The Effects of Fuel Standards on Air Pollution: Evidence from China Pei Li, Yi Lu and Jin Wang JEL Codes: L51; Q53; Q58

ABSTRACT This paper examines the causal relationship between China's fuel standards, which specify lower sulfur content, and air pollution. Combining measures of prefectures' regulations with hourly station-level pollution data from 2013 to 2015, we show that the enforcement of high-quality gasoline standards significantly improved air quality, especially in terms of fine particles and ozone. The average pollution across all pollutants was reduced by 4%. The new gasoline standard's net benefit is estimated to be about US$6 billion annually. These findings demonstrate the efficacy of precise standards in reducing air pollution in a developing country setting. Pei Li Department of Public Finance and Wang Yanan Institute for Studies in Economics, Xiamen University, Xiamen 361005, China. Email: bengcome@gmail.com. Yi Lu School of Economics and Management, Tsinghua University, Beijing, 100084, China; Department of Economics, National University of Singapore, 1 Arts Link, Singapore, 117570. Email: luyi@sem.tsinghua.edu.cn. Jin Wang Division of Social Science, Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong. Email: sojinwang@ust.hk. Data and codes for replication are available at:   https://www.dropbox.com/s/3qdukzz04c5997i/ReplicationData.rar?dl=0

The Effects of Fuel Standards on Air Pollution:

Evidence from China∗

Pei Li†

Xiamen U

Yi Lu‡

Tsinghua and NUS

Jin Wang§

HKUST

December 2018

Abstract

This paper examines the causal relationship between China’s fuel standards, which

specify lower sulfur content, and air pollution. Combining measures of prefectures’reg-

ulations with hourly station-level pollution data from 2013 to 2015, we show that the

enforcement of high-quality gasoline standards significantly improved air quality, espe-

cially in terms of fine particles and ozone. The average pollution across all pollutants

was reduced by 4%. The new gasoline standard’s net benefit is estimated to be about

US$6 billion annually. These findings demonstrate the effi cacy of precise standards in

reducing air pollution in a developing country setting.

Key words: Fuel standards; Sulfur content; Vehicle Emissions; Pollution; China

JEL Codes: L51; Q53; Q58

∗We would like to thank the Editor, Prof. Hunt Allcott and two anonymous referees for their commentsand suggestions that helped to substantially improve this paper. We also thank participants at the 2016Green Cities in Asia workshop and the 2018 China meeting of the Econometric Society for helpful discussions.Wang acknowledges the financial support of a research grant from Hong Kong University of Science andTechnology (no. SBI18HS06). All errors remain our own.†Department of Public Finance and Wang Yanan Institute for Studies in Economics, Xiamen University,

Xiamen 361005, China. Email: bengcome@gmail.com.‡School of Economics and Management, Tsinghua University, Beijing, 100084, China; Department of Eco-

nomics, National University of Singapore, 1 Arts Link, Singapore, 117570. Email: luyi@sem.tsinghua.edu.cn.§Division of Social Science, Hong Kong University of Science and Technology, Clear Water Bay, Hong

Kong. Email: sojinwang@ust.hk.

1

1 Introduction

92% of the world’s population lives in places where the average air quality is beyond the

World Health Organization’s suggested limit for pollutants. Those in Asia, Africa and the

Middle East are disproportionately exposed to high concentrations of air pollution (WHO,

2015). In part due to low quality fuel and numerous old and poorly maintained vehicles, road

transport has come to contribute from 12% to 70% of the air pollution in major developing

cities, imposing alarming health costs on the public (WHO, 2011). The sheer size of the

population and particularly the growing urbanization in less-developed countries have made

the welfare consequences of targeted environmental regulations—for example, to improve fuel

quality—of great importance. Yet, systematic empirical evidence on the extent to which such

policies can effectively address automobile pollution remains scant.

In an important contribution, Auffhammer and Kellogg (2011) examine the effects of

gasoline content regulations on ambient pollutant concentrations in the United States. But

evidence from developed economies may not readily transfer to the developing world, given

the different contexts and institutions. For example, developing countries face constraints

on their regulatory capacity to enforce fuel standards. There is also industry opposition,

and governments have limited financial resources and staff (International Council on Clean

Transportation, 2010). This study attempts to bridge this gap by documenting novel empir-

ical evidence and measuring the benefits of fuel standards in the largest developing country,

China.

More precisely, we investigate the relationship between China’s fuel standards, which

specify lower sulfur content, and levels of air pollution. There are three reasons for this

focus. First, China has become the world’s largest automobile market. There has been

an unprecedented increase in the country’s vehicle ownership. Intensive oil consumption

is associated with a large number of externalities (Parry, Walls, and Harrington, 2007), as

shown in Figure 1. China’s cities rank among the most polluted in the world (World Bank,

2007; Greenstone and Hanna, 2014; Chen et al., 2013; Cropper, 2010). That makes regulating

fuel content of great policy relevance. Second, uniform fuel standards have gradually been

introduced in Chinese cities. Their introduction provides a new opportunity to understand

the environmental implications of such initiatives. The step change in standards together

with accurate data allow overcoming methodological obstacles that have impeded progress

in the field.

[Insert Figure 1 Motor Vehicle Ownership and PM Emissions in China here]

Third, Chinese fuel standards generally follow those in Europe, Japan and the United

2

States. Very similar regulations facilitate meaningfully comparing policy effectiveness in-

ternationally. Apart from any insights into the role of policy in the Chinese setting, some

general lessons could emerge that enrich our views of fuel content regulations in a broader

context.

The study’s analyses exploit a compelling quasi-experiment: changes in standards for fuel

sold in Chinese cities. According to offi cial statistics from China’s Ministry of Environmental

Protection (MEP), exhaust frommotor vehicles contributes a quarter to a third of particulate

matter (PM) air pollution throughout the country (MEP, 2013). Against this background,

China has gradually been tightening its gasoline standards since 2013, followed by diesel

fuel standards upgrading in certain cities. The introduction of low-sulfur fuel has aimed at

reducing emissions from the motor vehicle fleet substantially, and the regulations have been

strictly enforced by the retailers.

The study’s analyses involve merging prefecture-level data on regulations with hourly

pollution data from 1,492 air quality monitoring stations in 337 prefectures for the three

years 2013 to 2015. In addition to the air quality index (AQI), a composite measure of

pollution, data on suspended particulates less than 10 µm in diameter (PM10) and less

than 2.5 µm (PM2.5) and on ozone (O3) concentrations are analyzed. Those pollutants

are particularly related to fuel composition, and are among the most harmful to human

health. To our knowledge, this study has been the first time that high-quality data on fine

particulates and ozone have been used to research Chinese environmental issues (Pope and

Dockery, 2013).

The study focuses on precisely estimating the effect on air pollution of the first change

in gasoline standards implemented during the period studied. Two estimation strategies are

applied. First, both temporal and geographic variations in the implementation of the new

gasoline standard are exploited to identify its effects. The empirical analysis compares daily

changes in the local concentrations of air pollutants between prefectures implementing the

new standard earlier (the treatment group) and later (the control group). The validity of the

difference-in-differences (DD) methods applied and the causal interpretation of the results

rely on the assumption that prefectures that adopted the new standard later are proper

counterfactuals for what would have happened to earlier adopters in the absence of the

change. A large number of other variables are included in the analyses, including monitoring

station fixed effects, day fixed effects, station-specific seasonality, weather conditions, winter

heating, fuel prices and subsequent changes in fuel quality standards. It is also necessary to

remove the confounding influence of other on-going policies aimed at curbing air pollution by

directly controlling for them. To further address the selected nature of the enforcement dates,

we control for any differences in the outcome trends between early adopters and later ones

3

depending on the key determinants in the timing of the fuel standard changes, a technique

used by Gentzkow (2006). Beyond that, a robustness test verifies that the treatment and

control prefectures are comparable in terms of their pre-regulation trends in the outcomes.

The study’s second set of analyses follows the lead of Davis (2008) and of Auffhammer

and Kellogg (2011). A regression discontinuity (RD) framework is used to identify the

effect of gasoline standards on air pollution. After the new regulations came into force, all

the retailers in the treatment prefectures were assumed to have immediately switched to

supplying only gasolines meeting the new standards. That is assumed to have created a

sharp discontinuity in tailpipe emissions. In effect, the sharp change makes other factors

relatively smooth on the implementation day, so the day just before the new regulations

serves a good counterfactual to the day the new regulations came into force.

The analyses yield several main results. First, the enforcement of more stringent fuel

standards significantly improved air quality. The average level of pollution across all pollu-

tants was reduced by 4%. The progression from a lower to a higher standard led to a 4.3%

fall in the concentration of PM2.5, an average reduction of 2.7 µg/m3. Such improvement

points to the importance of fuel standards in mitigating vehicles’environmental externalities.

Second, there are some intra-day fluctuations in pollution levels, but they are well ex-

plained by the atmospheric chemistry of the pollutants, which tends to validate the main

results. Local governments’overall environmental protection efforts are shown to be an im-

portant factor in achieving the environmental benefits of a better fuel. The new standards

are more effective in cities with more active government pollution reduction efforts.

Third, a back of the envelope analysis suggests that the annual net benefit of adopting

the new gasoline standard could have been about US$6 billion. This estimate is based

on epidemiological studies from China which suggest that a 2.7 µg/m3 reduction in PM2.5

concentration translates into a predicted 47,527 lives saved annually. Applying statistical

life values for China suggests that the improved air quality yields US$7.67 billion annually

in health benefits from reduced mortality. There are also US$0.13 billion annually in health

benefits from reduced morbidity. The resulting gain in labor productivity is US$2.2 billion

annually. The cost of the upgrading to the consumers was about US$3.99 billion.

This paper contributes to several strands of literature assessing the impact of environ-

mental regulations. It relates to a growing body of scholarly work that emphasizes the

technological aspects of environmental policies (Copeland and Taylor, 2004). Scholars have

previously demonstrated negative relationships between regulatory measures that tightened

vehicle emission standards and air pollution outcomes (Kahn, 1996; Kahn and Schwartz,

2008; Greenstone and Hanna, 2014).1 While the impact of vehicle emission standards on

1Kahn (1996) as well as Kahn and Schwartz (2008) found that both government regulation and auto

4

air pollution takes effect only gradually through turnover in the vehicle fleet, the results of

this study show that the new fuel standards in China can affect all vehicles on the road and

influence air quality dynamics immediately as well as (presumably) over the long term.

Closest in spirit to this work is that published by Auffhammer and Kellogg (2011). They

showed that the effectiveness of American gasoline content standards depends on flexibility

in choosing a compliance mechanism. Flexible federal gasoline standards did not improve air

quality, but accurately targeted, inflexible regulations in California did yield significant im-

provements. These results from China echo theirs and highlight the role of precise standards

in reducing air pollution. Auffhammer and Kellogg note that recent federal and Californian

gasoline regulations began to restrict the sulfur content, but they chose to focus on study-

ing the early regulation of volatile organic compounds and nitrous oxides produced through

burning gasoline. This study extends their work to sulfur content requirements, a key fuel

quality parameter. It also extends their work to China and includes for the first time PM2.5

levels as a main outcome.

Another strand of research has focused on regulations designed to reduce the scale of

pollution. In particular, several studies have examined regulations targeting gasoline con-

sumption and emissions (Parry, Walls and Harrington, 2007; Jacobsen, 2013; Anas, Timilsina

and Zheng, 2009).2 Others have looked at the relationship between driving restrictions and

air quality, including studies by Davis (2008), Wolff (2014), as well as Viard and Fu (2015).3

While administrative restrictions on the use of the vehicle fleet are known to be very costly

and compliance is known to be a critical issue, perhaps more emphasis should be placed on

cleaner technologies, of which cleaner fuel is a prominent example. The findings of this study

suggest that improving fuel standards could be an effi cient policy tool, since compliance can

be more strictly enforced.

