
Standard errors are clustered at the monitor level
***, **, and * represent significance at 1%, 5% and 10%, respectively.
Reported estimates denote the average daily change in Visibility Index and natural logarithm of Visibility Index collected at
NOAA monitoring stations during and after the APEC Summit.
We control for weather conditions including wind speed, temperature, humidity and air pressure and their second polynomials,
and include station fixed effects and date fixed effects.

127
Table S1.18 Effects of policy on straw burning
Panel B: Straw Burning
Fire Points ln(0.1+Fire Points)
(1) (2)
During Ban -7.048*** -0.078
(1.245) (0.088)
Post Ban -1.399* -0.154
(0.659) (0.089)
Observations 369,015 369,015
Notes:
Standard errors are clustered at the monitor level for the visibility index analysis and at the city level for the fire points analysis.
***, **, and * represent significance at 1%, 5% and 10%, respectively.
In model (2) we use “ln(0.1+Fire Points)” because fire points can be 0.
Reported estimates denote the average daily change in the number of fire points aggregated at city level during and after the
APEC Summit.
In these regressions, we control for weather conditions including wind speed, temperature, humidity and air pressure and their
second polynomials, and include station fixed effects and date fixed effects
Table S1.19 Effects of the policy on emissions from firms under Continuous Emission
Monitoring (CEM) system
Pollutants
TSP SO2 NOx
(1) (2) (3)
Panel A. Effects on Emission Density under CEM
During Ban -1.065** -11.278*** -13.896***
(0.338) (3.195) (1.773)
Post Ban -0.565 -7.148* -7.211**
(0.312) (3.507) (2.134)
Observations 2,510,240 2,610,672 2,578,541
Notes:
Standard errors are clustered at the firm and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.

128
Figure S1.7 Effect of the policy on Continuous Emission Monitoring (CEM) firms’ emission
Table S1.20 Effects of policy on the emission of State-Owned and non-State-Owned CEM
firms.
Non-SOE SOE
TSP SO2 NOx TSP SO2 NOx
(1) (2) (3) (4) (5) (6)
During Ban -1.401 -16.448** -11.850** -0.958** -9.587*** -14.390***
(0.808) (6.947) (4.356) (0.322) (2.123) (1.574)
Post Ban -1.692* -8.020 -1.343 -0.199 -6.786** -9.042***
(0.801) (7.256) (4.351) (0.276) (2.426) (2.110)
Observations 602,276 634,025 631,200 1,907,964 1,976,647 1,947,341
R2 0.524 0.539 0.577 0.519 0.548 0.603
Notes:
Standard errors are clustered at the firm level and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.

129
Table S1.21 Effects of policy on the emission of CEM firms by different industry categories.
Pollutants
TSP SO2 NOx
(1) (2) (3)
During - Iron/Steel -3.437*** -36.126*** -7.648***
(0.905) (7.565) (2.096)
Post Ban - Iron/Steel -0.960 -20.860** -6.785**
(0.859) (7.087) (2.721)
During Ban - Coal Power Plant -0.614 -6.522 -6.485*
(0.378) (4.793) (3.087)
Post Ban - Coal Power Plant -0.264 -3.688 3.503
(0.360) (5.011) (3.576)
During Ban - Cement/Lime/Glass -1.558** -15.917*** -3.253
(0.487) (3.806) (7.165)
Post Ban - Cement/Lime/Glass -0.899* -7.304 1.467
(0.433) (3.962) (6.879)
During Ban - Chemical -2.668 -8.799 -29.859*
(1.536) (6.318) (13.242)
Post Ban - Chemical -2.791 -31.867*** -43.939**
(1.686) (7.653) (16.358)
During Ban - Oil/Petrol 2.801** -23.681** -24.415***
(0.848) (7.685) (6.885)
Post Ban - Oil/Petrol -1.296 -7.807 -2.786
(0.830) (6.744) (7.740)
During Ban - Others -0.096 2.150 -20.439**
(1.164) (6.450) (6.349)
Post Ban - Others -0.505 -9.296 -23.500**
(1.039) (6.042) (7.856)
Observations 2,510,240 2,610,672 2,578,541
Notes:
Standard errors are clustered at the firm and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.

