Air pollution, stock returns, and trading activities in China

Abstract

The relationship between air pollution and stock pricing of locally headquartered firms is explored using firm-level data in China. Severe air pollution results in low returns, turnover, volatility, and high illiquidity, mainly through a home bias. The results remain robust after a series of checks. The relation between air pollution and local firm performance is insignificant, implying that the air pollution effects can be attributed to investor mood bias rather than to economic effects. The sensitivity of stock returns to air pollution is significantly large for high-growth stocks, distressed stocks, and stocks with high volatility.

Highlights

• •We examine how air pollution affects stock pricing of locally headquartered firms.
• •We validate that air pollution is one of factors inducing pessimistic moods.
• •Severe air pollution decreases local stock returns, liquidity and volatility.
• •Air pollution effect is more pronounced for stocks hard to value and arbitrage.

1. Introduction

With the rapid development of the economy, China has paid a high environmental cost. As highlighted by Xu et al. (2013), rapid economic growth improved the living conditions among the Chinese, but the accompanying air pollution continues to threaten citizens' health. In the last few years, air quality in major Chinese cities has received increasing attention. Nanshan Zhong, a leading Chinese specialist at the Chinese Academy of Engineering, emphasized that air pollution was more frightening than severe acute respiratory syndrome (SARS) because no one could escape it (Li, 2013). Since 2008, the US Embassy in China has been posting the fine particulate matter (PM_2.5) index for Beijing. On January 12, 2013, the PM_2.5 index of Beijing hit a record 886, beyond the limit of the measuring device, which led to the cancellation of hundreds of flights, warnings to avoid all outdoor activities, and a spike in respiratory disease cases (Ji et al., 2014). The literature has shown that air pollution can cause negative physiological and psychological reactions, but few studies have focused on how air pollution affects financial markets. This study links the negative effects of air pollution to stock pricing and trading activities based on firm-level data.

Levy and Yagil (2011) posit that air pollution induces negative moods and risk-aversion behaviors among investors, leading to a negative relationship between air pollution and stock returns. The authors' findings indicate that in the US, air pollution near the areas where stock exchanges are located negatively correlates with market index returns. Studies by Bilgin and Danis (2016) and Heyes et al. (2016) validate Levy and Yagil's (2011) conclusions. However, Lepori (2016) confirms that this negative effect only exists when stock exchange facilities use trading floor technology.

In an electronic order-driven stock market such as China, the trading process is driven by orders from all over the market. Considerable evidence shows that investors prefer to hold local stocks (Ivković and Weisbenner, 2005; Seasholes and Zhu, 2010; Bodnaruk, 2009; Becker et al., 2011). Therefore, if the investors' portfolios are largely composed of local stocks and the air quality where the investors are located affects their sentiments, risk attitudes, and investment decisions, then the stock pricing of these companies should be significantly affected by the air quality where the firms are headquartered; this expected outcome will provide physiological and psychological evidence regarding the home bias of equity investments. If we use only the air quality where the stock exchange is located to study its impact on return of stock index, there would be a significant bias. Research regarding the impact of air pollution on stock pricing in order-driven markets shows inconsistent results. Lepori's (2016) study found that the returns of the Shanghai Stock Exchange 50 Index (SSE 50 Index) and the Shanghai Stock Exchange 180 Index (SSE 180 Index), both market returns of order-driven markets, were not affected by the air quality in Shanghai. In another study, Li and Peng (2016) show that the market index of the Shanghai Stock Exchange (SSE) is significantly, positively affected by local air quality. After carefully replicating the studies of Lepori (2016) and Li and Peng (2016), our results verify that “air pollution effects” in an order-driven market are primarily driven by air pollution where firms are headquartered instead of where the stock exchange is located.

China provides a better setting to test how air pollution impacts the asset pricing of local firms through home bias. As China is a vast territory and there is a large spatial distance between major cities, each city's air quality is not fully synchronized with the air quality in other cities. From a global perspective, the contiguous US (despite having a large territory) is artificially divided into four time zones, with major investment institutions and individuals gathered in the eastern region. As the New York Stock Exchange (NYSE) uses a market-maker system, analyzing the impact of city pollution levels on firm-level stock trading would be difficult based on US stock market data. Canada and Australia have larger territories, transacting in the market-maker quotation system with smaller stock markets. In European countries, some stock exchanges employ an order-driven mechanism (e.g., Bolsa de Madrid) but these countries are generally smaller in territory and have shorter spatial distances between cities. These characteristics lead to similar air quality among cities in the same country. Thus, conducting a similar study in such countries would be difficult.

In this study, the works of Levy and Yagil (2011) and Lepori (2016) are extended by exploring how air pollution affects the firm-level stock returns and trading activities of firms headquartered in the same city in China. 1548 firms headquartered in 33 major cities in China comprise the sample.1

First, the effects of air pollution on asset pricing in the Chinese stock market are explored by examining home bias. A significant decrease in trading volume for firms in cities with a sudden deterioration in air quality is observed, whereas firms in cities without a sudden air quality deterioration do not show a decrease in trading volume on the same day. Additionally, no significantly negative relationship is observed between air pollution in Shanghai and the returns of the SSE 50 and SSE 180 indexes. However, air pollution is significantly and negatively related to the firm-level returns of locally headquartered firms on the SSE 50 and SSE 180 indexes. The results are similar using data from the Shenzhen Stock Exchange 100 (SZSE 100 Index) and Shenzhen Stock Exchange 500 indexes (SZSE 500 Index).

Second, using firm-level data, we conducted a “placebo” experiment to verify that air pollution effects are caused by local air pollution rather than by air pollution where the stock exchange is located or by chance. For the experiment, we randomly selected an air pollution time series in a city from the series sample, assigned the given returns time series of a firm not headquartered in the city, and ran regressions. After repeating this experiment 100 or more times, a significant negative proportion of coefficients were <10%, supporting previous results.

Third, the effects of air pollution on the stock returns and trading activities of locally headquartered firms are investigated. As mediated by the moods of investors, air pollution decreases daily stock returns. A single standard-deviation increase in the air quality index (AQI) significantly lowers daily returns by 0.0422%. Air pollution also decreases firm-level liquidity and volatility. Investors reduce their demand for stocks and trade less when air pollution is serious. Air pollution induces investor pessimism, impacting their decisions in financial markets.

Fourth, after considering the existence of endogeneity issue, we obtain similar results. Air quality also affects daily labor productivity (Chang et al., 2016; Lavy et al., 2014; Zivin and Neidell, 2012), and thus, the local economy could change firm fundamentals. To test whether this potential issue exists, we examined the relation between air pollution and firm performance. We found that air quality was not significantly associated with firm performance, verifying that the adverse effects of air pollution on local firm-level returns, liquidity, and volatility are not caused by a reduction in local labor productivity due to heavy air pollution.

Fifth, the sensitivity of stock returns to air pollution is larger for distressed firms, extreme growth firms, and firms with high volatility. This finding could be a result of by the negative moods induced by air pollution, which reduce the demand for portfolio risk; the demand for portfolio risk is difficult to determine, and arbitrage would decrease much more (Baker and Wurgler, 2006).

This study makes several contributions to the literature. First, we study the impact of air pollution on the stock market from the perspective of home bias based on firm-level data, which provides evidence for home bias theory and contributes to behavioral biases triggered by ambient air pollution. Our study clarifies how air pollution affects the stock market through investor moods in an order-driven market (e.g., China). This study elucidates the negative effects of air pollution on the stock returns of locally headquartered firms. Unlike the studies of Bilgin and Danis (2016), Levy and Yagil (2011), Lepori (2016), Li and Peng (2016), Heves et al. (2016), and Hu et al. (2014), which analyze air pollution effects using stock market indexes, we employ a firm-level analysis. Using intraday data as the sample, Zhang et al. (2017) found that air pollution significantly decreased returns of stocks headquartered in Beijing. Using a regression discontinuity model, Wu et al. (2018) indicated that only severe air pollution is associated with the stock returns of firms in six heavy pollution industries in China. Compared to the studies of Zhang et al. (2017) and Wu et al. (2018), who employ a special sample, we focus on the firms located in 33 main cities in China to provide a more general result. Our findings add evidence that air pollution reduces firm-level returns through investor moods. Additionally, the present study confirms that investors prefer to trade and hold locally headquartered stocks (Coval and Moskowitz, 1999; French and Poterba, 1991; Grinblatt and Keloharju, 2001; Ivković and Weisbenner, 2005). Second, to the best of our knowledge, this study is the first to examine how investor moods induced by air pollution affect firm-level trading activities. Hu et al. (2014) found that the air pollution in Shanghai relative to that in Beijing significantly decreased the total trading volume in the SSE and the Shenzhen Stock Exchange (SZSE). Focusing on firm-level data, we use the illiquidity from Amihud (2002) and turnover and volatility from Andersen et al., 2001a, Andersen et al., 2001b as trading activity variables; the results indicate that the impacts of air pollution on turnover and volatility are negative, whereas its impact on illiquidity is positive. Third, Baker and Wurgler (2006) show that investor mood exhibits a relatively sensitive effect on returns for small stocks, young stocks, high-volatility stocks, unprofitable stocks, non-dividend-paying stocks, extreme growth stocks, and distressed stocks. Using firm-level data, we find that the effects of negative moods induced by air pollution are strong on stock returns for firms sensitive to investor mood, supporting the findings of Baker and Wurgler (2006).

The remainder of this paper is organized as follows. Section 2 presents a literature review. Section 3 introduces the data and descriptive analysis. Section 4 tests for the existence of home bias in Chinese markets. Section 5 investigates how air pollution influences the stock returns and trading activities of locally headquartered firms and Section 6 examines the effects of air pollution on the local returns of firms that are difficult to value and arbitrage. Section 7 concludes the paper.

2. Literature review

2.1. Air pollution and moods

A growing body of research suggests that environmental conditions affect the psychological status of individuals, resulting in changes in decision making in financial markets. Saunders (1993), who conducted one of the earliest studies on this subject, determined that cloud cover in New York City significantly affected the index returns of the NYSE because a high cloud cover ratio induced depression among investors. Since then, several scholars determined that weather conditions mediated by moods influence investor behavior in financial markets, resulting in changes in stock returns and trading activity (Chang et al., 2008; Goetzmann et al., 2015; Hirshleifer and Shumway, 2003; Keef and Roush, 2007; Loughran and Schultz, 2004; Lu and Chou, 2012; Vijayakumar and Dharani, 2015). Kamstra et al. (2000) identified that the effect of weekends on daylight-saving time was related to stock market losses. Kamstra et al. (2003) observed a significant effect of seasonal affective disorder (SAD) on the stock market worldwide, supported by findings that daylight is correlated with risk aversion (Cohen et al., 1992). These studies verify that environmental conditions influence the psychological status and behavior of investors in financial markets.

Other environmental conditions caused by human activity may also influence this type of behavior. Air pollution is a typical example. Human activities to develop a regional economy, such as lumber milling, mining, and oil refining, significantly transform the natural environment and produce air pollution. Studies have determined that air pollution adversely affects human health (Brook et al., 2010; Maheswaran et al., 2005; Sagar et al., 2007). According to WHO statistics (World Health Statistics, 2016), air pollution is a major risk factor for people with noncommunicable diseases and leads to a significant increase in morbidity and mortality. Conducting research on behalf of the American Heart Association, Brook et al. (2004) identified a consistent increase in the risk of cardiovascular events in relation to both short- and long-term exposure to present-day concentrations of ambient air pollution. For example, increasing the 24-h PM_2.5 exposure by 10 μg/m³ could increase the relative risk for cardiovascular death by 0.4%–1.0%. Based on the National Morbidity, Mortality, and Air Pollution Study (NMMAPS) of 90 cities in the US, an analysis showed that mortality rates increase by 0.41% for every 10 μg/m³ increase in the PM₁₀ (Dominici et al., 2005). Similarly, Bell et al. (2005) used data from 95 cities in the NMMAPS study and found that an increase of 10 ppb ozone resulted in a 0.87% increase in mortality. Even a short exposure to pollution in the environment increases the incidence of cardiovascular diseases. Particulate matter also exacerbates respiratory diseases; it results in respiratory morbidity and mortality by creating oxidative stress and inflammation, leading to pulmonary anatomic and physiologic remodeling (Anderson et al., 2012).

According to a report from the WHO (2014), among the top 10 causes of death, ischemic heart disease, stroke, respiratory infection, and lung cancer are ranked as the top four. All four are related to air pollution. Sickness is inevitable for humans. However, if the risk of death results from a high risk of disease, people become extremely depressed. Psychology literature provides extensive evidence that air pollution induces depression (Horton et al., 2009; Lim et al., 2012; Power et al., 2015). Specifically, Cho et al. (2014) explores the short-term effects of ambient air pollution and finds that air pollution significantly increases the risk of depressive disorders. Using data from daily emergency department visits for depression, Szyszkowicz et al. (2009) show that ambient air pollution is positively related to the number of emergency department visits. Therefore, air pollution adversely affects people's moods in the short term.