The rest of the paper proceeds as follows. Section 2 lays out the fuel standard reform

background, followed by a description of the empirical strategy and data in section 3. Section

4 presents the empirical results. The last section concludes.

manufacturers’innovations significantly reduced regional air pollution caused by driving in the United States.The environmental impact of vehicle regulation should increase with time as the share of pre-regulationvehicles on the roads declines. Greenstone and Hanna (2014) have shown that India’s air pollution regulationsrequiring catalytic converters on new vehicles improved air quality and reduced death rates.

2Parry, Walls and Harrington (2007) have shown that among a set of American policy initiatives, in-creasing gasoline taxes more effectively reduced a greater number of important externalities than did fueleconomy standards. Further advances have been made by Jacobsen (2013), who has specifically studied themechanisms and welfare implications of fuel economy standards. Anas,Timilsina and Zheng (2009) com-pared the effectiveness of a congestion toll and a fuel tax in reducing traffi c congestion as well as gasolineconsumption and emissions in Beijing.

3They have shown that the policy outcome varies depending on the context, from no effects in Mexico,to significant effects in Germany and Beijing. Such differences crucially depend on the behavioral responsesof the drivers infiuenced (compliance versus compensating responses).

5

2 Fuel Standards in China

2.1 Fuel Content, Vehicle Emissions and Pollutant Formation

It has been well documented that gasoline’s sulfur concentration is among the most relevant

determinants of vehicle emissions.4 Higher-sulfur gasoline generates more sulfur compounds

in the exhaust and significantly impairs emission control systems (EPA, 2000; 2014). Gaso-

line vehicles depend crucially on catalytic converters to reduce tailpipe emissions of harmful

pollutants such as nitrous oxides (NOx), carbon monoxide (CO), and volatile organic com-

pounds (V OCs) that include precursors for ozone and secondary PM in the atmosphere.

There is ample evidence that the presence of even a tiny amount of sulfur in fuel has a measur-

able impact on catalyst effi ciency, known as sulfur inhibition.5 The consequent degradation

of catalyst effectiveness causes a vehicle’s emissions to significantly exceed its full useful life

emission standards. Moreover, this negative effect is known to be irreversible for one or

more pollutants, implying a profound and long-term impact on emissions of high sulfur fuels

(EPA, 2000, 2014).6

A large number of epidemiological studies have linked air pollutants with adverse health

effects. PM2.5 in particular pose a great risk to humans because the particles can penetrate

deep into the lungs and remain there for long periods. They diffuse readily into indoor

environments, and are transported over long distances (Pope and Dockery, 2006). They are

associated with increased incidence of lung cancer and with respiratory and heart disease

mortality, and are known to aggravate asthma seriously (Pope and Dockery, 2013; Green-

stone, 2004; Parry, Walls and Harrington, 2007). Ozone is another important air pollutant

which damages both human health and agricultural crops.

It is noteworthy that the air pollutant concentrations observed at different locations

depend on more than the quantity of various emissions. Extensive studies have shown that

wind speed, temperature and rainfall all play important roles in determining local pollutant

levels. Indeed, most air pollutants, including ozone and fine particulates, exhibit pronounced

seasonal patterns because of weather (Bharadwaj et al., 2017). It is therefore important to

control for weather and seasonality in any attempt to identify the impact of fuel standards

4Sulfur content is also one of the most important characteristics affecting diesel vehicles’NOx and PMemissions (International Council on Clean Transportation, 2010). During combustion, sulfur in diesel fuelsconverts into direct particulate matter and sulphur dioxide emissions, which can lead to secondary particleformation.

5Sulfur and sulfur compounds attach or "adsorb" to the precious metal catalysts that are required toconvert these emissions. Sulfur also blocks sites on the catalyst designed to store the oxygen that is necessaryto optimize the conversion of emitted NOx (EPA, 1998, 1999, 2000, 2014).

6The sulfur build-up on the catalysts due to past exposure to high sulfur fuels may be irreversible evenwhen current fuels have lower sulfur levels.

6

on local pollutant concentrations.

2.2 Fuel Content Regulations

China has primarily followed the European Union’s fuel standards since the late 1990s. After

a decade of practice in addressing worsening air pollution, especially in urban areas, China

decided to strengthen the hazardous materials control standards for vehicle gasoline and

diesel. In May 2011 the China IV gasoline standard was issued specifying a maximum sulfur

content of 50ppm (parts per million) of sulfur in gasoline. It had been phased in by the end

of 2013. In early 2013 the State Council issued a further directive calling for the nationwide

introduction of ultra-low-sulfur fuels (10ppm) by the end of 2017 (State Council, 2013).

China’s eastern coastal cities and mega-cities served as the starting point, before extending

the directive to other areas. That directive was translated into formal regulations over the

course of 2013 (International Council on Clean Transportation, 2013, 2014). Ultimately,

three new standards were issued: China IV diesel (50ppm) in February 2013 to be phased in

by December 31, 2014; China V diesel (10ppm) in June 2013 to be phased in by December

31, 2017; and China V gasoline (10ppm) in December 2013 to be phased in by December 31,

2017.7 Together, those standards constituted a road map for improving the fuel burned in

road vehicles nationwide, as shown in Table 1.

[Insert Table 1 Evolution of China’s Fuel Standards here]

China’s provinces have since been revising their regulations to implement the new fuel

standards. The provincial authorities (including the Department of Environmental Pro-

tection and Transport, the Development and Reform Commission, and the Economic and

Information Technology Commission) were deeply involved in this process. They consulted

various stakeholders such as the provincial branches of Sinopec and PetroChina, the major

state-owned oil companies which dominate the Chinese market and run both refineries and

retail stations, and also automobile manufacturers, research institutes and quality inspection

institutions. In deciding on local implementation, limiting exhaust emissions from road vehi-

cles was an important consideration, but local refining capacity was also a critical factor. The

new, higher-quality fuel would have to be refined and supplied without shortages.8 Some

7In addition to sulfur reductions, the progression from China III to China V gasoline standards has in-volved a reduction in maximum permitted manganese levels and reductions in minimum octane requirements.The progression from China III to V diesel standards has involved changes in the required cetane content.

8For example, the Xinjiang Autonomous Region started to implement the gasoline IV standard fromDecember 21st, 2013. By the end of November, five local Sinopec and PetroChina refineries had finishedupgrading their facilities and could supply Xinjiang’s local retailers after the regulations became effective.

7

provinces have moved faster than others. Each of China’s twenty-nine provinces decided

its own effective date for the standards and then applied them simultaneously in all of its

prefectures. Jiangsu and Guangdong provinces chose to extend the new standards to their

prefectures gradually. A few major prefectures were selected to apply the new standards first,

followed by the rest in a second stage. Although the prefectures and regions were allowed

to implement the fuel quality standards according to their own timelines, fuel price changes

are tightly regulated by the National Development and Reform Commission (NDRC), the

country’s top economic planner. To compensate for the required refinery upgrades and the

increased cost of producing the cleaner fuels, the NDRC increased the wholesale prices of

China IV gasoline and diesel by U290 and U370/ton, respectively. The prices of China Vgasoline and diesel were raised by a further U170 and U160/ton. When implementing thenew standards, retail stations raised their gasoline and diesel prices accordingly.

Figure 2 shows the gradual upgrading. By the end of 2013, 25.5% of prefectures had

implemented the gasoline IV standard, and 1.5% of prefectures had adopted diesel IV. Those

proportions had increased to 100% and 22%, respectively, by the end of 2014. By 2015 all of

China’s prefectures were supplying only gasoline IV and diesel IV. As for the ultra-low-sulfur

fuels, the gasoline V and diesel V standards are still in the process of implementation. By

2013, 3% of prefectures had started to supply gasoline V, while only 0.3% of prefectures had

adopted diesel V. Those proportions increased to 12.5% and 3.9% over the following year,

and by 2015, 14.5% of prefectures had adopted gasoline V and 14.2% of prefectures had

adopted diesel V.

[Insert Figure 2 Evolution of Gasoline and Diesel Content Regulations in China here]

2.3 Chinese Standards in a Global Context

Table 2 displays China’s gasoline standards in an international context. Specifically, the

Chinese standards are compared with those in Europe, California and the rest of the United

States.

[Insert Table 2 International Gasoline Standard Comparison here]

The European standards include specifications for lead and sulfur content in gasoline,

and diesel’s cetane number and sulfur content. The standards are periodically updated,

usually to reduce the allowed sulfur contents. A timeline for achieving sulfur level reductions

in Europe was established in 1999. In 2000, the Euro 3 standard mandated a maximum of

150ppm of sulfur in gasoline and 350ppm in diesel. From 2005, Euro 4 mandated a maximum

of 50ppm of sulfur in both gasoline and diesel. In 2009, Euro 5 mandated a maximum of

8

10ppm of sulfur in gasoline.

The US has been particularly concerned about reducing ozone pollution. The national

Environmental Protection Agency (EPA) created a federal reformulated gasoline (RFG)

standard, and California’s Air Resources Board (CARB) has since the 1990s had its own

regulations dealing primarily with V OCs and NOx emission limits. The RFG standard

caps the benzene content of gasoline at 1 percent by volume. Limits on total evaporation of

V OCs, toxic air pollutants and NOx emissions from gasoline were set based on a complex

model, but the RFG standard grants refiners flexibility in deciding which specific V OCs to

remove from their gasoline (Auffhammer and Kellogg, 2011). By contrast, the CARB has

mandated reductions in the concentrations of highly reactive V OCs that are more strin-

gent than the RFG standard. It established limits for sulfur, benzene, olefins, aromatic

hydrocarbons, oxygen, Reid vapor pressure and other parameters. Since the early 2000s, the

EPA regulations have started to shift their focus to sulfur requirements, slightly later than

Europe. Under the Tier 2 program, refiners and importers of gasoline have been given an

overall sulfur cap of 300ppm, with an annual average sulfur level of 120ppm in 2004. The

standard was reduced to a 30ppm average with an 80ppm cap in 2006. Since 2017, the Tier

3 program lowers the sulfur content of gasoline to a maximum of 10ppm. Evidence has yet

to be developed on the effectiveness of these restrictions.

Table 2 shows that China’s fuel quality standards mostly, but not entirely, follow Euro-

pean precedents. Both the Chinese and European fuel specifications pay particular attention

to sulfur content. A central question is how well low-sulfur fuel addresses air pollution prob-

lems. A prospective examination of their environmental impact in China is the main goal of

this study.

3 Estimation Strategy

3.1 Data

Analyzing the observed changes in pollution levels in response to changes in fuel standards in-

volved assembling a data set containing matched indicators of fuel standards implementation,

fuel prices, air pollution, weather conditions, socioeconomic conditions in the prefectures and

vehicle registration records.

The fuel standards information was assembled from circulars issued by the provincial

Development and Reform Commissions implementing standards IV and V in the various

provinces. The fuel price data were compiled from NDRC circulars.9 The fuel market is

9Source: http://www.sdpc.gov.cn/zcfb/zcfbtz (in Chinese).

9

not fully integrated in China due to geography, distance, and different levels of regional

development. There are regional price variations. Twenty-four provinces apply a province-

wide price in all of their prefectures. Another six provinces apply prefecture-specific prices.

Nationally, however, fuel price changes are tightly regulated by the NDRC. During the

period studied, the national planners adjusted fuel prices 52 times in response to changes

in international crude prices. In the circulars they issued they specified incremental price

changes and effective dates for all of the provinces.

Air pollution data are published hourly and daily by the MEP. The data for 2013 to 2015

came from 1,492 monitoring stations in 337 prefectures. Following the implementation of

new ambient air quality standards (MEP, 2012), data on fine particulates and O3 became

publicly available in for the first time in 2013. An AQI was developed based on the hourly

and daily observations of SO2, NO2, CO, PM10, PM2.5, and ozone. This was a notable shift

from the previous index which considered only SO2, NO2, and PM10. All 337 prefectures

were required to disclose their once-classified air quality data beginning in 2015. The AQI

scale ranges from 0 to 500. It is further divided into six ranges: 0—50, 51—100, 101—150, 151—

200, 201—300 and 301—500. In public reports these are termed good, moderate, unhealthy

for sensitive groups, unhealthy, very unhealthy, and hazardous, respectively.