130
i. Allowing city-specific daily pollution patterns for the hour-period analysis
On top of the baseline hour-period analysis, we add a robustness check that controls
for the fixed effects of cities interacted with the 6-hour-periods, allowing city-specific
daily patterns of pollution levels. Table S1.22 shows the results, which is to be
compared with Table S1.11. The results align with those of the baseline hour-period
analysis
j. Monitor quartiles based on area total sales instead of plant density
As robustness check to the density quartile analysis, we test for the competing
theory that the recoup emission intensity and stronger rebound is driven by higher total
sales in higher plant density quartiles rather than less average inspection. We regroup
the monitors into quartiles of total sales within the 20-mile radius, and run the
regression for the quartiles specific policy effects. Figure S1.8 shows the results, which
is to be compared with Figure 1.3 and Table S1.12. The figure does not show a constant
pattern of stronger rebound effect associated with higher area total sales, which denies
the total sales theory as the driver of stronger recoup behavior and pollution rebound.
k. Internal consistency of the hour-period and density quartile analyses
To validate the internal consistency of the supporting analyses, we construct 16
combinations for hour periods and density quartiles (4*4), and run analysis for hourperiod-quartile specific results. Table S1.23 coupled with Figure S1.9 show the results,
which aligns with our previous findings. As we can see within each quartile group, the
dark periods generally experience stronger rebound effect than the day periods.

131
Similarly, within each hour-period, the higher plant density quartiles experience
stronger rebound effect than the lowest density quartile.
Table S1.22 Effects of the policy by 6-hour periods, allowing city-specific daily patterns of
pollution
Pollutants
AQI PM10 PM2.5 SO2 NO2
(1) (2) (3) (4) (5)
During Ban (6am to 12pm) -10.32*** -8.07*** -11.64*** -3.50* -2.92***
(0.95) (1.35) (0.84) (1.93) (0.39)
During Ban (12pm to 6pm) -10.35*** -10.78*** -10.13*** 0.08 -0.82
(1.16) (1.93) (1.00) (1.49) (0.83)
During Ban (6am to 12pm) -8.74*** -7.92*** -11.51*** -2.20 0.39
(1.23) (1.78) (1.14) (1.99) (0.72)
During Ban (12am to 6am) -9.37*** -9.57*** -10.68*** -2.33 -1.61***
(1.04) (1.36) (0.99) (1.50) (0.36)
Post Ban (6am to 12pm) 16.86*** 24.86*** 12.49*** 11.61*** 4.36***
(1.42) (1.99) (1.25) (2.17) (0.50)
Post Ban (12pm to 6pm) 16.77*** 22.21*** 12.43*** 7.70*** 7.92***
(1.62) (2.49) (1.38) (1.73) (0.90)
Post Ban (6pm to 12am) 24.68*** 32.62*** 17.83*** 8.56*** 6.58***
(1.65) (2.38) (1.46) (2.23) (0.80)
Post Ban (12am to 6am) 22.49*** 29.12*** 15.90*** 5.07** 4.16***
(1.52) (1.95) (1.36) (1.70) (0.49)
Observations 3,924,189 3,577,963 3,890,458 3,877,828 3,844,817
Notes:
Standard errors are clustered at the station and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.