2.2. Moods and decision making

In economics, the role of mood in decision making was rarely focused on for most of the 20th century (Lerner et al., 2015). However, a veritable revolution in the science of mood has evolved in the 21st century. According to Lerner et al. (2015), yearly research studies on mood and decision making doubled from 2007 to 2011. Numerous psychologists recognize that mood is the dominant driver of most meaningful decisions in life (Ekman, 2007; Keltner et al., 2014; Keltner and Lerner, 2010; Loewenstein et al., 2001). Stock pricing involves the trade-off between long-term benefits (future net cash flows) and costs (the riskiness of future cash flows); thus, it is expected that an investor's mood influences stock pricing (Loewenstein et al., 2001; Lucey and Dowling, 2005). Mood primarily influences investor decision making through mood misattribution. This concept builds on research from psychology, which considers decision making to be guided by moods, even if the moods are not induced by environmental factors but by other factors. In cases with complex decision making involving risk and uncertainty, individuals in bad moods are more pessimistic regarding stock pricing than those in neutral moods whereas individuals in good moods are more optimistic regarding stock pricing. Given that social interaction is an important aspect of the decision-making process, an aggregate optimistic/pessimistic mood of society is transmitted through this interaction, which influences all types of decision making, including stock investment (Nofsinger, 2005). The stock market reaches a high (low) when the social mood is high (low), and a rising (falling) stock market indicates an increasingly positive (negative) social mood trend.

There are various limits to arbitrage, such as implementation costs, noise trader risk, model risk, and fundamental risk. These phenomena have resulted in equities that occasionally remain mispriced, even when arbitrageurs suspect mispricing. This implies that even if only a small group of investors are influenced by mood, this influence may lead to identifiable patterns in stock returns (Lucey and Dowling, 2005).

Levy and Yagil (2011) and Lepori (2016) propose two possible ways that air pollution affects the trading decisions of investors in stock markets. First, an increase in the level of air pollution could induce negative moods. Thus, investors could be pessimistic and less willing to buy or hold stocks. Second, an increase in air pollution could increase bodily levels of cortisol, which may lead to less risk-taking behavior among individuals (Rosenblitt et al., 2001); investors could be cautious and risk averse. Thus, air pollution could decrease stock returns under these circumstances. To the best of our knowledge, very few studies explore the relation of air pollution-induced moods and stock trading activities. However, air pollution adversely affects investors' physical and psychological health. Therefore, investors may be unwilling to actively trade if they are feeling pessimistic (Liu, 2015; Loughran and Schultz, 2004). Negative moods could decrease trading volume (Gervais and Odean, 2001; Statman et al., 2006) and increase market illiquidity (Baker and Stein, 2004; Liu, 2015). There is a heated argument on the topic of volatility (Nofsinger, 2005). One perspective states that a high level of pessimism causes more investor difference of opinion and that this investor difference reflects itself in the market through volatility (Gao et al., 2006; Chang et al., 2008; Nofsinger, 2005). Thus, low moods induced by air pollution are associated with large return volatility. The other perspective states that investor moods can cause systematic risk (Brown, 1999). Investors could be overconfident and trade too aggressively when moods are rising, which could cause more extreme return volatility (Gervais and Odean, 2001; Nofsinger, 2005; Statman et al., 2006). Thus, mood is positively related to return volatility.

2.3. Home bias in stock pricing

The majority of studies about local bias in finance indicate that investors prefer to buy or hold locally headquartered stocks (Coval and Moskowitz, 1999; Ivković and Weisbenner, 2005; Shive, 2012). French and Poterba (1991) found that US investors generally allocated 94% of their funds for investment in US assets, even though the proportion of the US capital market to the global market was only 48%. Shan and Gong (2012) determined that a negative mood among investors significantly lowered the stock returns for firms headquartered nearer the epicenter of the Wenchuan earthquake in China relative to areas farther away. Edmans et al. (2007) observed that a loss in the World Cup decreased the next-day abnormal stock returns of the country who loses and the results were also robust for international cricket, rugby, and basketball games. Chang et al. (2012) supported this finding using firm-level data and showed that the effects of loss were higher for small, non-dividend-paying firms and for firms with high return volatility and low profitability.

Thus, the relationships between air pollution and mood and between mood and decision making (documented in the literature) have implications for the relationship between air pollution and investor behavior in the stock market. Air pollution may lead to a collective change in the level of risk aversion for investors with a home preference for stocks, resulting in lower locally headquartered stock returns. Additionally, negative moods induced by air pollution could cause investors to trade less, which could lead to changes in local trading activity.

3. Data and summary statistics

3.1. Dependent variables

The firms listed on the SSE and SZSE were analyzed to examine the relations between air pollution, stock returns, and trading activities. We chose the daily stock returns, turnover, illiquidity and volatility for each firm as dependent variables. Daily stock returns (in percentage) were measured as

$${\mathit{Return}}_{i,t}=\left({\mathit{Price}}_{i,t}-{\mathit{Price}}_{i,t-1}\right)/{\mathit{Price}}_{i,t-1}\times 100$$ (1)

where Price _i,t is the closing price adjusted by the dividend and share splitting for firm i on day t. Because firms in the same city are likely to have similar characteristics, we also employ the returns adjusted using the CAPM model (Ret_adj). The adjusted return for each firm on day t is the residual of the CAPM regression during days t-1 to t-60. Daily turnover (Turn) is the percentage of daily trading volume divided by the outstanding shares. According to Amihud (2002), illiquidity (Illiq) is the absolute daily returns divided by the daily dollar trading volume scaled by 10⁸. Due to the existence of 10% daily price limits in Chinese stock markets, a stock could continue to be traded, but only those orders at or within the limits are executed. Thus, we follow the method of Andersen et al., 2001a, Andersen et al., 2001b to calculate the daily volatility (Vol) for each stock. First, we constructed five-minute returns. Chinese stock markets are open from 9:30 to 11:30 in the morning and from 1:00 to 3:00 in the afternoon, for a total of 48 five-minute returns each day. We calculated the variances of the five-minute returns per day per stock to represent the volatility. All firm-level variable data were obtained from the Wind Info and CSMAR databases.

3.2. Key explanatory variables

Prior to 2013, the Ministry of Environmental Protection of the People's Republic of China (MEPC) provided the air pollution index (API), evaluated according to particulate matter (PM₁₀), sulfur dioxide (SO₂), and nitrogen dioxide (NO₂), to quantitatively describe the air quality status. Subsequently, the AQI was employed by the MEPC to replace the API measure. Relatively, the AQI is based on PM_2.5, PM₁₀, SO₂, carbon monoxide (CO), NO₂, and ozone (O₃). Thus, the AQI is more comprehensive than the API as it includes more air pollution indicators. The air pollution data in this study were collected from the website of the Chinese Air Quality Study Platform (www.aqistudy.cn), which has provided daily air pollution data for 190 cities in China since December 2013. The website reports the AQI index and pollutant concentrations. To make use of this, our research period was from December 1, 2013, to December 31, 2015. In the Technical Regulation on Ambient Air Quality Index published by the MEPC, the AQI is divided into six ranges, 0–50, 51–100, 101–150, 151–200, 201–300, and over 300, which correspond to six levels (excellent, good, lightly polluted, moderately polluted, heavily polluted, and severely polluted, respectively) of air quality. When the air quality is poor and presents a serious health effect, the AQI is large. We use the AQI and other air pollution indicators (PM_2.5, PM₁₀, SO₂, CO, NO₂, and O₃) to obtain robust results.

3.3. Main control variables

Previous literature verified that weather conditions, including cloud cover (Hirshleifer and Shumway, 2003; Loughran and Schultz, 2004; Saunders, 1993), temperature (Cao and Wei, 2005; Keef and Roush, 2007), humidity (Yoon and Kang, 2009), and wind speed (Keef and Roush, 2002), all crucially affect stock markets. Thus, in our study, weather factors are controlled for. The weather condition data were obtained from the Weather Underground Corporation website (WUC: www.wunderground.com), which provides hourly temperature, humidity, wind speed, air pressure, visibility, and weather status for major cities worldwide. The present study assumes that weather conditions could affect the mood of investors and thus influence their behaviors in financial markets. The weather data comprise daily average weather variables, including the relative humidity (Hum), temperature in degrees Celsius (Temp), air pressure in kPa (Pressure), visibility in km (Visibility), wind speed in km/h (Wind), and cloud cover (Cloud). The WUC website does not directly disclose the cloud cover ratio, so we created a dummy variable. The variable is set to one when the weather condition is rain, snow, fog, and other weather events that mostly or entirely cover the sky,2 and zero otherwise. After matching the air pollution and weather conditions data, 33 cities in China constituted the sample. From a total of 2724 listed firms at the end of 2015, 1548 firms located in 33 cities were included in the final sample.

Past market returns could also affect investors' expectations, as past market returns are a crucial determinant of sentiment (Brown and Cliff, 2004). Thus, the average returns of the Shanghai Composite Index (R _m) over the past 30 days were controlled for. Several well-known anomalies were also controlled for based on the work of Chang et al. (2012), Levy and Yagil (2011), and Lu and Chou (2012). The dummy for the Monday effect was set to one on Mondays and zero on other days. The seasonal affective disorder (SAD) effect was measured based on the method of Kamstra et al. (2003), but the value was not replaced by zero in spring and summer. Additionally, monthly dummy variables were also included.

3.4. Summary statistics

Table 1 summarizes the statistical information of the key variables.3 A summary analysis of the dependent variable is shown in Panel A. Panel B of Table 1 indicates that the proportion of unhealthy trading days (AQI > 100) is at least 25%, which is significantly higher than that in the findings of Levy and Yagil (2011).4 Panel D reports the correlations between air pollution variables and market sentiment index. We hypothesize that air quality affects investors' behavior through investor moods. Thus, we expect that air quality to be negatively related to a market sentiment index such as the BW sentiment index (Baker and Wurgler, 2006), which is extensively used in behavioral finance research. The dividend premium and the share of equity issues in total equity and debt issues in monthly data are not available in the Chinese market, so we used a consumer confidence index (CCI) and the number of new investors who open trading accounts for the first time (NA) to substitute for them. Following Baker and Wurgler (2006), we constructed a sentiment index as the first principal component of the turnover, the number of IPOs, the average first-day returns of IPOs, the close-end fund discount, the CCI and the NA. As seen in the first row in Panel D, we averaged the daily air pollution proxy in one month for each city.5 Six of seven monthly air pollution proxies have a significantly negative relation with the market sentiment index. As seen in the second row in Panel D, we calculated the daily average air pollution proxy of 33 cities for each day, and similar results were obtained. Panel E reports the correlations between firm-level raw and adjusted returns, illiquidity, turnover, volatility, and air pollution proxy variables. It shows that all the air pollution proxies, except O ₃, have significantly negative correlations with firm-level returns, verifying the results of Levy and Yagil (2011) and Lepori (2016). The correlations of air pollution proxies with firm-level turnover indicate that the worse the air quality in the city is, the lower the trading volume of local firms is, which is consistent with the positive relationship between air pollution and illiquidity. Regarding the correlations with volatility, poor air quality appears to decrease firm-level volatility. We divided the sample into three groups according to the air quality status. The air quality was good (bad) when its status was excellent (moderately, heavily or severely polluted). We tested for differences in the dependent variables of the bad and good air quality groups. The results are shown in Panel F and are consistent with those in Panel E.

Table 1.

Summary statistics.

Variables	Mean	Std	Min	P25	P50	P75	Max	Observations
Panel A: Firm-level variables
Return (%)	0.2629	3.6927	−10.0027	−1.5616	0.2271	1.9868	10.0124	594,414
Ret_adj (%)	−0.0151	2.7549	−8.2380	−1.4684	−0.1441	1.1742	9.5798	579,306
Illiq	2.7713	3.9459	0.0000	0.4740	1.3606	3.3605	32.7825	600,337
Turn (%)	3.4690	3.3813	0.1054	1.1290	2.3923	4.5950	21.0768	594,305
Vol (%)	0.2959	0.3535	0.0011	0.0816	0.1639	0.3652	2.5032	570,735
Panel B: Air pollution variables
AQI	88.3249	57.4362	12.0000	51.0000	72.0000	107.0000	500.0000	721,047
PM2.5 (μg/m3)	60.3256	50.5023	4.5000	27.6000	46.1000	75.7000	648.0000	721,047
PM10 (μg/m3)	91.4417	64.3074	0.0000	47.2000	74.6000	116.6000	977.3000	721,047
SO2 (μg/m3)	21.7118	25.1073	1.5000	8.6000	14.2000	24.7000	387.3000	721,047
CO (mg/m3)	1.0725	0.6198	0.1410	0.7240	0.9280	1.2120	10.6740	721,047
NO2 (μg/m3)	44.5805	20.8796	4.4000	29.8000	40.4000	54.9000	183.9000	721,047
O3 (μg/m3)	100.7102	56.7304	2.0000	59.0000	89.0000	133.0000	532.0000	721,047
Panel C: Control variables
SAD	−0.0302	1.5981	−3.7055	−1.4061	−0.024	1.2896	3.4526	721,047
Hum	0.6602	0.1821	0.0788	0.5508	0.6925	0.7971	0.9979	721,047
Temp (°C)	17.0644	10.1308	−21.7500	9.8750	19.1667	25.5417	34.5833	721,047
Pressure (kPa)	101.5736	0.8816	98.3792	100.8167	101.5833	102.2167	104.4958	721,047
Visibility (km)	8.8470	5.1387	0.2875	5.2222	7.8571	11.3750	30.0000	721,047
Wind (km/h)	10.0375	4.7058	1.3500	6.7500	8.8500	12.0000	45.0857	721,047
Cloud	0.3793	0.4852	0	0	0	1	1	721,047