Those Chinese data were supplemented by weather data from the China Meteorological

Data Sharing Service System (http://cdc.cma.gov.cn/). Those data contains daily readings

from 820 weather stations in China during the 2013 to 2015 period. The meteorological vari-

ables were aggregated to the prefecture level by averaging the daily readings from all of the

weather stations within a prefecture. The indicators used were temperature, precipitation,

humidity, duration of sunshine and wind speed. Weather stations for which there were fewer

than 100 records per year were winsorized.

A comprehensive data set of monthly new car registrations by model is compiled by the

State Administration of Industry and Commerce. It documents the month and prefecture

of sales, the brand and model of each vehicle registered, as well as major attributes such as

fuel type, nominal fuel consumption rate, power, and engine size. The total number of new

cars registered from 2013 to 2015 was 65.42 million. We aggregate the individual data to the

month-prefecture level for the purposes of this study and use them to check the robustness

of the findings of the RD analysis.

The prefecture-level variables in Table 3 were extracted from the China Statistical Year-

book for Regional Economy 2008, supplemented by the China Urban Construction Statistical

Yearbook for 2008. The data about the refineries were obtained from China’s Annual Survey

of Industrial Firms for 2007, which covers all state-owned enterprises (SOEs) and non-SOEs

with annual sales of more than five million yuan. The indicator for local government’s overall

10

environmental efforts was constructed from 3,432 government annual work reports from the

286 prefecture-level cities from 2001 to 2012.

Information about exhaust emission regulations were obtained from the offi cial docu-

ments published by prefectural transportation and public security bureaus. Those documents

clearly state the exact dates when the prefectural governments began enforcing exhaust emis-

sion standard IV for gasoline vehicles and required the scrapping of yellow label vehicles,

which are defined as gasoline vehicles not meeting Chinese gasoline standard I and diesel

vehicles not meeting Chinese diesel standard III.

Detailed variable definitions and descriptive statistics are presented in Table 3.

[Insert Table 3 Variable Definition and Descriptive Statistics here]

3.2 Estimation Framework

To quantify the effects of the fuel content regulations on air pollution, we employ two esti-

mation techniques: a DD specification, and an RD framework.

3.2.1 The DD Framework

Temporal and regional variations in the regulations are exploited to conduct a DD estimation.

Specifically, air pollution outcomes in prefectures which had implemented new fuel content

regulations are compared with those in prefectures which had not (the first difference) before

and after the implementation of the regulations (the second difference).

We choose to focus on the shift from gasoline standard III to standard IV for the analysis

for several reasons. First, standard III had been implemented throughout China by 2009,

so there had been two years without any changes in the fuel content regulations before the

upgrading to gasoline standard IV. That provides a relatively long window to check the

comparability between treatment and control groups in the pre-treatment period. Second,

the sequence of later fuel standard reforms was highly correlated with the gasoline standard

IV reform. As a result, even if the treatment and control groups were comparable before

the gasoline standard IV reform, that change would make the two groups different given

the potential policy impact. That would violate the parallel trend assumption for the later

reforms, making it more diffi cult to quantify any air pollution effects.

To contain any possible contamination of the gasoline standard IV reform’s effect from

the later fuel content reforms, those later reforms are included as explicit controls in the DD

11

estimations. Specifically, the DD specification is:

yscd = β ·Gasoline4cd + ρLPolicycd + λs + λd + γXscd + εscd, (1)

where s, c, and d denote monitoring stations, cities, and days, respectively. yscd is the loga-

rithm of daily average pollutant concentrations: the AQI, PM2.5, PM10, andO3. Gasoline4cdis a dummy variable indicating whether city c has upgraded to gasoline standard IV by day d.

λs is the set of station fixed effects, controlling for all time-invariant variations across moni-

toring stations within a city, including topographic features; λd is the set of day fixed effects

controlling for the daily shocks common to all cities (e.g., monetary policy and exchange

rate changes); and εscd is the error term. The standard errors are clustered by station.

LPolicycd = {Gasoline5cd, Diesel4cd, Diesel5cd}, where Gasoline5cd is a dummy vari-able indicating whether city c has upgraded to gasoline standard V by day d; Diesel4cd and

Diesel5cd are dummy variables analogously defined.

The parameter of interest is β. A negative value would reflect better quality gasoline’s

reducing air pollution. We first estimate equation (1) separately for AQI, PM2.5, PM10, and

O3. To capture the average effect of the gasoline standard IV reform across pollutants, we

then stack the four pollutants along with the pollutant dummies as controls.

Such identification exploits daily variations among cities, so any potential bias could

only have arisen from omitted variables on the day level. One primary threat is seasonal-

ity. Specifically, if the implementation of new gasoline and diesel content regulations corre-

sponded with specific weather conditions, the estimates could be mistakenly attributed to

policy effects. While the day fixed effect has effectively controlled for all national average sea-

sonality, station-specific seasonality is addressed by including three sets of controls in Xscd:

λs×Day of Weekd, where Day of Weekd = {Monday, Tuesday, ..., Sunday}; λs×Weekof Monthd, whereWeek of Monthd = {Week1, ...,Week5}; and λs ×Month of Yeard,whereMonth of Yeard = {January, ..., December}. As weather conditions are well knownto significantly influence pollution levels, concerns about station-specific weather conditions

are further dealt with by adding a series of weather variables in Xscd: Temperaturecd,

Rainfallcd, Wind Speedcd, Sunshine T imecd and Humiditycd. In addition, winter heating

in Northern China is known to significantly increase air pollution (e.g., Almond et al., 2009;

Chen et al., 2013; Ebenstein et al., 2017), so an indicator for the winter heating period in

each city, WinterHeatingcd is included. Finally, as fuel prices changed a number of times

during the period studied, a Fuel Pricecd term is included in Xscd to isolate the effects of

the regulatory changes.

A second threat to the identification is that if there were other on-going reforms around

12

the time of the gasoline standard IV reform the estimates might also reflect those confounding

factors. To address this concern, government documents setting out environmental protec-

tion policies around the time of the gasoline standard IV reform are examined carefully. A

prominent policy change was the change in vehicle emission regulations, with two measures

targeting new and old vehicles separately. Exhaust emission standard IV was imposed for

new gasoline vehicles. For old vehicles, China adopted one of the world’s most ambitious

voluntary scrapping programs aimed at gasoline vehicles not meeting Chinese gasoline stan-

dard I and diesel vehicles not meeting Chinese diesel standard III (termed “yellow label

vehicles”). Two additional controls were included to remove any confounding effect of those

changes. Gasoline Car4cd indicates whether city c had adopted exhaust emission standard

IV for gasoline vehicles by day d, and Y ellow Carcd indicates whether city c had started to

phase out yellow label vehicles.

Third, provinces differed in their speeds of implementing gasoline standard IV based

on certain factors such as a city’s location and size, pre-existing exhaust emissions from

road vehicles, and the capacity of local refineries. If such factors differentially affected the

treatment and control cities in the post-gasoline standard IV period, the estimates of β would

be biased. To address this concern, we use the approach adopted by Gentzkow (2006), which

explicitly controls for flexible time trends in air pollution caused by the early fuel upgrading

to isolate the treatment effect. Specifically, we consider a vector of potential influencing

factors Zc, which includes the distance to the coast, city size, road density, cars per capita,

and the number of local refineries. We then include the interactions of that vector with the

day dummies λd as additional controls in the analysis.

To further verify the identifying assumptions of the DD analysis, we examine whether

the parallel trends assumption holds in this setting. Specifically, we use the stacked data

and conduct a flexible DD estimation based on regression specification (1). We estimate the

coeffi cients for a set of bi-monthly dummy variables indicating every two months within the

event window and extending 8 months prior to and 10 months after the implementation of

the gasoline IV regulation.

3.2.2 The RD Framework

The DD estimates might still have suffered from some incomparability between the treatment

and control cities, so the RD technique of Davis (2008) and of Auffhammer and Kellogg

(2011) is also applied. Regression discontinuity analysis focuses on a narrow window around

the policy change in which unobservables are allowed to act nonlinearly so long as they move

smoothly at the time of the reform. The RD technique is applicable here because gasoline

standard IV was imposed on all vehicles simultaneously, generating a discontinuous change

13

in their emissions, and the formation of pollutants responds quickly to changes in emissions.

Hahn, Todd and Van der Klaauw (2001) have shown that β can be identified as

β = limd↓dc0

E [yscd|dc = d]− limd↑dc0

E [yscd|dc = d] = β̂RD. (2)

We estimate β̂RD using a nonparametric approach, specifically, local linear regression, as

suggested by Hahn, Todd and Van der Klaauw (2001). β̂RD is estimated from

minα,β,δ,τ

N∑s=1

K

(dc − dc0

h

)[yscd − δ − τ (dc − dc0)− βEc − αEc (dc − dc0)]2 , (3)

where Ec takes a value of 1 if dc ≥ dc0 and 0 otherwise; h is the bandwidth; and K (.) is a

rectangular kernel function.

Essentially, RD compares for each station the outcomes just before the gasoline content

IV reform with those just after the reform. Since the reform was carried out in different

locations at different times, station fixed effects must also be controlled for (which then

restricts comparisons within the same station, before and after the reform), along with

station-specific seasonality (i.e., λs×Day of Weekd, λs×Day of Weekd, λs×Month ofYeard), station-specific weather conditions and fuel prices that potentially changed abruptly

with the reform. The estimation involved two steps. First the outcome yscd is regressed

against these controls to obtain a residual∼yscd. That value is used in the second step, the

nonparametric estimation (3) to obtain the parameter of interest β̂RD.

We calculate the optimal bandwidth h using the method developed by Imbens and Kalya-

naraman (2012). To check whether the results are sensitive to the optimal bandwidth se-

lected, alternative bandwidths are tested (see, e.g., Carneiro et al., 2015 for details). To

account for any serial correlation within a city over time, we first estimate standard errors

clustered by station. We also report the standard errors adjusted to reflect spatial depen-

dence as modelled by Conley (1999).

4 Findings

4.1 DD Estimates

Table 4 reports the results from the DD estimation of equation (1) showing the effect of fuel

standard regulations on the logarithm of daily average AQI, the concentrations of PM2.5,

PM10, and O3 and the stacked pollutants. The coeffi cients for Gasoline4cd are precisely

estimated. Column 1 indicates that the gasoline IV standard improved the AQI by about

14

4.1% on average (a drop of 3.8 units). Table A1 of the online Appendix presents the estima-

tion results with fewer controls (just station fixed effects, day fixed effects, station-specific

month seasonality and gasoline standard V). The results are quantitatively the same but

with smaller R-squared values.

[Insert Table 4 DD Estimates of Effects of Fuel Standard Regulations here]

Columns 2 to 4 confirm the contribution of cleaner gasoline fuels to reducing particulates

and ozone, the major components of smog and haze in Chinese cities. As column 2 shows,

the progression from gasoline standard III to IV caused an average 4.3% reduction in PM2.5

concentration. Given that the average PM2.5 concentration in the pre-treatment period

was 63.15 µg/m3, the estimate implies that implementing standard IV caused an average

2.7 µg/m3 reduction in PM2.5 concentration. Similarly, column 3 suggests that the higher

quality gasoline reduced PM10 concentrations by 4.1%, equivalent to a total reduction of

4.3 µg/m3 on average from the mean pre-treatment PM10 concentration of 104.6 µg/m3. In

column 4, ozone concentration has reduced by 5.5% (3.0 µg/m3). The findings for ozone

are consistent with those of Auffhammer and Kellogg (2011) who find that only precisely-

targeted, inflexible regulations requiring the removal of particularly harmful compounds can

significantly improve air quality.10

While columns 1—4 estimate the effects of fuel content regulations separately for each

pollutant, pooling all the pollutants and additionally controlling for pollutant dummies al-

low capturing the average effect of the gasoline standard change across multiple pollutants.