132
Figure S1.8 Effects of policy by different manufacturing total sales levels within 20 miles
from monitors

133
Table S1.23 Effects of policy by combinations of different 6-hour-periods and different
plant density quartiles
Pollutants
AQI PM10 PM2.5 SO2 NO2
(1) (2) (3) (4) (5)
During Ban 6am-12pm Q1 -8.133*** -2.566 -8.936*** -3.600 -4.647***
(1.578) (3.041) (1.354) (2.097) (0.675)
During Ban 12pm-6pm Q1 -6.518** -6.935* -5.951*** -3.955 -2.030*
(2.291) (3.328) (1.882) (2.597) (1.051)
During Ban 6pm-12am Q1 -3.682* -0.485 -6.978*** -4.623 -3.399***
(1.779) (2.522) (1.484) (2.745) (0.894)
During Ban 12am-6am Q1 -6.927*** 1.325 -8.750*** -1.704 -4.427***
(1.124) (2.278) (1.053) (1.975) (0.456)
Post Ban 6am-12pm Q1 8.205*** 26.804*** 2.504 9.926*** 1.573**
(1.893) (3.771) (1.466) (2.470) (0.702)
Post Ban 12pm-6pm Q1 8.737*** 19.613*** 4.572** 4.987 4.458***
(2.361) (3.454) (1.955) (3.088) (1.173)
Post Ban 6pm-12am Q1 17.275*** 30.385*** 8.903*** 5.137 1.334
(2.102) (3.053) (1.707) (3.061) (1.040)
Post Ban 12am-6am Q1 13.836*** 32.973*** 5.440*** 4.808* 0.602
(1.553) (2.776) (1.456) (2.394) (0.668)
During Ban 6am-12pm Q2 -10.332*** -10.592*** -11.163*** -1.854 -1.641*
(0.961) (1.843) (0.689) (2.502) (0.813)
During Ban 12pm-6pm Q2 -9.329*** -10.928*** -8.129*** 2.056 1.572
(1.800) (2.746) (1.425) (1.996) (0.934)
During Ban 6pm-12am Q2 -3.356** -1.979 -6.319*** 0.487 3.255***
(1.329) (1.687) (1.261) (2.645) (0.926)
During Ban 12am-6am Q2 -7.521*** -8.707*** -8.434*** -1.171 -0.120
(1.073) (1.818) (1.041) (1.956) (0.514)
Post Ban 6am-12pm Q2 16.566*** 21.713*** 12.289*** 14.752*** 5.337***
(1.571) (2.453) (1.345) (2.831) (0.932)
Post Ban 12pm-6pm Q2 16.741*** 21.120*** 11.845*** 11.957*** 9.864***
(2.200) (3.263) (1.884) (2.426) (1.042)
Post Ban 6pm-12am Q2 23.464*** 27.490*** 17.339*** 13.249*** 6.406***
(1.708) (2.186) (1.660) (2.874) (0.972)
Post Ban 12am-6am Q2 20.330*** 24.112*** 15.462*** 5.403** 4.185***
(1.604) (2.487) (1.377) (2.225) (0.611)
During Ban 6am-12pm Q3 -11.958*** -9.085*** -14.314*** -1.408 -3.845***
(1.372) (2.042) (0.969) (1.533) (0.739)
During Ban 12pm-6pm Q3 -13.726*** -15.367*** -13.141*** 2.151 -1.997*
(2.392) (3.646) (1.859) (1.864) (1.031)
During Ban 6pm-12am Q3 -12.922*** -15.209*** -15.936*** -1.551 -2.060**

134
(1.877) (2.674) (1.624) (1.983) (0.848)
During Ban 12am-6am Q3 -11.021*** -14.544*** -12.227*** -1.049 -3.378***
(1.625) (2.063) (1.444) (1.347) (0.496)
Post Ban 6am-12pm Q3 21.702*** 30.300*** 16.268*** 9.002*** 3.659***
(2.079) (3.202) (1.620) (1.793) (0.867)
Post Ban 12pm-6pm Q3 20.761*** 26.184*** 15.511*** 6.661** 8.479***
(2.815) (4.271) (2.226) (2.258) (1.146)
Post Ban 6pm-12am Q3 29.960*** 40.567*** 20.762*** 6.257** 6.948***
(2.343) (3.631) (2.074) (2.219) (0.948)
Post Ban 12am-6am Q3 29.854*** 37.531*** 21.070*** 2.788 3.488***
(2.070) (2.869) (1.724) (1.581) (0.707)
During Ban 6am-12pm Q4 -10.251*** -6.532*** -11.460*** -7.557** -2.406***
(1.071) (1.650) (1.056) (2.527) (0.754)
During Ban 12pm-6pm Q4 -10.159*** -9.519*** -11.175*** -1.674 -0.484
(2.002) (2.778) (1.736) (1.727) (1.167)
During Ban 6pm-12am Q4 -13.329*** -11.981*** -15.432*** -5.153** 1.107
(1.630) (2.184) (1.623) (2.317) (0.914)
During Ban 12am-6am Q4 -11.121*** -10.172*** -12.717*** -5.060** -0.349
(1.428) (2.116) (1.430) (2.094) (0.582)
Post Ban 6am-12pm Q4 17.658*** 22.947*** 14.980*** 10.730*** 5.472***
(1.654) (2.204) (1.632) (3.002) (0.766)
Post Ban 12pm-6pm Q4 17.967*** 20.037*** 15.268*** 6.008** 8.509***
(2.509) (3.366) (2.274) (2.100) (1.254)
Post Ban 6pm-12am Q4 25.228*** 31.850*** 20.612*** 7.196** 8.472***
(2.165) (2.854) (2.173) (2.576) (1.026)
Post Ban 12am-6am Q4 22.366*** 25.602*** 17.313*** 6.530** 5.863***
(1.878) (2.705) (1.729) (2.456) (0.727)
Observations 3,924,189 3,577,964 3,890,458 3,877,828 3,844,817
Notes:
Standard errors are clustered at the station and year-quarter level.
***, **, and * represent significance at 1%, 5% and 10%, respectively.