Panel D: Correlations between air pollution variables and market sentiment index
	AQI	PM2.5	PM10	SO2	CO	NO2	O3
Sentm	−0.0742⁎⁎	−0.1014⁎⁎⁎	−0.1180⁎⁎⁎	−0.1374⁎⁎⁎	−0.0905⁎⁎⁎	−0.0773⁎⁎	0.0246
	(0.0331)	(0.0036)	(0.0007)	(0.0001)	(0.0093)	(0.0264)	(0.4802)
Sentd	−0.1789⁎⁎⁎	−0.1601⁎⁎⁎	−0.2011⁎⁎⁎	−0.2370⁎⁎⁎	−0.1601⁎⁎⁎	−0.1255⁎⁎⁎	0.0351
	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0005)	(0.3341)

Panel E: Correlations between firm-level variables and air pollution variables
	AQI	PM2.5	PM10	SO2	CO	NO2	O3
Return (%)	−0.0085⁎⁎⁎	−0.0113⁎⁎⁎	−0.0073⁎⁎⁎	−0.0232⁎⁎⁎	−0.0152⁎⁎⁎	−0.0048⁎⁎⁎	0.0078⁎⁎⁎
	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0002)	(0.0000)
Ret_adj (%)	−0.0079⁎⁎⁎	−0.0080⁎⁎⁎	−0.0.0081⁎⁎⁎	−0.0108⁎⁎⁎	−0.0095⁎⁎⁎	−0.0094⁎⁎⁎	−0.0092⁎⁎⁎
	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Illiq	0.0402⁎⁎⁎	0.0400⁎⁎⁎	0.0508⁎⁎⁎	0.0945⁎⁎⁎	0.0334⁎⁎⁎	0.0330⁎⁎⁎	0.0018
	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)	0.1591
Turn(%)	−0.0764⁎⁎⁎	−0.0734⁎⁎⁎	−0.0776⁎⁎⁎	−0.0933⁎⁎⁎	−0.0665⁎⁎⁎	−0.0632⁎⁎⁎	0.0298⁎⁎⁎
	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Vol (%)	−0.0999⁎⁎⁎	−0.0975⁎⁎⁎	−0.1060⁎⁎⁎	−0.1131⁎⁎⁎	−0.0954⁎⁎⁎	−0.1012⁎⁎⁎	0.0994⁎⁎⁎
	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)	(0.0000)

Panel F: Groups
	Return	Ret_adj	Illiq	Turn	Vol
Bad air quality	0.1608	−0.0467	3.1095	2.9265	0.2286
Good air quality	0.2715	0.0244	2.5922	3.7932	0.3476
Bad-Good	−0.1107⁎⁎⁎	−0.0711⁎⁎⁎	0.5173⁎⁎⁎	−0.8667⁎⁎⁎	−0.1190⁎⁎⁎

Panels A to C in this table report the summary statistics of main variables. Return is the daily raw returns in percentage for each firm. Ret_adj is the daily adjusted returns for each firm using 60-day rolling regressions. Illiq is the absolute daily returns divided by the daily dollar trading volume scaled by 10⁸. Turn is the daily trading volume divided by its outstanding shares. Vol is measured as variance of five-minute return within one day for each firm. AQI, PM_2.5, PM₁₀, SO₂, CO, NO₂, and O₃ are the air pollution proxy variables. Hum, Temp, Pressure, Visibility, Wind, and Cloud are relative humidity, temperature, air pressure, visibility, wind speed, and cloud cover ratio, respectively. SAD is the seasonal affective disorder effect for cities where the firm's headquarters are located and is calculated using the method of Kamstra et al. (2003). Sent is the first principal component of the turnover, the number of IPOs, the average first-day returns of IPOs, the close-end fund discount, the CCI and the NA. Panel D reports the correlation coefficients between air pollution proxies and market sentiment index. Panel E reports correlation coefficients between firm-level variables and air pollution proxies. Panel F reports average daily returns and trading activities of groups sorted by AQI. *, ** and *** present significance at the 10%, 5%, and 1% levels, respectively.

3.5. Baseline model

The baseline model in this study is the following

$$\begin{array}{l}{\mathit{Dep}}_{\mathit{i}\mathit{,}\mathit{t}}=\alpha +{\beta }_1{\mathit{Dep}}_{\mathit{i}\mathit{,}\mathit{t}\mathit{-}\mathit{1}}+{\beta }_2{\mathit{AQ}}_{\mathit{c}\mathit{,}\mathit{t}}+{\beta }_3{\mathit{SAD}}_{\mathit{c}\mathit{,}\mathit{t}}+{\beta }_4{\mathit{Hum}}_{\mathit{c}\mathit{,}\mathit{t}}+{\beta }_5{\mathit{Temp}}_{\mathit{c}\mathit{,}\mathit{t}}+{\beta }_6{\mathit{Pressure}}_{\mathit{c}\mathit{,}\mathit{t}}+{\beta }_7{\mathit{Visibility}}_{\mathit{c}\mathit{,}\mathit{t}}\\ +{\beta }_8{\mathit{Wind}}_{\mathit{c}\mathit{,}\mathit{t}}+{\beta }_9{\mathit{Cloud}}_{\mathit{c}\mathit{,}\mathit{t}}+{\beta }_{10}{R}_m+{\beta }_{11}\mathit{Monday}+\sum _{j=1}^{11}{\gamma }_j{\mathit{Month}}_j+{\epsilon }_{i,t},\end{array}$$ (2)

where Dep is the dependent variable magnified by 100 times (e.g., raw and adjusted returns, illiquidity, turnover, and volatility).6 c indicates the city where the firm i is headquartered. AQ is the city-level air pollution proxy, including API, AQI, PM _2.5, PM ₁₀, SO ₂, CO, NO ₂, and O ₃. SAD is city-level seasonal affective disorder. Hum, Temp, Pressure, Visibility, Wind, and Cloud are city-level humidity, temperature, air pressure, visibility, wind speed, and cloud cover, respectively. R _m is average return of the Shanghai Composite Index over the past 30 days. Monday and month effects are also included in the model. All the models are regressed by the cross-sectional firm-specific fixed effects method with robust standard errors clustered by firm, as in Angrist and Pischke (2009).

4. Test of home bias

As a test for home bias, we examined firm trading volume surrounding sudden air quality deterioration occurrences. Poor air quality would cause a pessimistic mood among investors, leading to less investor risk-taking and unwillingness to trade stocks. Therefore, if investors are prone to buy or hold local stocks, we would expect to see a decrease in trading volume for firms in cities suffering a sudden air quality deterioration, whereas firms in cities not suffering a sudden air quality deterioration should not see a decrease in the trading volume on the same day.

Table 2 reports the average trading volume for firms headquartered cities with and without a sudden air quality deterioration during December 2013 to December 2015. To obtain a clear setting, we selected the windows for which the AQI was >150 on day t, <100 during days t-5 to t-1 and during days t + 1 to t + 2. There were 957 windows that had a sudden air quality deterioration in at least one of the 33 cities. Panel A of Table 2 reports the average trading volume in thousand shares for firms during the 957 windows. On the days when a sudden air quality deterioration occurs, the average trading volume for firms in cities suffering a sudden air quality deterioration was 11,990 thousand shares, a 10.13% decline in trading volume from the previous trading volume. However, firms in cities without a sudden air quality deterioration show no significant difference in the average trading volume between day t and during days t-5 to t-1 (17,115 thousand shares vs. 16,896 thousand shares). Firms located cities without a sudden air quality deterioration experienced an increase in trading volume on day t + 1. In addition, the average trading volume for firms located in cities with a sudden air quality deterioration reverted to normal levels on day t + 2. Panel B of Table 2 reports the average trading volume in thousand Yuan for firms during the 957 windows, and the results are similar.

Table 2.

Trading volume for firms headquartered cities with and without a sudden air quality deterioration.

Cities	Average firm volume during days t-5 to t-1	Average firm volume on day t	Average firm volume on day t + 1	Average firm volume on day t + 2
Panel A: Trading volume in thousand shares
Cities without a sudden air quality deterioration	16,896	17,115	18,987	18,017
Cities with a sudden air quality deterioration	13,341	11,990	12,862	13,772
Panel B: Trading volume in thousand Yuan
Cities without a sudden air quality deterioration	197,436.1631	205,980.7172	224,934.2993	211,432.7595
Cities with a sudden air quality deterioration	134,119.5383	122,670.2597	134,455.9405	139,764.2691

This table reports the average trading volume for firms headquartered cities with and without a sudden air quality deterioration. The research period is limited to the 957 days in which a sudden air quality deterioration on a trading day for 33 Chinese cities.

Lepori (2016) provides evidence of the relation between air pollution and the market index returns from international data and determines that air pollution does not significantly affect the asset pricing of the SSE 50 and SSE 180 indexes in China. The Chinese A-share market adopts an order-driven mechanism in which investors in different cities can directly submit trading orders to the stock exchanges. Thus, air pollution effects may not exist for the market index. However, Lepori (2016) fails to consider that the Chinese A-share market adopts an order-driven mechanism in which investors in different cities can directly submit trading orders to the SSE. According to the 2016 SSE Annual Statistic Report, the trading orders are mainly from Shanghai, Beijing, Shenzhen, Hangzhou, and Guangzhou, and the trading volume from these cities account for 15.89%, 8.67%, 7.74%, 3.95%, and 3.88% of the total volume, respectively. Even if most orders are from Shanghai, if the trading orders reflect the influence of air pollution on investor moods where they are located, only 15.89% of the SSE index would be affected by the air quality in Shanghai, whereas 84.11% would be affected by the air quality in other cities. Therefore, the returns of the stock index reflect the level of air pollution in Shanghai and other cities where the firms in the market index are located. However, the air pollution in Shanghai may not significantly affect the returns of the stock index.

Home bias is expected to exist in a “pollution effect.” Accordingly, we investigated the effects of air pollution on market index returns and firm-level returns. First, the returns of the SSE and SZSE market indexes were regressed using air pollution in Shanghai City and Shenzhen City, respectively. Second, the effects of air pollution on the firm-level returns of locally headquartered firms on the SSE 50, SSE 180, SZSE 100, and SZSE 500 indexes were tested. Thirty-three major cities in China were selected to limit the data. Of the 1548 firms mentioned above, 43, 144, 68, and 309 firms belonged to the SSE 50, SSE 180, SZSE 100, and SZSE 500 indexes, respectively; these firms are headquartered in 10, 26, 24, and 28 cities, respectively.

The results are reported in Table 3 . Panel A of Table 3 shows that only three air pollution proxies of 14 significantly affected the SSE market index returns at the 10% significance level, indicating that the effect of air pollution in Shanghai on market index returns is insignificant. This finding is consistent with the results of Lepori (2016), who provides international evidence of the effects of air pollution on market index returns. However, Panel C indicates that 12 air pollution proxy variables of 14 exhibit a significant and negative effect on the firm-level returns of locally headquartered firms on the Shanghai indexes. Regarding the results on the SZSE, the association between air pollution in Shenzhen and market index returns was also insignificant (Panel B). Using firm-level data, 12 of 14 air pollution proxies were significantly negative at the 5% or higher level (Panel C). However, the absolute values of the coefficients on the air pollution proxies are higher in Panels A and B than those in Panel C. One possible reason why the relevant coefficients are statistically insignificant in Panels A and B yet are statistically significant in Panel C is that the air quality in Shanghai and Shenzhen is much better than the average air quality of the 33 cities in the study, and thus, the values of the air pollution proxies in Panels A and B are relatively smaller.7 Another likely reason is that the number of observations in Panel C is overwhelmingly greater than those in Panels A and B; thus, the relevant coefficients can be estimated more precisely. Consequently, in the Chinese stock market, with a vast territory and order-driven trading, the effect of air pollution in the city where the stock exchange is located on returns of the market indexes appears to be offset, necessitating a better method for including the home bias perspective.

Table 3.

Analysis on home bias.