The estimation results are reported in column 5. The coeffi cient remained negative and

statistically significant, with an approximately 4% reduction.

Parallel Time Trends. Despite the exhaustive set of controls included in the analyses,

there may still be some concern about the comparability between the treatment and control

groups central to validity of the DD estimation. One validity check commonly used is to

examine whether the treatment and control groups show parallel pre-treatment trends. To

this end, we use the pre-treatment data for the 8 months leading up to the implementation

10Note that these estimates may just reflect the short-run impact of improved fuel quality. Given thepossible complementarity between fuels and engines, lower sulfur fuel might generate longer-run air qualityimprovements via vehicle emissions control equipment working better and lasting longer and through turnoverin the new vehicle fleet, an indirect impact which will gradually build up. Meanwhile, the fuel standardsmay have also spurred consumer demand, creating a larger market for higher-quality catalytic converters andimproved engine controls. This, in turn, might have generated endogenous technological innovation in thecar market and the adoption of more advanced emission technologies (Bresnahan and Yao, 1985; Acemogluand Linn, 2004). Any such effects are likely to have further enhanced the pollution reductions in the longterm.

15

of the gasoline IV regulation, when there were about 400 monitoring stations. We also

explore the post-treatment data up to 10 months after the implementation of the gasoline

IV regulations, up to about the time that cities started to further upgrade their standards

to gasoline V. Specifically, we replace β in the DD specification (1) with a series of βts for

every two months within the event window.

Estimated coeffi cients along with the 95% confidence intervals are presented in Figure 3.

There were some fluctuations, but no clear signs of a deterministic trend in the air pollution

before the new gasoline IV standard took effect. Moreover, there was a significant reduction

in air pollution within the six months after the policy implementation. Afterwards, the

pollution mitigation effects of the policy were still significant, though smaller.

[Insert Figure 3 Tests for Parallel Trends here]

One potential concern with the pre-treatment trends of Figure 3 is the sudden drop in

air pollution two months before the gasoline IV was implemented. Note that a negative

policy effect on air pollution has already been detected, so an abrupt decrease in pollution

levels pre-treatment may lead to an under-estimation of the true policy effect. To contain

any potential estimation bias from that drop, we control for the flexible time path of the

dependent variables whose correlation with gasoline IV regulation is driven by the pattern of

the pre-treatment pollution. Specifically, following Gentzkow (2006), we include in the DD

estimation (1) terms representing an interaction of the average pollution value during the

two-month period before the policy implementation with a cubic function in time f(t). t is

defined as 0 on days before the implementation, as 1 on the implementation day, as 2 on the

first day after the implementation, and so on until the gasoline V standard was implemented.

The results with the inclusion of the interactions between the pre-treatment air pollution

and a third-order polynomial f(t) are presented in Figure A1 of the online Appendix. The

treatment and control groups show smooth parallel trends before the implementation of

the gasoline IV standard. There was a significant drop in air pollution within six months

after the standard’s implementation, consistent with Figure 3. The coeffi cients beyond six

months after the gasoline IV regulation are close to zero. That could be due to the standard’s

complete implementation and the first upgrading to gasoline standard V. Regression results

controlling for the trends are reported in column 6. The negative and statistically significant

policy effect persists, with a larger magnitude.

Selection of implementation timing. The speed of implementing the gasoline standard

IV differed among the provinces (and cities). If the factors determining the timing of the

implementation also affected the treatment and control cities differently, that too would

16

bias the estimation. To address this potential concern, interactions between the treatment-

determining factors Zc and the day dummies λd are included in the analysis. The estimation

results are reported in column 7. The estimated coeffi cient remains unchanged, suggesting

that the results are not biased due to the timing of the standard’s implementation.

4.2 RD Estimates

Consider first the visual relationship between a normalized time variable (the assignment

variable; d̃c = dc−dc0) and a set of pollutant measures. Imbens and Lemieux (2008) and Leeand Lemieux (2010) suggest that the selection of bin size trades off precision in calculating

average outcome values against proximity to the cutoff point. Figures 4a-4e report the RD

figures using a bin width of two weeks.11 Specifically, we collapse the sample to bi-weekly,

and use the residuals of the regression on the controls described in Section 3.2.2. The circles

in the figures represent mean values for each two-week bin; the lines indicate the fitted values

from a local linear regression with the optimal bandwidth calculated using the method of

Imbens and Kalyanaraman (2012). The grey areas are the 95% confidence intervals, and

the vertical line is the cutoff point for the assignment variable. With this bin width, all five

plots are smooth on either side of the cutoff value, but at the same time show a clear drop

in pollutants at the cutoff point, consistent with the DD estimates.

[Insert Figure 4 Bi-weekly RD Estimates of the Effects of Fuel Standards here]

The regression results using the daily data are reported in Table 5.12 The estimates of β,

the coeffi cient of interest, are all negative and statistically significant in the five regressions,

further confirming the patterns of Figure A2 and the DD estimates in Table 4. The DD

and RD estimations use different control groups and different identifying assumptions, but

the consistent results between two estimation techniques lend support to the conclusion that

the new gasoline content regulations achieved their primary goal of reducing air pollution.

Regression results using the collapsed bi-weekly data are reported in Table A3 of the online

Appendix. Those results too are consistent with the patterns of Figure 4 and the RD results

using the daily data in Table 5.

[Insert Table 5 RD Estimates of Effects of Fuel Standards here]

11RD results using a bin width of 1 day are shown in Figure A2 of the online Appendix. Although thedata have on average of 991 observations per day, the plots are not smooth on either side of the cutoff value.12RD estimations with fewer controls (just the station dummies and station-specific month seasonality)

are reported in the online Appendix, Table A2. The results are similar but with smaller R-squared values.

17

Alternative bandwidth. Nonparametric estimation requires calculating an optimal band-

width. The method developed by Imbens and Kalyanaraman (2012) is used. To check

whether the findings might be sensitive to the optimal bandwidth selected, alternative band-

widths from h∗− 10 to h∗+10 in intervals of 2 are tested. Estimates using those alternativebandwidths are plotted in Figures 5a through 5e. The estimates remain stable, suggesting

that the results are not driven by a particular bandwidth.

[Insert Figure 5 Sensitivity Test on the Choice of Bandwidth here]

Price increase. Wholesale fuels prices were increased by the NDRC when each new fuel

was introduced to reflect the costs of fuel upgrading. This raises the possibility that the

discontinuity in fuel prices might itself constitute an omitted variable in the RD estima-

tion. For example, vehicle owners might have driven less in response to the price increase,

generating less air pollution. However, the price increased by only about 3% from gasoline

standard III to IV. Such a small change may not have had a significant impact on driving

behavior. Nonetheless, to check for any potential bias, we include fuel prices in all the RD

estimations. Also, we have collected monthly new vehicle registration records by prefecture

covering 2013 to 2015 (because there is no comprehensive data on daily driving and fuel

consumption). We then assess whether the fuel upgrade had any effect on the attributes

of the cars purchased to shed light on the regulations’effects on drivers’preferences about

driving and fuel consumption.

[Insert Table 6 Effects of Fuel Standard Regulations on Car Model Sales here]

As Table 6 shows, there is no significant decrease in total car purchase, in the rated fuel

consumption of the models purchased or in the other attributes modelled. These results

tend to confirm that reduced driving as a result of the price changes is not important in the

context studied.

4.3 Heterogeneous Effects

Some temporal and spatial differences in the results will now be used to discuss based on

the DD estimation, but the RD estimates display similar patterns (see Figure A3 and Table

A4 in the online Appendix).

Temporal heterogeneity. Hourly data on air pollution allow examining intra-day fluctu-

ations. To do so, we divide each day into eight time periods: 12—2 am, 3—5 am, ..., 9—11

pm. Figure 6 reports estimated coeffi cients for each time period separately. The coeffi cients

18

are quite stable; that is, the pollution reduction effects of the fuel content regulations per-

sist throughout the day. There are, however, some interesting intra-day fluctuations. For

ozone, the effect of fuel standards on its concentration seems to follow an inverse U-shaped

curve. Pollution mitigation starts to decline after daybreak and is at its weakest in the

mid-afternoon. The effect picks up again late in the afternoon. Sunlight and warm temper-

atures are essential elements in forming ground-level ozone, and they usually peak during

mid-afternoon. The regulations are also somewhat more effective in reducing particulates

in the afternoon and least effective in the evening. This is probably associated with daily

variations in the depth of the boundary layer and with other anthropogenic emissions (Zhang

and Cao, 2015). Taken together, the hourly estimates are consistent with the atmospheric

chemistry of the pollutants. This further validates the research design and reinforces the

credibility of the baseline results.

[Insert Figure 6 The Hourly Effects of Fuel Standards here]

Spatial heterogeneity. With such a vast territory, China exhibits significant regional dif-

ferences in local governance. Local authorities were deeply involved in implementing the

fuel standards, so their overall environmental efforts and regulatory attitudes were also im-

portant. Investigating them involves a thorough content analysis of each city government’s

annual work report from 2001 to 2012. The report is usually presented by the city’s mayor

at a Communist Party meeting early in the year. It sets goals and objectives for the city

governments’work for the upcoming year and the goals’ accomplishment is meant to be

supervised by local residents. Manually searching China Statistical Yearbooks, city govern-

ments’websites and local newspapers yields 3,432 reports from 286 prefecture-level cities.

The key words used are “environmental protection” (huan bao or huanjing baohu in Chi-

nese). We express the number of times these terms were mentioned as a fraction of the total

number of words in the report. A larger fraction is assumed to indicate a more active local

government role with respect to enforcing environmental regulations. The 286 prefectures

are ranked according to their fractions. The effects of the fuel standard on air quality are

then investigated for the top 30% and the bottom 30% separately.

The regression results are reported in Table 7. The stations in cities which mentioned

environmental protection more often did generally show larger effects of the higher-quality

fuels. Attitudes and efforts did apparently enhance the effectiveness of the fuel standard

regulations. This points to the importance of local governance in realizing the benefits of

environmental initiatives. This within-country exercise also helps enrich our understanding

of the differences in environmental policy outcomes often observed between developed and

19

developing countries. They can at least partially be attributed to differences in institutional

effectiveness.

[Insert Table 7 Spatial Variation in the Effects of Fuel Standards on Air Quality here]

4.4 Cost-Benefit Analyses

We apply the estimates in conducting “back of the envelope”benefit-cost analyses for the

fuel quality upgrading.

Benefits. The standards’primary benefits are assumed to be health improvements associ-

ated with air pollutant reductions, especially reductions in particulate matter (Wolff, 2014).

There is evidence from the European Union that PM is the most lethal air pollutant, with

an impact much greater than that of the second most deadly air pollutant, ozone (Watkiss et

al., 2005). The analysis therefore focuses on inferring any health benefits related to reducing

PM2.5 pollution.13

A number of recent epidemiological studies have examined the health impacts of PM2.5 on

both mortality and morbidity. Lei (2013) and Lei and Ho (2013) have systematically reviewed

the published studies conducted in mainland China, Hong Kong and Taiwan, with additional

reference to estimates from other countries.14 Those reviews contain rich evidence concerning

the effects of PM2.5 on both acute and chronic mortality, as well as cardiovascular hospital

admissions, respiratory hospital admissions and outpatient visits. Indeed, Lei (2013) and

Lei and Ho (2013) have used concentration-response functions to estimate the relative risks

of mortality and morbidity in response to a 10 µg/m3 change in the PM2.5 concentration.

They estimate the ratio of the probability that the event will occur in the group exposed

to the higher concentration to that of a baseline group. Along with the base incidences in

the population of those health endpoints, the health outcomes of a policy can be calculated

by the following formula: (the relative risk − 1)× (the change in concentration of an airpollutant ÷ 10) × (the population exposed to the pollutant) × (baseline incidence of the

health endpoint).