135
Figure S1.9 Estimated effect of the policy by plant density quartile and hour period
combinations

136
Appendix B – For Chapter III
Contents:
1. Policy scheme coverage area
Table S2.1 List of cities covered by the “2+26 Cities” policy scheme
Figure S2.1 “2+26 Cities” location
2. Robustness checks
2.1 Alternative industrial diversity indices
Table S2.2 Summary statistics for alternative indices and green patents
Table S2.3 Alternative indices: Industrial diversity impact on environmental regulation
performance
Table S2.4 Alternative indices: Industrial diversity impact on the industry-pollution relationship
2.2 Endogeneity threat of the technology level differences
2.3 Restricting sample pool
Table S2.5 Robustness Checks: Industrial diversity impact on environmental regulation
performance
Table S2.6 Robustness checks: Industrial diversity impact on the industry-pollution relationship
Table S2.7 Summary statistics with major cities excluded

137
1. Policy schemes coverage areas
Table S2.1 List of cities covered by the “2+26 Cities” policy scheme
Provinces Cities Provinces Cities
Beijing 北京 Beijing 北京 Shandong 山东 Jinan 济南
Tianjin 天津 Tianjin 天津 Zibo 淄博
Hebei 河北 Shijiazhuang 石家庄 Heze 菏泽
Baoding 保定 Dezhou 德州
Langfang 廊坊 Liaocheng 聊城
Xingtai 邢台 Binzhou 滨州
Handan 邯郸 Jining 济宁
Hengshui 衡水 Henan 河南 Zhengzhou 郑州
Cangzhou 沧州 Xinxiang 新乡
Tangshan 唐山 Jiaozuo 焦作
Shanxi 山西 Taiyuan 太原 Hebi 鹤壁
Jincheng 晋城 Anyang 安阳
Yangquan 阳泉 Kaifeng 开封
Changzhi 长治 Puyang 濮阳

138
Figure S2.1 “2+26 Cities” location

139
2. Robustness checks
2.1 Alternative industrial diversity indices
While the literature uses different measurements for industrial diversity, the entropy by
branch employment (employment entropy) is one of the most popular methods. Given most of
these studies focus mainly on industrial diversity’s properties in regards of local economic
performance, such as total employment rate and productivity, the employment entropy is suitable
considering the potential mechanism of industrial diversity’s interaction with economy is likely
through the labor force, whether by knowledge spillover or relocation of workers. Also, the
calculation of entropy has the benefit of being able to separate into entropy at different levels of
industry aggregation, which helps theoretically clarifying the different types of industrial diversity.
In this paper, the employment entropy calculation is also adopted for the main analysis.
In complement to the main analysis, this paper also tests the analysis results for alternative
industrial diversity indices: (1) entropy by branch product, (2) HHI (Herfindahl-Hirschman Index)
by branch employment, and (3) HHI by branch product. Together with the entropy by branch
employment calculation, the four measurements form two dimensions of measurement choice: the
calculation of entropy or HHI, and the measurement by employment or by product. For this paper,
one novel distinction from the literature is the outcome variable of interest being the environmental
factor of air pollution level, which is closely related with industrial production output, therefore it
is also reasonable to use productivity instead of employment as the basis for industrial diversity
calculation. When the branches of the industries function as a risk spreading portfolio, diversity of
branch gross product is arguably more influencing for the air pollution outcome in face of
environmental regulations than that calculated by employment. To see whether product-based
industrial diversity index show clearer effects in the case of this study or deviate from the