Variables	AQI	PM2.5	PM10	SO2	CO	NO2	O3	Observations
Panel A: Effects of air pollution in Shanghai City on returns of the market index
SSE 50 Index	−0.2913	−0.5361⁎	−0.2685	−1.3406	−68.6005⁎	−0.2103	−0.0950	417
	(0.2711)	(0.3134)	(0.2762)	(1.3499)	(38.4434)	(0.6378)	(0.2955)
SSE 180 Index	−0.1729	−0.4162	−0.1280	−0.9621	−62.8389⁎	−0.2789	0.0078	417
	(0.2606)	(0.2984)	(0.2647)	(1.2832)	(36.9332)	(0.6158)	(0.3060)
Panel B: Effects of air pollution in Shenzhen City on returns of the market index
SZSE 100 Index	0.3359	0.4337	0.3599	1.7012	59.6622	0.8462	−0.0191	435
	(0.6606)	(0.7805)	(0.5763)	(3.0610)	(59.5743)	(0.9179)	(0.3823)
SZSE 500 Index	0.3606	0.4391	0.3772	2.3226	64.1579	1.0431	0.0186	435
	(0.6926)	(0.8101)	(0.6006)	(3.0945)	(60.6069)	(0.9471)	(0.3970)
Panel C: Effects of air pollution on firm-level returns of locally headquartered firms belonging to the market index
SSE 50 Index	−0.1824⁎⁎⁎	−0.2583⁎⁎⁎	−0.1084⁎⁎⁎	−0.9638⁎⁎⁎	−18.9867⁎⁎⁎	−0.4507⁎⁎⁎	−0.0314	17,603
	(0.0372)	(0.0458)	(0.0386)	(0.1399)	(3.1313)	(0.1395)	(0.0452)
SSE 180 Index	−0.1446⁎⁎⁎	−0.2149⁎⁎⁎	−0.0809⁎⁎⁎	−0.4089⁎⁎⁎	−19.8125⁎⁎⁎	−0.4104⁎⁎⁎	0.1362⁎⁎⁎	57,550
	(0.0205)	(0.0252)	(0.0183)	(0.1200)	(2.0713)	(0.0618)	(0.0345)
SZSE 100 Index	−0.2602⁎⁎⁎	−0.3281⁎⁎⁎	−0.1325⁎⁎⁎	−0.3577⁎⁎⁎	−20.2766⁎⁎⁎	−0.3498⁎⁎	0.0396	25,967
	(0.0488)	(0.0538)	(0.0429)	(0.1045)	(3.7570)	(0.1343)	(0.0596)
SZSE 500 Index	−0.1823⁎⁎⁎	−0.2366⁎⁎⁎	−0.1097⁎⁎⁎	−0.2844⁎⁎⁎	−14.9347⁎⁎⁎	−0.1741⁎⁎⁎	0.0756⁎⁎	115,708
	(0.0223)	(0.0253)	(0.0197)	(0.0572)	(1.9696)	(0.0636)	(0.0293)

This table reports the effects of air pollution in Shanghai (Shenzhen) where the SSE (SZSE) is located on the market index returns and returns of locally headquartered firms belonging to the market index, respectively. The estimated model is Eq. (2). AQI, PM_2.5, PM₁₀, SO₂, CO, NO₂, and O₃ are the air pollution proxy variables. All the models control for one-lagged returns, relative humidity (Hum), temperature (Temp), air pressure (Pressure), visibility (Visibility), wind speed (Wind), cloud cover ratio (Cloud), and SAD effects (SAD). Average return over past 30 trading days of the Shanghai Composite Index (R_m) is also controlled. In addition, Monday and month effects are included in the models. Models in Panels A and B are estimated by OLS method with robust standard errors in White (1980). Models in Panels C are estimated by cross-sectional firm-specific fixed-effect method with robust standard errors clustered by firm in Angrist and Pischke (2009). Robust standard errors are in parentheses. To save space, the coefficients of control variables are not reported. *, ** and *** present significance at the 10%, 5%, and 1% levels, respectively.

Li and Peng (2016) examined how air quality affects the returns of the Shanghai Composite Index and the Shenzhen Composite Index.8 The authors found no significant relationship over their study period but found a positive relation from 2005 to 2009. Using the same air pollution variable (API) from the MEPC9 as in Li and Peng (2016), from January 2005 to January 2013, the influence of air pollution on the market index returns and firm-level returns were re-run. The results are shown in Table 4 . Table 4 indicates that air pollution has no effect on the returns of the Shanghai Composite and the Shenzhen Composite indexes, which is consistent with the results of Li and Peng (2016). In the last two columns of Table 4, the firm-level data present the effects of air pollution on the firm-level returns of locally headquartered firms in 33 cities. The air pollution is negatively correlated with firm-level returns, and this remains robust from 2005 to 2009.

Table 4.

Regressions of air pollution index on stock returns.

Variables	Returns of the Shanghai Composite Index	Returns of the Shenzhen Composite Index		Firm-level Returns
				2005–2013	2005–2009
Returnt-1	−0.0792⁎⁎⁎	−0.0049		0.0112⁎⁎⁎	0.0127⁎⁎⁎
	(0.0283)	(0.0306)		(0.0020)	(0.0022)
API	−0.0803	−0.0297		−0.0264⁎⁎⁎	−0.1371⁎⁎⁎
	(0.1859)	(0.2174)		(0.0091)	(0.0114)
SAD	12.7129	25.0886		3.8156⁎⁎⁎	0.3386
	(20.4021)	(23.0266)		(0.3980)	(0.4468)
Monday effects	Yes	Yes		Yes	Yes
Month effects	Yes	Yes		Yes	Yes
Observations	1,528	1,528		1,373,808	817,474

This table reports the impacts of API on returns of market index and firm-level stocks, respectively, by using data in Li and Peng (2016). Models control for one-lagged returns, SAD effects, Monday and month effects. The second and third columns are estimated by OLS method with robust standard errors in White (1980). The last two columns are estimated by cross-sectional firm-specific fixed-effect method with robust standard errors clustered by firm in Angrist and Pischke (2009). Robust standard errors are in parentheses. *, ** and *** present significance at the 10%, 5%, and 1% levels, respectively.

To examine whether air pollution effects are caused by local air pollution rather than by air pollution where the stock exchange is located or by chance, we conducted a “placebo” experiment as follows: given a firm headquartered in a city, such as Beijing, we randomly created an air pollution time series from the series sample, excluding Beijing, assigned a returns time series for the firm, and ran regressions. After repeating this experiment a sufficient number of times, we counted the significant negative coefficients of air pollution variables at the level of 10%. If only a small percentage of the coefficients of the air pollution proxies was significantly negative, our opinion would be supported. Table 5 reports the results of the placebo experiment for 100, 500, and 1000 iterations, respectively. We found that most coefficients of the AQI were not significant, and some were even positive. The proportion of significant negative coefficients was not >10%, verifying the existence of home bias. These results indicate that the research methodology of Lepori (2016) and Li and Peng (2016) in Chinese markets could have been biased. Thus, firm-level data may be more appropriate to analyze the effects of air pollution on the Chinese stock market.

Table 5.

Placebo experiment.

	Mean	P5	P25	P50	P75	P95	Proportion of significantly negative coefficients
Panel A: simulating 100 times
AQI	−0.0987	−1.5501	−0.4303	−0.0659	0.3076	1.0933	6.00%
	(0.7597)	(1.2207)	(0.3564)	(0.3770)	(0.6837)	(1.1667)
PM2.5	−0.0998	−1.7187	−0.421	−0.0022	0.3607	1.2513	3.00%
	(0.8982)	(1.4275)	(0.5171)	(0.5407)	(0.5378)	(5.0644)
PM10	−0.0019	−0.9902	−0.3456	−0.0066	0.4077	1.0664	4.00%
	(0.6246)	(0.7241)	(0.3272)	(0.2289)	(0.4397)	(1.0553)
SO2	−0.2972	−5.2938	−1.2260	−0.2939	0.7195	4.4998	5.00%
	(2.9219)	(5.3182)	(1.3702)	(1.1765)	(1.1984)	(9.8664)
CO	−13.7747	−193.257⁎	−52.2318	0.5731	28.7115	116.6565	5.00%
	(79.1935)	(104.7011)	(60.5055)	(35.7918)	(51.5116)	(153.4080)
NO2	−0.1975	−3.6077	−1.2054	−0.0180	0.9896	2.9543	6.00%
	(2.0036)	(3.0371)	(2.1109)	(1.0829)	(1.2677)	(3.3220)
O3	−0.0891	−1.2323	−0.4209	−0.1316	0.3101	1.3104	5.00%
	(0.8053)	(0.8378)	(0.4777)	(0.6049)	(0.5079)	(1.0470)
Panel B: simulating 500 times
AQI	−0.0296	−1.3695	−0.2922	−0.0221	0.3220	1.2880	3.00%
	(0.7023)	(1.1117)	(0.6107)	(0.8460)	(0.9383)	(0.9311)
PM2.5	0.0079	−1.3998	−0.3281	−0.0142	0.3921	1.5570⁎	2.60%
	(0.8263)	(1.2553)	(0.4523)	(0.3677)	(0.8393)	(0.8511)
PM10	−0.0145	−1.0122⁎	−0.2529	−0.0015	0.2449	1.0579⁎	3.60%
	(0.5866)	(0.5535)	(0.3887)	(0.2704)	(0.6648)	(0.6047)
SO2	−0.2256	−4.8092	−1.2263	−0.1232	0.9607	4.6251	6.00%
	(2.7192)	(4.2046)	(0.9436)	(1.8463)	(1.7725)	(3.9954)
CO	−3.1066	−142.456	−34.6898	−0.4783	36.2847	122.5629	3.20%
	(76.8834)	(97.2988)	(59.2918)	(49.1646)	(61.3932)	(103.3992)
NO2	−0.1149	−2.8211	−1.0725	0.0099	0.9791	2.8883⁎⁎	3.40%
	(1.9754)	(2.2614)	(2.9138)	(1.0505)	(2.0329)	(1.2777)
O3	−0.0741	−1.3114	−0.5392	−0.0839	0.3486	1.3365⁎	3.6%
	(0.8014)	(1.2367)	(0.8720)	(0.6079)	(0.7580)	(0.7305)
Panel C: simulating 1000 times
AQI	−0.1016	−1.3039	−0.3832	−0.0589	0.2944	1.0692	6.30%
	(0.6822)	(1.7162)	(0.6243)	(0.8147)	(0.7269)	(0.8161)
PM2.5	−0.0788	−1.4946	−0.4146	−0.0531	0.3747	1.3429	5.00%
	(0.8110)	(1.1134)	(0.4239)	(0.9008)	(0.5193)	(1.3749)
PM10	−0.0383	−1.0212⁎	−0.2716	−0.0221	0.2659	0.8648	5.30%
	(0.5648)	(0.5762)	(0.2936)	(0.5377)	(0.3964)	(0.6412)
SO2	−0.2044	−4.3740	−1.2844	−0.2789	0.7359	4.1665	6.70%
	(2.7181)	(4.0342)	(1.0678)	(1.8398)	(1.3291)	(2.9814)
CO	−5.7266	−142.566	−40.6200	−2.5603	32.1341	126.5916⁎	5.40%
	(76.0466)	(87.6906)	(85.9604)	(32.5372)	(67.2112)	(71.1274)
NO2	−0.1144	−3.3572	−1.1213	−0.0391	0.9537	2.7946⁎	5.10%
	(1.8825)	(1.9271)	(1.9153)	(1.5830)	(1.4626)	(1.5559)
O3	−0.0551	−1.2107	−0.4955	−0.0729	0.3515	1.2892⁎	4.70%
	(0.7805)	(1.0336)	(1.6427)	(0.8480)	(0.8201)	(0.6967)

This tables reports the summary statistics of coefficients of AQI by repeating placebo experiment for 100, 500 and 1000 times, respectively. The placebo experiment as follows: given a firm headquartered in a city, such as Beijing, we randomly draw an air pollution time series from the sample of series except for that in Beijing to assign returns time series of the firm, and then run regressions. The dependent variable is stock returns of locally headquartered firms. AQI, PM_2.5, PM₁₀, SO₂, CO, NO₂, and O₃ are the air pollution proxy variables. All the models control for one-lagged returns, relative humidity (Hum), temperature (Temp), air pressure (Pressure), visibility (Visibility), wind speed (Wind), cloud cover ratio (Cloud), and SAD effects (SAD). Average returns over past 30 trading days of the Shanghai Composite Index (R_m) is also controlled. In addition, Monday and month effects are included in the models. All the models are estimated by OLS method with robust standard errors in White (1980). Robust standard errors are in parentheses. *, ** and *** present significance at the 10%, 5%, and 1% levels, respectively.

5. Air pollution, stock returns and trading activities

5.1. Overall empirical results

The results above indicate that home bias exists in the relationship between air pollution and equity pricing of locally headquartered firms in the Chinese stock market. To obtain comprehensive evidence, the research sample was expanded to include the A-shares in 33 major Chinese cities. The proportion of the sample in A-shares is 55.11% and the ratio of the market value of the sample to that of all A-shares is 69.24%. Thus, the sample is sufficient for the study and verifies the “pollution effect.” To examine whether air pollution could lead to risk-aversion behavior among local investors and thus influence the stock pricing of firms located in the same geographical area, seven air pollution proxies were adopted to obtain credible results. Daily stock returns, illiquidity, turnover, and volatility on air pollution were regressed where the firms were headquartered. A cross-sectional firm-specific fixed-effect estimation method was adopted after controlling for weather conditions, SAD, Monday, and month effects. Table 6 presents the results of the effects of air pollution on stock returns and adjusted returns, and Table 7 presents the effects of air pollution on illiquidity, turnover, and volatility.

Table 6.

Effects of air pollution on the stock returns.