Applying the estimates from column 2 of Table 2 (i.e., an average 2.7 µg/m3 reduction

13Many studies have examined the health effects of PM10 exposure in China, but much less attention hasbeen paid to PM2.5, largely due to data availability (see He et al., 2016; Viard and Fu, 2015). In fact, PM2.5

is known to be a better predictor of PM-driven acute and chronic health effects than the levels of coarserparticles (Schwartz et al., 1996; Cifuentes et al., 2000; Pope and Dockery, 2006; Matus et al., 2012).14Since disease prevalence, age structures, exposure patterns, health care systems, average ambient con-

centrations, and the chemistry of pollutants (in the case of PM) are different in various countries, one shouldideally use C-R functions from epidemiological studies done in the country or region concerned (Levy andGreco, 2007; Lei, 2013).

20

in PM2.5 concentration), the implementation of the gasoline IV standard can be estimated

to have been associated in terms of acute effects with 6,871 lives saved (among the urban

population of 711.82 million). There is a lack of cohort studies relating long-term exposure to

PM with the risk of premature mortality in China, so Lei (2013) used American data (which

may still overestimate the risk in China). We conduct a similar analysis in this study and

estimate that the fuel standard led to a predicted 40,656 fewer deaths annually associated

with chronic illness. The total lives saved (assuming the acute and chronic estimates are

additive) are estimated to be 47,527 annually.

The value of a statistical life (VSL) can be used to monetize such savings. The VSL is

the sum the average person would pay to reduce their risk of dying by small amounts that,

together, add up to one statistical life (World Bank, 2007). Multiplying the VSL by the

predicted number of lives saved through PM2.5 reduction monetizes what is assumed to be

the fuel policy’s primary benefit. Values from a set of studies of VSLs are used in Table 8

to calculate the estimated health benefits associated with air quality improvement. A 2.7

µg/m3 reduction in PM2.5 concentration implies US$7.67 billion in health benefits for China

from reduced mortality assuming the World Bank’s baseline value for a statistical life–

US$0.1613 million based on a contingent valuation approach. However, the World Bank’s

value of a life tends to be conservative. As incomes have increased, the willingness to pay for

risk reduction is likely to have risen as well (Costa and Khan, 2004). The Chinese economy

has been growing rapidly during the past few decades, so it might be preferable to use an

alternative VSL of US$0.6194 million suggested by Qin (Qin et al., 2013) based on a hedonic

wage approach. The health benefits related to adopting the gasoline IV standard would

then be US$29.44 billion. Using alternative estimates of health effects inferred from natural

experiments (e.g., the Huai River policy of Chen et al. (2013) or a study by He et al. (2016)

exploiting the 2008 Beijing Olympic Games) would yield even larger estimates.15 Yet even

with conservative ones, the policy seems to have significantly improved health as well as air

quality.

[Insert Table 8 Valuation of the Health Benefits here]

The implementation of the gasoline IV standard is associated with 11,916 fewer cardiovas-

15Chen et al. (2013) provide useful benchmark estimates of the effect of total suspended particles (TSP)exposure on mortality risk. Their analysis suggests that long-term exposure to an additional 100µg/m3 ofTSP is associated with a 14% increase in the overall mortality rate and a reduction in life expectancy at birthof about 3 years. Using a 32.5% ratio to convert TSP to PM2.5 based on current and historical Chinese data(Cao et al., 2011), Chen et al.’s estimates imply that a 32.5 µg/m3 reduction in PM2.5 is associated with adecrease of 0.000893 in the mortality rate. Applying the estimates from column 2 of Table 2 (an average 2.7µg/m3 reduction in PM2.5 concentration), the implementation of gasoline standard IV is associated with a0.000038 reduction in the mortality rate. With 711.82million living in China’s cities in 2012, that translatesinto a predicted 52,807 lives saved.

21

cular hospital admissions, 21,633 fewer respiratory hospital admissions, and 2,607,522 fewer

outpatient visits. There are no published willingness-to-pay values associated with morbid-

ity in China, so Lei and Ho (2013) used a simple cost of illness (COI) to calculate direct

medical costs with data from the Chinese Health Statistics Yearbook. A 2.7 µg/m3 reduc-

tion in PM2.5 concentration yields US$0.13 billion in health benefits from reduced morbidity

estimated in that way.

Pollution also affects labor productivity. Empirical evidence reveals that pollution has

subtle effects on workers by reducing how much is produced on the job across agriculture,

manufacturing and the service sectors. Estimating the productivity gains from reducing air

pollution involves three steps. Work by Graff Zivin and Neidell (2012) has shown that in

the agricultural sector, worker productivity increases by 0.28% when ozone is reduced by 1

µg/m3. In manufacturing, productivity increases by 0.1—0.6% (He et al., forthcoming; Chang

et al. 2016) when PM2.5 is 1 µg/m3 lower. In the services sector, worker productivity rises

by 0.04% when the API decreases by 1 unit (Chang et al., forthcoming). The potential

productivity gains associated with the new fuel standards are estimated building on that

work. The gasoline IV standard is estimated to have reduced PM2.5 by 2.7 µg/m3, ozone

by 3.0 µg/m3 and the AQI by 3.8 units. That would predict productivity increases of

0.84% in agriculture, 0.27—1.62% in manufacturing and 0.13% in the services sector. We

provide a back of the envelope calculation for the aggregate impact in monetary terms based

on these figures. Consider that the wage bills of the three sectors in China were U76.08billion, U3066.1 billion and U3949.2 billion in 2012. The improved air quality would thentranslate into US$0.1 billion, $1.3—7.9 billion and $0.8 billion increases in workers’earnings

(assuming an exchange rate of 6.31). So progression from gasoline standard III to IV in

China could have resulted in productivity gains of US$2.2 to 8.8 billion. It is noteworthy

that the mitigation of PM2.5 contributes most of the benefits.

[Insert Table 9 Valuation of the Labor Productivity Benefits here]

Costs. The new standards of course also have their costs. A study by the International

Council on Clean Transportation in 2012 considered refinery capacity, capital investment

requirements and refining costs and found that the additional cost to produce standard IV

gasoline in China would be just U0.02 (US$0.003) per litre. In 2012, gasoline consumptionwas 86.84 million tons (1 ton of gasoline = 1379 litres). A progression from gasoline standard

III to IV would then have induced cost increases of approximately U2.27 billion (or US$0.36billion at an exchange rate of 6.31).

On the other hand, the NDRC presents a more conservative view of the cost position

22

of the oil refineries. It announced that the wholesale price of China IV gasoline would be

increased by U290/ton (U0.21/litre).16 The costs related to fuel upgrading were to be sharedby the oil companies and consumers (roughly, 30% vs. 70%), but the cost increase was not

fully passed through to consumer prices, at least according to the NDRC. With a simplified

assumption that the price changes applied to all gasoline consumed without considering the

gradual expansion of the use of higher quality fuels across China, a progression from gasoline

standard III to IV would then have induced cost increases of approximately U25.18 billion(US$3.99 billion) for the consumers.

The envelope theorem can be applied to Chinese fuel consumers, but no simple envelope

theorem applies to producers, in part due to the market structure of Chinese oil refineries

being neither perfectly competitive nor purely monopolies (Weyl and Fabinger, 2013). For-

mally quantifying the loss to producers from fuel upgrading would be very diffi cult and we

choose not to pursue it (Ganapati et al., 2017).

Note that the NDRC’s announced price increases were significantly larger than the

ICCT’s carefully-crafted engineering estimates of costs. Were the price adjustments more

than adequate, or did they fall short of the required refinery upgrades and increased produc-

tion costs for China IV gasoline? The answer critically depends on which better reflects the

economic realities of the oil refineries (often criticized for being non-transparent) in China.

Nonetheless, these rough but conservative estimates suggest that the net benefits of adopting

the new gasoline standard are on the order of US$6 annually. The benefits are significantly

higher than the engineering costs, although the price burdens imposed on consumers are

more notable, possibly due to the current gasoline pricing regime in China.

Policy Comparisons. We compare the impact of more stringent fuel standards with

that of the driving restrictions applied in some Chinese cities (Viard and Fu, 2015), and

with that of California’s gasoline regulations (Auffhammer and Kellogg, 2011). In Beijing,

restricting the number of vehicles driven each day has led to a 21% reduction in PM10 (a

30.8 µg/m3 drop from an average level of 147 µg/m3), translating into 1,114 fewer deaths

and 15.3 million fewer restricted activity days annually. The estimated benefits are U3.05billion (upper bound) as opposed to costs of U519 million annually. Gasoline standard IVdecreased PM10 by 4.1% (4.3 µg/m3). That implies that while restricting the number of

vehicles driven each day is known to be very costly and compliance is diffi cult to enforce,

perhaps more emphasis should be placed on developing and implementing cleaner fuels and

more effective emission control technologies.

16The NDRC explained such increases were based on comprehensive investigations and audits of Sinopecand PetroChina refineries which had already been upgraded to produce China IV and V fuel, as well asexperience from Beijing and Shanghai.

23

The CARB’s gasoline regulations, by reducing ozone, translate into 660 lives saved in

California each year. Valued in 2008, the CARB imposed a cost of $1.2—1.6 billion per year,

or $1.8—2.4 million per life saved. That is much lower than the EPA’s VSL (US$6.45 million).

Overall, the sheer size of China’s population yields a much larger effect from improved fuel

standards compared to CARB’s efforts, confirming the importance of precise regulation in a

large developing country.

5 Conclusion

Air pollution is currently China’s most severe environmental problem, with the population

increasingly experiencing prolonged and dangerous smog events. Extreme concentrations

of fine particles and ground-level ozone degrade human health. With the rapid growth of

the private motor vehicle fleet, vehicle exhaust has become a major source of ambient air

pollution in Chinese cities. This study has shown quantitatively the importance of fuel

quality and shed light on how low-sulfur fuels may help address air pollution.

Taking advantage of the roll-out and strict enforcement of new gasoline standards across

China, this study has shown that cleaner fuels do indeed translate into better air quality. The

adoption of higher gasoline standards significantly reduced local air pollutant concentrations,

including those of PM and ozone. Using published estimates, the study further showed

that improved air quality translates into significant health and productivity benefits. Local

governments’ environmental efforts and regulatory attitudes influence the effectiveness of

fuel standards in reducing pollution. These findings constitute the first compelling evidence

about the benefits of Chinese fuel standards.

This study focuses on the technical aspects of regulations in mitigating vehicle pollution.

Government offi cials’political incentives for enforcing environmental regulations are also an

important topic, but that diffi cult enquiry must be the subject of future research.

24

References

[1] Acemoglu, D., and Joshua Linn. 2004. “Market Size in Innovation: Theory and Evidence

from the Pharmaceutical Industry.” Quarterly Journal of Economics 119(3):. 1049—

1090.

[2] Almond, Douglas, Yuyu Chen, Michael Greenstone, and Hongbin Li. 2009. “Winter

Heating or Clean Air? Unintended Impacts of China’s Huai River Policy.”American

Economic Review 99(2): 184—90.

[3] Anas, Alex, Govinda Timilsina, and Siqi Zheng. 2009. “An Analysis of Various Policy

Instruments to Reduce Congestion, Fuel Consumption and CO2 Emissions in Beijing.”

World Bank Policy Research Working Paper 5068. World Bank: Washington, DC.

[4] Auffhammer, M., and R. Kellogg. 2011. “Clearing the Air? The Effects of Gasoline

Content Regulation on Air Quality.”American Economic Review 101(6): 2687—2722.

[5] Bharadwaj, P., Matthew Gibson, Joshua GraffZivin, and Christopher A. Neilson. 2017.

“Gray Matters: Fetal Pollution Exposure and Human Capital Formation.”Journal of

the Association of Environmental and Resource Economists 4: 505—542.

[6] Bresnahan, T.F., and D.A. Yao. 1985. “The Nonpecuniary Costs of Automobile Emis-

sions Standards.”Rand Journal of Economics 16: 437—55.