140
employment-based result, the product-based calculations are added as a parallel analysis for
comparison.
The index of choice provides another dimension. The usage of HHI for economic diversity
calculation is as prevalent as that of entropy in the literature. (Kemeny and Storper, 2015; Simon,
1988; Izraeli and Murphy, 2001) It essentially measures the level of specialization in the industrial
structure of a region, which is closely related to the entropy measurement of diversity: the more
specialized the local industrial sector is, the less diverse. While the entropy method has been
acknowledged for its property of being able to separate the entropy calculation at different
grouping levels (Attaran, 1986), with the focus of this paper dedicated to the branch level
(“Industrial Branch Level I”), the calculation is performable in the HHI style as well. Given the
nuanced differences between specialization and “adverse diversity”, for easier comparison, the
HHI calculation in this paper is defined as:
𝐻𝐻𝐼𝑖 = 1 − ∑𝑆𝑖𝑙
2
𝐿
𝑙=1
where 𝐻𝐻𝐼𝑖
stands for the HHI style measurement of industrial diversity for city 𝑖. 𝑆𝑖𝑙 stands for
the employment or product share of the industrial branch 𝑙 in prefecture 𝑖 , among a total of 𝐿
existing branches at the industrial branch level I for the entirety of the secondary sector. The
standard calculation of HHI uses firm-level shares, but here the index is calculated at the branch
level in attempt to catch the unrelated diversity. Also, standard HHI range from 0 to 1, with 1
indicating one-firm domination and bigger values for higher specialization; this paper reverses the
0 to 1 mapping to align with the entropy measurement of diversity and for easier comparison of
regression results.

141
With the two dimensions for the industrial diversity index, in addition to the entropy by
branch employment calculation for the main analysis, the regressions with entropy by branch
product, HHI by branch employment, and HHI by branch product as the industrial diversity
treatment (𝑇𝑖
in Equations (1) and (3) of the main text) are run as robustness checks. The results
corresponding respectively to the two steps of the main analysis are presented in Tables S2.2 and
S2.3. In Table S2.2, Models 1, 3, and 5 corresponds with Model 2 of Table 2.2 in the main text,
and Models 2, 4, and 6 with Model 4 of Table 2.2. The three models of Table S2.3 corresponds
with Model 2 of Table 3 of the main text. The regression results in general align with the main
analysis results with entropy by branch employment, indicating the consistency of the findings of
the industrial diversity’s impact on the environmental regulation performance and on the industrypollution relationship.
Table S2.2 Summary statistics for alternative indices and green patents
Policy Scheme Regions
Regression variables 2+26 No-Scheme
Entropy by branch product 2.551 2.453
(0.042) (0.014)
HHI by branch employment 0.880 0.881
(0.010) (0.002)
HHI by branch product 0.852 0.845
(0.011) (0.003)
Green patents 1.191 1.061
(0.135) (0.059)
Observations 140 1249