Variables	Return							Ret_adj
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model9	Model 10	Model 11	Model 12	Model 13	Model 14
Returnt-1(Ret_adjt-1)	0.0844⁎⁎⁎	0.0844⁎⁎⁎	0.0844⁎⁎⁎	0.0843⁎⁎⁎	0.0842⁎⁎⁎	0.0845⁎⁎⁎	0.0845⁎⁎⁎	0.0761⁎⁎⁎	0.0761⁎⁎⁎	0.0761⁎⁎⁎	0.0760⁎⁎⁎	0.0760⁎⁎⁎	0.0761⁎⁎⁎	0.0760⁎⁎⁎
	(0.0021)	(0.0021)	(0.0021)	(0.0021)	(0.0021)	(0.0021)	(0.0021)	(0.0019)	(0.0019)	(0.0019)	(0.0019)	(0.0019)	(0.0019)	(0.0019)
AQI	−0.0734⁎⁎⁎							−0.0111
	(0.0093)							(0.0089)
PM2.5		−0.0956⁎⁎⁎							−0.0213⁎⁎
		(0.0105)							(0.0102)
PM10			−0.0630⁎⁎⁎							−0.0125⁎
			(0.0081)							(0.0075)
SO2				−0.2696⁎⁎⁎							−0.1709⁎⁎⁎
				(0.0228)							(0.0254)
CO					−9.7169⁎⁎⁎							−2.8185⁎⁎⁎
					(0.9181)							(0.9154)
NO2						−0.1147⁎⁎⁎							−0.0758⁎⁎⁎
						(0.0286)							(0.0253)
O3							0.0811⁎⁎⁎							0.0303⁎⁎⁎
							(0.0129)							(0.0093)
SAD	9.2199⁎⁎⁎	9.4923⁎⁎⁎	9.0469⁎⁎⁎	11.9285⁎⁎⁎	11.0257⁎⁎⁎	8.9262⁎⁎⁎	10.7322⁎⁎⁎	3.8823⁎⁎⁎	4.0129⁎⁎⁎	3.8901⁎⁎⁎	5.7154⁎⁎⁎	4.4748⁎⁎⁎	4.1122⁎⁎⁎	4.6268⁎⁎⁎
	(0.9836)	(0.9847)	(0.9796)	(1.0330)	(1.0323)	(0.9882)	(1.0916)	(0.7517)	(0.7561)	(0.7476)	(0.8028)	(0.7983)	(0.7521)	(0.7658)
Cloud	−9.6917⁎⁎⁎	−9.5475⁎⁎⁎	−9.8754⁎⁎⁎	−9.7723⁎⁎⁎	−10.0609⁎⁎⁎	−10.3274⁎⁎⁎	−11.0042⁎⁎⁎	−5.1854⁎⁎⁎	−5.0875⁎⁎⁎	−5.1758⁎⁎⁎	−4.7594⁎⁎⁎	−5.1955⁎⁎⁎	−5.1069⁎⁎⁎	−5.4955⁎⁎⁎
	(1.0707)	(1.0701)	(1.0662)	(1.0618)	(1.0520)	(1.0625)	(1.0407)	(0.8652)	(0.8643)	(0.8631)	(0.8573)	(0.8542)	(0.8609)	(0.8503)
Temp	2.0559⁎⁎⁎	2.0538⁎⁎⁎	2.0221⁎⁎⁎	1.8749⁎⁎⁎	1.9523⁎⁎⁎	1.9907⁎⁎⁎	1.8570⁎⁎⁎	1.3471⁎⁎⁎	1.3510⁎⁎⁎	1.3441⁎⁎⁎	1.3084⁎⁎⁎	1.3320⁎⁎⁎	1.3379⁎⁎⁎	1.2855⁎⁎⁎
	(0.1219)	(0.1215)	(0.1209)	(0.1198)	(0.1203)	(0.1202)	(0.1183)	(0.1023)	(0.1022)	(0.1018)	(0.1015)	(0.1016)	(0.1017)	(0.1023)
Visibility	−1.4119⁎⁎⁎	−1.4510⁎⁎⁎	−1.3785⁎⁎⁎	−1.3096⁎⁎⁎	−1.4388⁎⁎⁎	−1.1969⁎⁎⁎	−0.9060⁎⁎⁎	−0.2771⁎⁎⁎	−0.3047⁎⁎⁎	−0.2855⁎⁎⁎	−0.3359⁎⁎⁎	−0.3183⁎⁎⁎	−0.3036⁎⁎⁎	−0.1640⁎
	(0.1236)	(0.1232)	(0.1208)	(0.1166)	(0.1209)	(0.1172)	(0.1227)	(0.0988)	(0.0984)	(0.0970)	(0.0932)	(0.0958)	(0.0934)	(0.0936)
Hum	40.1457⁎⁎⁎	41.0935⁎⁎⁎	37.8524⁎⁎⁎	35.1457⁎⁎⁎	43.7013⁎⁎⁎	41.2028⁎⁎⁎	51.7629⁎⁎⁎	16.3474⁎⁎⁎	16.3482⁎⁎⁎	15.7354⁎⁎⁎	12.1903⁎⁎⁎	17.0964⁎⁎⁎	15.3660⁎⁎⁎	20.1659⁎⁎⁎
	(3.0877)	(3.0658)	(3.1716)	(3.1813)	(3.0707)	(3.1246)	(3.3208)	(2.7417)	(2.7178)	(2.8116)	(2.8460)	(2.6958)	(2.7682)	(2.8745)
Wind	0.6040⁎⁎⁎	0.5756⁎⁎⁎	0.5881⁎⁎⁎	0.5128⁎⁎⁎	0.5409⁎⁎⁎	0.5382⁎⁎⁎	0.7263⁎⁎⁎	0.3629⁎⁎⁎	0.3538⁎⁎⁎	0.3584⁎⁎⁎	0.2964⁎⁎⁎	0.3401⁎⁎⁎	0.2900⁎⁎⁎	0.3951⁎⁎⁎
	(0.1202)	(0.1205)	(0.1198)	(0.1197)	(0.1215)	(0.1269)	(0.1214)	(0.0877)	(0.0879)	(0.0878)	(0.0884)	(0.0881)	(0.0933)	(0.0883)
Pressure	−2.3220⁎	−2.3488⁎	−2.6490⁎⁎	−3.0099⁎⁎	−3.1632⁎⁎	−2.2922⁎	−2.0662⁎	10.5985⁎⁎⁎	10.5539⁎⁎⁎	10.5237⁎⁎⁎	10.1225⁎⁎⁎	10.2992⁎⁎⁎	10.4428⁎⁎⁎	10.6313⁎⁎⁎
	(1.2380)	(1.2393)	(1.2384)	(1.2436)	(1.2438)	(1.2480)	(1.2456)	(1.0176)	(1.0168)	(1.0214)	(1.0244)	(1.0144)	(1.0219)	(1.0194)
Rm	79.8345⁎⁎⁎	79.5521⁎⁎⁎	79.9329⁎⁎⁎	79.8042⁎⁎⁎	79.9959⁎⁎⁎	80.0520⁎⁎⁎	80.4053⁎⁎⁎	−10.9440⁎⁎⁎	−11.0134⁎⁎⁎	−10.9406⁎⁎⁎	−10.6781⁎⁎⁎	−10.9355⁎⁎⁎	−11.0044⁎⁎⁎	−10.8391⁎⁎⁎
	(1.2837)	(1.2858)	(1.2834)	(1.2873)	(1.2829)	(1.2831)	(1.2835)	(0.9682)	(0.9685)	(0.9685)	(0.9677)	(0.9681)	(0.9684)	(0.9690)
Monday	7.1639⁎⁎⁎	7.1007⁎⁎⁎	7.0771⁎⁎⁎	7.1919⁎⁎⁎	7.3493⁎⁎⁎	6.9473⁎⁎⁎	6.9107⁎⁎⁎	5.3138⁎⁎⁎	5.3219⁎⁎⁎	5.3131⁎⁎⁎	5.3822⁎⁎⁎	5.3760⁎⁎⁎	5.2448⁎⁎⁎	5.2641⁎⁎⁎
	(1.0184)	(1.0179)	(1.0184)	(1.0197)	(1.0173)	(1.0201)	(1.0191)	(0.8825)	(0.8822)	(0.8824)	(0.8821)	(0.8812)	(0.8828)	(0.8825)
Intercept	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Month effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm-specific fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	581,001	581,001	581,001	581,001	581,001	581,001	581,001	568,138	568,138	568,138	568,138	568,138	568,138	568,138
F statistic	380.1884	380.9404	380.3269	385.4419	381.2302	380.1669	379.3902	166.0603	165.9980	165.7542	166.5752	165.7816	165.9335	165.6090
Adj-R2	0.0197	0.0197	0.0197	0.0198	0.0198	0.0197	0.0197	0.0114	0.0114	0.0114	0.0115	0.0115	0.0115	0.0115

This table reports the effects of air pollution on the stock returns (Return and Ret_adj) of locally headquartered firms. AQI, PM_2.5, PM₁₀, SO₂, CO, NO₂, and O₃ are the air pollution proxy variables. The estimated model is Eq. (2). All the models control for one-lagged returns, relative humidity (Hum), temperature (Temp), air pressure (Pressure), visibility (Visibility), wind speed (Wind), cloud cover ratio (Cloud), and SAD effects (SAD). Average returns over past 30 trading days of the Shanghai Composite Index (R_m) is also controlled. In addition, Monday and month effects are included in the models. All the models are estimated by cross-sectional firm-specific fixed-effect method with robust standard errors clustered by firm in Angrist and Pischke (2009). Robust standard errors are in parentheses. *, ** and *** present significance at the 10%, 5%, and 1% levels, respectively.

Table 7.

Effects of air pollution on illiquidity, turnover, and volatility.

Variables	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
Panel A: Illiquidity
AQI	0.1045⁎⁎⁎
	(0.0120)
PM2.5		0.0920⁎⁎⁎
		(0.0132)
PM10			0.1100⁎⁎⁎
			(0.0108)
SO2				0.8453⁎⁎⁎
				(0.0569)
CO					5.3902⁎⁎⁎
					(1.1286)
NO2						0.3392⁎⁎⁎
						(0.0378)
O3							0.0159
							(0.0147)
SAD	−10.5048⁎⁎⁎	−10.3979⁎⁎⁎	−10.4826⁎⁎⁎	−20.2987⁎⁎⁎	−10.8062⁎⁎⁎	−10.8873⁎⁎⁎	−8.9016⁎⁎⁎
	(1.5470)	(1.5527)	(1.5461)	(1.6379)	(1.5689)	(1.5672)	(1.5872)
Cloud	5.2759⁎⁎⁎	5.6121⁎⁎⁎	5.2723⁎⁎⁎	3.7902⁎⁎⁎	6.3712⁎⁎⁎	5.5977⁎⁎⁎	6.6826⁎⁎⁎
	(1.2735)	(1.2726)	(1.2758)	(1.2654)	(1.2532)	(1.2716)	(1.2584)
Temp	−2.8223⁎⁎⁎	−2.7881⁎⁎⁎	−2.7867⁎⁎⁎	−2.3798⁎⁎⁎	−2.7049⁎⁎⁎	−2.7379⁎⁎⁎	−2.7488⁎⁎⁎
	(0.2569)	(0.2572)	(0.2559)	(0.2516)	(0.2553)	(0.2569)	(0.2590)
Month effects Firm-specific fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	594,220	594,220	594,220	594,220	594,220	594,220	594,220
F statistic	967.4660	967.3922	965.7472	959.0008	965.2903	965.6434	965.0934
Adj-R2	0.2038	0.2037	0.2039	0.2052	0.2037	0.2038	0.2036

Panel B: Turnover
AQI	−0.0483⁎⁎⁎
	(0.0050)
PM2.5		−0.0510⁎⁎⁎
		(0.0056)
PM10			−0.0524⁎⁎⁎
			(0.0046)
SO2				−0.1345⁎⁎⁎
				(0.0191)
CO					−2.9050⁎⁎⁎
					(0.5687)
NO2						−0.1470⁎⁎⁎
						(0.0172)
O3							−0.0122⁎
							(0.0063)
SAD	−4.2313⁎⁎⁎	−4.1835⁎⁎⁎	−4.2252⁎⁎⁎	−3.0086⁎⁎⁎	−3.9855⁎⁎⁎	−4.1037⁎⁎⁎	−5.1068⁎⁎⁎
	(0.7205)	(0.7209)	(0.7193)	(0.7538)	(0.7220)	(0.7201)	(0.7308)
Cloud	−1.9785⁎⁎⁎	−2.0290⁎⁎⁎	−1.9553⁎⁎⁎	−2.1783⁎⁎⁎	−2.4523⁎⁎⁎	−2.1562⁎⁎⁎	−2.6030⁎⁎⁎
	(0.6494)	(0.6496)	(0.6487)	(0.6463)	(0.6420)	(0.6486)	(0.6431)
Temp	0.5265⁎⁎⁎	0.5168⁎⁎⁎	0.5111⁎⁎⁎	0.4254⁎⁎⁎	0.4707⁎⁎⁎	0.4870⁎⁎⁎	0.5002⁎⁎⁎
	(0.0935)	(0.0937)	(0.0935)	(0.0932)	(0.0935)	(0.0937)	(0.0944)
Month effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm-specific fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	586,020	586,020	586,020	586,020	586,020	586,020	586,020
F statistic	6047.8033	6045.1600	6046.3436	6113.7227	6041.1413	6047.6738	6046.9971
Adj-R2	0.5839	0.5839	0.5839	0.5839	0.5839	0.5839	0.5838