[7] Cao, J., C. Yang, J. Li, R. Chen, B. Chen, D. Gua, and H. Kan. 2011. “Association

between Long-term Exposure to Outdoor Air Pollution and Mortality in China: A

Cohort Study.”Journal of Hazardous Materials 186: 1594—1600.

[8] Carneiro, Pedro, Katrine V. Løken, and Kjell G. Salvanes. 2015. “A Flying Start?

Maternity Leave Benefits and Long-Run Outcomes of Children.” Journal of Political

Economy 123(2): 365—412.

[9] Chang, Tom, Joshua Graff Zivin, Tal Gross, and Matthew Neidell. 2016. “Particulate

Pollution and the Productivity of Pear Packers.”American Economic Journal: Eco-

nomic Policy 8(3): 141—69.

[10] Chang, Tom, Joshua GraffZivin, Tal Gross, and Matthew Neidell. “The Effect of Pollu-

tion on Worker Productivity: Evidence from Call Center Workers in China.”American

Economic Journal: Applied Economics forthcoming.

25

[11] Chen, Y., A. Ebenstein, M. Greenstone, and H. Li. 2013. “Evidence on the Impact

of Sustained Exposure to Air Pollution on Life Expectancy from China’s Huai River

Policy.”Proceedings of the National Academy of Sciences 110: 12936—12941.

[12] Cifuentes, L.A., J. Vega, K. Kopfer, and L.B. Lave. 2000. “Effect of the Fine Fraction of

Particulate Matter versus the Coarse Mass and Other Pollutants on Daily Mortality.”

Journal of the Air and Waste Management Association 50: 1287—1298.

[13] Conley, T. G. 1999. “GMM Estimation with Cross Sectional Dependence.”Journal of

Econometrics 92: 1—45.

[14] Copeland, Brian R., and M. Scott Taylor. 2004. “Trade, Growth, and the Environment.”

Journal of Economic Literature 42(1): 7—71.

[15] Costa, Dora L., and Matthew E. Kahn. 2004. “Changes in the Value of Life, 1940—1980.”

Journal of Risk and Uncertainty 29(2): 159—180.

[16] Cropper, Maureen L. 2010. “What Are the Health Effects of Air Pollution in China?”

In Is Economic Growth Sustainable? edited by Geoffrey M. Heal, 10—46. Palgrave

Macmillan: Houndmills, Basingstoke, Hampshire.

[17] Davis, L.W. 2008. “The Effect of Driving Restrictions on Air Quality in Mexico City.”

Journal of Political Economy 116(1): 38—81.

[18] Ebenstein, Avraham, Maoyong Fan, Michael Greenstone, Guojun He, and Maigeng

Zhou. 2017. “New Evidence on the Impact of Sustained Exposure to Air Pollution on

Life Expectancy from China’s Huai River Policy.”Proceedings of the National Academy

of Sciences 114(39): 10384—10389.

[19] Environmental Protection Agency. 1998. U.S. EPA StaffPaper on Gasoline Sulfur Issues

EPA 420-R-98-005.

[20] Environmental Protection Agency. 1999. Tier 2/Sulfur Regulatory Impact Analysis.

Appendix B: Irreversibility of Sulfur’s Emission Impact.

[21] Environmental Protection Agency. 2000. Final Rule for Control of Air Pollution From

New Motor Vehicles: Tier 2 Motor Vehicle Emissions Standards and Gasoline Sulfur

Control Requirements.

[22] Environmental Protection Agency. 2014. Tier 3 Motor Vehicle Emission and Fuel Stan-

dards.

26

[23] Ganapati, Sharat, Joseph S. Shapiro, and Reed Walker. 2017. “The Incidence of Car-

bon Taxes in U.S. Manufacturing: Lessons from Energy Cost Pass-Through.”NBER

Working paper no. 22281.

[24] Gentzkow, M. 2006. “Television and Voter Turnout.”Quarterly Journal of Economics

121(3): 931—972.

[25] Graff Zivin, Joshua, and Matthew Neidell. 2012. “The Impact of Pollution on Worker

Productivity.”American Economic Review 102(7): 3652—73.

[26] Greenstone, Michael. 2004. “Did the Clean Air Act Cause the Remarkable Decline in

Sulfur Dioxide Concentrations?”Journal of Environmental Economics and Management

47(3): 585—611.

[27] Greenstone, Michael, and Rema Hanna. 2014. “Environmental Regulations, Air and

Water Pollution, and Infant Mortality in India.”American Economic Review 104(10):

3038—72.

[28] Hahn, Jinyong, Petra Todd, and Wilbert van der Klaauw. 2001. “Identification and Es-

timation of Treatments Effects with a Regression-Discontinuity Design.”Econometrica

69(1): 201—209.

[29] He, G., M. Fan, and M. Zhou. 2016. “The Effect of Air Pollution on Mortality in China:

Evidence from the 2008 Beijing Olympic Games.”Journal of Environmental Economics

and Management 79: 18—39.

[30] He, Jiaxiu, Haoming Liu, and Alberto Salvo. “Severe Air Pollution and Labor Pro-

ductivity: Evidence from Industrial Towns in China.”American Economic Journal:

Applied Economics forthcoming.

[31] Imbens, G., and Thomas Lemieux. 2008. “Regression Discontinuity Designs: A Guide

to Practice.”Journal of Econometrics 142(2): 615—635.

[32] Imbens, Guido, and Karthik Kalyanaraman. 2012. “Optimal Bandwidth Choice for the

Regression Discontinuity Estimator.”Review of Economic Studies 79(3): 933—959.

[33] International Council on Clean Transportation. 2010. “Overview of China’s

Vehicle Emission Control Program: Past Successes and Future Prospects.”

http://www.theicct.org/sites/default/files/publications/Retrosp_final_bilingual.pdf.

27

[34] International Council on Clean Transportation. 2012. “Technical and Economic Analy-

sis of the Transition to Ultra-low Sulfur Fuels in Brazil, China, India, and Mexico.”

http://http://www.theicct.org/sites/default/files/publications/ICCT_ULSF_refining

_Oct2012.pdf.

[35] International Council on Clean Transportation. 2013. “China Announces Break-

through Timeline for Implementation of Ultra-Low Sulfur Fuel Standards.”

http://www.theicct.org/sites/default/files/publications/ICCTupdate_CH_fuelsulfur_

mar2013_rev.pdf.

[36] International Council on Clean Transportation. 2014.

“China V Gasoline and Diesel Fuel Quality Standards.”

http://www.theicct.org/sites/default/files/publications/ICCTupdate_ChinaVfuelquality

_jan2014.pdf.

[37] Jacobsen, Mark R. 2013. “Evaluating US Fuel Economy Standards in a Model with

Producer and Household Heterogeneity.”American Economic Journal: Economic Pol-

icy 5(2): 148—87.

[38] Kahn, Matthew E. 1996. “New Evidence on Trends in Vehicle Emissions.”RAND Jour-

nal of Economics 27(1): 183—96.

[39] Kahn, Matthew E., and Joel Schwartz. 2008. “Urban Air Pollution Progress Despite

Sprawl: The ‘Greening’of the Vehicle Fleet.”Journal of Urban Economics 63(3): 775—

87.

[40] Lee, David S., and Thomas Lemieux. 2010. “Regression Discontinuity Designs in Eco-

nomics.”Journal of Economic Literature 48(2): 281—355.

[41] Levy, J. I., and S. L. Greco. 2007. “Estimating Health Effects of Air Pollution in China:

An Introduction to Intake Fraction and the Epidemiology.” In Clearing the Air: The

Health and Economic Damages of Air Pollution in China edited by M. S. Ho and C. P.

Nielsen. MIT Press: Cambridge, MA.

[42] Matus, K., K.M. Nam, N.E. Selin, L.N. Lamsal, J.M. Reilly, and S. Paltsev. 2012.

“Health Damages from Air Pollution in China.”Global Environmental Change 22(1):

55—66.

28

[43] Ministry of Environmental Protection. 2013. “National Environmental Statistic Re-

port.” http://zls.mep.gov.cn/hjtj/ qghjtjgb/201503/t20150316_297266.htm (in Chi-

nese).

[44] Ministry of Environmental Protection. 2012. “Technical Regulation on Ambient Air

Quality Index Daily Report.”GB3095-2012.

[45] Lei, Y. 2013. “Benefits to Human Health and Agricultural Productivity of Reduced

Air Pollution.” In Clearer Skies Over China: Reconciling Air Quality, Climate, and

Economic Goals edited by M. S. Ho and C. P. Nielsen. MIT Press: Cambridge, MA.

[46] Lei, Y. and Mun S. Ho. 2013. “The Valuation of Health Damages.” In Clearer Skies

Over China: Reconciling Air Quality, Climate, and Economic Goals edited by M. S.

Ho and C. P. Nielsen. MIT Press: Cambridge, MA.

[47] Parry, Ian W. H., Margaret Walls, and Winston Harrington. 2007. “Automobile Exter-

nalities and Policies.”Journal of Economic Literature 45(2): 373—399.

[48] Pope , C.A., R.T. Burnett , M.J. Thun, E.E. Calle, D. Krewski, K. Ito, and G.D.

Thurston. 2002. “Lung Cancer, Cardiopulmonary Mortality, and Long-term Exposure

to Fine Particulate Air Pollution.”Journal of the American Medical Association 287:

1132—1141.

[49] Pope, C.A., and D.W. Dockery. 2006. “Health Effects of Fine Particulate Air Pollution:

Lines That Connect.”Journal of the Air and Waste Management Association 56: 709—

742.

[50] Pope, C.A., and D.W. Dockery. 2013. “Air Pollution and Life Expectancy in China and

Beyond.”Proceedings of the National Academy of Sciences 110(32): 12861—2.

[51] Qin, X., L. Li, and Y. Liu. 2013. “The Value of Life and its Regional Difference in

China.”China Agricultural Economics Review 5: 373—390.

[52] Schwartz, J., D.W. Dockery, and L.M. Neas. 1996. “Is Daily Mortality Associated Specif-

ically with Fine Particles?”Journal of the Air and Waste Management Association 46:

927—939.

[53] State Council, China. 2013. “Ten Measurements of the Atmospheric Pollution Pre-

vention Action Plan.” http://www.gov.cn/ldhd/2013-06/14/content_2426237.htm (in

Chinese).

29

[54] Viard, V. Brian, and Shihe Fu. 2015. “The Effect of Beijing’s Driving Restrictions on

Pollution and Economic Activity.”Journal of Public Economics 125: 98—115.

[55] Watkiss, P.S., S. Pye, and M. Holland. 2005. “CAFE CBA: Baseline Analysis 2000–

2020.”Report to the European Commission’s Director General for the Environment,

Brussels.

[56] Weyl, E. Glen, and Michal Fabinger. 2013. “Pass-Through as an Economic Tool: Prin-

ciples of Incidence under Imperfect Competition.”Journal of Political Economy 121(3):

528—583.

[57] Wolff, H. 2014. “Keep Your Clunker in the Suburb: Low-emission Zones and Adoption

of Green Vehicles.”The Economic Journal 124(578): 481—512.

[58] Wong, T.W., T.S. Lau, T.S. Yu, A. Neller, S.L. Wong, W. Tam, and S.W. Pang. 1999.

“Air Pollution and Hospital Admissions for Respiratory and Cardiovascular Diseases in

Hong Kong.”Occupational and Environmental Medicine 56: 679—683.

[59] World Bank. 2007. Cost of Pollution in China: Economic Estimates of Physical Dam-

ages. World Bank: Washington, D.C.

[60] World Health Organization. 2011. Urban Transport and Health. World Health Organi-

zation: Geneva.

[61] World Health Organization. 2015. Ambient Air Pollution: A Global Assessment of Ex-

posure and Burden of Disease. World Health Organization: Geneva.