142
Table S2.3 Alternative indices: Industrial diversity impact on environmental regulation
performance
Note: Standard errors are clustered at the city level. ***, **, and * represent significance at 1%, 5% and 10%, respectively.
Model Specification
Product Entropy Employment HHI Product HHI
Key estimators (1) (2) (3) (4) (5) (6)
Effect of policy scheme
"2+26 Cities" × Post 10.473 13.245 12.250
(7.571) (13.077) (9.004)
"2+26 Cities" × 2016 17.838*** 33.944*** 23.567***
(6.763) (7.636) (6.758)
"2+26 Cities" × 2017 21.095** 30.455** 27.739***
(8.299) (13.759) (10.180)
"2+26 Cities" × 2018 19.748** 28.202* 24.742**
(9.938) (16.682) (12.141)
"2+26 Cities" × 2019 18.722* 32.907* 20.838*
(10.369) (19.044) (11.329)
Effect of industrial diversity
Post × Employment entropy -2.472*** -14.888*** -8.133**
(0.776) (4.603) (3.530)
2016 × Employment entropy -0.802 -3.081 -2.052
(0.804) (4.125) (3.341)
2017 × Employment entropy -2.992*** -16.831*** -8.580*
(1.080) (5.776) (4.380)
2018 × Employment entropy -4.483*** -24.396*** -14.473**
(1.250) (7.084) (5.852)
2019 × Employment entropy -1.333 -8.508 -4.964
(1.026) (5.903) (4.152)
"2+26 Cities" × Post × Employment entropy -8.168*** -26.799* -26.498***
(2.857) (14.570) (10.210)
"2+26 Cities" × 2016 × Employment entropy -8.248*** -42.259*** -31.449***
(2.748) (9.033) (8.131)
"2+26 Cities" × 2017 × Employment entropy -11.700*** -44.670*** -42.920***
(3.279) (15.551) (11.784)
"2+26 Cities" × 2018 × Employment entropy -12.951*** -47.160** -44.613***
(3.837) (18.790) (13.902)
"2+26 Cities" × 2019 × Employment entropy -13.126*** -53.895** -41.446***
(3.968) (21.318) (12.953)
Year FE Yes Yes Yes Yes Yes Yes
City FE Yes Yes Yes Yes Yes Yes
Observations 1,180 1,180 1,180 1,180 1,180 1,180
Adjusted R2 0.939 0.941 0.939 0.941 0.938 0.939
Adjusted R2 within 0.131 0.163 0.130 0.162 0.117 0.142

143
Table S2.4 Alternative indices: Industrial diversity impact on the industry-pollution
relationship
Note: “LogSec” stands for the natural logarithm of the city secondary sector gross product, same as in Equation [5]. Standard errors
are clustered at the city level. ***, **, and * represent significance at 1%, 5% and 10%, respectively.
2.2 Endogeneity threat of the technology level differences
The main analysis results are threatened by the potential endogeneity of the technological
levels. It is an applaudable alternative explanation that local industrial diversity is strongly
correlated with the local technological capabilities to shift to cleaner production in face of
introduced regulation policies. This mechanism can also reduce the abatement costs and buffer the
economy-environment trade-off, while industrial diversity does not actually have a play in this
process. In such case, the omission of the techno-logical factors in the model would grant
significant effects for industrial diversity in the regression due to the strong correlation.
However, technological levels and the capability to upgrade to cleaner production is
complex at the city level, making it hard to factor and define. In this paper, I choose from China’s
Urban Yearbooks data the closest available factor as the proxy for the technological level factor –
the “per capita number of green technology patents granted.” (referred to as “green patents”) It is
Industrial Diversity Index
Product Entropy Employment HHI Product HHI
Key estimators (1) (2) (3)
LogSec 15.639*** 23.113*** 17.644***
(4.813) (7.785) (5.264)
"2+26" × LogSec 62.868 106.049 102.269
(55.645) (72.810) (70.576)
LogSec × Employment entropy -3.441** -18.081** -12.233**
(1.742) (8.682) (5.684)
"2+26" × LogSec × Employment entropy -31.627 -140.835 -139.034
(24.349) (88.016) (88.026)
Year FE Yes Yes Yes
City FE Yes Yes Yes
Observations 988 988 988
Adjusted R2 0.925 0.925 0.925
Adjusted R2 within 0.040 0.040 0.040

144
added to the main analysis models parallel to the industrial diversity index to check for collinearity.
If the endogeneity issue exists, the estimators for the effects of the green patents would deprive
the significance for the industrial diversity effect estimators. The results are shown in Model 1 of
Table S2.4 for the first step analysis and Model 1 of Table S2.5 for the second step analysis. From
these tables, the inclusion of the green patents factor as parallel to industrial diversity does not
deprive the statistical significance from the industrial diversity effect estimators, supporting the
validity of the original results and conclusions. However, in Table S2.5 Model 1, the addition of
the green patents factor brings significance to the second step model estimators of the “2+26 Cities”
specific industry-pollution relationship and effect of employment entropy, without changing the
signs or magnitudes of the coefficients. The green patents factor itself also show modestly
significant correlation with the secondary sector size as well as with the policy region cities’
secondary sector size. These results suggest that: 1) while the buffering effect of industrial
diversity for the industry-pollution reduction universally exist for all cities, the effect may be
enhanced or pronounced in areas of pollution reducing policies; and 2) although the green patents,
proxying for local green technology accessibility and capabilities, is not the pathway through
which industrial diversity is correlated with air pollution levels and the industry-pollution
relationship, the green patents factor itself shows also a buffering effect that eases the economyenvironment trade-off and enhances environmental policy performance.