Panel C: Volatility
AQI	−0.0102⁎⁎⁎
	(0.0008)
PM2.5		−0.0090⁎⁎⁎
		(0.0008)
PM10			−0.0125⁎⁎⁎
			(0.0007)
SO2				−0.0103⁎⁎⁎
				(0.0024)
CO					−0.3553⁎⁎⁎
					(0.0830)
NO2						−0.0317⁎⁎⁎
						(0.0026)
O3							−0.0086⁎⁎⁎
							(0.0011)
SAD	−0.9083⁎⁎⁎	−0.9183⁎⁎⁎	−0.8938⁎⁎⁎	−0.8852⁎⁎⁎	−0.9238⁎⁎⁎	−0.8739⁎⁎⁎	−1.2727⁎⁎⁎
	(0.1102)	(0.1104)	(0.1098)	(0.1158)	(0.1115)	(0.1101)	(0.1135)
Cloud	0.1303	0.0991	0.1501	0.0276	0.0143	0.0994	0.0241
	(0.1050)	(0.1052)	(0.1044)	(0.1045)	(0.1050)	(0.1054)	(0.1051)
Temp	0.1694⁎⁎⁎	0.1661⁎⁎⁎	0.1665⁎⁎⁎	0.1554⁎⁎⁎	0.1587⁎⁎⁎	0.1619⁎⁎⁎	0.1732⁎⁎⁎
	(0.0126)	(0.0126)	(0.0126)	(0.0127)	(0.0126)	(0.0126)	(0.0128)
Visibility	0.2600⁎⁎⁎	0.2717⁎⁎⁎	0.2468⁎⁎⁎	0.2984⁎⁎⁎	0.2939⁎⁎⁎	0.2712⁎⁎⁎	0.2913⁎⁎⁎
	(0.0190)	(0.0189)	(0.0194)	(0.0183)	(0.0187)	(0.0192)	(0.0179)
Month effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm-specific fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	560,632	560,632	560,632	560,632	560,632	560,632	560,632
F statistic	4855.0080	4853.4050	4852.4970	4853.9704	4865.7690	4861.3504	4850.5166
Adj-R2	0.3390	0.3389	0.3391	0.3388	0.3388	0.3390	0.3389

This table reports the effects of air pollution on the illiquidity (Illiq), turnover (Turn), and volatility (Vol) of locally headquartered firms. AQI, PM_2.5, PM₁₀, SO₂, CO, NO₂, and O₃ are the air pollution proxy variables. The estimated model is Eq. (2). All the models control for one-lagged dependent variables, relative humidity (Hum), temperature (Temp), air pressure (Pressure), visibility (Visibility), wind speed (Wind), cloud cover ratio (Cloud), and SAD effects (SAD). Average returns over past 30 trading days of the Shanghai Composite Index (R_m) is also controlled. In addition, Monday and month effects are included in the models. All the models are estimated by cross-sectional firm-specific fixed-effect method with robust standard errors clustered by firm in Angrist and Pischke (2009). Robust standard errors are in parentheses. To save space, some coefficients of control variables are not reported. *, ** and *** present significance at the 10%, 5%, and 1% levels, respectively.

Model 1 to Model 7 of Table 6 show that all air pollution proxies, except for ozone, significantly and negatively influence stock returns at the 1% significance level, which is consistent with the results of Lepori (2016) and Levy and Yagil (2011). One change in standard deviation in the AQI (57.4362 from Table 1) decreases stock returns by 0.0422% (0.000734 × 57.4362),10 which is approximately 1.1417% of one standard deviation (3.6927% from Table 1) of the stock returns. The influence of other air pollution proxies on stock returns is economically insignificant. Thus, the study confirms that a “pollution effect” exists, as proposed by Levy and Yagil (2011), despite its negligible economic magnitude. Model 8 to Model 14 of Table 6 indicates that the adjusted returns show similar results. Table 7 reports the effects of air pollution on firm-level illiquidity, turnover, and volatility. The results indicate that air pollution increases the illiquidity of locally headquartered firms whereas it decreases firm-level turnover. Loughran and Schultz (2004) note that pessimistic moods induced by extreme weather conditions could lead to a dramatic decrease in the trading volume of local firms. Similarly, a high level of air pollution could dampen local investors' desires to trade, reducing turnover (Lepori, 2016). Panel C conflicts with the opinion that return volatility only reflects investor disagreement (Gao et al., 2006; Shalen, 1993), however, the panel supports the opinion that investors may trade actively when moods are increasingly positive, creating a high level of return volatility (Gervais and Odean, 2001; Nofsinger, 2005; Statman et al., 2006).

The coefficients of the SAD effects positively influence local stock returns. This finding is consistent with that of Kamstra et al. (2003), who determined that SAD effects cause significant and positive returns in markets at higher latitudes relative to those closer to the equator. Kramer and Weber (2012) determine that investors affected by SAD effects trade cautiously and carefully and prefer safe investment portfolios, unlike investors unaffected by SAD effects. The results indicate that SAD effects are negatively associated with local turnover, implying investors suffering from SAD effects may trade less, which is consistent with the findings of Kramer and Weber (2012). Here, SAD effects reduce local illiquidity. Illiquidity is the response of price to order flow, resulting from adverse selection costs and inventory costs (Amihud and Mendelson, 1980). Therefore, the information in the illiquidity is different from that in turnover. If investors affected by SAD effects trade more cautiously and carefully, they would obtain enough information before executing a market order, reducing information asymmetry in the market and the possibility of adverse selection and thus reducing the illiquidity. In addition, the results show that investors suffering from SAD effects would not trade actively, decreasing return volatility.

Regarding weather conditions, a high cloud cover level significantly decreases the returns of locally headquartered firms, which is consistent with the results of Saunders (1993) and Hirshleifer and Shumway (2003). Cloud cover is negatively related with turnover and volatility and positively related with illiquidity. This relation indicates that cloud cover leads to bad moods in investors, and thus, they are likely to trade much less. Cao and Wei (2005) note that people could be more both aggressive and apathetic in higher temperature conditions. Aggression reduces investor risk-taking whereas does apathy not. Therefore, the relationship between stock returns and temperature depends on competition between these two effects. Our results show that a high temperature level increases local returns, turnover and volatility, whereas it decreases illiquidity, supporting the view that aggression caused by high temperature plays a key role. Other weather variables, including visibility, humidity, wind speed and air pressure, also affect local returns and trading activities, which confirms previous studies on weather effects.

5.2. Robustness checks

In this research, firms located in the same city are affected by the same factors (including local air quality), which may lead to cross-sectional correlations in the error terms of the regression models. Thus, we used the Fama-MacBeth regression method to obtain robustness results (Fama and MacBeth, 1973), shown in Panel A in Table 8 . For brevity, we only report the coefficients of the air pollution proxies. We found that the air pollution effects are generally consistent with the results above. The coefficients of the AQI and PM_2.5 on local returns were significantly negative at the 10% and 5% levels, respectively. Air pollution negatively influences the turnover of locally headquartered firms, and five of seven coefficients were significant at the 1% level. The coefficients of the AQI, PM_2.5, PM_10, and SO₂ on return volatility were significantly negative, which verifies that market volatility decreased when the investor mood is pessimistic. PM₁₀, SO₂, NO₂ and O₃ were positive and related to local illiquidity, which reveals that low moods induced by air pollution reduce local liquidity. In addition, we were concerned that the error terms in the regression models appeared to be time-series autocorrelated because the dependent variables are very persistent. Thus, we ran the regressions with Driscoll-Kraay standard errors (Driscoll and Kraay, 1998) and obtained similar results to those in Panel B, although the significance was relatively weak.

Table 8.

Robustness checks.