[62] Xu, X.P., B.L. Li, and H.Y. Huang. 1995. “Air Pollution and Unscheduled Hospital

Outpatient and Emergency Room Visits.”Environmental Health Perspectives 103(3):

286—289.

[63] Zhang, Y.L., and F. Cao. 2015. “Fine Particulate Matter (PM2.5) in China at a City

Level.”Scientific Reports 5, 14884.

30

Figures

Figure 1: Motor Vehicle Ownership and PM Emissions in China

Notes: Using the data extracted from China’s annual Vehicle Emission Control reports for 2010–2016, this figure plots the total car stock (millions) and related PM emissions (10,000 tons). The PM emissions from cars are as estimated in the Ministry of Environmental Protection’s nationwide air pollution source analysis, which is aimed at identifying sources of ambient air pollution and quantifying their contributions.

050

100

150

200

To

tal C

ar

Sto

ck(m

illio

n)

1980 1985 1990 1995 2000 2005 2010 2015

Year

1020

3040

5060

PM

Em

issi

on

s fr

om

Ca

rs (

10

,00

0 to

ns)

1980 1985 1990 1995 2000 2005 2010 2015

Year

 Figure 2 Evolution of Gasoline and Diesel Content Regulations in China

Notes: This figure displays the gradual implementation of higher fuel standards in China. From 2013 to 2015, the gasoline and diesel fuel standards have been upgraded from III to IV and then to V. The fuel standards information was assembled from circulars issued by the provincial Development and Reform Commissions on implementing gasolines (or diesel) standards in the various provinces.    

 Figure 3 Tests for Parallel Trends

Notes: This figure plots the bi-monthly βt coefficient estimates of specification (1) (and their 95 percent confidence intervals), using pre-treatment data 8 months leading up to the implementation of the gasoline IV standard and post-treatment data for 10 months after the implementation. Regressions control for any station-specific seasonality, the weather condition, the gasoline V standard, the diesel IV standard, the diesel V standard, exhaust emission standard IV for gasoline vehicles, the voluntary scrapping programs, fuel prices, pollutant dummies as well as winter heating policies. The dependent variable is the logarithm of the level of the stacked pollutants: the AQI, PM2.5, PM10 and O3.

-.2

-.1

0.1

-8 -6 -4 -2 0 2 4 6 8 10+Months after the implementation of the gasoline IV standard

Figure 4 Bi-weekly RD Estimates of the Effects of Fuel Standards on Air Quality

Notes:  Collapsing the sample to bi-weekly, this figure plots the residuals of the pollutant measures against a normalized time variable based on the time of the gasoline IV standard’s implementation. The estimation method involves two steps. First the outcome is regressed against the controls to obtain a residual. That value is used in the second step, a nonparametric estimation of equation (3) to obtain the parameter of interest βRD. The circles represent mean values for each two-week bin. The lines indicate the fitted values from a local linear regression with the optimal bandwidth calculated using the method of Imbens and Kalyanaraman (2012). The grey areas are the 95% confidence intervals, and the vertical line is the cutoff point for the assignment variable.

AQI

Bi-week-.2

0.2

Res

idua

l

-10 -5 0 5 10Gasoline IV

(a)PM2.5

Bi-week-.2

0.2

Res

idua

l

-10 -5 0 5 10Gasoline IV

(b)PM10

Bi-week-.2

0.2

Res

idua

l

-8 -4 0 4 8Gasoline IV

(c)

O3

Bi-week-.3

-.1

.1R

esid

ual

-5 0 5Gasoline IV

(d)

Bi-week-.2

0.2

Res

idua

l

-12 -6 0 6 12Gasoline IV

(e)Stack

Figure 5 Sensitivity Test on the Choice of Bandwidth

 

Notes: This figure plots βRD coefficients estimated using bandwidths from h*-10 to h*+10 in intervals of 2 for a set of pollutant measures. The vertical line is the optimal bandwidth h* calculated using the method of Imbens and Kalyanaraman (2012). The dashed lines show the 95 percent confidence intervals for each coefficient. The estimation method is described in full in Section 4.2.

AQI

-.2

-.15

-.1

-.05

0.0

5C

oef

ficie

nt

15415

615

816

016

216

416

616

817

017

217

4

Optimal Bandwidth

(a)

PM2.5

-.2

-.15

-.1

-.05

0.0

5C

oef

ficie

nt13

713

914

114

314

514

714

915

115

315

515

7

Optimal Bandwidth

(b)

PM10

-.2

-.15

-.1

-.05

0.0

5C

oef

ficie

nt

78 80 82 84 86 88 90 92 94 96 98

Optimal Bandwidth

(c)

O3

-.2

-.15

-.1

-.05

0.0

5C

oef

ficie

nt

57 59 61 63 65 67 69 71 73 75 77

Optimal Bandwidth

O3

(d)Stack

-.2

-.15

-.1

-.05

0.0

5C

oef

ficie

nt

10710

911

111

311

511

711

912

112

312

512

7

Optimal Bandwidth

(e)

  

Figure 6 The Hourly Effects of Fuel Standards on Air Quality

Notes: This figure plots the impact of the gasoline IV standard on a set of pollutants through the day: 12-2am, 3-5am, …, 9-11pm. The coefficients β in equation (1) are estimated for each time period separately.      

-.15

-.1

-.05

0.0

5E

stim

ate

d C

oef

ficie

nt

0~2

3~5

6~8

9~11

12~1

4

15~1

7

18~2

0

21~2

3

Hours

AQI PM2.5 PM10 O3 Stack

Table 1Evolution of China's Fuel Standards

China Gasoline III GB 17930-2006 150 Dec 6, 2006 Phased-in by Dec 31, 2009Gasoline IV GB 17930-2011 50 May 12, 2011 Phased-in by Dec 31, 2013

Gasoline V GB 17930-2013 10 Dec 18, 2013 Phased-in by Dec 31, 2017 China Diesel III GB 19147-2009 350 June 12, 2009 Phased-in Jan 1, 2010–Jul 1, 2011

Diesel IV GB 19147-2013 50 Feb 7, 2013 Phased-in by Dec 31, 2014Diesel V GB 19147-2009 10 June 8, 2013 Phased-in by Dec 31, 2017

Source : Intl. Council on Clean TransportationNotes : This table presents the key dates for changes to China 's nationwide gasoline and diesel standardsfrom III to IV and then to V. A maximum sulfur content is specified in each new standard.

Stage Standard Maximum sulfur (ppm)

Standard Issued on Implementation

Table 2 International Gasoline Standard Comparison

Fuel Property/Year implemented by 2009 by 2013 by 2017 2000 2005 2009 1995-2000s 2004 2006 2017 2003Research Octane, min. 97-90 97-90 95-89 95–91 95–91 95–91 NS NS Motor Octane, min. 88-85 88-85 90-84 85–81 85–81 85–81 NS NS Aromatics, vol%, max. 40 40 40 42 35 35 25 25Olefins, vol%, max. 30 28 24 18 18 18 6Benzene, vol%, max. 1 1 1 1 1 1 1 0.8

RVP, kPa

Lead, mg/L, max. 5 5 5 5 5 5 ND

Oxygen, % m/m 2.7 (max.) 2.7 (max.) 2.7 (max.) 2.7 (max.) 2.7 (max.) 2.7 (max.) 2 1.8-2.2Sources : Intl. Council on Clean Transportation (2010); Intl. Council on Clean Transportation (2014); U.S. Environmental Protection Agency (EPA) websiteNS = Not specified; ND = Nondetectable

EPA Tier 2

48.2/47.6 max (7 psi)

Sulfur, ppm, max. 20

CARB Phase III

Fuel Standards China III China IV China V Euro III Euro IV Euro V EPA RFG EPA Tier 3

150 50 10 150

88 Winter 72 Summer max.

42-85 Winter 40-68 Summer max.

45-85 Winter 40-657 Summer max.

60/70 max. 60/70 max. 60/70 max.

120/300 (average/max)

30/80(average/max)

10

48 kPa (7.0 psi) max

50 10

Notes : This table displays the gasoline standard of China in an international context. In terms of the key parameters, the Chinese gasoline standards III, IV and V are compared with those inEurope and the United States (including California). EPA RFG is the US EPA' s reformulated gasoline. EPA Tier 2 and Tier 3 are the US EPA's Gasoline Sulfur Programs. CARB is theCalifornia Air Resources Board. RVP is Reid vapor pressure.

NS MMT<6 (by 2011) MMT<2(by 2014)

NDManganese, mg/L, max. 16 8 2 NS ND

Table 3Variable Definition and Descriptive StatisticsVariable Definition Mean SDFuel Standard Regulations and PricesGasoline IV 0.714 0.452

Gasoline V 0.148 0.355

Diesel IV 0.522 0.500

Diesel V 0.069 0.253

Fuel_Price Average retail price of gasoline and diesel (thousand yuan/ton) 8.410 0.972Vehicle Emission RegulationsGasoline_Car IV 0.898 0.303

Yellow Car 0.613 0.487

Air Pollutant VariblesAQI Air Quality Index 87.732 57.002PM2.5 PM2.5 concentration (μg/m3) 58.764 49.085

PM10 PM10 concentration (μg/m3) 99.344 74.952

O3 O3 concentration (μg/m3) 55.347 35.825

Weather VariablesSunshine_time Prefecture daily average sunshine hours (0.1 hour) 54.454 40.538Temperature Prefecture daily average temperature (0.1 degree) 149.545 105.889Wind_Speed Prefecture daily average wind speed (0.1 m/s) 22.122 27.349Rainfall Prefecture daily average rainfall (0.1 mm) 56.214 186.933Humidity Prefecture daily average humidity (1%) 69.117 17.449

Prefecture Socio-economic Variables

Road_Density2007 Road density in year 2007 (km/km 2) 0.968 0.444

Car_Percapita2007 No. of cars per capita in year 2007 0.647 0.685MegaCity =1 if a prefecture has a population of over 5 million; =0 otherwise. 0.457 0.498Refinery2007 =1 if a prefecture has at lease one large oil refinery; =0 otherwise. 0.331 0.471GovEnvironmental_effort 0.271 0.087

Car Model Sales VariablesTotal_CarSales Prefecture montly total sale of car models (thousands) 4.692 6.621Horsepower Prefecture average horsepower of cars sold (kw) 81.318 5.462FuelConsumption 7.330 0.343

Displacement Prefecture average displacement of cars sold (litres) 1.513 0.115CurbWeight Prefecture average curb weight of cars sold (1,000 kilograms) 1.298 0.129Other Variables

Coast_Distance Distance between a pollution monitoring station and the coastline of China (km) 530.997 644.761

WinterHeating 0.182 0.386

Station_Latitude Latitude of pollution monitoring station 33.054 6.435Station_Longitude Longitude of pollution monitoring station 114.230 7.709

=1 if a prefecture has upgraded its gasoline content regulation from standard IIIto standard IV; =0 otherwise.=1 if a prefecture has upgraded its gasoline content regulation from standard IVto standard V; =0 otherwise.

=1 if a prefecture has upgraded its diesel content regulation from standard IV tostandard V; =0 otherwise.

=1 if a prefecture has upgraded its diesel content regulation from standard III tostandard IV; =0 otherwise.

Notes: This table summarizes the variables. Definitions, means and standard deviations are reported. The data sources are describedin full in Section 3.1. Fuel standard regulations, vehicle emission regulations, fuel prices and weather variables are all at theprefectural level. "Green" keywords include "environmental protection" (huan-jing-bao-hu or huan-bao ). The air pollutant variablesare by monitoring station.

=1 if a prefecture has upgraded its gasoline vehicle emission regulation fromstandard III to standard IV; =0 otherwise.

=1 if a prefecture has started to phase out yellow label vehicles; =0 otherwise.

Prefecture average fuel consumption of cars sold (litres/100 kilometres)

=1 if a prefecture had been experiencing the coal-based winter heating period; =0 otherwise.