145
Table S2.5 Robustness Checks: Industrial diversity impact on environmental regulation
performance
Model Specification
Green Patents Major Cities Excluded
Key estimators (1) (2) (3)
Effect of policy scheme
"2+26 Cities" × Post 12.710 -9.824*** 10.793
(9.917) (2.151) (9.062)
Effect of industrial diversity
Post × Employment entropy -3.409*** -3.071***
(0.895) (0.930)
"2+26 Cities" × Post × Employment entropy -7.685** -7.884**
(3.664) (3.349)
Effect of green patents
Green patents 0.452
(0.424)
"2+26 Cities" × Green patents -3.496*
(1.923)
Post × Green patents -0.097
(0.207)
"2+26 Cities" × Post × Green patents 0.039
(0.717)
Year FE Yes Yes Yes
City FE Yes Yes Yes
Observations 1,180 1,052 1,046
Adjusted R2 0.940 0.936 0.938
Adjusted R2 within 0.145 0.087 0.135
Note: Standard errors are clustered at the city level. ***, **, and * represent significance at 1%, 5% and 10%, respectively.

146
Table S2.6 Robustness checks: Industrial diversity impact on the industry-pollution
relationship
Model Specification
Green Patents Major Cities Excluded
Key estimators (1) (2) (3)
LogSec 19.623*** 5.358** 17.025***
(6.288) (2.168) (6.511)
"2+26" × LogSec 118.052** -17.176 62.164
(52.329) (13.990) (56.782)
LogSec × Employment entropy -4.572** -4.453*
(2.223) (2.306)
"2+26" × LogSec × Employment entropy -45.338** -32.069
(19.665) (23.973)
Effect of green patents
Green patents 5.071
(3.688)
"2+26" × Green patents -0.250
(0.205)
LogSec × Green patents -109.143*
(58.405)
"2+26" × LogSec × Green patents 5.757*
(3.310)
Year FE Yes Yes Yes
City FE Yes Yes Yes
Observations 988 878 872
Adjusted R2 0.931 0.925 0.925
Adjusted R2 within 0.119 0.025 0.036
Note: Standard errors are clustered at the city level. ***, **, and * represent significance at 1%, 5% and 10%, respectively.
2.3 Restricting sample pool
To further eliminate the endogenous threat for the estimated industrial diversity effects
coming from other city characteristics, the main analysis is run for a restricted sample pool of cities,
excluding the major cities of China. The four only Administrative Level 1 cities – Beijing, Tianjin,
Shanghai, and Chongqing – are excluded from the sample cities, together with all of province
capital cities. In addition, the only major city not included in the above set, Shenzhen, is also

147
excluded as one of the biggest and most economically and technologically developed cities in
China. This restriction to the sample pool is due to the consideration of the tendency in China that
resources crowd into the biggest city in the region. In each province, the province capital is often
the most developed city, and much more so than the rest of the cities. They would have much
better labor and financial resources, more complete industries, along with better institutions. This
pattern also applies nationwide, as the Admin. Level 1 cities and the Tier 1 cities enjoy the best
national resources. Excluding these cities from the analysis would reduce the possibility of the
estimated industrial diversity effects coming from the heterogeneous features of the big cities’
industries, and allow an observation into the less outlying cities.
The results with the restricted samples are shown in Models 2 and 3 of Tables S2.5 and S2.6. These
results align with the main analysis, supporting the validity and applicability of the original
conclusion in a more stable sample set.