	AQI	PM2.5	PM10	SO2	CO	NO2	O3	Observations
Panel A: Fama-MacBeth regressions
Return	−0.0370⁎	−0.0581⁎⁎	−0.0150	−0.0769	−1.5189	0.0376	0.0057	594,414
	(0.0219)	(0.0269)	(0.0178)	(0.0773)	(2.1160)	(0.0512)	(0.0194)
Ret_adj	−0.0075	−0.0340⁎	0.0054	−0.0245	−0.4275	−0.0037	0.0016	579,306
	(0.0237)	(0.0204)	(0.0176)	(0.0745)	(2.0151)	(0.0463)	(0.0173)
Illiq	−0.0068	−0.0047	0.1046⁎⁎⁎	0.7093⁎⁎⁎	10.4260⁎⁎⁎	0.0021	0.0845⁎⁎⁎	600.337
	(0.0239)	(0.0308)	(0.0210)	(0.0743)	(3.0314)	(0.0596)	(0.0233)
Turn	−0.1628⁎⁎⁎	−0.1736⁎⁎⁎	−0.1131⁎⁎⁎	−0.0192	−10.5633⁎⁎⁎	−0.2728⁎⁎⁎	−0.0131	594,305
	(0.0233)	(0.0285)	(0.0172)	(0.0651)	(2.3257)	(0.0498)	(0.0202)
Vol	−0.0060⁎⁎	−0.0052⁎	−0.0048⁎⁎	−0.0303⁎⁎⁎	−0.1827	0.0049	0.0014	570,735
	(0.0026)	(0.0030)	(0.0019)	(0.0074)	(0.2712)	(0.0050)	(0.0020)
Panel B: Regressions with Driscoll-Kraay standard errors
Return	−0.0641	−0.1059⁎	−0.0527	−0.2995⁎⁎	−8.6043⁎	−0.0785	0.0573	581,001
	(0.0598)	(0.0635)	(0.0501)	(0.1230)	(5.0758)	(0.1684)	(0.0692)
Ret_adj	−0.0007	−0.0573⁎	−0.0457⁎	−0.0667	−3.6684⁎	−0.1553⁎⁎	0.0183	568,138
	(0.0322)	(0.0336)	(0.0275)	(0.0580)	(2.0843)	(0.0707)	(0.0296)
Illiq	−0.0125	−0.0272	0.0503	0.5299⁎⁎⁎	−1.7131	−0.0189	0.0005	594,220
	(0.0455)	(0.0470)	(0.0356)	(0.0877)	(3.2740)	(0.1076)	(0.0335)
Turn	−0.0360⁎⁎	−0.0397⁎⁎	−0.0395⁎⁎⁎	−0.0831⁎⁎⁎	−2.5595⁎	−0.0822⁎	−0.0134	581,001
	(0.0157)	(0.0176)	(0.0127)	(0.0305)	(1.4469)	(0.0440)	(0.0165)
Vol	−0.0081⁎	−0.0070	−0.0113⁎⁎⁎	−0.0121	−0.3921	−0.0181	−0.0074	560,632
	(0.0049)	(0.0056)	(0.0042)	(0.0077)	(0.4174)	(0.0125)	(0.0078)
Panel C: Regressions with robust error clustered by city
Return	−0.0734⁎	−0.0956⁎⁎	−0.0630⁎	−0.2696⁎⁎⁎	−9.7169⁎⁎⁎	−0.1147	0.0811	581,001
	(0.0371)	(0.0416)	(0.0367)	(0.0886)	(3.1069)	(0.1622)	(0.1029)
Ret_adj	−0.0284⁎	−0.0363⁎⁎	−0.0362⁎⁎⁎	−0.1709⁎⁎	−3.0276⁎⁎	−0.0758	0.0303	568,138
	(0.0157)	(0.0172)	(0.0132)	(0.0635)	(1.4231)	(0.0681)	(0.0405)
Illiq	0.1045⁎	0.0920	0.1100⁎⁎⁎	0.8453⁎⁎⁎	5.3902	0.3392⁎⁎	0.0159	594,220
	(0.0578)	(0.0676)	(0.0377)	(0.1817)	(7.6222)	(0.1256)	(0.0660)
Turn	−0.0483⁎⁎⁎	−0.0510⁎⁎⁎	−0.0524⁎⁎⁎	−0.1345⁎⁎⁎	−2.9050	−0.1470⁎⁎	−0.0122	586,020
	(0.0153)	(0.0158)	(0.0135)	(0.0446)	(3.3205)	(0.0543)	(0.0187)
Vol	−0.0102⁎⁎	−0.0090⁎⁎	−0.0125⁎⁎⁎	−0.0103	−0.3553	−0.0317⁎⁎	−0.0086⁎⁎	560,632
	(0.0039)	(0.0039)	(0.0038)	(0.0088)	(0.6654)	(0.0154)	(0.0040)
Panel D: Regressions with both firm-specific and time-specific fixed effects
Return	−0.0134	−0.0200⁎	−0.0091	−0.0337⁎	−1.9496⁎⁎	−0.0670⁎⁎⁎	−0.0126	581,001
	(0.0090)	(0.0103)	(0.0075)	(0.0194)	(0.8763)	(0.0254)	(0.0097)
Ret_adj	−0.0131	−0.0218⁎⁎	−0.0041	−0.0334	−2.3002⁎⁎	−0.0460⁎	0.0006	568,138
	(0.0096)	(0.0110)	(0.0081)	(0.0244)	(0.9631)	(0.0274)	(0.0090)
Illiq	0.0262⁎⁎	0.0224⁎	0.0184⁎	0.0530	0.5200	0.0205	0.0209	59,4220
	(0.0121)	(0.0135)	(0.0110)	(0.0497)	(1.1368)	(0.0392)	(0.0135)
Turn	−0.0050	−0.0072	−0.0042	−0.0221	−0.8955	−0.0301⁎	−0.0057	586,020
	(0.0054)	(0.0059)	(0.0047)	(0.0209)	(0.5759)	(0.0182)	(0.0063)
Vol	−0.0018⁎⁎⁎	−0.0022⁎⁎⁎	−0.0013⁎⁎	0.0010	−0.0667	−0.0047⁎⁎	−0.0018⁎	560,632
	(0.0007)	(0.0008)	(0.0006)	(0.0025)	(0.0754)	(0.0024)	(0.0009)
Panel E: Deseasonalized air pollution proxies
Return	−0.0093	−0.0309⁎⁎⁎	−0.0451⁎⁎⁎	−0.3595⁎⁎⁎	−5.3147⁎⁎⁎	0.0422	0.1112⁎⁎⁎	581,001
	(0.0094)	(0.0105)	(0.0085)	(0.0281)	(0.9538)	(0.0295)	(0.0155)
Ret_adj	−0.0169⁎	−0.0249⁎⁎	−0.0270⁎⁎⁎	−0.3352⁎⁎⁎	−3.5778⁎⁎⁎	−0.0241	0.0427⁎⁎⁎	568,138
	(0.0090)	(0.0104)	(0.0080)	(0.0334)	(0.9806)	(0.0262)	(0.0105)
Illiq	0.0894⁎⁎⁎	0.0731⁎⁎⁎	0.1024⁎⁎⁎	1.2149⁎⁎⁎	2.9673⁎⁎⁎	0.2496⁎⁎⁎	0.0474⁎⁎⁎	594,220
	(0.0120)	(0.0132)	(0.0111)	(0.0681)	(1.1195)	(0.0386)	(0.0172)
Turn	−0.0696⁎⁎⁎	−0.0732⁎⁎⁎	−0.0759⁎⁎⁎	−0.2741⁎⁎⁎	−4.1295⁎⁎⁎	−0.2054⁎⁎⁎	−0.0231⁎⁎⁎	586,020
	(0.0052)	(0.0058)	(0.0049)	(0.0218)	(0.6041)	(0.0180)	(0.0071)
Vol	−0.0154⁎⁎⁎	−0.0139⁎⁎⁎	−0.0175⁎⁎⁎	−0.0341⁎⁎⁎	−0.9058⁎⁎⁎	−0.0449⁎⁎⁎	−0.0102⁎⁎⁎	560,632
	(0.0008)	(0.0008)	(0.0008)	(0.0031)	(0.0873)	(0.0026)	(0.0012)
Panel F: Control local economic conditions
Return	−0.0675⁎⁎⁎	−0.0900⁎⁎⁎	−0.0572⁎⁎⁎	−0.2449⁎⁎⁎	−9.4685⁎⁎⁎	−0.1004⁎⁎⁎	0.0821⁎⁎⁎	581,001
	(0.0093)	(0.0106)	(0.0082)	(0.0230)	(0.9184)	(0.0287)	(0.0128)
Ret_adj	−0.0005	−0.0246⁎⁎	0.0016	−0.1015⁎⁎⁎	−2.2219⁎⁎	−0.0407	0.0309⁎⁎⁎	568,138
	(0.0089)	(0.0102)	(0.0076)	(0.0250)	(0.9161)	(0.0251)	(0.0092)
Illiq	0.0202⁎	0.0122	0.0271⁎⁎⁎	0.3237⁎⁎⁎	1.9733⁎	0.1766⁎⁎⁎	0.0059	594,220
	(0.0110)	(0.0123)	(0.0100)	(0.0463)	(1.0384)	(0.0335)	(0.0125)
Turn	−0.0185⁎⁎⁎	−0.0227⁎⁎⁎	−0.0228⁎⁎⁎	0.0751⁎⁎⁎	−1.5146⁎⁎⁎	−0.0891⁎⁎⁎	−0.0075	586,020
	(0.0049)	(0.0055)	(0.0043)	(0.0201)	(0.5375)	(0.0163)	(0.0059)
Vol	−0.0021⁎⁎⁎	−0.0013⁎	−0.0043⁎⁎⁎	0.0436⁎⁎⁎	0.0161	−0.0161⁎⁎⁎	−0.0068⁎⁎⁎	560,632
	(0.0007)	(0.0008)	(0.0006)	(0.0030)	(0.0719)	(0.0023)	(0.0010)
Panel G: Excluding firms belonging to finance industry
Return	−0.0702⁎⁎⁎	−0.0927⁎⁎⁎	−0.0593⁎⁎⁎	−0.2632⁎⁎⁎	−9.3990⁎⁎⁎	−0.1031⁎⁎⁎	0.0838⁎⁎⁎	562,093
	(0.0095)	(0.0107)	(0.0083)	(0.0229)	(0.9390)	(0.0292)	(0.0133)
Ret_adj	−0.0087	−0.0194⁎	−0.0092	−0.1744⁎⁎⁎	−2.5743⁎⁎⁎	−0.0719⁎⁎⁎	0.0344⁎⁎⁎	550,898
	(0.0091)	(0.0105)	(0.0077)	(0.0257)	(0.9376)	(0.0257)	(0.0094)
Illiq	0.1066⁎⁎⁎	0.0941⁎⁎⁎	0.1122⁎⁎⁎	0.8639⁎⁎⁎	5.6075⁎⁎⁎	0.3482⁎⁎⁎	0.0164	574,957
	(0.0124)	(0.0137)	(0.0111)	(0.0586)	(1.1704)	(0.0390)	(0.0152)
Turn	−0.0472⁎⁎⁎	−0.0497⁎⁎⁎	−0.0516⁎⁎⁎	−0.1335⁎⁎⁎	−2.8435⁎⁎⁎	−0.1421⁎⁎⁎	−0.0116⁎	568,145
	(0.0051)	(0.0057)	(0.0047)	(0.0195)	(0.5831)	(0.0176)	(0.0065)
Vol	−0.0101⁎⁎⁎	−0.0089⁎⁎⁎	−0.0124⁎⁎⁎	−0.0092⁎⁎⁎	−0.3621⁎⁎⁎	−0.0318⁎⁎⁎	−0.0084⁎⁎⁎	542,351
	(0.0008)	(0.0008)	(0.0008)	(0.0025)	(0.0856)	(0.0026)	(0.0011)
Panel H: Excluding firms headquartered in Shanghai and Shenzhen
Return	−0.0490⁎⁎⁎	−0.0594⁎⁎⁎	−0.0377⁎⁎⁎	−0.2222⁎⁎⁎	−5.5455⁎⁎⁎	−0.0182	0.0997⁎⁎⁎	429,576
	(0.0102)	(0.0115)	(0.0086)	(0.0226)	(0.9459)	(0.0313)	(0.0148)
Ret_adj	0.0062	−0.0002⁎	−0.0001⁎	−0.1258⁎⁎⁎	−0.8987	−0.0288	0.0505⁎⁎⁎	419,884
	(0.0098)	(0.0001)	(0.0001)	(0.0257)	(0.9743)	(0.0295)	(0.0105)
Illiq	0.0718⁎⁎⁎	0.0479⁎⁎⁎	0.0849⁎⁎⁎	0.7467⁎⁎⁎	−0.5315	0.2246⁎⁎⁎	0.0632⁎⁎⁎	439,179
	(0.0133)	(0.0148)	(0.0117)	(0.0571)	(1.1709)	(0.0459)	(0.0175)
Turn	−0.0393⁎⁎⁎	−0.0404⁎⁎⁎	−0.0409⁎⁎⁎	−0.1273⁎⁎⁎	−0.5342	−0.0847⁎⁎⁎	−0.0245⁎⁎⁎	432,709
	(0.0055)	(0.0062)	(0.0049)	(0.0202)	(0.5683)	(0.0200)	(0.0071)
Vol	−0.0082⁎⁎⁎	−0.0077⁎⁎⁎	−0.0101⁎⁎⁎	−0.0079⁎⁎⁎	0.1035	−0.0210⁎⁎⁎	−0.0075⁎⁎⁎	414,585
	(0.0008)	(0.0009)	(0.0008)	(0.0026)	(0.0745)	(0.0030)	(0.0012)
Panel I: Excluding firms belonging to polluting industry
Return	−0.0750⁎⁎⁎	−0.0982⁎⁎⁎	−0.0670⁎⁎⁎	−0.2654⁎⁎⁎	−10.4163⁎⁎⁎	−0.1121⁎⁎⁎	0.0763⁎⁎⁎	509,630
	(0.0099)	(0.0111)	(0.0086)	(0.0247)	(0.9808)	(0.0305)	(0.0138)
Ret_adj	−0.0157	−0.0271⁎⁎	−0.0146⁎	−0.1741⁎⁎⁎	−3.5207⁎⁎⁎	−0.0751⁎⁎⁎	0.0276⁎⁎⁎	498,945
	(0.0097)	(0.0110)	(0.0082)	(0.0286)	(0.9970)	(0.0271)	(0.0099)
Illiq	0.1035⁎⁎⁎	0.0937⁎⁎⁎	0.1079⁎⁎⁎	0.8720⁎⁎⁎	5.7753⁎⁎⁎	0.3454⁎⁎⁎	0.0115	521,306
	(0.0129)	(0.0142)	(0.0115)	(0.0613)	(1.1903)	(0.0392)	(0.0149)
Turn	−0.0490⁎⁎⁎	−0.0516⁎⁎⁎	−0.0548⁎⁎⁎	−0.1331⁎⁎⁎	−2.9975⁎⁎⁎	−0.1453⁎⁎⁎	−0.0143⁎⁎	513,950
	(0.0054)	(0.0060)	(0.0049)	(0.0195)	(0.6058)	(0.0182)	(0.0068)
Vol	−0.0102⁎⁎⁎	−0.0088⁎⁎⁎	−0.0126⁎⁎⁎	−0.0102⁎⁎⁎	−0.3550⁎⁎⁎	−0.0306⁎⁎⁎	−0.0086⁎⁎⁎	491,896
	(0.0008)	(0.0009)	(0.0008)	(0.0026)	(0.0926)	(0.0028)	(0.0012)

This table reports the results of robustness checks. The estimated model is Eq. (2). The dependent variables are stock returns (Return and Ret_adj), illiquidity (Illiq), turnover (Turn), and volatility (Vol) of locally headquartered firms, respectively. AQI, PM_2.5, PM₁₀, SO₂, CO, NO₂, and O₃ are the air pollution proxy variables. All the models control for one-lagged dependent variable except for models in Panel A. Relative humidity (Hum), temperature (Temp), air pressure (Pressure), visibility (Visibility), wind speed (Wind), cloud cover ratio (Cloud), and SAD effects (SAD) are controlled, too. Average returns over past 30 trading days of the Shanghai Composite Index (Rm) is also controlled except for models in Panel A. In addition, Monday and month effects are included in the models except for that in Panel A. Models in Panel A are estimated by Fama-MacBeth method (Fama and MacBeth, 1973). Panel B are regressed with Driscoll-Kraay standard errors. Models in Panel C to H are estimated by cross-sectional firm-specific fixed-effect method with robust standard errors clustered by firm in Angrist and Pischke (2009). Robust standard errors are in parentheses. To save space, the coefficients of control variables are not reported. *, ** and *** present significance at the 10%, 5%, and 1% levels, respectively.

In addition, we re-ran regressions with robust error clustered by city instead of robust error clustered by firm as a robustness check and found similar results, reported in Panel C. In addition to controlling for firm-specific fixed effects, we added time-specific fixed effects to control for the impacts of shocks in time trend. We found that air pollution reduced the returns, liquidity and volatility of locally headquartered firms in Panel D. Moreover, we used deseasonalized air pollution proxies as alternative air pollution variables. The seasonal air pollution was measured as the average of the raw values of air pollution proxies during the same calendar week over the entire sample period. Thus, the deseasonalized air pollution is the difference between the raw values and the seasonal values. We used the deseasonalized air pollution to reexamine the air pollution effects, and we found similar results to those shown in Panel E. Economic conditions may also impact investor opinion on stock market investment; thus, local economic conditions such as the population (Pop), population growth (Dpop), GDP (Gdp) and GDP (Dgdp) growth are controlled for in the models. The similar results are reported in Panel F. We re-ran our tests excluding firms in the finance industry and firms headquartered in Shanghai and Shenzhen, where the SSE and SZSE are located. The results are consistent, shown in Panels G and H. Finally, we were concerned that the negative relationship between air pollution and local returns were caused by local government intervention. When air pollution levels increase too much, seriously damaging health, local governments would force polluting firms to close due to public pressure. Thereby, local government hurts the local economy and local firms' profit, further reducing stock returns. To exclude this effect, we excluded firms belonging to polluting industries and ran our tests. Again, we found that air pollution reduced the returns, turnover, and volatility of locally headquartered firms and increased the illiquidity, as shown in Panel I. Overall, the results are similar.

5.3. Economic relation

We hypothesize that local air pollution affects local stock returns and trading activity through its impact on local investor moods. However, there appears to be an alternative potential explanation for these results. It could be possible that air quality affects daily labor productivity and thereby affects the local economy by changing firm fundamentals. For example, increases in local air pollution levels lead to significant increases in lost work days (Hausman et al., 1984), decreases in worker productivity for physically demanding tasks (Zivin and Neidell, 2012) and indoor white-collar work (Chang et al., 2016), and decreases in cognitive performance (Lavy et al., 2014).

Hu et al. (2014) note that air pollution could affect the performance of local firms if heavy air pollution prevents economic activities and changes firms' fundamental values. To test whether this issue exists, we used Hu et al.’s (2014) model to examine the relation between air pollution and firm performance. In every quarter, we calculated three accounting performance variables for each city. They were measured as follows.

$${\mathit{Fper}}_{c,q}=\sum _{i=1}^n{\mathit{Profit}}_{c,i,q}/\sum _{i=1}^n{\mathit{Asset}}_{c,i,q}\times 100$$ (3)

where Profit _c,i,q is the operating profit, total profit, or net profit for firm i in city c in quarter q and Asset _c,i,q is the total assets for firm i in city c in quarter q. The estimated model is

$${\mathit{Fper}}_{c,q}={\gamma }_0+{\gamma }_1{\mathit{Poor}}_{c,q}+{\gamma }_2{\mathit{AP}}_{c,q}+{\gamma }_3\mathit{Squ}\_{\mathit{AP}}_{c,q}+\sum _{j=1}^3{\lambda }_j{\mathit{Quarter}}_j+{\mu }_{c,q}$$ (4)

where AP is the air pollution proxies (including AQI, PM _2.5, PM ₁₀, SO ₂, CO, NO ₂, and O ₃), which are averaged in each quarter for each city. Squ_AP is the quadratic term of AP. Poor is a dummy variable set to one when the average AQI is >100, and zero otherwise. Quarter is a quarterly dummy variable.

Table 9 shows the results of the models with and without squared air pollution variables. For brevity, we only report the coefficients when using the AQI as air pollution proxies. We obtained similar results employing other air pollution proxies. As shown in Table 9, none of the coefficients are significantly negative, consistent with Hu et al. (2014). Our results verify that the adverse effects of air pollution on local firm-level returns, turnover, and volatility are not caused by reduced local labor productivity due to heavy air pollution.

Table 9.

Tests for economic relation.

Variables	Operate profit			Total profit			Net profit
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9
Poor	4.9464	−34.0956	−78.8622	−9.9552	−26.8933	−79.7456	0.1502	−28.3590	−76.9952
	(66.3902)	(77.8412)	(102.1039)	(61.8549)	(77.5879)	(106.1736)	(54.0009)	(69.5344)	(94.2743)
AQI		1.0605	6.1032		0.4601	6.4136		0.7744	6.2530
		(1.5207)	(6.9072)		(1.4736)	(6.6713)		(1.3054)	(5.7415)
Squ_AQI			−0.0162			−0.0191			−0.0176
			(0.0183)			(0.0189)			(0.0165)
Quarter effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	297	297	297	297	297	297	297	297	297
F statistic	10.3043	8.7161	7.8637	9.6701	8.1352	6.7442	8.8737	7.5758	6.2443
Adj-R2	0.1057	0.1052	0.1072	0.0555	0.0527	0.0559	0.0578	0.0561	0.0598

This table reports the effects of AQI on local firm performance. The estimated model is Eq. (4). The dependent variables are ratio of the sum of firm-level operating profit (or total profit, or net profit) in city c to that of firm-level total assets in city c. Poor is a dummy variable equal to one when the average AQI>100, and zero otherwise. AQI is the air pollution proxy variables. Squ_AQI is the quadratic term of AQI. All the models control for quarterly effects. All the models are estimated by cross-sectional city-specific fixed-effect method with robust standard errors clustered by city in Angrist and Pischke (2009). Robust standard errors are in parentheses. To save space, the coefficients of control variables are not reported. *, ** and *** present significance at the 10%, 5%, and 1% levels, respectively.

However, the firm performance variables are measured quarterly, much less frequent than daily air pollution proxies. Thus, the performance variables are realized consequences rather than ex-anti risks. If air pollution does not change firms' fundamental value, we would expect to see a decline in stock returns on days with more severe air pollution and an increase in stock returns on subsequent days.

According to the standard of AQI from the MEPC, air pollution is harmful to all people when the AQI is >200. Thus, we defined more severe air pollution as when the air quality status is heavily or severely polluted. We chose the windows for which the AQI was >200 on day t and <100 during days t-2 to t-1 and during days t + 1 to t + 2. We obtained 366 event windows. Then, we calculated the average stock returns for each day. Fig. 1 shows the average returns from day t-2 to t + 2. The left graph shows a large decline in raw stock returns on day t and a leap in raw stock returns in subsequent days. The right graph shows similar results for adjusted returns. Overall, the relations between air pollution and returns and trading activities are attributed to mood effects.

Fig. 1.

Average stock returns during event windows.

This figure reports the average returns among 366 event windows. Return is the daily raw return in percentage for each firm. Ret_adj is the daily adjusted return for each firm using 60-day rolling regressions.

6. Cross-sectional effects of air pollution on stock returns

Baker and Wurgler (2006) show that investor mood has a relatively sensitive effect on returns for small stocks, young stocks, high-volatility stocks, unprofitable stocks, non-dividend-paying stocks, extreme growth stocks, and distressed stocks. By adopting the previous method, the present work investigated whether air pollution strongly affects firms prone to the influence of investor moods. We divided the daily return observations into 10 groups according to firm characteristics at the beginning of each month and according to the daily AQI. As before, we defined the air quality as good if the air quality status was excellent and bad if the air quality status was moderately, heavily, or severely polluted. We calculated daily equal-weighted average returns for each group. To analyze the sensitivity of air pollution effects, we computed the conditional difference between the bad and good air quality across deciles of each firm characteristic. Table 10 reports the results.

Table 10.

Two-way sorts: Firm characteristics and air quality.

	Air quality	Decile										Overall
		1	2	3	4	5	6	7	8	9	10	10–1	10–5	5–1
ASSET	Good	0.5204	0.3398	0.3042	0.2535	0.2164	0.2147	0.2191	0.1850	0.1993	0.1648	−0.3556	−0.0515	−0.3041
	Bad	0.2874	0.2231	0.2445	0.1827	0.1785	0.1233	0.0957	0.0902	0.1044	0.0678	−0.2196	−0.1107	−0.1089
	Bad-Good	−0.2330	−0.1166	−0.0597	−0.0708	−0.0378	−0.0913	−0.1234	−0.0948	−0.0949	−0.0970	0.1360	−0.0592	0.1952
RVOL	Good	0.5191	0.1759	0.2351	0.2459	0.2102	0.2865	0.2593	0.2540	0.2721	0.2577	−0.2614	0.0475	−0.3089
	Bad	0.2348	0.0917	0.1859	0.1750	0.1708	0.2134	0.1235	0.1262	0.1480	0.0710	−0.1639	−0.0998	−0.0640
	Bad-Good	−0.2843	−0.0842	−0.0491	−0.0709	−0.0394	−0.0730	−0.1358	−0.1277	−0.1241	−0.1867	0.0976	−0.1473	0.2449
BE/ME	Good	0.2243	0.2484	0.2571	0.2847	0.2772	0.2645	0.2776	0.2801	0.2794	0.2181	−0.0062	−0.0591	0.0529
	Bad	0.1292	0.1011	0.2380	0.2257	0.1606	0.1580	0.1869	0.1763	0.0988	0.0949	−0.0343	−0.0657	0.0314
	Bad-Good	−0.0951	−0.1473	−0.0191	−0.0590	−0.1166	−0.1065	−0.0907	−0.1037	−0.1805	−0.1232	−0.0282	−0.0067	−0.0215
IPOAGE	Good	0.4598	0.2909	0.2446	0.2282	0.1780	0.1681	0.2028	0.2442	0.2995	0.2813	−0.1785	0.1033	−0.2818
	Bad	0.2919	0.2778	0.1587	0.1578	0.1067	0.1002	0.0919	0.1228	0.1128	0.1252	−0.1666	0.0185	−0.1851
	Bad-Good	−0.1680	−0.0131	−0.0859	−0.0705	−0.0713	−0.0678	−0.1109	−0.1214	−0.1867	−0.1561	0.0119	−0.0848	0.0967
GS	Good	0.2689	0.2426	0.2601	0.2432	0.2613	0.2565	0.2853	0.2843	0.3141	0.2919	0.0230	0.0306	−0.0076
	Bad	0.0429	0.0731	0.1159	0.1398	0.1877	0.1947	0.2171	0.2429	0.2490	0.1716	0.1287	−0.0161	0.1448
	Bad-Good	−0.2260	−0.1695	−0.1442	−0.1035	−0.0735	−0.0618	−0.0683	−0.0413	−0.0651	−0.1203	0.1057	−0.0467	0.1525

This table reports the equal-weighted average daily returns in percentage of each group, sorted by firm characteristics and air quality. At the beginning of each month, we divide the daily return observations into ten groups according to firm characteristics, and then according to the level of AQI on each day. Firm characteristics includes firm size (ASSET), return volatility (RVOL), book-to-market ratio (BE/ME), IPO age (IPOAGE) and sales growth (GS). The sort of good air quality if air quality status is excellent, whereas the sort of bad air quality if air quality status is moderately, heavily or severely polluted.

The first rows in Table 10 show the conditional effects of firm size (ASSET), which was measured as the log of total assets at the beginning of each month. These rows indicate that the daily average returns decrease by decile rank per firm size, regardless of whether the air pollution-induced moods are optimistic or pessimistic. These results are consistent with those of Banz (1981), and Fama and French (1992). The daily average return was 0.5204% for the bottom ASSET decile and 0.1648% for the top decile when the air quality was good. Similarly, the daily average return was 0.0678% for the top decile, which is much smaller than 0.2874% for the bottom decile, when the air quality is poor. However, the return difference between the first two rows reveals that returns on the same day are lower when the mood is low. This result supports the view that investors who are in poor moods (pessimistic) would have a lower desire to buy or hold stocks, leading to a decrease in returns.

The second row in Table 10 illustrates the cross-sectional effects of return volatility (RVOL), as measured by the standard deviation of daily returns in the previous month. The results show that low-risk stocks remain attractive regardless of the investor mood. The stocks earned relative high returns of 0.5191% (0.2348%) when air quality is good (bad). This result conflicts with findings that high volatile stocks earn larger returns when moods are lower, as in Baker and Wurgler (2006). However, they employed expected monthly returns, whereas we used daily returns on the same day.

The next rows in Table 10 show the pattern of the book-to-market ratio (BE/ME), as measured by the ratio of book value to market value at the beginning of each month. When investor moods are optimistic, the average daily returns are higher, with increasing BE/ME. Firms with the highest BE/ME earn an average daily return of 0.2181%, and firms with the lowest BE/ME earn 0.2243%. When investor moods are pessimistic, the average daily returns are lower for both high and low BE/ME stocks. Thus, an inverted U-shaped pattern occurs in the difference row for BE/ME. The fifth row of sales growth (GS), which is a percentage, also presents a similar pattern.

The fourth row in Table 10 reports the cross-sectional effects of the IPO age (IPOAGE), which is measured as the number of years since the firm went public. We found an unconditional pattern effect in which younger firms earned higher daily returns. However, the row of return difference did not show a pattern.

7. Conclusions

Environmental pollution has accompanied high-speed economic development in China, and the pollution, such as air pollution, adversely affects individuals' moods and decisions. In this paper, we explored whether such pollution creates negative effects in financial markets.

We investigated the relations among the air pollution, stock returns, and trading activity of locally headquartered firms using firm-level data. In an order-driven market, trading orders come from all over the market. As a result, examining only the relationship between the market index and air pollution in the city where the stock exchange is located may create a serious bias. China provides a clear and unique setting to examine this possibility. We verified that air pollution affects stock prices in Chinese stock markets mainly through home bias. We confirmed that air pollution has significant negative effects on local returns by decreasing volatility and trading. After excluding the potential phenomenon of air pollution influencing local firms' stock prices through effects on local economic activities, we conducted a series of robustness checks and obtained similar results. The returns for extreme growth stocks, distressed stocks, and stocks with high volatility were confirmed to be more sensitive to negative moods induced by air pollution. The process of how firm characteristics influence the relations between air pollution and trading activity will be examined in a future study.

Table A1.

Definitions of variables.

Variables	Definitions	Data source
Return	It is the daily returns in percentage of locally headquartered firms, and measured as the ratio of difference between closing price on day t and t-1 to closing price on day t-1	Wind Info database
Ret_adj	Ret_adj is the daily adjusted returns for each firm using 60-day rolling regressions	Wind Info and CSMAR database
Illiq	It is the daily illiquidity of locally headquartered firms, and measured as the absolute daily returns divided by the daily dollar trading volume scaled by 108	Wind Info database
Turn	It is the daily turnover in percentage of locally headquartered firms, and measured as daily trading volume divided by its outstanding shares	Wind Info database
Vol	It is the daily volatility in percentage of locally headquartered firms, and measured as the variances of five-minute returns within a day for each stock	CSMAR database
AQI	It is air quality index, evaluated according to fine particulate matter, particulate matter, sulfur dioxide, carbon monoxide, nitrogen dioxide, and ozone	www.aqistudy.cn
PM2.5	Fine particulate matter	www.aqistudy.cn
PM10	Particulate matter	www.aqistudy.cn
SO2	Sulfur dioxide	www.aqistudy.cn
CO	Carbon monoxide	www.aqistudy.cn
NO2	Nitrogen dioxide	www.aqistudy.cn
O3	Ozone	www.aqistudy.cn
Hum	Relative humidity	www.wunderground.com
Temp	Temperature in degrees Celsius	www.wunderground.com
Pressure	Air pressure in kPa	www.wunderground.com
Visibility	Visibility in km	www.wunderground.com
Wind	Wind speed in km/h	www.wunderground.com
Cloud	It is cloud cover, which is equal to one when the weather condition is rain, snow, fog, and other weather events that mostly or entirely cover the sky, and zero otherwise.	www.wunderground.com
SAD	It is the seasonal affective disorder effect, and measured based on the method of Kamstra et al. (2003), whereas the value is not replaced by zero in spring and summer	–
Rm	It is the average return of the Shanghai Composite Index over the past 30 days	Wind Info. database
Pop	City-level population	www.stats.gov.cn and local statistical yearbook
Dpop	City-level population growth	www.stats.gov.cn and local statistical yearbook
Gdp	City-level GDP	www.stats.gov.cn and local statistical yearbook
Dgdp	City-level GDP growth	www.stats.gov.cn and local statistical yearbook