Average count of "green" keywords in the government work reports for 2001 to2012 divided by total word count (×10 3)

Table 4DD Estimates of Effects of Fuel Standard Regulations on Daily Average Pollutant Concentrations

AQI PM2.5 PM10 O3(1) (2) (3) (4) (5) (6) (7)

Gasoline IV -0.041*** -0.043*** -0.041*** -0.055*** -0.040*** -0.072*** -0.040***(0.011) (0.011) (0.014) (0.022) (0.010) (0.027) (0.013)

Station dummies Yes Yes Yes Yes Yes Yes YesDay dummies Yes Yes Yes Yes Yes Yes YesStation dummies × Day of week dummies Yes Yes Yes Yes Yes Yes YesStation dummies × Week of month dummies Yes Yes Yes Yes Yes Yes YesStation dummies × Month of year dummies Yes Yes Yes Yes Yes Yes YesWeather condition controls Yes Yes Yes Yes Yes Yes YesGasoline V Yes Yes Yes Yes Yes Yes YesDiesel IV Yes Yes Yes Yes Yes Yes YesDiesel V Yes Yes Yes Yes Yes Yes YesGasoline_Car IV Yes Yes Yes Yes Yes Yes YesYellow_Car Yes Yes Yes Yes Yes Yes YesGasoline_Price Yes Yes Yes Yes Yes Yes YesWinterHeating Yes Yes Yes Yes Yes Yes YesPollutant dummies Yes Yes YesTime polynomial interactions YesRoad_Density2007 × Day dummies YesCar_Percapita2007 × Day dummies YesCoast_Distance × Day dummies YesMegaCity × Day dummies YesRefinery2007 × Day dummies Yes

Adjusted R2 0.596 0.588 0.609 0.597 0.389 0.402 0.409

No. of observations 917,653 915,178 901,055 903,146 3,637,066 2,009,597 3,637,066

Log(daily average pollutant concentration)Stack

Notes : This table presents DD estimates of the logarithm of daily average pollutant concentrations on implementation of thegasoline IV standard and other control variables. In columns 1–4 the dependent variables are the logarithms of AQI, PM2.5, PM10and O3. In columns 5-7 the dependent variable is the logarithm of the stacked pollutants. The weather condition controls are thedaily average temperature, rainfall, humidity, sunshine time and wind speed for each prefecture. All of the specifications control forany station-specific seasonality and the indicators of the gasoline V standard, the diesel IV standard, the diesel V standard, exhaustemission standard IV for gasoline vehicles, the voluntary scrapping programs, fuel prices, as well as winter heating policies. Column6 additionally includes time polynomial interactions, which are a third-order polynomial in time interacted with pre-treatmentpollution (measured by the average value of air pollution during the two-months period before the policy implementation), andcolumn 7 controls for the differences in the trends in outcomes between the treatment and control groups depending on keydeterminants in the selection of gasoline IV implementation timing. Reported in parentheses are robust standard errors clustered bymonitoring station. ***, ** and * represent significance at the 1%, 5% and 10% levels of confidence, respectively.

Table 5RD Estimates of Effects of Fuel Standards on Daily Average Pollutant Concentrations

AQI PM2.5 PM10 O3 Stack(1) (2) (3) (4) (5)

Gasoline IV -0.111*** -0.103*** -0.084*** -0.050*** -0.059***(0.009) (0.011) (0.010) (0.016) (0.008)[0.009] [0.011] [0.010] [0.016] [0.008]

Day 0.001*** 0.001*** 0.001*** 0.000** 0.001***(0.000) (0.000) (0.000) (0.000) (0.000)[0.000] [0.000] [0.000] [0.000] [0.000]

Gasoline IV × Day -0.001*** -0.001*** -0.002*** 0.001*** -0.001***(0.000) (0.000) (0.000) (0.001) (0.000)[0.000] [0.000] [0.000] [0.001] [0.000]

Optimal bandwidth 164 147 88 67 117Control group mean 0.041 0.062 0.063 -0.048 0.036

R2 0.007 0.009 0.010 0.001 0.003

No. of observations 193,078 173,004 103,788 80,011 554,053Notes : This table presents estimates of RD regressions of the logarithm of daily average pollutantconcentrations on implementation of the gasoline IV standard. In columns 1–4 the dependentvariables are the logarithm of AQI, PM2.5, PM10 and O3 respectively. In column 5 the dependentvariable is the logarithm of the stacked pollutant concentrations. Reported in paretheses are robuststandard errors clustered by monitoring station. Reported in brackets are standard errors adjusted toreflect spatial dependence as modelled by Conley (1999). ***, ** and * represent significance at the1%, 5% and 10% levels of confidence respectively. Day is the normalized time variable, i.e. , the dateof the gasoline IV standard's implementation (the assignment variable). The estimation method isdescribed in full in Section 3.2.2 and involves two steps. First the outcome is regressed against thecontrols to obtain a residual. That value is used in the second step, the nonparametric estimation (3) toobtain the parameter of interest βRD. In the first step, the controls include the set of station-specificseasonality variables, weather condition variables, fuel price variable and station dummies. Themethod developed by Imbens and Kalyanaraman (2012) is used to calculate the optimal bandwidth.

Table 6Effects of Fuel Standard Regulations on Car Model Sales

Log(Total_CarSales) Log(Horsepower) Log(OilConsumption) Log(Displacement) Log(CurbWeight)(1) (2) (3) (4) (5)

Gasoline IV 0.014* 0.002* 0.0004* 0.005* 0.0003*(0.024) (0.002) (0.002) (0.003) (0.005)

Prefecture dummies Yes Yes Yes Yes YesMonth dummies Yes Yes Yes Yes YesPrefecture dummies × Year dummies Yes Yes Yes Yes YesWeather condition controls Yes Yes Yes Yes YesGasoline V Yes Yes Yes Yes YesDiesel IV Yes Yes Yes Yes YesDiesel V Yes Yes Yes Yes YesGasoline_Car IV Yes Yes Yes Yes YesYellow_Car Yes Yes Yes Yes YesGasoline_Price Yes Yes Yes Yes YesWinterHeating Yes Yes Yes Yes Yes

Adjusted R2 0.976 0.906 0.836 0.648 0.335

No. of observations 9,241 9,241 9,241 9,241 9,241Notes : This table presents DD estimates of major attributes of car models sold by month and prefecture on implementation of the gasoline IVstandard. The data are from the State Administration of Industry and Commerce. The dependent variables are the logarithm of total car sales,power, fuel consumption, displacement, and curb weight of the new vehicles registered. Reported in parentheses are robust standard errorsclustered by prefecture. ***, ** and * represent significance at the 1%, 5% and 10% levels of confidence, respectively. The weather conditioncontrols are the daily average temperature, rainfall, humidity, sunshine time and wind speed by prefecture. All of the specifications control for theindicators of the gasoline V standard, the diesel IV standard, the diesel V standard, exhaust emission standard IV for gasoline vehicles, thevoluntary scrapping programs, fuel prices, as well as winter heating policies.

Table 7Spatial Variation in the Effects of Fuel Standards on Air Quality

AQI PM2.5 PM10 O3 Stack AQI PM2.5 PM10 O3 Stack(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Gasoline IV -0.059*** -0.068*** -0.047* -0.130*** -0.073*** 0.019* 0.041* -0.055* 0.012* 0.026*(0.020) (0.017) (0.026) (0.041) (0.018) (0.025) (0.038) (0.033) (0.050) (0.027)

Station dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes YesDay dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes YesStation dummies × Day of week dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes YesStation dummies × Week of month dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes YesStation dummies × Month of year dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes YesWeather condition controls Yes Yes Yes Yes Yes Yes Yes Yes Yes YesGasoline V Yes Yes Yes Yes Yes Yes Yes Yes Yes YesDiesel IV Yes Yes Yes Yes Yes Yes Yes Yes Yes YesDiesel V Yes Yes Yes Yes Yes Yes Yes Yes Yes YesGasoline_Car IV Yes Yes Yes Yes Yes Yes Yes Yes Yes YesOld_Car Yes Yes Yes Yes Yes Yes Yes Yes Yes YesGasoline_Price Yes Yes Yes Yes Yes Yes Yes Yes Yes YesWinterHeating Yes Yes Yes Yes Yes Yes Yes Yes Yes YesPollutant dummies Yes Yes

Adjusted R2 0.610 0.593 0.631 0.599 0.276 0.585 0.600 0.591 0.617 0.241

No. of observations 296,824 295,849 291,060 291,272 1,175,027 202,824 202,581 198,977 200,579 804,969Notes : This table presents DD estimates of the fuel standard's effect on air quality based on differences in prefectural emphasis on environmental protection. The286 prefectures are ranked based on mentions of environmental concern in their government work reports. The top 30% and bottom 30% of the cities arecompared. Reported in parentheses are robust standard errors clustered by monitoring station. ***, ** and * represent significance at the 1%, 5% and 10% levelsof confidence respectively. The weather controls are the daily average temperature, rainfall, humidity, sunshine time and wind speed by prefecture. All of thespecifications controlled for station-specific seasonality and indicators of the adoption of the gasoline V standard, the diesel IV standard, the diesel V standard,the exhaust emission standard IV for gasoline vehicles, the voluntary scrapping programs, fuel prices, as well as winter heating policies.

More Environmental Effort Less Environmental Effort

Table 8Valuation of the Health Benefits associated with Air Quality Improvement

161300 World Bank (2007) 7.67619400 Qin et al. (2013) 29.44

Acute mortality 1.0065 Lei and Ho (2013) 6871Chronic mortality 1.04 Pope et al. (2002) 40656Morbidity in which: 0.13

1.01 0.0062 Lei (2013) 11916 1648 Lei and Ho (2013) 0.02

1.0268 0.0042 Lei (2013) 21633 825 Lei and Ho (2013) 0.02

1.0039 3.4788 Lei (2013) 2607522 35 Lei and Ho (2013) 0.09

0.0055Mortality in which: 47527

Health Endpoint Relative risks if PM2.5

concentration increased by 10 μg/m3

Reference Baseline incidence for 2005 (cases per person per year)

The change in the number of cases of health endpoint with PM2.5

reduced by2.7 μg/m3

Valuation of VSLs and cost-of-illness in China ($US)

References on VSLs and COIs

Health benefits (US$ billions)

Reference

Lei (2013)

T. W. Wong et al. (1999)

Hospital admissions, cardiovascular Hospital admissions, respiratoryOutpatient visits, all causes Notes: This table presents estimates of the health benefits associated with air quality improvement. Using published epidemiological data, the healthimpacts of improved air quality as a result of fuel standards are inferred. The mortality effect includes both acute and chronic one. The morbidityeffect includes hospital admissions due to cardiovascular disease and respiratory disease, and outpatient visits for all reasons.

T. W. Wong et al. (1999)

Xu, Li, and Huang (1995)

Table 9Valuation of the Labor Productivity Benefits associated with Air Quality Improvement

Agriculture U.S. O3 0.28% 0.0084 76.1 0.1

U.S. PM2.5 0.60% 0.0162 3066.1 7.9

China PM2.5 0.10% 0.0027 3066.1 1.3

Service China API 0.04% 0.00133 3949.2 0.8

Total: 2.2-8.8Notes: This table presents estimates of the labor productivty benefits associated with air quality improvement using thefindings of studies previously published. The potential productivity gains associated with air pollution reductions from thefuel standards are inferred. 1ppb = 1.96µg/m3 is used to convert Graff Zivin and Neidell (2012)'s estimates. An exchange rateof 6.31 is assumed.

Manufacturing

ReferencesSector Location of Study

Pollutant Productivity change reducing Pollution by 1 unit

Productivity effects of reducing PM2.5 by 2.7 μg/m3, O3 by 3.0 μg/m3, AQI by 3.8 unit

Wage Bills in China, 2012 (Billion RMB)

Productivity Benefits (US$ Billions)

Graff Zivin and Neidell (2012)

a. Chang et al . (2016)

b. He et al . forthcoming

Chang et al . forthcoming