148
Table S2.7 Summary statistics with major cities excluded
Policy Scheme Regions
Regression variables 2+26 No-Scheme
AQI 104.113 67.473
(1.247) (0.492)
Entropy by branch employment 2.731 2.666
(0.039) (0.012)
Natural logarithm of second sector product 16.459 15.754
(0.068) (0.032)
GDP per capita 58966.386 56497.187
(2270.386) (969.133)
Precipitation 2.402 3.895
(0.403) (0.269)
Wind speed annual average 10.425 10.835
(0.952) (0.541)
Temperature annual average 57.911 59.398
(0.206) (0.287)
Observations 140 1249

Abstract (if available)

Abstract Industrializing, developing countries face strong domestic demands for economic growth alongsided emerging environmental issues. Given limited resources and weak institutions , these competing demands create a trade-off in the policy agenda, representing a common challenge for environmental policymaking in these nations. This dissertation seeks to deepen the understanding of the economy-environment trade-off in the developing country scenario by examining China's air pollution policies.
Chapter II explores the 2014 APEC short-term pollution ban in Beijing. The study identified a significant reduction in air pollution during the policy, followed by a larger pollution spike. The rebound is identified to be caused by unregulated recouping behavior of the industry. The finding imply how economic incentives and weak institutions can undermine the efficiency of environmental policy design.
Chapter III investigates the potential industrial diversity to provide economic resilience against policy shocks and easing the trade-off through China's "2+26 Cities" regional air pollution reduction policy scheme. The findings indicate that cities characterized by greater industrial diversity witness more substantial reductionsin air pollution levels in the policy area. Furthermore, the study exhibites the buffering effect of inudstrial diversity which mitigates the correlation between industrial sector productivity and air pollution levels.
Chapter IV investigates the economic costs of the "2+26 Cities" policy. The analysis reveals a significant contraction in the secondary sector productivity, supporting the hypothesis that environmental regulations can impose short-term economic costs. The study also finds that while the secondary sector experienced a notable decline, the tertiary sector remained relatively unaffected.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Climate change, air pollution and health: how policy can induce adaptation

PDF

Processes, effects, and the implementation of market-based environmental policy: southern California's experiences with emissions trading

PDF

A study of Chinese environmental NGOs: policy advocacy, managerial networking, and leadership succession

PDF

Three essays on the causes and consequences of China’s governance reforms

PDF

China's environmental reform: ecological modernization, regulatory compliance, and institutional change

PDF

Why go green? Cities' adoption of local renewable energy policies and urban sustainability certifications

PDF

The functions of the middleman: how intermediary nonprofit organizations support the sector and society

PDF

Governing regional collaboratives: institutional design, management and leadership

PDF

Governance and urban design: Omaha by design

PDF

The effects of interlocal collaboration on local economic performance: investigation of Korean cases

PDF

Institutional analysis of factors critical to facilitating project management and implementation of Metro ExpressLanes

PDF

Financial crises and trade policy in developing countries

PDF

From “squatters” to citizens? Slum dwellers, developers, land sharing and power in Phnom Penh, Cambodia

PDF

Does collaborative R&D policy work? The effect of collaborative R&D policy on innovative activities of firms in Korea

PDF

The perspectives of community organizers on advocating for environmental justice related to transportation air-pollution

PDF

Assessing the governance of the independent regulatory agencies in China

PDF

The social making of authoritarian environmentalism: protest-litigation nexus and policy changes in China

PDF

Mission match, reputation, and bureaucratic motivation

PDF

A comparative study of administrative corruption in ancient and modern China

PDF

Essays on fiscal outcomes of cities in California

Asset Metadata

Creator Shen, Renzhi (author)

Core Title The economy-environment trade-off in China's air pollution policies

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Public Policy and Management

Degree Conferral Date 2024-12

Publication Date 10/28/2024

Defense Date 10/10/2024

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag Air pollution,Chinese policy,developing country,environmental economics,Environmental policy

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Bento, Antonio (committee chair), McCarthy, T.J. (committee member), Tang, Shui Yan (committee member)

Creator Email renzhi1969@gmail.com,renzhish@usc.edu

Unique identifier UC11399CW4Q

Identifier etd-ShenRenzhi-13606.pdf (filename)

Legacy Identifier etd-ShenRenzhi-13606

Document Type Dissertation

Format theses (aat)

Rights Shen, Renzhi

Internet Media Type application/pdf

Type texts

Source 20241104-usctheses-batch-1220 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu